TECHNOLOGY DIGEST - Draper Laboratory

THE DRAPER 

TECHNOLOGY DIGEST 

2006 Volume 10 CSDL-R-3005 #*

The Draper Technology Digest (CSDL-R-3005) is published annually by the Education Office of 

The Charles Stark Draper Laboratory, Inc. Editorial questions and requests for reprints can be 

directed to: 

Media Services 

Phone: (617) 258-1491 

Fax: (617) 258-4535 

E-mail: techdigest@draper.com 

Mailing address changes can be forwarded to: 

communications@draper.com 

Editor-in-chief 

Dr. George Schmidt, Director, Education Office 

Creative Director 

Charya Peou 

Designer 

Pamela Toomey 

Editor 

Beverly Tuzzalino 

Photography Coordinator 

Drew Crete 

Photographers 

Jay Couturier 

Michael Brooks 

Copyright © 2006 by The Charles Stark Draper Laboratory, Inc. All rights reserved.

Introduction by Vice President, Engineering, Dr. Eli Gai 

Papers 

The Silicon Oscillating Accelerometer: A High-Performance MEMS 

Accelerometer for Precision Navigation and Strategic Guidance Applications 

RALPH HOPKINS, JOSEPH MIOLA, ROY SETTERLUND, BRUCE DOW, WILLIAM SAWYER 

Autonomous Guidance, Navigation, and Control of Large Parafoils 

DAVID CARTER, SEAN GEORGE, PHILIP HATTIS, LEENA SINGH, STEVEN TAVAN 

An Ultra-Low-Power Physics Package for a Chip-Scale Atomic Clock 

MARK MESCHER, ROBERT LUTWAK, MATHEW VARGHESE 

Detection of Biological and Chemical Agents Using Differential Mobility 

Spectrometry (DMS) Technology 

MELISSA KREBS, ANGELA ZAPATA, ERKINJON NAZAROV, RAANAN MILLER, 

ISAAC COSTA, ABRAHAM SONENSHEIN, CRISTINA DAVIS 

Autonomous Mission Management for Spacecraft Rendezvous Using an 

Agent Hierarchy 

MARK JACKSON, CHRISTOPHER D'SOUZA, HOBSON LANE 

Fluid Effects in Vibrating Micromachined Structures 

PETER KWOK, MARC WEINBERG, KENNETH BREUER 

2005 Published Papers 

Patents 

Patents Introduction 

Optically Rebalanced Accelerometer: Patent No. 6,867,411 B2 

Issued: March 15, 2005 

WILLIAM KELLEHER, STEPHEN SMITH, RICHARD STONER 

2005 Patents Issued 

The 2006 Charles Stark Draper Prize 

The Draper Distinguished Performance Awards 

The Howard Musoff Student Mentoring Award 

2005 Graduate Research Theses 

2005 Technology Exposition 

2 

4 

14 

26 

32 

42 

56 

70 

77 

78 

81 

82 

84 

85 

86 

88 

1

This year is the 10 th anniversary of Draper’s Technology 

Digest. For the past 10 years, the Digest has included 

over 60 of the best papers Draper staff members have published, 

demonstrating the breadth of Draper Laboratory 

technology. It has also included lists of all patents issued 

to Draper staff members. I would like to take this 

opportunity to thank the Media Services Group, who are 

responsible for the publication of the Digest, for continuously 

improving the visual look of every issue. 

This year’s issue includes six papers that were published 

and patents that were issued during the calendar year 

2005. Committees established by the Vice President of 

Engineering evaluated all papers and patents during this 

period and selected those to be included in this issue, as 

well as the winner of the annual Vice President’s Awards. 

The papers in this issue can be divided into two groups 

covering two growth areas of Draper technologies. Four 

papers describe applications of Microelectromechanical 

Systems (MEMS) for inertial instruments, the atomic 

clock, and biomedical instruments. The other two papers 

cover autonomous systems – one to guide a parachute, the 

other for spacecraft rendezvous. 

The first paper, “The Silicon Oscillating Accelerometer: A 

High-Performance MEMS Accelerometer for Precision 

Navigation and Strategic Guidance Applications” by Ralph 

2 

10th An 

D R A P E R T E C H 

Introduction by Vice President, Engineering, Dr. Eli Gai 

Hopkins, Joseph Miola, Roy Setterlund, William Sawyer, 

and Bruce Dow was selected for the Vice President’s Award 

for Best Paper published in 2005. This paper describes the 

theory of operations, performance goals, and the fabrication 

process for two silicon oscillating accelerometer 

(SOA) designs, one for strategic missile guidance and one 

for submarine navigation. Test data show that both 

versions of the SOA met the performance required for 

their respective applications. 

The second paper, “Autonomous Guidance, Navigation, and 

Control of Large Parafoils” by David Carter, Sean George, 

Philip Hattis, Leena Singh, and Steven Tavan deals with 

the design and flight testing of a guidance, navigation, and 

control (GN&C) software package that enables precision 

payload airdrop delivery using large parafoils. The 

modular software design is structured to accommodate 

payloads ranging from 2000 to 30,000 lb. The requirements 

for low components cost resulted in a limited 

onboard processor and a laptop personal computer. In 

spite of these limitations, flight test data results showed a 

miss distance of 200 m. 

The third paper, “An Ultra-Low-Power Physics Package for 

a Chip-Scale Atomic Clock” by Mark Mescher, Robert 

Lutwak, and Matthew Varghese reports on the design of a 

low-power, small-package, thermally-isolated physics 

package for an atomic clock. This physics package will

niversary 

N O L O G Y 

D I G E S T 

enable communication and navigation systems that 

require an atomic frequency standard with the above tight 

requirements. The design demonstrated a package requiring 

only single milliwatts of power, which represented a 

reduction of two orders of magnitude compared with the 

lowest of existing commercial technology. 

The fourth paper, “Detection of Biological and Chemical 

Agents Using Differential Mobility Spectrometer 

Technology” by Melissa Krebs et al. describes a new 

MEMS-based sensor, the Differential Mobility Spectrometer 

(DMS) that was developed to detect chemical or 

biological agents down to the parts per trillion within 

seconds. The sensor is small, robust, can detect multiple 

agents simultaneously, and is nondestructive during its 

operation. Several experiments were conducted to demonstrate 

the ability of the sensor to detect Bacillus subtilis 

spores. 

The fifth paper, “Autonomous Mission Management for 

Spacecraft Rendezvous Using an Agent Hierarchy” by 

Mark Jackson, Christopher D’Souza, and Hobson Lane is 

the second paper in this issue that deals with vehicle 

autonomy. In this case, it is a chaser satellite, and the 

mission management task is to rendezvous with another 

satellite. The software manager is based on a hierarchy of 

“activity agents,” each of which monitor, diagnose, plan, 

and execute the activity. The algorithms in this agent use 

low-thrust guidance to arrive in the neighborhood of 

the target, followed by high-thrust burns to arrive at a 

specified point at a specified time. 

The last paper by Peter Kwok, Marc Weinberg, and 

Kenneth Breuer entitled “Fluid Effects in Vibrating 

Micromachined Structures,” has resulted in a closed-form 

solution for fluid damping and lift that can be used to 

design accelerometers, tuning-fork gyroscopes (TFG), 

and other micromechanical devices. The closed-form and 

the numerical solutions are compared against experimental 

data over a range of pressures using TFG samples, and 

the results show agreement among the three. 

Three patents were nominated for the Vice President’s 

Award for Best Patent issued during calendar year 2005. 

The winning patent, “Optically Rebalanced Accelerometer,” 

was authored by Bill Kelleher, Steve Smith, and 

Richard Stoner. Acceleration is observed by measuring 

the position changes of a proof mass. The sensitivity of 

the accelerometer is predicted to be 1 µg with less than 1ppm 

scale factor and maximum acceleration of 100 g, 

thus making it a candidate for strategic guidance 

applications. 

3

The Silicon Oscillating Accelerometer: A High-Performance 

MEMS Accelerometer for Precision Navigation and Strategic 

Guidance Applications 

Ralph Hopkins, Joseph Miola, Roy Setterlund, Bruce Dow, William Sawyer 

Copyright © 2005 by The Charles Stark Draper Laboratory, Inc. 

Presented at the Institute of Navigation 61 st Annual Meeting, Cambridge, MA, June 27-29, 2005 

INTRODUCTION 

SOA Applications and Performance Goals 

The ICBM/submarine-launched ballistic missile (SLBM) has 

the most demanding requirements of any inertial guidance 

application. The high degree of accuracy required of the 

weapon system, combined with the high acceleration levels 

and large velocity at reentry body deployment place an especially 

stringent performance requirement on the guidance 

system accelerometers. The SSBN SINS system requires similarly 

precise inertial performance from the navigation 

system accelerometers, although compared with the missile 

application, the SINS accelerometers enjoy a more benign 

4 

The intercontinental ballistic missile (ICBM) and submarine-launched strategic missiles developed 

over the past 50 years have employed successive generations of increasingly accurate inertial guidance 

systems. The comparatively short time of guided flight and high acceleration levels characteristic of the 

ballistic missile application place a premium on accelerometer performance to achieve desired weapon 

system accuracy. Currently, the U.S. strategic missile arsenal relies on variants of the pendulous 

integrating gyro accelerometer (PIGA) to meet the high-performance, radiation-hard requirements of 

the weapon system. 

Likewise, precision navigation systems such as the currently deployed SSBN ship inertial navigation 

systems (SINS) employ highly specialized and complex electromechanical instruments that, like the 

PIGA, present a system life-cycle cost and maintenance challenge. 

The PIGA and the electromagnetic accelerometer (EMA) demonstrate unsurpassed performance, however, 

their life-cycle cost has motivated a search for a high-performance, solid-state, strategic 

accelerometer. 

Draper Laboratory is currently developing the silicon oscillating accelerometer (SOA), a microelectromechanical 

system (MEMS)-based sensor that has demonstrated in laboratory testing the part-per-million 

(ppm)/µg scale-factor and bias performance stability required of strategic and precision navigation 

applications. 

The ICBM and SSBN applications have significantly different environmental, acceleration dynamic 

range, and resolution requirements that are best satisfied by optimizing the SOA geometry for each 

application. The design flexibility and wafer-scale fabrication methods of the silicon MEMS process 

enable manufacturing both instrument designs with essentially zero incremental cost associated with 

the additional instrument assembly line. That is, the SOAs developed for the ICBM guidance and SSBN 

navigation applications share a common sensor package, electronics architecture, main housing, and 

instrument assembly process. This paper presents an overview of Draper’s SOA and compares and 

contrasts performance data taken to date on both versions of the SOA. 

operational environment and a smaller dynamic range 

requirement. [5] 

Although there are many system-derived performance parameters 

specified for inertial-grade accelerometers (see Table 1), in 

broad terms, accelerometer performance can be characterized 

with two parameters: bias and scale-factor (SF) stability. 

Accelerometer bias is the DC offset indicated from the instrument 

output under zero applied acceleration. Scale factor is the 

instrument gain or sensitivity that relates the applied acceleration 

to the instrument output signal (e.g., V/g, Hz/g, etc.).

Bias 

Table 1. Typical SOA Performance Goals. 

Parameter Units Missile High- 

Guidance Performance 

Navigation 

Long-Term Stability* µg 1 - 100 1 

Short-Term Stability** 

Scale Factor 

µg 1 - 5 0.5 

Long-Term Stability* ppm 1 - 100 10 

Short-Term Stability** ppm 1 - 5 5 

Asymmetry ppm TBD TBD 

g2 (Compensated) µg/g2 0.1 - 0.2 1 

g3 (Compensated) µg/g3 TBD TBD 

VRW ft/s√h 0.014 - 0.030 0.0014 

White Noise 

Misalignment 

µg/√(Hz) 10 - 22 1 

Long-Term Stability* arcsec 0.1 - 5 TBD 

Short-Term Stability** arcsec 0.1 - 2 0.4 

Vibration Rect. µg/(grms) 2

Figure 1. SOA schematic. 

The SOA input axis lies in plane as indicated in Figure 1; 

under acceleration, the proof mass axially loads the two 

resonator pairs. The vibration frequency of each resonator 

changes under the applied load. This frequency change is 

measured and serves as the indicated acceleration output 

of the SOA. Note that the resonators are arranged so they 

are loaded differentially by the proof mass. That is, one 

resonator is placed in tension, the other in compression. 

This differential design doubles the sensitivity or scale 

factor of the accelerometer and furnishes a cancellation of 

error sources common to both resonators. 

The resonators are excited by an electrostatic comb 

drive, [1],[2] similar to that used in Draper’s micromechanical 

tuning-fork gyro (TFG). The comb drive has both inner 

and outer motor stator combs that are fixed to the glass 

substrate. The outer motor combs apply the drive force, the 

inner motor combs sense the drive amplitude and frequency. 

A detail of the comb geometry is shown in Figure 2. 

Figure 2. SOA oscillator detail. 

The goal of the SOA design is to achieve a high scale-factor, 

high-Q resonator to achieve high performance. Large scale 

factor is desirable because it decreases the degree of frequency 

stability required to resolve a given acceleration 

level. For example, 0.1-mHz frequency stability is required 

of a 100-Hz/g SF unit to resolve 1 µg. A 10-Hz/g unit has a 

10 times more restrictive frequency stability requirement 

(10 µHz) to resolve the same 1-µg input. 

6 

Anchored 

Component 

Suspended 

Component 

Electrostatic 

Component 

The Silicon Oscillating Accelerometer 

It can be shown [3] that the lateral stiffness of an axially 

loaded beam with fixed ends is: 

where: 

K = stiffness 

E = Young’s modulus 

I = moment of inertia 

L = beam length 

P = axial load 

If a lumped mass is supported between two beams, the 

natural frequency of the mass-beam system as a function 

of axial load is given by: 

where: 

f = resonant frequency 

m = mass of lumped oscillator 

Rearranging Eq. (2) gives: 

where: 

(1) 

(2) 

(3) 

f = resonant frequency 

f0 = nominal unloaded (bias) resonant frequency 

= 

m= resonator mass 

L = beam length 

E = Young’s modulus 

I = beam inertia 

P = applied axial load 

Note that the frequency versus applied acceleration load 

relationship in the SOA is nonlinear, as indicated by Eq. 

(3) and shown in Figure 3. 

Frequency (Hz) 

Buckling Load 

Bias Frequency fo 

Linear SF (Hz/g) 

f = fo 

Input acceleration (g) 

1 + M g 

π2EI L2 Figure 3. SOA frequency vs. acceleration curve.

A series expansion of Eq. (3) can be used to determine the 

linearized SOA scale factor (the slope about zero acceleration 

in Figure 3) and higher-order g-coefficients: 

where S = L2/π2EI. Equation (4) can be rewritten as: 

where: 

Kn =K1bn(K1/f0) n-1 (Hz/gn) bn =bn-1 (3-2n)/n 

K1 = S/2 

b1 =1 

g = acceleration 

Note that the values of the g2 and g3 coefficients (K2, K3) 

are controlled by the linear SF (K1) and bias frequency (f0). 

The linearized SF is dependent on resonator dimensions, 

Young’s modulus, and the mass of the resonator (m) and 

proof mass (M): 

(6) 

The SF stability of the SOA will be largely controlled by 

the Young’s modulus sensitivity to temperature (∆E/E/∆T = 

-50 ppm/°C), as it is an order of magnitude larger than 

silicon’s thermal coefficient of expansion (TCE) (2.5 

ppm/°C), the parameter that would control the resonator 

dimensional stability. Given the square root relationship, 

the linear SF temperature coefficient is approximately 25 

ppm/°C, indicating that 0.01°C temperature control will 

maintain better than 1-ppm SF performance. 

From Eqs. (4) and (5), the values of the g2 and g3 coefficients 

(K2, K3) can be expressed by the linear SF (K1) and 

bias frequency (f0): 

For a 100-Hz/g (per side) SF, 20-kHz nominal bias frequency 

unit, Eqs. (7) and (8) project g2 and g3 coefficients 

of 0.25 Hz/g2 and 0.0125 Hz/g3. Normalizing these coefficients 

by dividing by the linear SF gives 2500 µg/g2 and 

12.5 µg/g3, respectively. 

Compensating these coefficients to the sub-µg level is feasible 

because their stability will be of the order of the linear 

SF and bias. Differentiation of Eqs. (7) and (8) gives: 


(4) 

(5) 

(7) 

(8) 

(9) 

(10) 

Note that 1-µg performance implies a resonator frequency 

stability (∆f/f) on the order of 5 parts per billion (ppb) 

(given a100-Hz/g per side SF and 20-kHz nominal frequency 

unit). Combined with 1-ppm SF, this implies that 

the above SF nonlinearity coefficients should be stable to 

approximately 2 to 3 ppm. This high degree of stability 

should enable compensating the raw nonlinear coefficients 

to sub-µg levels (although calibrating the higher-order gcoefficients 

will require precision centrifuge testing). 

Additionally, the net SOA g 2 coefficient and other evenorder 

terms will be reduced as the g 2 contribution from 

each resonator will be common mode differenced in the 

net SOA output. 

SOA Electronics 

Figure 4 shows the SOA electronics block diagram. As 

mentioned above, the SOA employs an electrostatic comb 

drive to excite the resonators, and to pick off the resonator 

displacement. The electrostatic drive force is given by: 

where: 

F = drive force 

dC/dx = comb position sensitivity 

V = applied voltage 

Proof 

mass 

Gain 

X1 

-X1 

Gain 

X1 

-X1 

90 

deg 

Drive 

gen 

90 

deg 

Drive 

gen 

Freq 

det 

Ampl 

det 

C(s) 

Freq 

det 

Ampl 

det 

C(s) 

(11) 

Figure 4. SOA electronics block diagram. 

The drive amplitude stability furnished by the electronics is 

critical to maintaining nominal resonator frequency stability. 

The resonator beams stiffen with lateral deflection, causing 

a dependence of the resonant frequency with drive amplitude. 

This nonlinear stiffening effect introduces a bias 

uncertainty from the resonator drive amplitude instability. 

It can be shown [4] that if the resonator is modeled as a 

linear plus cubic stiffness element, the resonator frequency 

dependence on amplitude is given by: 

Fa 

– + 

Fb 

– + 

REF 

REF 

7

(12) 

where: 

f = frequency at amplitude x 

f0 = nominal resonant frequency 

K = linear stiffness of resonator 

K3 = cubic stiffness coefficient 

x = drive amplitude 

The stability requirement on the drive amplitude can be 

determined by differentiating Eq. (12) to get: 

(13) 

From Eqs. (12) and (13), it can be seen that a small drive 

amplitude will minimize the resonator frequency variance 

from drive amplitude instability and noise. An alternate 

means of maximizing frequency stability is to minimize 

the amount of nonlinear stiffening, i.e., design a resonator 

with a low K3 coefficient. 

The resolution or noise floor of the SOA can be estimated 

by calculating the amplitude and phase noise associated 

with the sense comb frequency readout. From Ref. [1], the 

capacitance across a set of engaged comb drive fingers is 

given by: 

(14) 

where: 

C = capacitance 

εo = permittivity of air 

N = Number of teeth per side 

α = fringing factor 

t = comb finger thickness 

g = air gap between fingers 

L = engaged length of fingers 

The capacitance sensitivity to position (i.e., engaged 

length) is given by: 

(15) 

where dC/dx = sensitivity to position. 

Equation (12) gives the relationship between resonator 

amplitude and frequency, which gives the resonator frequency 

power spectral density (PSD) as: 

where: 

8 

φf = frequency PSD in Hz/√Hz 

x = nominal drive amplitude 

φA = amplitude noise PSD 

f0 = nominal resonant frequency 

K3/K = stiffness coefficient ratio 


(16) 

The contribution of phase noise in the drive frequency 

electronics can also be estimated. The PSD of the oscillator 

phase noise is approximately equal to the PSD of the 

amplitude noise divided by the peak amplitude. 

At resonance, the phase noise is related to frequency noise 

by: 

where: 

φf = frequency noise PSD 

φp = phase noise PSD 

(17) 

ωn = nominal resonant frequency 

Q = Q of resonator 

The high Q’s achieved in the SOA oscillators (~100,000) 

significantly reduce the frequency noise in the output from 

phase jitter. The net frequency noise in the SOA readout is 

dominated by oscillator amplitude noise. Consequently, 

frequency readout resolution is improved with increasing 

bias voltage and decreasing drive amplitude. 

SOA MEMS Sensor Design, Fabrication, and Screening 

The ICBM application and the SSBN SINS system have significantly 

different environmental, acceleration dynamic 

range, and resolution requirements, as mentioned above 

(Table 1). The SOA instrument can be easily adapted to 

either application by adjusting the SOA MEMS sensor element 

geometry for the specific operating acceleration 

range required. Figure 5 is similar to Figure 3 and shows 

the load vs. acceleration curve for two different SOA 

designs: a higher g-capable design suitable for 

ICBM/SLBM environments and a lower g design tailored 

for the more benign SINS environment. Note that the 

higher sensitivity (higher SF) SINS design has a correspondingly 

lower buckling load. This is a consequence of 

beam mechanics and is a fundamental design trade-off in 

selecting SOA acceleration sensitivity. 

f = fo 

1 + M g 

π2EI L2 Linear SF (Hz/g) 

Buckling Load 

Frequency (Hz) 

Bias Frequency fo 

Low-g 

High-g 

Input acceleration (g) 

Figure 5. SOA dynamic range. 

The geometries of the two different SOAs are optimized for 

each application by adjusting oscillator and proof mass 

geometry. The advantages of the silicon MEMS design 

approach is apparent as two different applications can be 

serviced with only the incremental expense of the MEMS

photo-maskset needed to fabricate the application-specific 

SOA sensor element. The design flexibility and wafer-scale 

fabrication methods of the silicon MEMS process enable 

manufacturing both instrument designs with essentially 

zero incremental cost associated with the additional sensor 

design. That is, both versions of the SOA share a common 

sensor package, electronics architecture, main housing, 

and instrument assembly process. The only difference 

between the two instrument assembly processes is the use 

of a different MEMS photo-maskset in the fabrication of 

the SOA sensor element. 

Draper uses the silicon-on-glass dissolved-wafer process to 

fabricate the SOA, [6] a mature MEMS fabrication process 

that Draper employs elsewhere in MEMS-based inertial 

technology. [2] 

The main process steps of the dissolved-wafer process 

are illustrated in Figure 6. The starting wafers used in 

processing are a boron-doped epitaxial silicon layer 

grown on an undoped silicon handle layer. The thickness 

of the epitaxial layer determines the final 

thickness of the silicon device layer. The first process 

step is to etch mesas in the epitaxial silicon layer; this 

step defines the eventual operating air gap between the 

suspended silicon sensor element structures and the 

glass substrate. 

Silicon 

p ++ Epitaxial Growth 

Form Mesas 

Deep Trench Etch 

Glass Wafer 

Metalization 

EDP Etch Anodic Bond 

Figure 6. SOA fabrication process. 

The device structural layer is then photolithographically 

patterned on the silicon, and the wafers are etched using 

high-aspect-ratio micromachining in a reactive ion etch 

(RIE) machine. This step forms the 2-D geometry of the 

SOA silicon sensor element. Recent advances in RIE etching 

technology have enabled a very high degree of feature 

size control (e.g., beam width uniformity, side wall perpendicularity, 

etc.) in MEMS processing, a critical factor 

affecting as-built SOA sensor performance. 


The patterned wafers are then anodically bonded to glass 

substrates that have been metalized with the SOA electrode 

pattern. Finally, the silicon wafer is dissolved in an 

anisotropic wet etchant such as ethylenediamine pyrocatechol 

(EDP), which removes the silicon handle layer. The 

boron doping in the epitaxial layer stops the chemical 

etching action of the EDP, leaving the finished device layer 

array on the glass wafer substrate. Individual SOA sensors 

are then diced and screened prior to packaging. 

The first screening step performed on the SOA MEMS sensor 

die is a particle and defect inspection, followed by 

electrical probing for proper resistance and continuity. 

Units passing this step are “wiggle tested” by energizing 

the oscillator elements to verify oscillator freedom at the 

expected drive frequency. 

After probe testing, SOAs with satisfactory performance 

are packaged and vacuum sealed in an aluminum oxide 

leadless ceramic chip carrier (LCCC). The residual pressure 

achieved in the LCCC after vacuum seal is less than 1 

mTorr, which ensures high oscillator Q factors. A plot of Q 

vs. pressure for a typical SOA is shown in Figure 7, indicating 

that oscillator drive Q’s on the order of 100,000 are 

achieved with this process. The packaged SOA is mated to 

a front-end preamplifier electronics module and subsequently 

mounted in an instrument main housing 

assembly. The final SOA instrument assembly is shown in 

Figure 8. 

100000 

10000 

Log Q 1000000 

1000 

0.01 0.1 1 10 100 1000 

Log Pressure (mTorr) 

Figure 7. Q vs. pressure relationship. 

Figure 8. SOA instrument. 

Missile Guidance and SINS SOA Performance Test Data 

Both versions of the SOA demonstrate the performance 

required for their respective applications. As mentioned 

9

above, the SINS SOA is tailored with a higher sensitivity to 

exploit the reduced dynamic range environment of the SSBN 

application. The SINS SOA sensor design also incorporates 

design features to maximize longer time scale drift stability. 

The contrast in the performance of both versions of the 

SOA is best captured by the acceleration resolution capability 

demonstrated via the noise PSD plots shown in 

Figure 9. Note that both units have a white noise floor on 

the order of 10-11 g2/Hz, however, the SINS design maintains 

the resolution capability over longer time scales (i.e., 

lower frequency bands). The missile guidance SOA shows 

a noise PSD that starts to break upward with a 1/f signature 

in the 10-1 Hz to 10-2 Hz decade. The SINS SOA 

shows the white noise floor extending another decade or 

two before exhibiting a 1/f signature. 

PSD Amplitude [(g) 2 /Hz] 

Figure 9. SOA acceleration noise PSD. 

This 1/f (or “flicker”) noise is indicative of an acceleration 

drift instability that limits the net resolution capability of 

the accelerometer. This is also characterized by the Allan 

variance charts calculated from the PSD measurements 

shown in Figure 10. The missile guidance SOA resolution 

10 

10 -8 

10 -9 

10 -10 

10 -11 

Acceleration Uncertainty (g) 

10 -5 

10 -6 

Missile Guidance SOA 

White Noise -1/2 slope 

φ = 4.5 µg/√Hz 

VRW = 0.006 ft/s/√h 


Guidance SOA 

SOA-SINS 

10 

Frequency (Hz) [Fs = 1 sample/s] 

-12 

10-6 10-5 10-4 10-3 10-2 10-1 100 1 µg 

0.5 µg 

is limited to 0.5 µg, which is achieved over 100-s averaging 

times. Longer averaging times do not further improve 

resolution in the missile guidance SOA because of the 1/f 

drift instability, but the low-frequency extension of the 

white noise floor of the SINS SOA enables 80 nano-g 

resolution, which is achieved after 1000 s of averaging. 

The Allan variance plots the standard deviation of indicated 

acceleration against data averaging time. The white noise 

floor of the corresponding PSD can be determined from the 

portion of the Allan variance having a minus-one-half slope 

on the log scale. The equivalent white noise PSD can be calculated 

from a point on the minus-one-half slope line from: 

where: 

σ = acceleration standard deviation 

φ = white noise PSD 

T = averaging time 

(18) 

The data from both versions of the SOA (Figure 10) in the 

minus-one-half slope region indicate (from Eq. (18)) a 

white noise PSD of 4.5 µg/√Hz, which is equivalent to a 

velocity random walk coefficient of 0.006 ft/s/√h. 

Figure 11 shows SOA bias and SF uncertainty measured 

on a missile guidance SOA in a 2-position test, and Figure 

12 shows SF and bias uncertainty measured on an SOA- 

SINS unit in a 4-position test. The SF and bias data shown 

in these figures are uncompensated SOA output and are 

the average values over a 30-min dwell as determined from 

1-s averaged data. The data shown extend over a roughly 

3-day period and show 1σ standard deviations in SF and 

bias of 0.56 ppm and 0.92 µg, respectively, for the missile 

guidance SOA. The quieter SOA-SINS unit demonstrates 

1σ uncertainty in SF and bias of 0.14 ppm and 0.19 µg, 

respectively. 

SOA-SINS 

10 

Time (s) 

-7 

10 

Time (s) 

-8 

10 

Figure 10. SOA Allan variance. 

0 101 102 103 104 105 100 101 102 103 104 105 10 -4 

10 -5 

10 -6 

10 -7 

White Noise (-1/2 slope) 

φ = 4.5 µg/√Hz 

VRW = 0.006 ft/s/√h 

1 µg 

0.08 µg

SF (ppm) & Bias (µg) 

2.0 

1.5 

1.0 

0.5 

0.0 

-0.5 

-1.0 

-1.5 

SF (1σ) = 0.56 ppm 

Bias (1σ) = 0.92 µg 

SF (ppm) 

Bias (µg) 

-2.0 

0.00 20.00 40.00 

Time (h) 

60.00 80.00 

Figure 11. Missile guidance SOA SF and bias stability. 

SF (ppm) & Bias (µg) 

0.5 

0.4 

0.3 

0.2 

0.1 

0 

-0.1 

-0.2 

-0.3 

-0.4 

4 Position Calibration 

SF (1σ) = 0.14 ppm 

Bias (1σ) = 0.19 µg 

-0.5 

0 10 20 30 40 50 60 70 

Time (h) 

Figure 12. SOA-SINS SF and bias stability. 

The SOA also demonstrates sub-ppm and sub-µg performance 

across multiposition tumble tests. The following error 

model was used to fit the tumble data: 

(19) 

where: 

aind = indicated acceleration output 

f = net (differenced) SOA frequency output 

SF = SOA scale factor (Hz/g) 

bias = SOA bias frequency 

ai = input axis applied acceleration 

am, ao = cross axis applied accelerations 

K2, K3 = second- and third-order SF nonlinearity 

Kim, Kio = cross axis coupling coefficients 

Kmm, Koo = cross axis nonlinearity coefficients 


Table 2 shows the nominal values of the above parameters 

for the SOA as determined in a 16-position tumble 

test. Note that a value for the Kmm and Koo coefficients 

can not be extracted from a single-orientation, multiposition 

tumble. The contribution of these terms is 

absorbed in the nominal values of bias and the K2 coefficient. 

Table 2 also shows the 1σ variation in the 

coefficients during the tumble, a test that takes 9 h to 

complete. 

Residual (µg) 

Table 2. Multiposition Tumble Results. 

Tumble about OA Tumble about MA 

Parameter Nominal Std. Dev. Nominal Std. Dev. 

Value (1σ) Value (1σ) 

Bias 797.965 0.354 797.724 0.295 

Hz µg Hz µg 

Scale Factor 126.697 0.989 126.722 0.822 

Hz/g ppm Hz/g ppm 

K2 121.33 0.652 117.49 0.542 

µg/g 2 µg/g 2 µg/g 2 µg/g 2 

K3 6.167 1.324 0.24 1.10 


KimKio -7.293 0.629 -6.65 0.523 


Misalignment 2.34 0.295 28.3 0.245 

mrad µrad mrad µrad 

Figure 13 is a plot of the residual (i.e., the difference 

between the actual SOA output and the fitted error 

model) for multiposition tumble tests about the instrument’s 

two cardinal axes. Note that the 1σ standard 

deviations of the residuals are less than 1 µg (0.95 and 

0.79 µg) and the residual plot does not show a systematic 

variation with table angle, an indication that there 

is not a major unmodeled error term present in the 

sensor. 

OA Residual (1σ) = 0.95 µg 

MA Residual (1σ) = 0.79 µg 

Table Angle (deg) 

OA Tumble 

MA Tumble 

2.0 

1.5 

1.0 

0.5 

0.0 

-0.5 

-1.0 

-1.5 

-2.0 

0 100 200 300 400 

Figure 13. Multiposition tumble residuals. 

11

Finally, Figures 14 and 15 show long-term (30-day) SF 

and bias stability measured on two SOA-SINS units (S/Ns 

028 and 063). The better of the two units demonstrates a 

1σ SF stability of 0.73 ppm and a 1σ bias stability of 

approximately 2 µg over 30 days. 

ppm 

µg 

12 

S/N 028 Avg. SF = 247.4 Hz/g 

S/N 063 Avg. SF = 258.8 Hz/g 

S/N 028 SF (1σ) = 0.73 ppm 

S/N 063 SF (1σ) = 1.22 ppm 

Days 

Figure 14. Long-term SOA-SINS SF stability. 

Figure 15. Long-term SOA-SINS bias stability. 


SF 063 

SF 028 

Linear (SF 063) 

Linear (SF 028) 

2.0 

1.5 

1.0 

0.5 

0.0 

-0.5 

dSF/SF = -0.102 ppm/day 

-1.0 

-1.5 

dSF/SF = -0.011 ppm/day 

-2.0 

0 5 10 15 20 25 30 35 40 

S/N 028 Bias (1σ) = 2.04 µg 

S/N 063 Bias (1σ) = 2.68 µg 

Days 

Bias 063 

Bias 028 

Linear (Bias 063) 

Linear (Bias 028) 

5.0 

4.0 

3.0 

2.0 

1.0 

0.0 

-1.0 

-2.0 

dBias = 0.220 µg/day 

-3.0 

-4.0 

dBias = 0.064 µg/day 

-5.0 

0 5 10 15 20 25 30 35 40 

CONCLUSIONS 

Draper Laboratory is currently developing the SOA, a 

MEMS-based inertial-grade sensor. Performance data 

acquired to date meet requirements for missile guidance 

missions and for high-performance navigation applications 

such as the SLBM SINS navigation systems. 

The missile and SSBN applications have significantly different 

environmental, acceleration dynamic range, and 

resolution requirements that are best satisfied by optimizing 

the SOA geometry for each application. The design 

flexibility and wafer-scale fabrication methods of the silicon 

MEMS process enable manufacturing both instrument 

designs with essentially zero incremental cost associated 

with the additional instrument assembly line. That is, both 

versions of the SOA share a common sensor package, electronics 

architecture, main housing, and instrument 

assembly process. 

The MEMS technology is low cost and offers a rapidly 

expanding commercial business base to leverage and sustain 

accelerometer production and deployment in 

next-generation strategic system applications. Integration of 

the SOA with mixed-signal, application-specific integrated 

circuit (ASIC) electronics offers the promise of a low-cost, 

small-size, low-power, and high-reliability strategic-grade 

instrument. 

REFERENCES 

[1] Tang, W., Electrostatic Comb Drive for Resonant Sensor and 

Actuator Applications, PhD Dissertation, University of California 

at Berkeley, November 21, 1990. 

[2] Kourepenis, A., J. Borenstein, J. Connelly, R. Elliott, P. Ward, M. 

Weinberg, “Performance of MEMS Inertial Sensors,” 24th Joint 

Services Data Exchange for Guidance, Navigation, and Control, 

November 16-20, 1998, Anaheim, CA. 

[3] Roark and Young, Formulas for Stress and Strain, 5th Edition, 

McGraw Hill, 1975. 

[4] Nayfeh, Mook, Nonlinear Oscillations, Wiley & Sons, New York, 

1979. 

[5] Hays, K.M., R.G. Schmidt, W.A. Wilson, J.D. Campbell, D.W. 

Heckman, M.P. Gokhale, A Submarine Navigator for the 21st Century, The Boeing Co., IEEE 0-7803-7251-4/02, 2002, pp. 

179-188. 

[6] Borenstein, J.T., N.D. Gerrish, R. White, M.T. Currie, E.A. 

Fitzgerald, “Silicon Germanium Epitaxy: A New Material for 

MEMS,” Mat. Res. Soc. Symp., Vol. 657, 2000, pp. 7.4.1.


(l-r) William Sawyer, Ralph Hopkins, 

Roy Setterlund, and Bruce Dow 

Ralph Hopkins is a Principal Member of the Technical Staff and Group Leader in the Guidance Hardware Division at 

Draper Laboratory. Currently, he is Technical Director of the SOA development program, a high-performance silicon 

MEMS VBA targeted for strategic-grade applications. He has also led and contributed to the development of navigation 

and tactical-grade MEMS gyroscopes and accelerometers, and high-performance electromechanical inertial sensors, 

such as floated instruments and dynamically-tuned gyros. He holds several patents, has authored several papers, 

and is an invited speaker for tutorials on inertial instruments and inertial technology. He is a current member of the 

AIAA Guidance Navigation and Control Technical Committee. Mr. Hopkins holds BS and ME degrees in Mechanical 

Engineering from Rensselaer Polytechnic Institute, an ME in Engineering Mechanics from Columbia University, and an 

MS in Engineering Management from The Gordon Institute of Tufts University. 

Joseph Miola joined the Laboratory in 1959 and was in the Systems Integration, Test, and Evaluation Division before 

his retirement in 2005. He worked in the development programs for Navy Polaris and Trident and for Air Force 

Minuteman and Peacekeeper guidance systems, primarily in accelerometer and gyro instrument design and test. 

Experience included over 25 years in management with Section Chief and Associate Director positions and Task Leader 

for the A-10 GPS/IDM Integration Test Program. Mr. Miola was responsible for the test and evaluation of the SOA 3 

accelerometer design. He received a BS in Engineering and an MS in Electrical Engineering from Northeastern University. 

Roy Setterlund is an Associate Director in Draper’s Strategic Systems Program Office and is responsible for Air Force 

ICBM and Navy/Air Force reentry programs. These programs consist of ICBM sustainment activities related to the 

Minuteman III guidance system, new strategic instrument development work as part of the Air Force’s Guidance 

Applications Program (GAP), various science and technology efforts sponsored by the Air Force Research Laboratory 

(AFRL), and reentry instrumentation and guidance development activities for the strategic Navy’s reentry branch, SP28. 

He also oversees Draper’s IR&D related to strategic systems technology. Prior to this, Mr. Setterlund was a Business 

Development Manager in Draper’s Space and Missile’s Program Office, where he concentrated on business development 

for space systems, including spacecraft attitude control systems, small satellites, and precision pointing and stabilization 

applications. He has published many papers for various professional journals and has worked at Draper Laboratory since 

graduating from MIT with an MS in 1973. 

Bruce Dow is a Program Manager in the Strategic Systems Directorate at Draper. He has over 15 years experience in 

systems engineering and test of advanced guidance, navigation, and control (GN&C) systems and over 3 years experience 

in technical director and program management positions supporting both Navy and Air Force ballistic missile 

guidance and reentry system applications. He has been a Systems Engineer for undersea and aerial autonomous vehicles 

and has also served as Technical Director and Program Manager for various ballistic missile guidance and reentry 

system programs in support of AFRL, Navy SP23, and Navy SP28. Mr. Dow has also served as Program Manager for 

the design and installation of DoD and DoJ intelligence network systems world-wide, including the White House 

Communications Agency, Navy Space Command, National Drug Intelligence Center, Office of Naval Intelligence, and 

Joint Analysis Center. Mr. Dow has spent 8 years on active military duty and is a Lieutenant Colonel in the U.S. Army 

Reserves, deployed from 2003 to 2004 for Operation Iraqi Freedom. He holds a BA from the U.S. Military Academy, 

and an MS in Systems Engineering and an MBA, both from Boston University. 

William Sawyer has been employed at Draper Laboratory for 7 years, during which time he has worked on the manufacturing 

process for several Draper inertial MEMS devices, including the Tuning-Fork Gyro (TFG)14 and TFG20. 

For the past several years, he has concentrated on the SOA. He developed the Bonded Etchback of Silicon (BESOI) 

process in collaboration with Jeff Borenstein. This work led to the patent for which Dr. Sawyer and Dr. Borenstein 

received Draper’s 2004 Best Patent Award. He received a BS in Physics from Colby College (1978), a Diplome (Masters) 

in Physics from Albert Ludwig University, Freiburg, Germany (1985), and a PhD in Physics from the University of 

Stuttgart, Stuttgart, Germany (1989). 

13

Autonomous Guidance, Navigation, and Control of 

Large Parafoils 

David Carter, Sean George, Philip Hattis, Leena Singh, Steven Tavan 

Copyright © 2005 by The Charles Stark Draper Laboratory, Inc. 

Published by the American Institute of Aeronautics and Astronautics, Inc., with permission 

INTRODUCTION 

Under the Joint Precision Airdrop System (JPADS) program, a Draper Laboratory autonomous 

guidance, navigation, and control (GN&C) software package that enables precision payload airdrop 

delivery using large parafoils has been developed in prototype form and successfully flight tested. 

The modular software design is structured to accommodate parafoil airdrop systems for payloads 

ranging from under 2,000 lb to over 30,000 lb. The initial GN&C software implementation has been 

demonstrated on the Para-Flite Dragonfly 10,000-lb class parafoil using an airborne guidance unit 

(AGU) provided by Wamore, Inc. and an avionics package provided by RoboTek. Among the 

primary avionics selection criteria was low component cost, resulting in the use of a processor with 

very limited data throughput capability. To accommodate the processor limits, the guidance 

algorithm includes table-driven trajectory data that guide the parafoil through precision finaldescent 

maneuvers while imposing very limited processor throughput burden. The GN&C 

algorithms and associated mission planning software have also been incorporated into the Precision 

Airdrop System (PADS) laptop personal computer. This accommodates easy, in the field, ground 

loading of the GN&C software onto the AGU and enables PADS updates of the airdrop system 

mission files during flight of the carrier aircraft to the airdrop release point. The details of the GN&C 

design and flight test results to date are discussed. 

A key element of the JPADS Advanced Concept 

Technology Demonstration (ACTD) is the development 

of GN&C software to autonomously fly the Dragonfly 

10,000-lb-capable parafoil. This software must guide the 

parafoil from deployment altitudes up to 25,000 ft above 

mean sea level (MSL) to landings within a 100-m circular 

error probable (CEP) of the target. Other key goals 

include robustness to a variety of failure modes, algorithms 

that are sufficiently generic to facilitate adaptation 

to both smaller and larger decelerators, efficient enough 

14 

to perform well on a very modest microprocessor, and 

capable of meeting system performance requirements 

with a navigation sensor suite limited by recurring system 

costs. Also important are government ownership of 

the resulting code, the ability to handle user-supplied 

waypoints, plus efficient and cost-effective integration 

with the previously developed Air Force and Army PADS 

mission planner. Draper Laboratory, one of the developers 

of the PADS planner, is implementing the 

autonomous GN&C.

The Flight Hardware Configuration 

As noted, an important goal of this program is to keep 

recurring system costs at a minimum. Hence, the avionics 

selected are the ultimate in simplicity. The sole navigation 

sensor is the CSI Wireless Vector dual-Global Positioning 

System (GPS) receiver, which provides not only system 

position and velocity, but also heading and heading rate. 

The Vector unit provides this by using two tightly coupled 

high-performance GPS receivers and two antennae, separated 

by 10 in, providing a good balance between heading 

accuracy and GPS acquisition time. This approach avoids 

including a magnetic compass for the heading reference, 

with its attendant difficulties due to local changes in the 

magnetic field and field perturbations from the various 

metal masses in the AGU. Other means of determining 

heading without additional heading sensors were examined 

and determined to be operationally unattractive. The 

flight processor selected is the Rabbit RCM3400 microcontroller, 

augmented by 8 MB of flash memory to hold 

guidance data tables and record GN&C parameters during 

flight for later analysis. A Freewave 902- to 928-MHz 

spread-spectrum modem is also utilized to receive remote 

control commands from the ground and to downlink mission 

data for post-flight analysis. This modem can be used 

to download mission files prior to airdrop system release, 

although 802.11-g wireless network components are 

being integrated as a replacement for operational use. 

Basis for the Initial Airdrop System Dynamics Models 

Draper has a long history of research in guided ram-air 

parafoils, beginning with work on the Precision Guided 

Airdrop System (PGAS). [1],[2] As part of this program, an 

engineering simulation was implemented based on 

numerous technical references detailing the appropriate 

structure for a 6-degree-of-freedom (DOF) parafoil 

dynamics model complete with high-fidelity environment 

models. The initial Dragonfly dynamics models were 

derived from this simulation framework and then updated 

substantially using the best available experimental and 

theoretical analyses of large-scale parafoil systems. In particular, 

a wealth of incredibly detailed flight performance 

data have been generated by the NASA X-38 program on 

two parafoil systems in the same size category as the 

Dragonfly system. [3],[4] In addition, an updated theoretical 

treatment of apparent mass effects was undertaken in 

Reference [5], the results of which were rolled into the initial 

Dragonfly simulation. It was understood from the start of 

the program that there would be opportunities for 

conducting performance estimation tests on the new 

canopy, therefore, most of the initial work focused on 

developing tools that could be used to quickly update the 

original models as performance data were collected. It 

should be understood, however, that due to the lack of 

inertial instrumentation onboard the Dragonfly system, as 

well as a program focus on cost effectiveness, no attempt 

has been made to complete a detailed system identification 

of all parameters in the Dragonfly parafoil model. Rather, 


the focus of the modeling effort has been on matching: (1) 

steady-state velocity and turn rate, (2) toggle line actuator 

dynamics, and (3) the basic turn rate time constant. Given 

the uncertainty in measurement data as well as wind 

conditions during flight testing, it is believed that engineering 

models suitable for GN&C development have 

been generated. 

Analysis tools have been developed that attempt to fit 

longitudinal lift-to-drag and velocity flight data to relatively 

simple mathematical expressions for the lift, drag, and 

pitch moment characteristics of the parafoil. The parafoil 

aerodynamics model is used to generate tabular force and 

moment coefficient data as a function of brake setting and 

angle of attack as inputs into the simulation. Values for 

various aerodynamic parameters were originally set using 

past historical data collected on PGAS as well as general 

trends seen in the X-38 program. The process to ascertain 

whether the parametric aerodynamics model had sufficient 

complexity was to attempt to match X-38 L/D and velocity 

performance. The aerodynamics model was able to reasonably 

produce the steady-state response for this large 

parafoil airdrop system; therefore, there was confidence 

that subsequent test data on the Dragonfly could be used 

to give a good engineering model of the flight performance. 

The main uncertainty in “fitting” aerodynamic 

parameters against the flight data was that, unlike X-38 

flights, angle-of-attack was not measured explicitly by the 

Dragonfly AGU. Instead, trends from the X-38 program, 

coupled with ensuring physically reasonable constraints 

on aerodynamic parameters, were used to help produce a 

longitudinal model for the Dragonfly. Detailed analysis of 

flight data is shown in the Flight Test Program section: 

“System Identification from Flight Test Data,” with comparisons 

from the parafoil model updated to match the 

steady-state response of the system. 

The lateral model parameters for the Dragonfly were based 

on both theoretical calculations as well as appropriately 

scaled historical data from the PGAS and X-38 programs. 

Originally, a very complicated yaw rate response model 

was used that tried to match the nonlinear steady-state 

turn characteristics seen with the X-38 at low brake 

settings. Subsequent flight tests did not show this type of 

behavior, therefore, the turn rate response was augmented 

to a more linear relationship with differential toggle (see 

the Flight Test Program section: “System Identification 

from Flight Test Data”). The turn rate dynamics are generally 

dominated by two factors: the inertial (both “real” and 

apparent) properties of the system and the toggle actuator 

response. Effort was given to an appropriate dynamics 

model for the toggle motors. Generally, the Dragonfly 

motors are capable of ~2 ft/s of maximum line pull rate 

and have relatively high acceleration characteristics. Fullscale 

toggle changes from 0 to 100 in general take ~5 s. In 

addition, yaw inertial and damping characteristics were 

originally chosen to give the parafoil turn rate response a 

5-s time constant. The overall sluggish turn behavior of 

15

the parafoil was well modeled by the original model 

parameters used, and generally corresponded to a fairly 

highly damped large inertia system. Subsequent flight 

tests were used to slightly modify the parameters, but in 

general, the original lateral model was sufficient to give the 

gross dynamic behavior of the Dragonfly system. 

The GN&C Design 

Top-Level Overview 

The GN&C flight software resides in the AGU. While the 

airdrop system is in flight, the software receives information 

from a navigation package about system position, velocity, 

and heading. Based on these inputs, it computes a trajectory 

toward the programmed target landing point. It then flies 

the system toward that point via a series of extensions 

and/or retractions of the parafoil’s two control lines. 

After a stable canopy opening, the GN&C software optionally 

enters a control line trim mode in which it adjusts the 

control line lengths for straight flight. After completing (or 

skipping) the trim mode, state estimates derived by filtering 

the position, velocity, and heading inputs from the navigation 

package are utilized by guidance to begin directing 

the control algorithms to home the airdrop system toward 

the target. When the airdrop system comes within about 

200 m horizontal distance from the target, it enters an 

energy management mode, flying figure-eight patterns in 

the vicinity of the target until ground-relative altitude 

drops below 500 m, at which point, it steers toward the 

target to establish the final approach. From this point until 

landing, the guidance software utilizes precomputed steering 

commands from a lookup table with a large family of 

trajectories that minimize final position and heading error. 

This table-driven approach minimizes the trajectory determination 

processing burden on the available, small, 

low-throughput flight processor. 

The navigation instrument is a CSI wireless, dual-GPS 

receiver-based navigation package that generates position, 

velocity, and heading data. The processed GPS data from 

the navigation package is filtered by an algorithm in the 

GN&C software, with the resulting state estimates sent to 

the guidance software in the form of airdrop system altitude, 

heading, and estimated wind velocity. The wind estimates 

are derived by comparing the GPS velocity data with a 

model of the expected parafoil air speed. 

Using gain scheduling, the control software attempts to 

maintain the heading rate commanded by the guidance 

software by extending or retracting the left and right 

parafoil control lines. The control software also attempts to 

maintain symmetric operation of the control lines about a 

base retraction length when determining the line positions 

that will achieve the desired heading rate. Also, a parafoil 

flare capability, achieved by fully retracting both control 

lines, achieves a stall condition that is used just before 

touchdown to reduce the parafoil’s ground impact velocity. 

Development of the GN&C software began in 

Government FY 2004. Draper developed, integrated, and 

16 


validated the autonomous GN&C flight software for initial 

use on the Para-Flite 10,000-lb-capable parafoil equipped 

with the Wamore, Inc. AGU and RoboTek-supplied avionics. 

All aspects of the GN&C design are modularized to readily 

accommodate eventual adaptation to other smaller and 

larger guided parafoil airdrop systems. Also, both the 

GN&C software and mission planning algorithms associated 

with the use of guided airdrop systems that apply the 

GN&C algorithms have been hosted onto the PADS 

mission planner laptop personal computer (PC). This will 

facilitate the use of the PADS PC as a common platform to 

download the GN&C software and data files to the AGU on 

the ground, as well as to generate airdrop load-specific 

mission files shortly before release onboard the carrier aircraft. 

GN&C Development, Integration, and Test Methodology 

The development, integration, and test methodology 

applied by Draper staff has followed a rigorous process 

tailored to the flight prototype demonstration objectives 

applicable to the current program. The key process steps 

are summarized as follows: 

• Algorithm conceptualization, preliminary design, and 

evaluation: This development phase involved preliminary 

algorithm design and off-line evaluation of the 

algorithms by the individual developers. This process 

included internal and external (customer) reviews of the 

proposed algorithms before their integration into a 

complete GN&C implementation. 

• GN&C integration and software simulation assessment: 

A 6-DOF model of the parafoil dynamics was formulated 

and validated against available parafoil test data. A software 

simulation was developed using the 6-DOF 

parafoil model that provided data interfaces for GN&C 

software equivalent to those on the AGU. The GN&C 

algorithms were then integrated together and into the 

simulation. Closed-loop GN&C assessments were then 

performed using this simulation, the results of which 

were subjected to internal and customer review. This 

simulation tool has subsequently supported the evaluation 

of flight test results. 

• Hardware-in-the-loop (HWIL) simulation assessment: 

An HWIL simulation was assembled at Draper, using 

actual AGU processors and memory as well as a set of 

control line actuation motors. This HWIL facility was 

used to evaluate proper code function with the real 

timing, space, and word-length limitations of the target 

processor and memory, and was used to demonstrate 

proper message handling by the control actuation 

motors. Subsequently, the CSI navigation package was 

added. The navigation package provided a means to 

evaluate proper message interfacing, although the static 

nature of the HWIL simulation assembly prevents the 

use of the actual navigation package during flight 

dynamics simulations. 

• Preflight code validation and freeze: Prior to each flight 

test cycle, the intended GN&C implementation was

integrated and validation tested, first on the software 

simulation, and then on the HWIL simulation. When 

the intended performance was realized, the GN&C software 

was frozen as a defined code release under a 

configuration control process, and it was then delivered 

for field test use. A final integration test was performed 

at the AGU vendor’s facility prior to the first flight of 

each new software release. 

Guidance Implementation Details 

The primary function of guidance is to compute steering 

commands for use by control, given position, heading, and 

estimated wind profile from navigation. As a part of the 

steering calculation, guidance is responsible for timing the 

flare (deep brakes) maneuver prior to landing, signaling 

mode control to transition to flare; mode control then 

commands the other parts of the system accordingly. 

Throughout flight, guidance accepts the system mode 

from mode control. The mode can be any of the following: 

initialize, preflight, trimflight, autoflight, manual, or terminal. 

Steering calculations, including timing of the flare maneuver, 

are performed only when system mode is autoflight. When 

the mode is trimflight, guidance monitors altitude and distance 

to the first waypoint (which could be the target); if 

the ratio of distance to altitude becomes sufficiently small, 

guidance requests that trim be abandoned, causing transition 

to autoflight mode. In the modes initialize, preflight, 

manual, and terminal, guidance monitors system position 

but is otherwise inactive. 

Guidance obtains MSL altitude, the north and east 

position coordinates with respect to the target, flight path 

azimuth (the direction of the air velocity vector), as well as 

a table of wind velocity vs. altitude from navigation. It also 

accepts a list of waypoints, given as north and east offsets 

from the target, from the mission loads file. During the 

autoflight mode, guidance computes the commanded rate 

of change of the flight path azimuth, which is referred to 

as “turning rate,” and sends this to control. Guidance 

interfaces are shown in Figure 1. 

Mode 

Control 

Mode 

Navigation Wind Table Guidance 

Mission 

Loads 

h, x, y, χ 

Waypoints 

Figure 1. Dragonfly guidance interfaces. 

Guidance does its calculations in a deweighted, wind-fixed 

frame; a portion of the anticipated displacement due to 

wind is added to the navigated position with respect to 

the next waypoint (or the target) so that steering calculations 

can be done as if winds were zero. A nominal, single 


χ, dχ/dt 

Submode 

Abandon 

Trim 

Control 

Mode 

Control 

waypoint trajectory in this wind-fixed frame is shown in 

Figure 2. The guidance strategy is best understood by 

considering this trajectory, which is marked to show the 

different flight phases (modes and submodes). 

Due North (>0) target relative position (ft) 

7000 

6000 

5000 

4000 

3000 

2000 

1000 

0 

-1000 

Table 

Lookup 

Trim 

Homing 

Flare 

Energy 

Management 

-6000 -5000 -4000 -3000 -2000 -1000 0 1000 2000 3000 4000 

Due East (

Navigation Implementation Details 

Navigation uses data from a dual-antenna GPS navigation 

package to compute position and flight path azimuth for 

use by guidance, and flight path azimuth and flight path 

azimuth rate (turning rate) for use by control. Using GPSmeasured 

velocity with respect to the ground together 

with a simple analytic model of system air speed and GPSmeasured 

heading, navigation estimates wind velocity. The 

wind velocity estimate is used to correct an a priori table 

of wind vs. altitude; the correction is largest for altitudes 

close to the altitude at which navigation’s wind estimate is 

valid. Navigation sends the corrected wind table to guidance. 

Early in the flight, navigation monitors the sink rate 

in order to detect canopy inflation; this information is 

important for system moding. Navigation interfaces are 

shown in Figure 3. 

Figure 3. Dragonfly navigation interfaces. 

Navigation receives standard National Marine Electronics 

Association (NMEA) data messages from the dual-antenna 

GPS navigation system. Coordinated Universal Time 

(UTC), latitude, longitude, and altitude MSL are 

obtained from the “GGA” data message. The GGA message 

also contains a quality indicator, the number of 

satellites used in the position calculation, and an estimate 

of horizontal dilution of precision (HDOP). Navigation 

considers the GPS position data to be valid provided the 

time has been updated from its previous value, the quality 

indicator takes values of 1 or 2, the number of 

satellites used in the position fix is at least 4, and HDOP 

does not exceed a configurable threshold whose current 

value is 10. 

Navigation receives ground velocity (speed and course) 

from the “VTG” data message. This message is considered 

to be valid whenever its numeric fields are fully populated. 

Navigation receives true heading from the “HDT” 

data message and heading rate (rate of turn) from the 

“ROT” message. These messages are considered valid 

when their numeric fields are populated; this happens 

only when the receiver has lock on both channels. 

When the HDT message is valid, navigation sets the 

flight path azimuth equal to the HDT heading. This 

amounts to asserting that sideslip is zero, an approximation 

that seems to work well for low bandwidth GN&C 

18 

Mode 

Control 

GPS 

Interface 

Mission 

Loads 

Mode 

h, x, y, χ 

t, h, lat, lon, v, 

Wind 

course 

Table 

ψ, dψ/dt, χ, dχ/dt, 

Validity Data 

Impact Point 

Navigation 

Submode 

Release Point 

a priori 

Wind Table 


Submode 

Sink Rate 

Stabilized 

Guidance 

Control 

Mode 

Control 

of large parafoil systems. Likewise, navigation identifies 

flight path azimuth rate with heading rate from the ROT 

message when this message is valid. Wind estimates are 

made only when valid GGA, VTG, and HDT data are 

available. Horizontal velocity with respect to the ground 

is obtained from the VTG message. Altitude is obtained 

from GGA and used together with data provided in the 

mission loads file to compute atmospheric density and 

nominal air speed (which depends on atmospheric density). 

Air velocity is estimated using computed nominal 

air speed, nominal flight path angle, and measured heading 

from HDT; the calculation assumes that sideslip is 

negligible. The air velocity estimate is low-pass filtered 

and subtracted from ground velocity to obtain the wind 

estimate. 

Control Implementation Details 

When formulating the control algorithm for Dragonfly, 

we stipulated the following control objectives: (1) the 

closed-loop system must track wind-relative lateral rate 

commands issued from guidance, and (2) the control 

algorithm must reject external disturbances, in particular, 

the frequency of the payload oscillation relative to the 

canopy. In this phase of the program, control only regulates 

lateral directional heading rate errors based on 

commands produced by guidance; flight speed is treated 

as a system parameter governed by the base (nominal 

brake setting) deflection during the flight. From initial, 

manually-controlled, system-identification parafoil flight 

drops, we identified the heading and sideslip rate 

dynamics in response to the lateral controls or differential 

toggle commands (right toggle – left toggle). We 

operate the vehicle at a 25% base deflection of 50 in 

since this allows the parafoil turning dynamics to be 

nearly linear and ensures that flight operations do not 

show the reflexive behavior that was observed in the 

small toggle deflection dynamics of the X-38 parafoil 

canopy. Around this base deflection, we collectively represented 

the heading and slideslip lateral dynamics mode 

shapes in terms of flight-path azimuth (χ = Ψ – β); 

experimental results showed that heading rate and 

slideslip had nearly identical natural modes, so we were 

able to collect the terms. We identified a second-order 

heading rate model shown in Eq. (1) using parameter 

identification techniques: 

(1) 

where ∆ is the control deflection difference between the 

left and right toggles and serves as the lateral control 

input. G represents the open-loop plant heading rate 

transfer function. Because we operate around a 25% base 

deflection, we achieve the desired deflection difference 

by symmetrically operating the line deflections about the 

base setting. Similarly, we identified the plant model 

parameters for symmetric toggle commands produced

about a 25% base deflection. The lateral control toggle 

commands, ∆, are issued by symmetrically deflecting the 

left and right toggles around the base deflection, b: δL = 

b - ∆/2; δR=b + ∆/2. 

Figure 4 shows the measured values of Dragonfly heading 

rate (dΨ/dt) and its sideslip rate in response to a 

100-in differential control toggle step from which the 

plotted flight path azimuth rate quantity was also 

derived. Based on simulations, it was determined that the 

natural open-loop bandwidth was 12.5 s with a damping 

coefficient of about 0.6. Figure 5 shows the derived 

azimuth rates obtained from the parafoil at two different 

toggle deflections as well as the corresponding response 

of the simulated second-order plant model that has the 

input gain, bandwidth, and damping characteristics 

determined from system identification. The second-order 

model can not capture very precisely the small overshoot 

combined with large settling times seen in the high 

deflection regimes. Nevertheless, the rise time, overshoot, 

and time constant are matched quite well. 

Heading Rate (deg/s) 

12 

10 

Azimuth Rate 

8 

6 

4 

Heading Rate 

2 

0 

-2 

-4 

Sideslip Rate 

-6 

0 10 20 30 40 50 60 

Time (s) 

Figure 4. Measured Dragonfly rate response for a 100-in 

differential control toggle step. 

Heading Rate (deg/s) 

12 

10 

8 

6 

4 

2 

Time (s) 

Figure 5. Comparison of predicted and flight-measured 

Dragonfly rate response at two differential 

control toggle settings. 


Azimuth Rate at 100-in deflection 

(from Flight Test Data) 

Azimuth Rate from Second-Order 

Matched Model at 100-in deflection 

Azimuth Rate from Second-Order 

Matched Model at 40-in deflection 

Azimuth Rate at 40-in deflection 

(from Flight Test Data) 

0 0 10 20 30 40 50 60 

Figure 6 shows the feedback control block interconnections. 

The controller rate errors are based on commands 

from guidance; it forms the left and right toggle commands 

to the motor controllers. Plant response is measured relative 

to the wind frame. Disturbances result from unexpected 

wind gusts, unknown and unmodeled plant dynamics, 

and payload oscillation frequencies that appear in the 

feedback signal. Based on this plant model, we identified 

control gains for a proportional, integral, derivative (PID) 

control structure that maximized the closed-loop bandwidth, 

provided at least 60 deg of phase margin, and 

minimized the gain of the auxiliary output-input loop to 

limit the impact of the payload oscillation frequency on the 

applied control. Since the payload oscillates with pendular 

motion beneath the AGU, these oscillations are picked up 

by navigation sensors and appear in the feedback signal 

even though the canopy itself does not exhibit significant 

oscillations. Therefore, one control objective was to synthesize 

the control gains so that the control line deflection 

command sent to the motors was desensitized to this 

disturbance signal and did not attempt to correct for this 

spurious measurement. Thus, we designed the control 

gains to manage the gain and phase margins of three significant 

transfer functions shown in Eq. (2). The transfer 

function P measures the gain between the commanded 

line deflection and the heading rate command error. S 

is the sensitivity function and T the closed-loop plant 

transfer function. 

Command 

from 

Guidance 

Plant Lateral 

Controller Dynamics 

r + 

- K G + 

ε 

Disturbance 

δleft 

δright 

Ψ Ψmeas 

Figure 6. Dragonfly feedback control block interconnections. 

We posed a multi-objective optimization problem and 

solved for the control gains that maximize three gain and 

phase quantities at suitable frequencies. This produced a 

control structure: , where (k1, k2, 

k3) are the outputs of the robust optimization problem. 

We used the following control gains in the controller: (k1 

= 1.65, k2 = 3.0, k3 = 1.07). The controlled-system transfer 

function characteristics are shown in Figures 7 and 8. We 

designed the control gains to maximize the bandwidth and 

phase margin of the controlled-system transfer function 

(T), minimized the gain of the input transfer function (D) 

at the payload natural frequency, and maintained low sensitivity 

at all frequencies. 

(2) 

19

Magnitude (dB) 

Phase (deg) 

-270 

10 

Frequency (rad/s) 

Figure 7. Dragonfly controlled system open-loop Bode plot. 

-1 100 Magnitude (dB) 

Phase (deg) 

0 

10 

Frequency (rad/s) 

Figure 8. Dragonfly controlled system transfer function 

(GK) sensitivity features. 

-1 100 20 

0 

-20 

-40 

-60 

-80 

0 

-90 

-180 

10 

0 

-10 

-20 

90 

60 

30 

Guided/Smart 

Airdrop 

Systems 

Navigator or 

Navigation 

Systems 

System Transfer Function 

Sensitivity Diagram 

Air Force Weather 

Agency Atmospheric 

Forecast Model - High: 

Resolution Nested Grid 

Surrounding Drop Zone (s) 

INTERNET/SIPRNET 

Mesoscale 

4D Field 


PADS Laptop Computer 

3-D Field - Wind, Density, 

Pressure from Drop Time 

Reference 

Ballistic 

Trajectory 

5-km Grid Domain within 

15-km Grid Domain 

Assimilation 

Processor 

Computed Air 

Release Point 

(CARP) 

Supporting PADS Mission Planning and File Download 

Capability 

The PADS program has developed a laptop PC-based airdrop 

mission planning capability to support the determination of 

proper airdrop release points and to enable updates to guided 

airdrop system mission files while aboard the carrier 

aircraft in transit to the drop zone. The PADS implementation 

architecture is shown in Figure 9. The PADS PC includes 

a means to access current meteorological information and 

uses the altitude-dependent wind and density data, combined 

with models of the release and flight dynamics of 

airdrop systems to derive optimized computed air release 

points (CARPs) for unguided airdrop systems and allowable 

release envelopes for guided airdrop systems. Also, for the 

guided airdrop systems, nominal CARPs are designated 

within the derived release envelope. Meteorological data can 

be collected by PADS from any combination of the following 

sources: forecasts loaded before takeoff; data received 

through an encrypted satellite link during transit flight to the 

drop zone; and sondes released from or near the carrier aircraft, 

with the data retrieved by PADS through the carrier 

aircraft UHF antenna and processed into suitable form by 

software within PADS. All the available meteorological data 

are assimilated within PADS into a best estimate of current 

conditions near the drop zone. 

PADS has an interface to connect to a remote terminal on 

the carrier aircraft 1553 data bus to acquire the current 

vehicle navigation state, to obtain the current aircraft ataltitude 

wind measurement, and to monitor various 

airdrop-related mission parameters. The vehicle navigation 

state is used to enable PADS to display the aircraft position 

relative to the CARP and/or release envelope on a 

Airdrop 

Dynamics 

Simulation 

Figure 9. The PADS planning system architecture. 

Wind Data Sources 

• Satellite-Derived 

• TACMET Radiosonde 

• Theater Pilot Reports 

Combat Track II 

Radio 

Receiver 

Secure 

Interface 

Dropsonde 

Processor 

Radio 

Receiver 

Aircraft 1553 

Data Bus 

Communications 

Satellite 

Aircraft 

Top 

Antenna 

Aircraft 

Bottom 

Antenna 

GPS 

Dropsonde

FalconView graphical map interface (GMI). The at-altitude 

wind measurement serves as an additional source of meteorological 

data that is assimilated with the other available 

data. The airdrop-related mission parameters are recorded 

during release operations to enable post-release assessment 

of the release condition accuracy. 

PADS also includes a wireless link to enable loading mission 

file updates generated by PADS to guided airdrop systems 

while onboard the carrier aircraft. These mission file 

updates reflect the PADS-computed meteorological state in 

a format uniquely designed to be compatible with each airdrop 

system class. 

More details about the PADS design, current features, and 

flight test experience are provided in References [6] 

through [9]. Limited quantities of the PADS units are in 

field use supporting mission planning with unguided airdrop 

systems. Updates to PADS to support field use for 

mission planning and mission file updates for several 

classes of guided airdrop systems are now in work. Incountry 

operational demonstrations of these systems are 

expected presently. 

A version of PADS has been updated to include models of 

the Para-Flite Dragonfly 10,000 lb-class parafoil. This 

enables the PADS computer to provide mission file 

updates to the AGU prior to flight test release of the 

parafoil in tests of the autonomous GN&C system. This 

version of the PADS PC also includes a load of the GN&C 

software to enable the use of the same PC in the field to 

load that software onto the AGU via the wireless link or a 

temporary cable connection. This added PADS capability 

demonstrates the intended use of PADS as a common 

platform for preflight and in-flight support of all future 

airdrop operations from Air Force carrier aircraft (e.g., C- 

17s and C-130s). 

Flight Test Program 

Flight Test Program Overview 

Flight testing relevant to the development of the 

autonomous GN&C software commenced in March 2004 

at Red Lake in Kingman, Arizona. Initial flights in March 

and April were remote controlled, executing planned 

maneuvers to establish the flight characteristics of the 

Dragonfly system. Draper used the results of these flights 

to conduct system identification and to establish GN&C 

parameters as described elsewhere in this paper. First 

flight of the autonomous flight software occurred in May 

2004. Testing has continued since then at approximately 

6-week intervals, with flights starting in October at the 

Corral Drop Zone (DZ) at Yuma Proving Ground (YPG), 

Yuma, Arizona. During this time, the GN&C software was 

matured in parallel with the evolution of the canopy, 

rigging, and airborne hardware, including a major 

upgrade to the AGU involving new actuation motors, 

necessitating revised flight software motor interfacing. The 

move to YPG was a milestone as this was the first time the 


system flew from a C-130 airplane, deploying at 130-kn 

indicated air speed (KIAS), considerably faster than the C- 

123 used in Kingman. Flights from military aircraft 

commenced in February 2005. As the flight test program 

proceeded, system weights were gradually increased up to 

the Dragonfly maximum of 10,000 lb, as were drop altitudes, 

heading toward a goal of flights from 18,000 ft MSL 

by spring 2005. Initial autonomous flights were deployed 

directly over the targeted impact point, and then gradually 

more offset from the target was introduced. GN&C software 

was initialized in early tests assuming no winds, then 

forecast winds were used, and eventually flight tests will 

include updates of the GN&C mission file while en route 

to the DZ with current winds estimates based on an assimilation 

of forecast and dropsonde wind and density data. 

System Identification from Flight Test Data 

As mentioned previously, flight tests were used to update 

the original models of the Dragonfly. A sequence of manual 

input tests were conducted to capture isolated flight 

conditions under varying brake and differential toggles. 

Longitudinal tests using nominally zero turn maneuvers 

under varying brakes were used to generate a map of the 

glide slope (lift-to-drag (L/D)) and speed characteristics 

of the parafoil over a prescribed flight envelope. Figures 

10 and 11 show some results from these tests, including 

a comparison with the current parafoil model. Figure 10 

demonstrates a clear reduction in glide slope with 

increasing brake setting; however, the loss in performance 

is very gradual up until more than 50 in of toggle. 

Tests have been planned to examine performance at toggles 

exceeding 100 in, to capture the stall characteristics 

of the parafoil, but at the limits tested thus far, no collapse 

of canopy cells is evident from video or flight data. 

Error bars on the L/D flight data are particularly large 

because of the large noise generated by the GPS navigation 

package in the vertical channel. The model 

comparisons show that a relatively simple aerodynamics 

model can be used to capture most of the steady-state 

performance of the system. Figure 11 shows the speed 

envelope of the parafoil over the range of tested brake 

settings. The original toggle limit of 100 in (it is now 200 

in) resulted in nearly 20-ft/s variation in flight speed. 

Lift-over-Drag 

5.0 

4.5 

4.0 

3.5 

3.0 

2.5 

2.0 

1.5 

1.0 

0.5 

Flight Data 

Model 

0.0 

0 20 40 60 80 100 120 

Brake Command (in) 

Figure 10. L/D ratio vs. brake setting. 

21

Sea-Level Airspeed (ft/s) 

0 

0 20 40 60 80 100 120 

Brake Command (in) 

Figure 11. Sea-level air speed vs. brake setting. 

Manual steady-state turn rate tests were also conducted on 

the Dragonfly system. Figure 12 shows the turn rate vs. 

differential toggle deflection. Several conclusions can be 

drawn from this plot: (1) maximum steady-state turn rate 

is ~9 deg/s at a toggle of 100 in, (2) there is a considerable 

amount of yaw oscillatory noise that permeates the heading 

data channel causing significant variance in the data, and 

(3) the data collected show a near linear relation between 

differential toggle and turn rate. The noise in the yaw 

channel appears to be caused by some relative motion 

between parafoil and payload (with the AGU that holds 

the GPS navigation system located close to the payload), 

combined with wind perturbations. Small yaw oscillations 

persisted throughout most flight tests and did not appear 

to correspond to changes in the commanded turn rate of 

the parafoil. These oscillations were not indicative of the 

highly damped dynamics of the parafoil’s general turn rate 

response. Earlier X-38 studies showed a nonlinearity in 

turn rate performance at low brake settings that resulted in 

nearly zero response up to almost 35-40% of the stall differential 

toggle limits. There is insufficient data at this time 

to ascertain the exact low-brake response characteristics of 

the Dragonfly system since we have conducted the majority 

of our autonomous flights near 50 in of brake to reduce 

risk and complexity. Future plans call for additional low 

brake setting turns that should help document if a turn 

rate deadband exists. Thus far, we have seen no evidence 

that the Dragonfly exhibits this nonlinear turn-rate behavior. 

Turn Rate (deg/s) 

22 

70 

60 

50 

40 

30 

20 

10 

15.0 

10.0 

5.0 

0.0 

-5.0 

-10.0 

Flight Data 

Model 

-15.0 -125 -100 -75 -50 -25 0 25 50 75 100 125 

Differential Toggle (in) 

Figure 12. Turn rate vs. differential toggle setting. 


The variability in the observed Dragonfly turn rate at 

zero toggle deflection (seen in Figure 12) results from a 

combination of some of the effects just noted (rotation of 

the payload relative to the canopy and wind-driven 

canopy rotation) as well as some asymmetric control 

response. The asymmetric control effect varied from 

flight to flight, possibly indicating some flight-specific 

rigging effects and/or individual control motor response 

variations. 

Flight data analysis also resulted in some changes to the 

toggle actuator model that included the effects of 

aeroloading. Data collected on commanded line toggle 

position vs. actual position indicated some response 

degradation at higher brake settings, indicative of a falloff 

in motor response. Revised torque limits and a more 

elaborate line acceleration sensitivity to load were added 

to the general actuator model to represent this behavior. 

The response characteristics of the first-generation AGU 

motor also resulted in a redesign of the toggle actuator 

models to increase torque limits on the lines. Insufficient 

data were collected on these motors from flight tests; 

however, updates to the simulated actuator models has 

already been completed based on manufacturer specifications. 

In general, the torque limits of the motors did not 

impact system response except during flare maneuvers. 

Some small modifications to the lateral model parameters 

were undertaken following test flights. In general, the 

focus was on ensuring proper steady-state response characteristics, 

but in addition, some small changes were 

made to ensure that the turn rate characteristics matched 

flight data, which showed a time of ~10 to 12 s from the 

commanded differential toggle until steady-state turn 

rate was reached (with the indicated time including actuator 

response). 

GN&C Flight Test Performance and Associated Design 

Evolution 

Table 1 gives a summary of the Dragonfly flight tests to date 

that have provided canopy dynamics data in response to 

scripted manual remote control (R/C) inputs or that enabled 

autonomous GN&C flight with onboard logging of system 

performance data. Other flight tests that experienced prototype 

avionics or canopy-related hardware failures that 

precluded GN&C-related testing or that inhibited essential 

data logging have not been included in the table. The five listed 

flight tests in March-April 2004 were performed under 

R/C to support system identification objectives. The first tests 

of the autonomous GN&C occurred in May 2004. Post-flight 

analysis revealed that large misses in these initial autonomous 

GN&C tests were due, at least in part, to unintended limiting 

of toggle commands by the flight software. This software error 

was corrected before the next test cycle in August 2004. 

Some of the initial flights in August experienced read/write 

failure of the flash memory chip used for in-flight data 

logging and for storage of the large data table applied by 

the lookup terminal guidance algorithm. Consequently, to

Table 1. Initial GN&C-Related Dragonfly Flight Test Results. 

Date Control Drop Drop Release GRW Miss Comments 

State Aircraft Speed Altitude (lb) Dist (m) 

(KIAS) (ft MSL) 

3/1/04 R/C C-123K 110 9K 8K n/a Good flight characterization data 



3/5/04 R/C C-123K 110 9K 8K n/a Good flight characterization data; hard landing but no damage 


5/20/04 Auto C-123K 110 12K 8K ~400 Excellent deployment and autoflight; very good landing 

5/21/04 Auto C-123K 110 12K 8K ~1000 Excellent deployment and autoflight; very good landing 

8/12/04 Auto C-123K 110 12K 8K 183 Guidance table mode disabled; very soft landing 

12/8/04 Auto C-130A 130 10K 8K 142 Extended drogue descent; good canopy opening; good energy 

management and final approach/flare 

12/10/04 Auto C-130A 130 10K 10K 170 Extended drogue; good canopy opening; auto, high offset, 

navigation to target; good final approach/flare 

2/2/05 R/C, Auto C-130H 130 14K 8K 256 Extended drogue phase; good canopy opening; auto, high 

offset, navigation to target; good final approach/flare 

avoid additional flash memory problems during that flight 

test cycle, the lookup algorithm was disabled for the flight 

on August 12, and the system was allowed to fly to the 

ground using the energy management technique (circles 

with adjustable radius, which was the baseline at the time, 

rather than figure-eights). This was moderately successful. 

Despite a series of deployment and avionics malfunctions 

in October and December 2004, two drops on December 

8 and 10 enabled reasonably successful GN&C tests; the 

system deployed well and flew autonomously to within 

140 m and 170 m, respectively, of the target. 

The February 2005 test series was the first to use Wamore’s 

second-generation AGU. Of the two drops performed in this 

series, on February 2, autonomous GN&C usage was 

enabled on one. While the apparent GN&C performance on 

this flight was somewhat poorer than anticipated, that deficiency 

was later traced to a communication glitch with the 

servo motors in the new AGU, which has since been rectified. 

Next Design and Development Steps 

Given the generic, modular architecture of the 

autonomous GN&C software, its adaptation to fly other 

parafoils will be straightforward. This year, flight characterization 

of two subscale models of 30,000-lb (30 klb) 

capable parafoils are planned, followed by adaptation of 

the existing GN&C software and then autonomous flight 

tests of these canopies. Following this, it is expected that 

the software will be adapted to fly the full-size 30-klb 

systems when they are developed. 


Through the Small Business Innovative Research (SBIR) 

program, the Natick Soldier Center (NSC) is developing 

two navigation sensors to enhance the landing precision of 

guided airdrop systems. The Dragonfly parafoil with the 

GN&C flight software described herein will be the initial 

flight test vehicle for these sensors. The most mature of 

these sensors, in the middle of Phase II at this time, is a 

precision ground-relative altitude sensor under development 

by Creare, Inc. of Hanover, NH. Utilizing primarily 

sound detection and ranging (SODAR) to detect the 

ground over varied terrain, including through foliage, this 

sensor will provide height precision of ±1 ft during the 

terminal phase of the flight, allowing GN&C to time the 

final braking maneuver precisely, thereby increasing landing 

accuracy. This sensor will be small and lightweight (less 

than 5 lb) with a recurring cost of less than $500 per unit, 

cheap enough to be installed in just about any airdrop 

system from the 2000-lb class and up. Testing of this sensor 

will take place this calendar year. 

Wind uncertainty remains a significant error source for airdrop 

systems despite the improvements provided by the 

PADS mission planner discussed above. For guided systems, 

which have varying degrees of wind penetration capability, 

the wind uncertainty at lower altitudes when there is less 

time for correction is a significant problem. Also under 

development, nearing completion of Phase I SBIR trade 

studies, are several competing light detection and ranging 

(LIDAR) wind sensors that will provide real-time wind 

23

speed and direction to the GN&C software along the sensor 

line of sight ahead of the vehicle. This will allow GN&C to 

refine its onboard wind estimate during final maneuvers, 

further improving overall system landing performance. 

Initial tests of one of these units should take place in about 

1 year. 

CONCLUSIONS 

A GN&C system that enables autonomous precision payload 

delivery using the Dragonfly 10,000-lb-payload-class 

parafoil has completed prototype development and has 

undergone initial flight testing. The guidance algorithm 

uses a proportional scheme for initial homing to the target, 

S-turns for energy management near the target, and a table 

lookup implementation of optimal terminal control for 

final approach. A terminal flare maneuver capability is provided 

for landing. The control algorithm is a proportional, 

integral, derivative design to account for control actuator 

deflection constraints. Navigation relies on a coupled pair 

of GPS receivers with two antennae to determine position, 

velocity, and heading. Wind velocity is also estimated inflight 

from the navigation data for use by the guidance 

algorithm. A Dragonfly mission planning capability has 

been integrated into the laptop-PC-based PADS that is 

used onboard the Dragonfly’s carrier aircraft to determine 

the desired aerial release point as well as to wirelessly 

transmit the mission plan file to the Dragonfly, including 

the best current estimate of the expected winds near the 

drop zone during descent. Flight testing of the Dragonfly 

has included system identification tests, the results of 

which have been analyzed and factored into the dynamics 

models used in the GN&C algorithm design. Autonomous 

GN&C flight tests have already demonstrated a delivery 

accuracy capability of about 200 m despite a variety of 

developmental problems with the prototype canopy, 

avionics, and actuation systems that have been experienced 

to date. Assessment of the simulation and flight 

test results suggest that significant improvement in the 

payload delivery accuracy will be realized once the canopy 

and actuator dynamics are more fully characterized, and 

the avionics/actuator developmental problems experienced 

to date are overcome by design refinements and/or 

component upgrades. 

ACKNOWLEDGMENTS 

The autonomous GN&C software described in this paper 

is one part of the Dragonfly program, developed through 

a team effort of many individuals. The tireless efforts of the 

rest of the contractor team, ParaFlite, Wamore, and 

RoboTek, provided the vehicle to be flown. Test support 

by C-123 pilot Jim Blumenthal of Kingman, Arizona, and 

his ground support team got us through the early months 

24 


of the program. We would also like to thank the large team 

of professional system testers at the U.S. Army Yuma 

Proving Grounds who helped bring the system so much 

closer to military utility. 

The authors gratefully acknowledge the funding support 

of Joint Forces Command JPADS ACTD, the U.S. Army 

30K Science and Technology Objective, and the Air Force 

Air Mobility Command. 

The material in this paper is based on work supported by 

the U.S. Army Natick Soldier Center under contract Nos. 

W9124R-04-C-0154, -0144, and -0118. Any opinions, 

findings, and conclusions or recommendations expressed 

in this material are those of the authors and do not necessarily 

reflect the views of the Natick Soldier Center. 

REFERENCES 

[1] Hattis, P.D. and R. Benney, “Demonstration of Precision Guided 

Ram-Air Parafoil Airdrop Using GPS/INS Navigation,” presented 

at the 52 nd Institute of Navigation Annual Meeting, Cambridge, 

Massachusetts, June 18-20, 1996. 

[2] Hattis, P., B. Appleby, T. Fill, and R. Benney, “Precision Guided 

Airdrop System Flight Test Results,” AIAA paper 97-1468 presented 

at the 14th AIAA Aerodynamic Decelerator Systems 

Conference, June 3-5, 1997. 

[3] Madsen, C.M. and C.J. Cerimele, “Updated Flight Performance 

and Aerodynamics from a Large Scale Parafoil Test Program,” 

presented at the AIAA Modeling and Simulation Conference, CP 

2000-4311, Denver, CO, August 14-17, 2000. 

[4] Madsen, C.M. and C.J. Cerimele, “Flight Performance, 

Aerodynamics, and Simulation Development for the X-38 

Parafoil Test Program,” presented at the AIAA Aerodynamic 

Decelerator Systems Technology Conference, CP 2003-2108, 

Monterey, CA, May 19-22, 2003. 

[5] Barrows, T., “Apparent Mass of Parafoils with Spanwise Camber,” 

presented at the AIAA Aerodynamic Decelerator Systems 

Technology Conference, CP 2001-2006, Boston, MA, May 22- 

24, 2001. 

[6] Hattis, P., T. Fill, D. Rubenstein, R. Wright, and R. Benney, “An 

Advanced Onboard Airdrop Planner to Facilitate Precision 

Payload Delivery,” presented at the AIAA Guidance, Navigation, 

and Control Conference, CP 2000-4307, Denver, Colorado, 

August 14-17, 2000. 

[7] Hattis, P., T. Fill, D. Rubenstein, R. Wright, R. Benney, and D. 

LeMoine, “Status of an Onboard PC-Based Airdrop Planner 

Demonstration,” presented at the AIAA Aerodynamic 

Decelerator Systems Conference, CP 2001-2066, Boston 

Massachusetts, May 22-24, 2001. 

[8] Hattis, P., K. Angermueller, T. Fill, R. Wright, R. Benney, and D. 

LeMoine, “An In-Flight Precision Airdrop Planning System,” presented 

at the 23nd Army Science Conference, Orlando, Florida, 

December 2-5, 2002. 

[9] Hattis, P., K. Angermueller, T. Fill, R. Wright, R. Benney, D. 

LeMoine, and D. King, “In-Flight Precision Airdrop Planner 

Follow-On Development Program,” presented at the AIAA 

Aerodynamic Decelerator Systems Conference, CP 2003-2141, 

Monterey, California, May 19-22, 2003.


(l-r) Leena Singh, Phil Hattis, 

David Carter and Sean George 

David Carter is a Principal Member of the Technical Staff. His interests are algorithm development and 

design and specification in the areas of guidance, control, and estimation theory. Before joining Draper, he 

was an Associate Professor of Mathematics at the University of Virginia. Dr. Carter holds a BA in Physics 

from Haverford College, an MSE in Electrical Engineering from Princeton, and a PhD in Mathematics from 

Columbia University. 

Sean George is a Senior Member Technical Staff in the Vehicle Systems Group. He has over 8 years experience 

at Draper Laboratory as an aerodynamicist on engineering projects relating to the analysis, simulation, 

and design of a wide range of vehicle platforms, including projectiles, rotorcraft, aerodecelerators, 

and airplanes. He has experience with a range of computational modeling tools, including Navier-Stokes 

and potential flow solvers. He was chief configuration design engineer for the Wide Area Surveillance 

Projectile (WASP) gun-launched unmanned aerial vehicles (UAVs) as well as several folded-flyer variants 

proposed for alternative missions. He has also provided analysis support for a number of Draper internal 

and external rotorcraft projects. His principal role on Draper’s airdrop programs is in the areas of vehicle 

modeling, flight data analysis, and simulation development for both ballistic and guided parachute systems. 

Mr. George holds a BS in Applied Physics from Harvey Mudd College (1996) and an SM in 

Aeronautics and Astronautics from MIT (1998). 

Philip Hattis has been employed at Draper Laboratory since 1974, and currently holds the position of 

Laboratory Technical Staff in the Mission Design and Analysis Group of the GN&C Systems Division. 

Responsibilities have included technical leadership for projects involving launch and space vehicle GN&C 

system development, precision delivery airdrop systems, precision Mars landing systems, ballistic missile 

defense systems, ground warrior systems, helicopter fire control systems, autonomous aerial vehicles, 

autonomous space systems, and advanced satellite navigation systems. He now serves as Technical Lead 

for the Crew Exploration Program (CEV) GN&C system development as well as Technical Director for precision 

airdrop programs and for missile defense programs. Dr. Hattis is a Fellow in the American Institute 

of Aeronautics and Astronautics (AIAA). He has a BS in Mechanical Engineering from Northwestern 

University, an MS in Aeronautics from Caltech, and a PhD in Aeronautics and Astronautics from MIT. 

Leena Singh is a Senior Member of the Technical Staff in the Autonomous Control Group. Dr. Singh has 

9 years of research experience exploring new autonomous GN&C algorithms for systems, which have 

included UAVs, helicopter fire control systems, and aircraft engines. Her recent research focuses on envelope-extending 

advanced autonomous guidance and control algorithms. Dr. Singh received a BS in Physics 

and Mathematics from Mt. Holyoke College, an MS from Rensselaer Polytechnic Institute (1993), and a 

PhD from MIT (1997). 

Steven Tavan is the Precision Airdrop GN&C Research Leader at the U.S. Army Natick Soldier Center, 

Airdrop/Aerial Delivery Directorate, Natick, MA. For the last 3 years, he has been the government sponsor 

of Draper Laboratory’s activities in airdrop mission planning and autonomous guidance, navigation, 

and control of parafoils. Prior to his government service, Mr. Tavan worked on aerospace projects at the 

MITRE Corporation and Draper Laboratory. He has a BS in Earth Sciences from MIT. 

25

An Ultra-Low-Power Physics Package for a Chip-Scale 

Atomic Clock 

Mark Mescher, Robert Lutwak, and Mathew Varghese 

Copyright © 2005 by The Charles Stark Draper Laboratory, Inc. Presented at Institute of Electrical and Electronics 

Engineers (IEEE) Transducers ’05, 13 th International Conference on Solid-State Sensors, Actuators, and Microsystems, 

Seoul, Korea, June 5-9, 2005 

INTRODUCTION 

26 

We report the design and measured thermal and mechanical performance of an ultra-lowpower 

physics package for a chip-scale atomic clock (CSAC). This physics package will enable 

communications and navigation systems that require a compact, low-power atomic frequency standard. 

The physics package includes a unique combination of thermal isolation, mechanical stability and 

robustness, and small package volume. We have demonstrated temperature control at a nominal 

operating temperature of 75°C in a room-temperature, vacuum ambient requiring only 7 mW of heating 

power. This represents a power reduction of over two orders of magnitude compared with the lowestpower 

existing commercial technology [1] and more than an order of magnitude improvement over 

other CSAC development efforts. [2] 

Atomic clocks play an essential role in the timing and synchronization 

of modern communications and navigation 

systems. To date, however, their relatively large size and 

power consumption (125 cm3 and 6 W) have prevented 

their application in portable devices. We are developing a 

chip-scale atomic clock, with size

opposite side of the cell back to a photodiode that is integrated 

on the VCSEL die. The VCSEL is tuned to the D1 

optical transition of cesium at λ = 894 nm and directly 

modulated at νHF/2 = 4.6 GHz, where νHF is the hyperfine 

transition frequency of cesium. The cell and VCSEL are 

temperature-stabilized at approximately 75°C to optimize 

clock performance. [4] 

The cell and VCSEL are supported by a novel suspension 

system that relies on the strength and low thermal conductivity 

of polyimide to minimize conductive thermal losses 

from the heated resonance cell while providing mechanical 

stability and isolation from external vibration. The suspension 

structure consists of upper and lower halves (Figure 1). 

Mirror 

6 mm 

Heater/Sensor Elements 

Vapor Cell 

VCSEL/PD 

Figure 1. Left: cross-section view of physics package without 

leadless chip carrier (LCC); right: exploded view. 

Each half consists of a silicon frame that supports multiple 

polyimide tethers converging on a central attachment 

plate. During assembly, the two central attachment plates 

mount to opposite sides of the vapor cell, while the two 

frames are mounted on opposite sides of a spacer. The 

spacer and the cell thickness are such that the tethers are 

stretched out of plane in the assembly process. The introduction 

of a small angle in the tethers greatly stiffens the 

suspension structure by eliminating low-frequency bending 

modes in the suspension beams; thus, z-axis motion of 

the cell causes the tethers to stretch rather than bend. 

The superior thermal isolation and mechanical stiffness 

provided by the suspension structure is made possible 

through appropriate material choice and packaging techniques. 

Polyimide has extremely low thermal conductivity 

(0.2 W/m°C) that minimizes conduction heat loss and permits 

compact suspension designs by reducing required 

beam lengths. In addition, similar to other polymers, it has 

a high yield strain (3%), which enables the angled suspension 

system to be built from components that are 

fabricated using planar, batch-fabrication processes. In 

addition, the electrical leads to the heated components 

may be directly patterned onto the polyimide. 

Fabrication 

Cell Fabrication 

The 2-mm 3 cesium resonance cell is fabricated of silicon 

with anodically-bonded Pyrex ® windows. Details of the 

cell fabrication and cesium loading have been described 

previously. [5] A quarter-wave plate necessary for circular 


Frame 

Spacer 

Vapor 

Cell 

Thermal Control Suspension 

VCSEL 

Suspension 

VCSEL/PD 

polarization is attached to the tablet after cesium filling but 

prior to dicing (Figure 2). 

Figure 2. Left: cross-section model of cesium vapor cell 

including wave plate; right: 4 x 4 tablet (1 cm 2 ) of 

silicon cells before cesium loading and tablet dicing. 

Suspension Fabrication 

The upper (thermal control) and lower (VCSEL/photodiode 

(PD)) portions of the suspension structure are 

fabricated similarly on separate silicon wafers. The frame 

spacer is conventionally machined from aluminum. The 

patterned polyimide layers that form the suspension tethers 

are identical except for a 1.5-mm hole in the center 

plate of the VCSEL suspension that allows optical transmission 

of the laser source and collected light. The process 

sequences are identical except for the metallization. These are 

outlined in Table 1. Figure 3 shows a fabricated assembly. 

Table 1. Suspension Structure Fabrication Sequence. 

Grow etch stop for backside silicon etch: 

thermal SiO2 (1.0 µm) 

Spin on suspension structural material: 

photodefined polyimide (5 µm) 

Sputter thermal control suspension metal: 

titanium (Ti)/platinum (Pt) (0.03 µm/0.25 µm) 

Sputter VCSEL/PD suspension metal: 

Ti/Pt/gold (Au)/Ti (0.03 µm/0.4 µm/0.4 µm/0.1 µm) 

Sputter bond pad metal: 

Ti/Au (0.03 µm/0.5 µm) 

Etch silicon (Si) deep reactive ion etching (DRIE) for suspension 

release: 

through wafer (500 µm) 

Plasma-etch thermal oxide etch-stop layer 

Figure 3. Fabricated physics package assembly mounted 

in an LCC (thermal control side facing up). 

27

Assembly and Packaging 

The cell and suspension structures are assembled with epoxy. 

EPO-TEK ® 353ND is used for its optical transmission characteristics 

and its low outgassing properties. An exploded 

view before assembly is depicted in Figure 1. Individual cells 

are mounted in the suspension structure. The thermal control 

suspension structure is placed face-down in an 

alignment fixture. The cell is mounted mirror-side down on 

the central attachment plate of the suspension via manual 

dispensing of epoxy. After curing, the adhesive interface 

thickness is 5 µm. The aluminum frame spacer is then 

aligned via fiducial marks and adhered to the silicon frame 

similarly. Finally, the VCSEL/PD suspension is aligned using 

another fiducial mark on the frame. Epoxy attaches the center 

plate to the cell and the frame to the frame spacer. 

The VCSEL/PD die is attached to the cell via solder reflow. 

First, 0.018-in solder balls are attached to the VCSEL/PD 

pads. The VCSEL/PD solder balls are aligned to pad locations 

on the center plate of the VCSEL/PD suspension and 

attached by reflowing the solder. Simultaneously, solder balls 

are attached to pads on the frame of the VCSEL/PD suspension 

to enable mounting of the frame assembly to a ceramic 

LCC. The LCC is then aligned and a final reflow is done to 

connect the frame pads to those of the LCC. The heater and 

temperature sensor are connected via wire bond to corner 

pads in the LCC. Finally, an alumina lid containing an activated 

getter is attached in vacuum. A solder preform is 

mounted onto the seal-ring of the LCC and reflowed to seal 

the device. Including the vacuum package, the overall size 

of the physics package is approximately 0.6 cm3 . 

Performance Characteristics 

Thermal Control 

The cell temperature is maintained through integrated single-element 

resistive platinum temperature-sensing and 

heating elements, both of which are patterned directly onto 

the central polyimide plate of the thermal control side of the 

suspension structure. Measured temperature coefficients of 

resistance for the sputtered films are 2300 ppm/°C with a 

4% variation across a 4-in wafer. The temperature sensor is 

distributed uniformly across the cell face to provide an accurate 

average temperature measurement of the vapor cell. 

The portion of the trace that runs along the suspension tether 

from the center plate to outer frame bond pad is designed 

to have relatively low resistance (0.9% in current devices) to 

prevent partial sensing of the frame (i.e, ambient) temperature. 

The sense resistor is sized (10 kΩ) to achieve 

approximately 0.1°C temperature sensitivity with lowpower 

readout electronics. The heater resistance is 

distributed around the periphery of the cell rather than uniformly 

across the plate surface. This provides a more 

uniform temperature because most of the heat transfer away 

from this surface is through the silicon walls of the cell to all 

cell faces, which then radiate heat to the LCC package. The 

heater resistance is sized (400 Ω) such that power is 

delivered efficiently from the control electronics (3-V supply) 

while still maintaining sufficient voltage overhead to 

28 


provide fast device turn-on and good control response. In 

the case of the heater, the tether traces contribute to inefficiency 

by heating the tethers and the frame (7.5% in current 

devices). Both the heating and sensing resistors are configured 

such that current flowing through one segment of the 

resistor is balanced by a current in another segment in close 

proximity that is flowing in the opposite direction, which 

minimizes magnetic fields in the vicinity of the vapor cell. 

Heat loss to the frame and package occurs through radiation, 

gas conduction and convection, and conduction in 

the suspension tethers. Gaseous conduction and convection 

are eliminated by vacuum packaging the device. 

However, even at atmospheric pressure, convection is suppressed 

due to the small size of the gap between the heated 

device and the LCC package. Radiation is driven by surface 

area and emissivity. Gas conduction is driven by gap 

dimensions, gas composition, and pressure. The bulk conduction 

is determined by thermal conductivity and the 

shape and size of the suspension tethers and their metal 

interconnect. Detailed radiation models must account for 

the multiple surface emissivities and shape factors of the 

system. However, the heated cesium cell will have radiative 

heat loss with a temperature dependence given by 

P = ε . σ . Acell . (Tcell 4 – Tamb 4 ) (1) 

where ε is a parameter that embodies the relative emissivities 

of the cell and heat-reflecting package, Acell is the surface 

area of the cell, and σ is the Stefan-Boltzmann constant. As 

a general rule, heat loss by thermal radiation is reduced by 

minimizing the cell’s surface area and average emissivity. The 

cell has four saw-diced silicon surfaces, one polyimide-andplatinum-covered 

surface, and one surface of gallium 

arsenide (the VCSEL/PD). These materials all have emissivities 

around 0.5, with variation depending on surface finish. 

Some materials, such as polished aluminum, can have emissivities 

as low as 0.03. While we have not yet implemented 

surface coatings on the cell sidewalls to reduce radiation, we 

have conducted experiments with test aluminum cells in 

place of the standard cell that indicate that even lower heater 

power requirements are possible (Figure 4). 

12 

Power (mW) 

10 

8 

6 

4 

2 

Silicon/Pyrex Cell 

AI Cell 

0 

0 30 60 90 120 

Temperature (˚C) 

Figure 4. Measured steady-state heater power as a function 

of cell temperature in a vacuum ambient (24°C, 

2 mTorr) for a silicon/Pyrex cell and a low-emissivity 

polished aluminum test cell.

Heat loss due to gas conduction was characterized to determine 

how package vacuum level affects required heater 

power. Figure 5 shows heater power for a cell held at 80°C 

in 24°C ambient as a function of gas (air) pressure. For low 

pressures (less than about 20 mTorr), the power is dominated 

by radiation. For intermediate pressures, heat transfer is 

molecular and is proportional to pressure. At high pressures, 

the heat transfer is essentially independent of pressure 

and depends on the thermal conductivity of the gas. 

Power (mW) 

80 

60 

40 

20 

0 

0.001 0.1 1.0 1000 

Pressure (Torr) 

Figure 5. Heater power required to maintain cell temperature 

at 80°C in a 24°C ambient of varying 

pressure. Experiment was performed in a bell jar 

with the device packaged as shown in Figure 3. 

The solid material conduction losses in our CSAC physics 

package are not measurable currently and are small compared 

with radiation losses. The predicted power loss at 

75°C by conduction is 0.9 mW, or 13% of the total. The 

specifications and predicted thermal performance characteristics 

of the suspension system are shown in Table 2. 

Table 2. Tether System Design Values. 

Total tether count 16 

Polyimide tether 5 x 375 x 1000 µm 

VCSEL/PD metal widths 10 µm 

Heater metal trace widths 80 µm 

Sensor metal trace widths 30 µm 

Thermal Resistances: 

Polyimide tethers (total)* 167°C/mW 

Metal traces (total)† 87°C/mW 

Total 57°C/mW 

* 0.2 W/m°C polyimide conductivity assumed 

† bulk conductivity values assumed 

The heat capacity of the system was also estimated from a 

time constant. The step response used in the estimate is 

shown in Figure 6. The measured time constant was 63 s. 

From the steady-state measurements shown in Figure 4, the 

effective total thermal resistance (radiation-dominated) at 

75°C is approximately 6.3°C/mW, which yields a heat capacity 

of 0.010 J/C. This is within 5% of the calculated value. 


Temperature (˚C) 

80 

78 

76 

74 

72 

70 

0 200 400 

Time (s) 

600 

Figure 6. Measured thermal step response of physics package. 

Mechanical Design 

As described earlier, the mechanical design takes advantage 

of several material properties of polyimide to meet the 

conflicting design requirements of low heat loss and 

mechanical rigidity while permitting planar MEMS batch 

fabrication techniques for the suspension system. We 

describe the design in the context of two mechanical specifications: 

fundamental mechanical resonance frequency 

and maximum sustainable steady acceleration (“g” load) 

without plastic deformation of the polyimide tethers. 

Figure 7 shows a cross section of one half of the suspension 

system. As is suggested by the figure, the tethers are 

dominated by axial rather than bending stresses. They are 

more appropriately thought of as cables rather than bending 

beams. If one considers only z-axis forces, the system 

is equivalent to the mass-spring model shown in Figure 7, 

where d is the spring’s displacement from equilibrium after 

assembly (both springs are nominally in tension). For large 

z-displacements, these springs are highly nonlinear. 

θ 

d 

Mcell 

Figure 7. Left: depiction of a cross section of half-device 

showing exaggerated tether angle; right: linearized 

mass-spring model. 

In order for the tethers to act as springs, they must be in 

tension at all times. Under this condition, the small signal 

resonant frequency of the suspension system is independent 

of the magnitude of the tensile stress. It is the angle of 

the tethers rather than the pre-tension that sets the resonant 

frequency. However, if the stress in the tethers turns 

compressive at any point, they buckle and do not provide 

any significant mechanical support. Such a condition is 

encountered under a high static “g” load and is described 

later. The restoring force of a single tether for a displacement 

z is 

+z 

∆z 

11 

10 

9 

8 

7 

6 

Heater Power (mW) 

29

(2) 

This equation is nonlinear in z, but for small motions ∆z 

around the equilibrium displacement, d, a linear equation 

for the total restoring force in the z-direction can be used 

FT = -N . k . ∆z (3) 

Here N is the total tether count and k is the linearized 

spring constant of a single tether 

(4) 

E is elastic modulus, A is tether cross-sectional area, L is 

tether length, and θ0 is the as-built tether angle (Figure 7). 

This spring constant greatly exceeds that of the bending 

stiffness for even modest angles. The resonant frequency is 

now given by the familiar relation 

It should be noted that x- and y-axis resonance frequencies 

can be calculated simply by replacing cos (θ0) with sin 

(θ0) in the equation for the spring constant. It follows from 

this that resonant frequencies and maximum g loads are 

higher in the x and y directions. 

Test units with a range of built-in tether angles were built 

and tested to validate the model. A small magnetically permeable 

disk was attached to the cell in these test units and 

excited by a miniature electromagnet. Displacement was 

measured with an interferometer as the excitation frequency 

was swept manually. Fundamental mode resonances 

were calculated by fitting the data to a standard massspring-damper 

model. Measured resonant frequency data 

were estimated to be accurate to better than 1%. The 

results are compared to the model in Figure 8. The elastic 

modulus assumed for the model was 3.5 GPa. The mass of 

the cell and attached magnetic disk was estimated at 27 

mg, significantly greater than the cell alone (18 mg). Thus, 

the resonant frequency of real devices is expected to be 

higher by the square root of the mass ratio of the test unit 

and real device (27 mg/18 mg) 0.5 for the same tether 

angle. This predicts a resonant frequency of 2450 Hz for a 

real device with a tether angle of 10 deg. 

In order to calculate the maximum g load of the devices, 

we estimate the forces present at the yield strain, εmax, of 

the tethers. Under steady acceleration, the maximum g 

load, Amax, can be expressed as 

30 


(5) 

(6) 

Resonant Frequency (Hz) 

3000 

2500 

2000 

1500 

1000 

500 

Model 

Measured 

0 

0 5 10 15 

Tether Angle (deg) 

Figure 8. Measured and modeled mechanical resonance frequency 

as a function of as-built tether angle. The error 

bars are attached to the theoretical curve; they are 

based on estimates of fabrication tolerances, including 

cell mass, tether dimensions, and tether angle. 

We assume that the cell has been displaced such that only 

the upper or lower halves of the suspension system remain 

in tension and thus provide mechanical support. For εmax 

= 0.03, Amax = 2200 g (1500 grms). 

CONCLUSIONS 

We reported on the design and measured thermal and 

mechanical performance of an ultra-low-power physics 

package for a CSAC. We have demonstrated temperature 

control at a nominal operating temperature of 75°C in a 

room-temperature, vacuum ambient requiring only 7 mW 

of heating power. Lowering the cell’s emissivity has been 

demonstrated experimentally to provide further power 

reduction. Measured resonance frequency corresponding 

to 2.5 kHz for a physics package was presented. In addition, 

models of the suspension system indicate that the 

physics package can sustain maximum g loads in excess of 

2000 g without damage. 


The authors wish to thank Mike Garvey of the 

Symmetricom Technology Realization Center (TRC) and 

John McElroy of Draper Laboratory for supporting and 

directing this effort. We are also thankful for the technical 

and theoretical support provided by Gary Tepolt, 

John LeBlanc, and Amy Duwel of Draper Laboratory, 

Don Emmons and Peter Vlitas of the TRC, and Darwin 

Serkland of Sandia National Laboratories. We thank Yen- 

Wah Ho, Greg Romano, Ryan Prince, Maria Cardoso, 

Maria Holmboe, Katherine Ashton, and Bessy Silva for 

assistance in the fabrication and assembly of physics 

package components. We also thank V.M. Montano and 

A.T. Ongstad for assistance in the fabrication of the 

VCSEL/RCPD chips, and T.W. Hargett for assistance in 

the epitaxial semiconductor growth. This work is supported 

by the Defense Advanced Research Projects 

Agency, Contract No. NBCHC020050.

REFERENCES 

[1] X-72 Data Sheet, http://www.symmetricom.com/media/ 

pdf/documents/ds-x72.pdf. 

[2] Kitching, J., “Chip Scale, Microfabricated Atomic Clocks (An 

Emerging Technology),” 29th Annual Time and Frequency 

Metrology Seminar, National Institute of Standards & 

Technology (NIST), Boulder, CO, June 14-17, 2004. 

[3] Lutwak, R. et al., “The Chip-Scale Atomic Clock – Coherent 

Population Trapping vs. Conventional Interrogation,” 

Proceedings of the 34th Annual Precise Time and Time Interval 

(PTTI) Systems and Applications Meeting, December 3-5, 2002, 

Reston, VA, pp. 539-550. 


[4] Lutwak, R. et al., “The Chip-Scale Atomic Clock – Recent 

Development Progress,” Proceedings of the 35th Annual Precise 

Time and Time Interval (PTTI) Systems and Applications 

Meeting, December 2-4, 2003, San Diego, CA, pp. 467-478. 

[5] Lutwak, R., J. Deng, W. Riley, M. Varghese, M. Mescher, D.K. 

Serkland, G.M. Peake, “The Chip-Scale Atomic Clock – Recent 

Development Progress,” Proceedings of the 36th Annual Precise 

Time and Time Interval (PTTI) Systems and Applications 

Meeting, December 7-9, 2004, Washington, DC. 

[6] Liew, L-A. et al., “Microfabricated Alkali Atom Vapor Cells,” 

Applied Physics Letters, Vol. 84, 2004, p. 2694. 

(l-r) Mark Mescher, Mathew Varghese 

Mark Mescher is a Principal Member of Technical Staff in the Advanced Hardware Development Division. 

He won the Draper 2005 Distinguished Performance Award with fellow team members for their work on 

the chip-scale atomic clock. Past or current project leadership includes transcleral drug delivery development, 

MEMS acoustic arrays for underwater imaging, sensor systems for biological sensing applications, 

and microfluidic system development. Dr. Mescher received a BS in Computer Engineering from Wright 

State University (1993), and MS and PhD degrees in Electrical and Computer Engineering from Carnegie 

Mellon University (1995 and 1999, respectively). 

Robert Lutwak is a Senior Scientist in the Research group of the Symmetricom TRC. His areas of expertise 

include the engineering of atomic instrumentation, particularly the design of laser oscillators and laser 

systems, ultra-high vacuum and atomic beam design, and electronic and computer control of complex systems. 

He is currently a Principal Investigator on the DARPA-sponsored CSAC program, developing atomic 

clocks two orders of magnitude smaller and lower power than any existing technology. He also leads the 

Symmetricom Advanced Technology Atomic Frequency Standard (ATAFS) program to develop next-generation, 

high-performance atomic clocks for deployment onboard the GPS satellite constellation. Dr. 

Lutwak is a member of the American Physical Society and the IEEE Ultrasonics, Ferroelectrics, and 

Frequency Control (UFFC) Society. He serves on the Technical Program Committee for the IEEE 

Frequency Control Symposium. Dr. Lutwak received a BS in Physics, summa cum laude, from Miami 

University (1988) and a PhD in Atomic and Optical Physics from MIT (1997). 

Mathew Varghese currently heads the Microsystems Integration Group and is a Principal Member of 

Technical Staff at Draper Laboratory. His research interests focus on the fabrication, design, and analysis 

of microsystems. He has led projects to build microphones, drug delivery devices, MEMS RF filters, and 

CSAC. He won the 2005 Draper Distinguished Performance Award for leading the CSAC development 

effort at Draper. Dr. Varghese received a BS in Electrical Engineering and Computer Science with a minor 

in Physics from the University of California, Berkeley and SM and PhD degrees in Electrical Engineering 

and Computer Science from MIT (1997 and 2001, respectively). 

31

Detection of Biological and Chemical Agents Using 

Differential Mobility Spectrometry (DMS) Technology 

Melissa Krebs, Angela Zapata, Erkinjon Nazarov, Raanan Miller, Isaac Costa, Abraham Sonenshein, Cristina Davis 

Copyright © 2005 by The Charles Stark Draper Laboratory, Inc. Printed with permission in IEEE Sensors Journal, Vol. 5, No. 4, August 

2005, pp. 696-703 

INTRODUCTION 

With international concern growing over the potential for chemical and biological terrorism, 

there is an urgent need for a sensor that can detect chemical and biological agents quickly and accurately. 

Such a sensor must be portable, robust, and sensitive, with fast sample analysis time. We will 

demonstrate the use of a micromachined differential mobility spectrometer (DMS) with all these characteristics 

that is able to detect multiple agents simultaneously on a time scale of seconds. In this study, 

we have demonstrated the ability of the DMS to detect Bacillus subtilis spores, a surrogate for Bacillus 

anthracis spores, the causative agent of anthrax. Pyrolysis was used as the sample introduction method 

to volatilize the spores before introducing material into the DMS. In addition, we examined the effect 

of pyrolysis on B. subtilis spores suspended in sterile water using SDS-PAGE. These experiments 

showed that the spores must be heated at 650°C or greater for 5 s or at 550°C for at least 10 s to be 

fragmented into particles considerably smaller than 10 kilodaltons (kDa), which the DMS is able to 

detect. Several major biomarkers can be distinguished easily above the background of the sterile water 

in which the spores are suspended, and we hypothesize that additional biomarkers could be liberated 

by further optimizing conditions. The DMS also has shown promise as a detector for chemical weapon 

agents, and we have also demonstrated the ability of the DMS to detect nerve and blister agent simulants 

at clinically relevant levels. 

The use of microbes for bioterrorism is well documented. 

As long ago as the Middle Ages, warriors catapulted 

bodies of plague victims over the walls of villages under 

siege in an attempt to hasten their victory. [1] In the 18 th 

century, blankets that had been used by smallpox victims 

were intentionally distributed to the Native American 

population. [1] Within the last century, there are many 

more examples of the use of biological weapons. Japan 

32 

conducted biological warfare experiments in the 1930s 

and 1940s, to which outbreaks of plague, cholera, and 

typhus were credited. [2] In 1979, there was an epidemic 

of inhalation anthrax in Sverdlovsk resulting from a 

release of weapons-grade B. anthracis from a military 

facility. [3] In 1993, the Aum Shinrikyo cult attempted to 

disperse anthrax spores from a high-rise building in 

Tokyo. [4] There have also been instances of the intentional

Detection of Biological and Chemical Agents Using Differential Mobility Spectrometry (DMS) Technology 

release of biological agents in the United States. In 1985, 

the Bhagwan Shree Rajneesh cult used Salmonella to 

infect 752 patients in Oregon. [5] In 2001, the United 

States experienced the first intentional anthrax release in 

this country. [6] Bacillus anthracis has the ability to form 

resilient spores, making it a prime candidate for a biological 

weapon; the spores are highly resistant to extreme 

conditions and can be dispersed easily. [7],[8] 

The use of chemical agents as weapons of mass destruction 

has also been well documented. Chemical weapons 

include nerve agents, blister agents, asphyxiants, and 

pulmonary irritants. [9] In 1936, Japan used chemical 

weapons (hydrogen cyanide, mustard gas, phosgene) 

during their invasion of China. [10] Between 1980-1988, 

Iraq used mustard gas and nerve agents against Iran and 

Iraqi Kurds. [10] In 1995, the Aum Shinrikyo cult released 

sarin, a toxic nerve agent, in the subway system in 

Tokyo. [4] 

The development of early detection technologies that are 

able to identify both biological and chemical warfare 

agents quickly is important to homeland defense agencies, 

national health care systems, and first responders. 

Such a technology should be robust, field-deployable, 

and inexpensive. It should yield fast and sensitive identification 

of an unknown sample on location, and it should 

function autonomously. [11] 

Scientists have adapted molecular biology techniques to 

detect B. anthracis using DNA-based, antibody-based, 

and mass spectrometry analysis approaches. These tests 

vary greatly in sensitivity, response time, cost, availability, 

and complexity of use. Polymerase chain reaction 

(PCR) and other nucleic acid-based methods have been 

widely examined for the detection of anthrax. [12]-[22] 

Several antibody-based methods have also been examined 

for anthrax spore recognition. [23]-[29] 

Currently, several chemical detectors are being investigated 

to rapidly identify Bacillus spores. Virtually all gas 

chromatograph (GC) detectors, such as the widely used 

flame ionization detector (FID), produce a signal indicating 

the presence of a compound eluted from the column; 

however, they lack the specific information required for 

unambiguous compound identification. An expedient 

and simple method to identify unknown analytes 

requires a detector to provide an orthogonal set of information 

for each chromatographic peak. The mass 

spectrometer (MS) is generally considered one of the 

most definitive detectors for compound identification, as 

it generates a fingerprint pattern of fragment ions for 

each GC elutant. Mass spectrometric information is often 

sufficient for sample identification through comparison 

to compound libraries, and has been used to identify 

some Bacillus species. [30]-[34] 

While GCs are continuously being miniaturized and 

reduced in cost, [35] mass spectrometers are still very 

expensive, between $50,000 and $75,000. They are relatively 

large, making field deployment difficult. They 

also require operation at low pressures and their spectra 

can be difficult to interpret. An alternative miniaturized 

technology developed at Draper Laboratory, known as 

the differential mobility spectrometer (DMS), is available 

as a hand-held mobile unit that operates at ambient temperature 

and pressure. Our micromachined DMS sensor 

is capable of rapid, sensitive detection of many chemicals. 

[36]-[39] For example, it has been shown to 

distinguish o-, p-, and m-xylene isomers; [40] these isomers 

cannot be separated and distinguished using 

traditional chemical analysis techniques (such as GC- 

MS). We tested the ability of the DMS to detect simulants 

of chemical and biological weapons agents. We have previously 

shown its capability to distinguish dipicolinic 

acid, picolinic acid, and pyridine – three chemical components 

present in high concentrations in spores as 

verified by GC-MS. [41] Here, we show evidence of spectrum-wide 

biological markers that distinguish pyrolyzed 

spores from background materials, and the ability to 

detect chemical weapons agent simulants. 

Material and Methods 

DMS Operating Principles 

Draper Laboratory developed the micromachined DMS 

as a novel Microelectromechanical System (MEMS) sensor 

for biological and chemical agent detection. When 

incorporated into a detection system, the DMS provides 

quantitative, sensitive detection down to the parts-pertrillion 

(ppt) range. [40] It offers the advantage of 

operating in air at atmospheric pressure and ambient 

temperature. It filters ions based on their nonlinear 

mobility in high-strength RF electric fields. The miniaturized 

DMS has enhanced sensitivity and resolution 

due to the close spacing of the RF plates in the drift 

tube region. These features make the device suitable for 

use as a quantitative, relatively low-cost, portable 

detector. 

DMS operation measures the differential mobility of 

ions as they travel through the air in response to an 

applied electric field, as shown in Figure 1. Differential 

mobility is dependent on the charge, size, and mass of 

an ion as it experiences high and low electric fields. A 

solid or liquid sample is first pyrolyzed, breaking apart 

the chemical bonds by thermal energy, and converting 

the sample to a gaseous form of smaller and more 

volatile fragments. These gaseous sample fragments are 

then introduced into the spectrometer where they are 

ionized by either a radioactive source or ultraviolet 

(UV) radiation. This paper presents data using a 63Ni 

33

ionization source. Once ionized, the sample is transported 

by a carrier gas through an ion filter toward the 

detecting electrodes (Faraday plates). The electric field 

that is applied between the parallel ion filter electrodes 

can be adjusted to tune the ion filter. Two fields are 

applied across the parallel plates: an asymmetric waveform 

electric field alternating between high- and 

low-strength fields, and a low-strength DC compensation 

voltage. The asymmetric field amplitude is held 

constant while the compensation voltage levels are 

adjusted to allow a particular ion species to pass 

through the ion filter. Once the ion species passes 

through the ion filter, it is detected upon collision with 

the detector electrodes as an ion current. Uncompensated 

ions are scattered toward the ion filter 

electrodes, neutralized, and swept out by the carrier 

gas. By noting the applied field conditions (voltages) 

and the current level at the detector electrode, various 

ion species can be identified. The DMS can also be programmed 

to sweep through a range of compensation 

voltages over a specified amount of time (seconds). This 

is beneficial as it allows simultaneous detection of various 

ion species that originate from the same sample. An 

example of a DMS chip is shown in Figure 2(a); a development 

platform (Sionex Corporation, Waltham, MA) 

designed for proof of principle and user evaluation is 

shown in Figure 2b. 

34 

Sample 


Pyrolysis 

∆H 

67-keV 

Ionization 

63Ni 

Figure 1. Schematic of pyrolysis-DMS operation. A solid or liquid sample is pyrolyzed, and the fragments are then ionized 

as they flow past a radioactive nickel source. The ions are then swept into the drift tube of the DMS where they 

are separated based on their mobility in the applied electric fields. Under the appropriate compensation voltage, 

an ion will leave the drift tube and strike a Faraday plate that detects the strike based on the charge transfer. 

+ 

+ 

Drift Tube 

+ 

Compensation 

Electric Field 

Drift Tube 

V 

Detector 

RF 

Electric Field 

Detector 

(+) 

Figure 2. (a) Picture of a DMS chip. The drift tube region 

and detectors are shown with arrows. (b) 

Photograph of a hand-held Sionex Corporation, 

Inc. DMS prototype unit. 

Bacillus Spore Preparation 

B. subtilis spores were selected as a surrogate for B. 

anthracis to evaluate the ability of the DMS to detect these 

spores. B. subtilis strain SMY, a wild-type, prototrophic, 

Marburg strain (obtained from P. Schaeffer), [42] was pregrown 

overnight at 30°C on a plate of tryptose blood agar 

base (Difco Laboratories, Franklin Lakes, NJ) and used to 

inoculate 2-L of DS medium [43] in a 6-L Erlenmeyer flask. 

The flask was incubated with shaking (200 rpm) at 37°C 

for 48 hours. The cells were harvested by centrifugation at 

13,000 x g for 20 min at 4°C, washed four times with 100ml 

sterile, deionized water, and resuspended in 20-ml 

sterile water. The suspension was estimated to contain 

95% mature, refractile spores by phase contrast 

(-)


microscopy. The spore titer was determined by assaying 

colony formation on DS agar plates after heating to 80°C 

for 10 min. Spores were diluted in sterile water when 

lower concentrations were required for testing. 

The Effect of Pyrolysis on B. subtilis Spores as 

Determined by SDS-PAGE 

Pyrolysis was used as the sample introduction technique 

tested to volatilize the spores before entering the DMS. The 

Pyroprobe 1000 from CDS Analytical, Inc. (Oxford, PA) was 

used. This unit is fabricated to interface with a GC inlet 

port. A slurry containing spores resuspended in sterile water 

was loaded into a quartz tube (2.5 mm outer diameter, 1.95 

mm inner diameter, 25.5 mm length) that was placed in the 

pyrolysis probe. The probe was then loaded into the pyrolysis 

unit, which has a nitrogen environment. The sample 

can then be pyrolyzed, and if there is gas flow, it will be carried 

out of the pyrolysis unit and into a GC column. For 

these experiments, we used a 0.5-m deactivated fused silica 

column only (Agilent Technologies, Palo Alto, CA). 

To better understand the effect of pyrolysis on B. subtilis 

spores, the spores were pyrolyzed using various conditions 

and examined using denaturing polyacrylamide gel electrophoresis 

(SDS-PAGE). Fifteen µg of spores 

(approximately 1.5e8 spores) resuspended in water were 

exposed initially to 110°C interface temperature and then 

pyrolyzed at a ramping rate of 0.01°C/ms under each of the 

following conditions: 250°C, 5 s; 250°C, 10 s; 350°C, 5 s; 

350°C, 10 s; 450°C, 5 s; 450°C, 10 s; 550°C, 5 s; and 

550°C, 10 s; 600°C, 5 s; 650°C, 5 s; 700°C, 5 s; 750°C, 5 s; 

800°C, 5 s; 850°C, 5 s; and 900°C, 5 s. The pyrolyzed samples 

were recovered with Laemmli sample buffer (Bio-Rad), 

and dithiothreitol (DTT) was added to a final concentration 

of 300 mM to denature the proteins prior to SDS-PAGE 

analysis. The samples were then placed in boiling water for 

5 min and run on a gradient 4-20% tris-glycine polyacrylamide 

gel (Bio-Rad, Hercules, CA). A sample of 1.5 µg of 

spores that were not pyrolyzed or exposed to the 110°C 

interface temperature was included as a control. 

As the gradient gels consist of a gradually increasing density 

of polyacrylamide matrix, proteins can often stain 

unevenly. We found that the larger proteins in the more permeable 

matrix in the top area of the gel are able to stain 

more efficiently than those that are smaller and found in the 

thicker polyacrylamide matrix at the bottom of the gel. To 

minimize the disproportionate staining, we found that it 

was best to prestain the gel initially with Coomassie Blue, 

destain the gel, and then stain with silver to produce uniform 

results. Other groups have used the technique of 

consecutive Coomassie Blue and silver staining in the examination 

of proteins separated by electrophoresis in gradient 

gels, [44] and silver staining reveals proteins at lower concentration 

levels than other analytical methods. 

The gels were stained in a 40% methanol/10% acetic acid/ 

0.01% Coomassie Blue R-250 solution for 1 h, and then 

destained in a 40% methanol/10% acetic acid solution 

overnight. The gels were then prepared for silver staining by 

washing twice in deionized water for 30 min. The Silver 

Stain Plus Kit (Bio-Rad) was used to silver stain the gels for 

20 min, with the silver stain solution made per manufacturer 

instructions. The staining reaction was stopped by 

exposing the gel to a 5% acetic acid solution for 20 min, and 

the gels were then washed with deionized water for an additional 

20 min before recording the results by digital 

photography. 

Bacillus Spore Analysis Using DMS 

Pyrolysis was the sample introduction method for DMS 

analysis of spores. The pyrolysis unit was attached to an 

inlet port of an HP 5890 GC. The sample was pyrolyzed and 

swept into 0.5 m of a fused silica guard column with the 

carrier gas, helium (He), at 150 ml/min flow. This flow exited 

the GC oven and joined 50 ml/min nitrogen (N2) flow to 

enter the DMS. Mass flow controllers (MKS Instruments, 

Andover, MA) were used to ensure constant volumetric flow 

of the carrier gases. The GC oven temperature was set to a 

250°C constant temperature. 

The DMS was programmed so that the compensation electric 

field swept through a range of voltages from -40 to 10 V 

every 1.6125 s. The RF field was set to 1200 V. The detection 

of ions at the various compensation voltages over time 

is recorded on a laptop computer that is connected to the 

DMS unit. Ten 4-µl sterile water samples were pyrolyzed 

and their spectra from the DMS recorded. These spectra give 

the background signal of the water in which the spores were 

resuspended. Next, 10 samples of 120,000 spores (3 x 108 spores/ml slurry) were pyrolyzed and the DMS spectra 

recorded. Finally, 10 samples of 40,000 spores (1 x 107 spores/ml slurry) were pyrolyzed. The pyrolysis unit was 

held at a 150°C interface temperature, and the samples were 

pyrolyzed at 550°C for 100 s, with a 20°C/ms ramping rate. 

The spectra were analyzed using OriginPro 7.0 and the data 

graphed as the signal intensity vs. the compensation voltage. 

Each plotted line is the signal average of 25 voltage sweeps 

after spore pyrolysis in a single experiment. The baseline of 

all data signals was shifted to zero, and the maximum 

absolute value across each scan was calculated. Each spectral 

sweep was then normalized at the highest peak. 

Simulant Nerve and Blister Agent DMS Detection 

Solutions of methyl salycilate (MS) (Kintek, La Marque, 

TX), a simulant nerve agent, were made at the following 

concentrations: 344 parts-per-billion (ppb), 57 ppb, 9.5 

ppb, 1.6 ppb, 0.27 ppb, and 0.045 ppb. These solutions 

were analyzed with the DMS. The intensity of the signal was 

examined as a function of concentration. Next, solutions of 

MS and dimethyl methyl phosphonate (DMMP) (Kintek, La 

35

Marque, TX), a simulant blister agent, were analyzed 

using the DMS. The spectrum for each compound was 

first recorded individually. Then, a mixture of DMMP 

and MS was prepared, and the spectrum of this mixture 

was recorded. 

RESULTS 


Pyrolysis was used to volatilize the samples for introduction 

into the DMS. The SDS-PAGE results show that 

when the spores are pyrolyzed at higher temperatures, 

they are fragmented into molecules smaller than the 

limit of detection of this assay (Figure 3). The lower 

limit of band resolution on the 4-20% tris-glycine gels is 

10 kDa, although proteins as small as 1-2 kDa can still 

be detected. At the lower pyrolysis temperatures (lanes 

2-4), the band patterns of the pyrolyzed spore samples 

look similar to the spores that did not undergo pyrolysis 

(lanes 1 and 10), indicating little, if any, disruption to 

the spores when exposed to these temperatures. The 

bands seen here are likely due to slight fragmenting of 

the spores when exposed to the denaturing conditions 

during SDS-PAGE sample preparation. The intermediate 

temperatures (lanes 5-7) cause proteins of high molecular 

weight to be released, indicating that the spores have 

been disrupted but not efficiently volatilized. At temperatures 

of 650°C and greater held for 5 s (lanes 13-18) 

and at 550°C when held for 10 s (lane 9), the recovered 

material from the pyrolyzed spores does not show up on 

the gel, indicating that the fragments in the sample are 

not retained in the gel and so are well under 10 kDa. 

Figure 3. Effect of pyrolysis on B. subtilis spores. 4-20% 

tris-glycine SDS-PAGE of Bacillus subtilis spores, 

stained with Silver Stain Plus Kit (Bio-Rad). All 

pyrolyzed samples are 15 µg of B. subtilis spores 

exposed initially to 110°C interface temperature 

and with a ramping rate of 0.01 C/ms. Lanes: 1, 

1.5 µg B. subtilis spores; 2, pyrolyzed spores at 

250°C, 5 s; 3, pyrolyzed spores at 250°C, 10 s; 4, 

pyrolyzed spores at 350°C, 5 s; 5, pyrolyzed 

spores at 350°C, 10 s; 6, pyrolyzed spores at 



spores at 550°C, 10 s. Middle: Precision Plus 

Protein Standard (Bio-Rad). Lanes: 10, 1.5 µg B. 

subtilis spores; 11, pyrolyzed spores at 550°C, 5 s; 

36 

1 2 3 4 5 6 7 8 9 MW 

kDa 

250 

150 

100 

75 

50 

37 

25 

20 

15 

10 

10 11 12 13 14 15 16 17 18 

12, pyrolyzed spores at 600°C, 5 s; 13, pyrolyzed 





900°C, 5 s. 

B. subtilis spore samples were then analyzed using the 

DMS. They were pyrolyzed and carried though a 0.5-m 

deactivated fused silica column with helium as a carrier 

gas, and then joined with nitrogen to flow into the DMS. 

The compensation field, time, and abundance of ions 

striking the detector were recorded. The results for the 

sterile water and for the two different concentrations of 

spores are shown in Figure 4. The plots show DMS signal 

amplitude on the y-axis and the compensation 

voltage on the x-axis. Each line represents a different 

sample’s spectra averaged over time. Major biomarker 

peaks associated with the presence of spores are seen at 

-29 V, -12 V, and -3 V (peaks 1, 3, and 4, respectively). 

The peaks are concentration-dependent and are not 

detected in the water controls. Many smaller biomarkers 

are seen between -12 V and -3 V, although they are not 

consistent in amplitude between experimental trials. 

(a) 

(b) 

DMS Signal Amplitude (V) 

2 

1 

1 2 3 4 5 

0 

-40 

2 

-30 -20 -10 0 10 

1 

(c) 2 

0 -40 -30 -20 -10 0 10 

1 

0 

-40 -30 -20 -10 0 10 

Compensation Voltage (V) 

Figure 4. Major and minor DMS spectral biomarkers from 

B. subtilis spore pyrolysis: (a) water only controls, 

(b) 40,000 pyrolyzed spores, (c) 120,000 

pyrolyzed spores. Biomarkers 1, 3, and 4 appear 

to correlate with the presence of spores, and the 

amplitude is concentration-dependent. 

MS, a nerve agent simulant, was used in six concentrations 

ranging from 344 ppb to 45 ppt to determine the 

ability of the DMS to detect the compound, as well as its 

limit of detection. Figures 5(a) and 5(b) show these 

results; the DMS detected part-per-trillion concentrations 

of MS with very high signal-to-noise ratios at 

concentrations well below those of clinical significance.

Intensity (a.u.) 


Methyl Salycilate 

(a) 389 ng (344 ppb) 

(b) 

65 ng (57 ppb) 

10.8 ng (9.5 ppb) 

1.8 ng (1.6 ppb) 

0.3 ng (0.27 ppb) 

0.05 ng (0.045 ppb) 

0 

-25 -20 -15 -10 -5 0 5 10 -12 -7 


Log of Sample Amount (gr) 

Figure 5. Simulant chemical weapon detection with the DMS. (a) DMS spectra of varying concentrations of MS. The signal 

intensity is shown as a function of the compensation voltage. (b) DMS signal peak area as a function of mass of 

MS introduced. Amount of compound introduced corresponds to concentrations ranging from 344 ppb to 45 ppt. 

Also shown is the molecular structure of MS. 

Additionally, the ability of the DMS to simultaneously 

detect two chemical weapon agent simulants was examined. 

For these studies, single and dual-compound 

solutions of DMMP and MS were employed. Figure 6(a) 

shows the DMS spectra collected for the single compounds. 

As can be seen, the DMS produces a unique 

spectral signature for each compound. Figure 6(b) shows 

the spectrum for a DMMP-MS mixture (red line) superimposed 

on the spectrum for MS (green line). The resultant 

spectrum shows features of each of the two compounds. In 

the positive ion spectrum, the DMMP peak at about -5 V 

and the MS peak at about -2 V are observed. In the negative 

ion spectrum, only the MS peak at -5 V shows up 

clearly. This demonstrates that simultaneous detection of 

the two simulants can be achieved with the DMS. 

(a) 

DMMP 

DMMP 

MS 

MS 

(b) 


DISCUSSION 

Bacillus anthracis is an example of a biological agent that 

could be used as a biological weapon, as the spores can be 

dispersed easily, are highly resistant to extreme conditions, 

and are highly pathogenic. In this study, we demonstrate 

the ability of the DMS to detect biomarkers of B. subtilis, 

a surrogate for B. anthracis, chosen for its ready availability 

and lack of pathogenicity. As we have now 

demonstrated that pyrolysis coupled with DMS is a useful 

method for the analysis of spore components, we can 

hypothesize that different markers would be found in B. 

anthracis spores compared with B. subtilis spores, as the 

spores differ considerably in their coat proteins and in 

some cytoplasmic proteins. In fact, for this reason, it is 

unlikely that the same peaks would be found if examining 

Figure 6. Simulant chemical weapon detection with the DMS. (a) DMS spectra of MS and DMMP. Spectra were collected 

sequentially. (b) Simultaneous detection of nerve and blister agent simulants DMMP and MS. A mixture of DMMP 

and MS is graphed with signal intensity as a function of compensation voltage (green line). The spectrum for MS 

(red line) is shown again for reference. 

Log of Peaks Area 

2.5 

2 

1.5 

1 

0.5 

45 ppt 

MS 

DMMP+MS 

MS 

DMMP+MS 

344 ppb 


0 

0 

H0 

Positive Ions 

Negative Ions 

-30 -20 -10 0 10 -30 -20 -10 0 10 

37

B. anthracis spores. This would be beneficial as it would 

allow distinction between B. subtilis or other spores from 

B. anthracis. Additionally, we further demonstrate the ability 

of the DMS to detect chemical weapon agent simulants. 

Pyrolysis was used as the sample introduction method for 

the DMS, which uses thermal energy to break apart chemical 

bonds. [45] The fragmentation of the sample will be 

characteristic of the relative strengths of the bonds in the 

original molecule. The maximum known detectable particle 

size for the DMS is about 500 Daltons (0.5 kDa) based 

on previous results (data not shown). We tested several 

pyrolysis methods to ensure that the spores were broken 

down sufficiently for detection by the DMS. Based on the 

SDS-PAGE results shown, the spores were fragmented 

completely at temperatures above 650°C. However, it is 

important to note that the spores were also completely 

fragmented at the lower temperature 550°C when this 

temperature was held for 10 s. The fragmentation thus 

seems to be dependent not only on temperature, but also 

on the time at which this temperature is held. This allows 

flexibility in determining an optimal pyrolysis condition to 

use for DMS analysis. The knowledge of various pyrolysis 

conditions that ensure adequate fragmentation for DMS 

analysis is important not only for these experiments, but 

for any field setups in the future that use pyrolysis as a 

sample introduction method. 

We examined the response of the DMS to pyrolyzed 

spores. We chose to use a lower pyrolysis temperature and 

extend it over a longer period of time in an attempt to 

ensure fragmentation of the spores to particles well under 

the detection limit of the gel, but also without complete 

destruction of any characteristics of these molecular fragments. 

The DMS spectra shown in Figure 4 demonstrate the ability 

of the DMS to distinguish the spores from the sterile 

water in which they are suspended. Significant peaks are 

marked with numbers 1-5. Peaks 1, 3, and 4 appear to be 

correlated with the presence of spores and could potentially 

mark the presence of specific biomarkers. In fact, it 

seems that these peaks may also be concentration-dependent, 

as their magnitude appears to increase proportionally 

to the number of spores present in the sample. The spectra 

for the spores can be distinguished from that of the 

water in which they were resuspended. 

We also demonstrate the use of the DMS for chemical 

weapons agents. MS was used as a simulant nerve agent. 

The DMS produces a clear spectrum for MS at concentration 

levels as low as 45 ppt, as shown in Figures 5a and 5b. 

Another compound, DMMP, was used as a simulant blister 

agent. A unique DMS spectrum is obtained for this compound 

when compared with the one obtained for MS. 

When a mixture of these two compounds was analyzed, the 

resultant spectrum showed features of both compounds. 

38 


This experiment shows that the specificity of the DMS 

enables the detection of multiple chemical warfare agents 

simultaneously. 

CONCLUSIONS 

In view of the growing concern over terrorism, there is a 

need for a sensor that can detect trace amounts of chemical 

or biological warfare agents to minimize the potential 

impact of such a release on the public. Such a sensor 

would aid the government in the defense against chemical 

and biowarfare attacks; an early alert to such an attack 

could save many lives. It would also aid the public health 

sector in the treatment of biowarfare victims by speeding 

diagnosis based on the identification of the particular 

agent delivered. To service both industries, such a sensor 

must be portable, inexpensive, sensitive, and able to detect 

multiple biological or chemical agents. 

Draper Laboratory has designed a sensor that fits these 

characteristics, the DMS. Sionex Corporation is a Draper 

spin-off company created to commercialize and market 

this new device. The DMS has several advantages over 

other types of detectors. It is small, extremely sensitive, 

and relatively inexpensive. Not only are conventional ion 

mobility spectrometers much larger and more expensive, 

they also operate with short pulses of ions. In comparison, 

the DMS analyzes continuously-introduced ions with 

nearly 100% of the “tuned” ions reaching the detector. 

Also, as the electric fields required to filter the ions are on 

the order of 10,000 V/cm, the DMS actually benefits from 

miniaturization; by reducing the gap dimensions to the 

order of 500 µm, the voltages required for ion filtering are 

easily achievable. Mass spectrometers are larger and more 

expensive. They also require an inert environment for 

detection, whereas the DMS operates at atmospheric pressure. 

A portable hand-held DMS unit is already a reality 

and further miniaturization is possible. 

We have demonstrated the utility of the DMS to detect 

potential biological and chemical warfare agents. The DMS 

is a sensitive, hand-held device that can detect multiple 

biological and chemical agents, even in the presence of 

interferents. Furthermore, these devices can be manufactured 

using mass production techniques, significantly 

lowering their cost. The DMS identified a unique, repeatable 

spectrum for B. subtilis spores, used as a surrogate for 

B. anthracis spores. The concentrations shown here as 

being detected easily are very low for the reagentless and 

fast (less than 5 min) class of sensors. Even at such low 

concentrations, the spores can be distinguished from the 

water in which they are resuspended. This sensor with the 

detection capacity currently shown could be used initially 

as a trigger device, where a warning would be given if a 

signal that looks similar to that resulting from the presence 

of spores is discovered. In the future, however, if we can 

show the ability to detect lower concentrations than

shown here, we could move toward a detector that can 

actually make class decisions based on species and amount 

present, and would therefore offer alarms only in the event 

of an actual release. Furthermore, we have shown the 

capabilities of the DMS in the detection of chemical 

weapons agents. The sensitivity of the DMS enabled detection 

of a simulant nerve agent at a concentration of 45 ppt. 

The specificity of the sensor was demonstrated by the 

simultaneous detection of two chemical weapons simulants. 

Future experiments will test the ability of the DMS to distinguish 

between closely-related biological and chemical 

agents. We also intend to try these experiments using various 

interferents to further demonstrate the specificity of 

the sensor. 


We would like to express our thanks to the following 

Draper employees for their assistance on this project: 

Joung-Mo Kang, Andrew D. Dineen, Henry A. 

Raczkowski, and J. Ryan Prince. The authors would also 

like to thank Drs. Jeffrey A. Gelfand and Michael V. 

Callahan for fruitful discussions on chemical and biological 

weapons detection. 

This project was sponsored by the Department of the 

Army, Cooperative Agreement DAMD-99-2-9001. The 

content of this paper does not necessarily reflect the position 

or the policy of the government, and no official 

endorsement should be inferred. 

REFERENCES 


[1] Klietmann, W.F. and K.L. Ruoff, “Bioterrorism: Implications for 

the Clinical Microbiologist,” Clinical Microbiology Reviews, Vol. 

14, 2001, pp. 364-81. 

[2] Harris, S., “Japanese Biological Warfare Research on Humans: A 

Case Study on Microbiology and Ethics,” Ann. N.Y. Acad. Sci., 

Vol. 666, 1992, pp. 21-52. 

[3] Abramova, F.A., L.M. Grinberg, O.V. Yampolskaya, D.H. Walker, 

“Pathology of Inhalation Anthrax in 42 Cases from the 

Sverdlovsk Outbreak of 1979,” Proc. Natl. Acad. Sci., Vol. 90, 

1993, pp. 2291-2294. 

[4] Olson, K.B., “Aum Shinrikyo: Once and Future Threat?” Emerg. 

Infect. Dis., Vol. 5, 1999, pp. 513-6. 

[5] Torok, T.J., R.V. Tauxe, R.P. Wise, J.R. Livengood, R. Sokolow, S. 

Mauvais, K.A. Birkness, M.R. Skeels, J.M. Horan, L.R. Foster, “A 

Large Community Outbreak of Salmonellosis Caused by 

Intentional Contamination of Restaurant Salad Bars,” JAMA, 

Vol. 278, 1997, pp. 389-95. 

[6] Jernigan, J.A., D.S. Stephens, D.A. Ashford, C. Omenaca, M.S. 

Topiel, M. Galbraith, “Bioterrorism-Related Inhalational 

Anthrax: the First 10 Cases Reported in the United States,” 

Emerg. Infect. Dis., Vol. 7, 2001, pp. 933-44. 

[7] Mock, M. and A. Fouet, “Anthrax,” Annu. Rev. Microbiol., Vol. 

55, 2001, pp. 647-71. 

[8] Broussard, L.A., “Biological Agents: Weapons of Warfare and 

Bioterrorism,” Molecular Diagnostics, Vol. 6, 2001, pp. 323-33. 

[9] Bozeman, W.P., D. Dilbero, J.L. Schauben, “Biological and 

Chemical Weapons of Mass Destruction,” Emerg. Med. Clin. 

North Am., Vol. 20, 2002, pp. 975-93. 

[10] Evison, D., D. Hinsley, P. Rice, “Chemical Weapons,” BMJ, Vol. 

324, 2002, pp. 332-5. 

[11] Parker, M., P. Emanuel, J. Wilson, P. Estacio, M. Callahan, D. 

Cullin, Proceedings from The Consensus Conference on the Role 

of Biosensors in the Detection of Agents of Bioterrorism, 

Telemedicine & Advanced Technology Research Center 

(TATRC), 2003, pp. 1-15. 

[12] Turnbull, P.C., R.A. Hutson, M.J. Ward, M.N. Jones, C.P. Quinn, 

N.J. Finnie, C.J. Duggleby, J.M. Kramer, and J. Melling, 

“Bacillus anthracis but Not Always Anthrax,” J. Appl. Bacteriol., 

Vol. 72, 1992, pp. 21-8. 

[13] Patra, G., P. Sylvestre, V. Ramisse, J. Therasse, J.L. Guesdon, 

“Isolation of a Specific Chromosomic DNA Sequence of Bacillus 

anthracis and Its Possible Use in Diagnosis,” FEMS Immunol. 

Med. Microbiol., Vol. 15, 1996, pp. 223-31. 

[14] Cooney, S., “Rapid Anthrax Test Development,” Nature 

Medicine, Vol. 7, pp. 1265. 

[15] Dubiley, S., E. Kirillov, A. Mirzabekov, “Polymorphism Analysis 

and Gene Detection by Minisequencing on an Array of Gel- 

Immobilized Primers,” Nucleic Acids Res., Vol. 27, 1999, pp. e19. 

[16] Higgins, J.A., M.S. Ibrahim, F.K. Knauert, “Sensitive and Rapid 

Identification of Biological Threat Agents,” Ann. N.Y. Acad. Sci., 

Vol. 894, 1999, pp. 130-48. 

[17] Beyer, W., S. Pocivalsek, R. Bohm, “Polymerase Chain Reaction- 

ELISA to Detect Bacillus anthracis from Soil Samples 

-Limitations of Present Published Primers,” J. Appl. Microbiol., 

Vol. 87, 1999, pp. 229-36. 

[18] Bruno, J.G. and J.L. Kiel, “In Vitro Selection of DNA Aptamers 

to Anthrax Spores with Electrochemiluminescence Detection,” 

Biosens. Bioelectron., Vol. 14, 1999, pp. 457-64. 

[19] Lee, M.A., G. Brightwell, D. Leslie, H. Bird, A. Hamilton, 

“Fluorescent Detection Techniques for Real-Time Multiplex 

Strand-Specific Detection of Bacillus anthracis Using Rapid 

PCR,” J. Appl. Microbio., Vol. 87, 1999, pp. 218-23. 

[20] Strizhkov, B.N., A.L. Drobyshev, V.M. Mikhailovich, A.D. 

Mirzabekov, “PCR Amplification on a Microarray of Gel- 

Immobilized Oligonucleotides: Detection of Bacterial Toxinand 

Drug-Resistant Genes and Their Mutations,” Biotechniques, 

Vol. 29, 2000, pp. 844-848. 

[21] Makino, S.I., H.I. Cheun, M. Watarai, I. Uchida, K. Takeshi, 

“Detection of Anthrax Spores from the Air by Real-Time PCR,” 

Lett. Appl. Microbiol., Vol. 33, 2001, pp. 237-40. 

[22] Uhl, J.R., C.A. Bell, L.M. Sloan, M.J. Espy, T.F. Smith, J.E. 

Rosenblatt, F.R. Cockerill, “Application of Rapid-Cycle Real- 

Time Polymerase Chain Reaction for the Detection of Microbial 

Pathogens: the Mayo-Roche Rapid Anthrax Test,” Mayo Clin. 

Proc., Vol. 77, 2002, pp. 673-80. 

[23] Phillips, A.P. and K.L. Martin, “Quantitative Immunofluorescence 

Studies of the Serology of Bacillus anthracis 

Spores,” Appl. Environ. Microbiol., Vol. 46, 1983, pp. 1430-2. 

[24] Phillips, A.P., K.L. Martin, M.G. Broster, “Differentiation 

Between Spores of Bacillus anthracis and Bacillus cereus by a 

Quantitative Immunofluorescence Technique,” J. Clin. 

Microbiol., Vol. 17, 1983, pp. 41-7. 

[25] Phillips, A.P. and K.L. Martin, “Investigation of Spore Surface 

Antigens in the Genus Bacillus by the Use of Polyclonal 

Antibodies in Immunofluorescence Tests,” J. Appl. Bacteriol., 

Vol. 64, 1988, pp. 47-55. 

39

[26] Yu, H., “Comparative Studies of Magnetic Particle-Based Solid 

Phase Fluorogenic and Electrochemiluminescent Immunoassay,” 

J. Immunol. Methods., Vol. 218, 1998, pp. 1-8. 

[27] Longchamp, P. and T. Leighton, “Molecular Recognition 

Specificity of Bacillus anthracis Spore Antibodies,” J. Appl. 


[28] De, B.K., S.L. Bragg, G.N. Sanden, K.E. Wilson, L.A. Diem, C.K. 

Marston, A.R. Hoffmaster, G.A. Barnett, R.S. Weyant, T.G. 

Abshire, J.W. Ezzell, T. Popovic, “A Two-Component Direct 

Fluorescent-Antibody Assay for Rapid Identification of Bacillus 

anthracis,” Emerg. Infect. Dis., Vol. 8, 2002, pp. 1060-5. 

[29] Zhou, B., P. Wirsching, K.D. Janda, “Human Antibodies Against 

Spores of the Genus Bacillus: a Model Study for Detection of 

and Protection Against Anthrax and the Bioterrorist Threat,” 

Proc. Natl. Acad. Sci. USA, Vol. 99, 2002, pp. 5241-6. 

[30] Shute, L.A., C.S. Gutteridge, J.R. Norris, R.C. Berkeley, “Curie- 

Point Pyrolysis Mass Spectrometry Applied to Characterization 

and Identification of Selected Bacillus Species,” J. Gen. 


[31] Fox, A., G.E. Black, K. Fox, S. Rostovtseva, “Determination of 

Carbohydrate Profiles of Bacillus anthracis and Bacillus cereus 

Including Identification of O-Methyl Methylpentoses Using Gas 

Chromatography-Mass Spectrometry,” J. Clin. Microbiol., Vol. 

31, 1993, pp. 887-94. 

[32] Hathout, Y., P.A. Demirev, Y.P. Ho, J.L. Bundy, V. Ryzhov, L. 

Sapp, J. Stutler, J. Jackman, C. Fenselau, “Identification of 

Bacillus Spores by Matrix-Assisted Laser Desorption Ionization- 

Mass Spectrometry,” Appl. Environ. Microbiol., Vol. 65, 1999, 

pp. 4313-9. 

[33] Demirev, P.A., J. Ramirez, C. Fenselau, “Tandem Mass 

Spectrometry of Intact Proteins for Characterization of 

Biomarkers from Bacillus cereus T Spores,” Anal. Chem., Vol. 

73, 2001, pp. 5725-31. 

[34] Elhanany, E., R. Barak, M. Fisher, D. Kobiler, Z. Altboum, 

“Detection of Specific Bacillus anthracis Spore Biomarkers by 

Matrix-Assisted Laser Desorption/Ionization Time-of-Flight 

Mass Spectrometry,” Rapid Commun. Mass Spectrom., Vol. 15, 

2001, pp. 2110-6. 

[35] Mowry, C., C. Morgan, Q. Baca, R. Manginell, R. Kotenstette, P. 

Lewis, G. Frye-Mason, “Rapid Detection of Bacteria with 

Miniaturized Pyrolysis-GC Analysis,” Proceedings of SPIE, 

Boston, MA, 2001. 

40 


[36] Miller, R.A., G.A. Eiceman, E.G. Nazarov, A.T. King, “A Novel 

Micromachined High-Field Asymmetric Waveform-Ion 

Mobility Spectrometer,” Sensors and Actuators, Vol. 67, 2000, 

pp. 300-6. 

[37] Eiceman, G.A., B. Tadjikov, E. Krylov, E.G. Nazarov, R.A. Miller, 

J. Westbrook, P. Funk, “Miniature Radio-Frequency Mobility 

Analyzer as a Gas Chromatographic Detector for Oxygen- 

Containing Volatile Organic Compounds, Pheromones, and 

Other Inset Attractants,” J. Chromatography, Vol. 917, 2001, 

pp. 205-17. 

[38] Miller, R.A., G.A. Eiceman, E.G. Nazarov, A.T. King, “A MEMS 

Radio-Frequency Ion Mobility Spectrometer for Chemical Agent 

Detection,” presented at Solid-State Sensor and Actuator 

Workshop, Hilton Head Island, SC, 2000. 

[39] Krylov, E., E.G. Nazarov, R.A. Miller, B. Tadjikov, G.A. Eiceman, 

“Field Dependence of Mobilities for Gas-Phase-Protonated 

Monomers and Proton-Bound Dimers of Ketones by Planar 

Field Asymmetric Waveform Ion Mobility Spectrometer 

(PFAIMS),” J. Phys. Chem., Vol. 106, 2002, pp. 5437-44. 

[40] Miller, R.A., E.G. Nazarov, G.A. Eiceman, A.T. King, “A MEMS 

Radio-Frequency Ion Mobility Spectrometer for Chemical Vapor 

Detection,” Sensors and Actuators, Vol. 91, 2001, pp. 307-18. 

[41] Davis, C.E., J.M. Kang, C.E. Dubé, J.T. Borenstein, E.G. 

Nazarov, R.A. Miller, A.M. Zapata, “Spore Biomarker Detection 

Using a MEMS Differential Mobility Spectrometer,” presented at 

Transducers, Solid-State Sensors, Actuators and Microsystems, 

2003. 

[42] Schaeffer, P., J. Millet, and J.P. Aubert, “Catabolic Repression of 

Bacterial Sporulation,” Proc. Natl. Acad. Sci. USA, Vol. 54, 

1965, pp. 704-711. 

[43] Fouet, A. and A.L. Sonenshein, “A Target for Carbon Source- 

Dependent Negative Regulation of the citB Promoter of Bacillus 

subtilis,” J. Bacteriol., Vol. 172, 1990, pp. 835-44. 

[44] Kirschstein, M., R. Jensen, R. Schroder, K. Sack, “Proteinuria 

after Renal Transplantation: Diagnosis with Highly Sensitive 

Silver Stain in Sodium Dodecylsulphate-Polyacrylamide 

Gradient Gel Electrophoresis,” Klin. Wochenschr, Vol. 69, 1991, 

pp. 847-52. 

[45] Wampler, T.P., Analytical Pyrolysis: An Overview, Marcel 

Dekker, Inc., New York, 1995.


(l-r) Angela M. Zapata, Melissa D. Krebs 

Melissa D. Krebs is a Member of the Technical Staff at Draper Laboratory. Her Masters thesis work involved the study of 

protein interactions in the cellulosome of Clostridium thermocellum to investigate the mechanism for the efficient degradation 

of cellulose. Her current research interests include various application areas for DMS sensor technology, including 

biowarfare detection and clinical diagnostics. She received BS and MS degrees in Chemical Engineering from the University 

of Rochester (2002 and 2003, respectively). 

Angela M. Zapata is a Senior Member of the Technical Staff in the Biomedical Engineering Group at Draper Laboratory. She 

is responsible for developing novel technologies for solutions in clinical diagnostics and therapeutics, environmental monitoring 

and remediation, and homeland defense. In 2002, she joined the Cambridge Rindge and Latin School, Cambridge, 

where she started a 4-year school-to-work biotechnology training program for high school students. She rejoined Draper 

on a full-time basis in the fall of 2004. She holds 2 patents and 14 publications in the areas of field-deployable and MEMSbased 

analytical devices. Dr. Zapata received a BS in Chemistry from Worcester State College (1992) and a PhD in Analytical 

Chemistry from Tufts University (2000). 

Erkinjon G. Nazarov is a Chief Scientist at Sionex Corporation, Waltham, MA. He is a pioneer in DMS technology and has 

extensive experience in the design, fabrication, and evaluation of differential mobility spectrometry-based sensors. He 

received a BS from Leningrad Polytechnic Institute, Leningrad, Russia, a PhD from Ioffe Physical Technical Institute, 

Leningrad, and Dr.Phys. and Math.Sci. degrees from St. Petersburg Polytechnic University, St. Petersburg, Russia (1981 and 

1992, respectively). 

Raanan A. Miller is the Founder, Vice President of Technology, and Chief Technical Officer of Sionex Corporation, 

Waltham, MA. He has over 12 years experience in MEMS technology development, including sensor and actuator design, 

modeling and process design, and fabrication. He has developed novel electromagnetic optical switches and beam-steering 

systems incorporating magnetic thin films and worked on incorporating novel materials such as lead zirconate titanate 

(PZT) thin films into MEMS acoustic sensors and actuators. A leading expert in differential mobility spectrometry, he is 

responsible for the MEMS microDMS sensor development and much of the Sionex intellectual property. Dr. Miller received 

a BS from Boston University and MS and PhD degrees from the California Institute of Technology, Pasadena. 

Isaac S. Costa was previously a member of the MEMS and Bioengineering groups at Draper Laboratory, where he worked 

for 5 years. His research included applying DMS to spore detection and process development for the manufacture of the 

micro-canary sensor, the nonvolatile residue sensor, and the all-silicon oscillating accelerometer. His primary interest is in 

advanced process development for silicon MEMS devices based on silicon-on-insulator technology. He received a BS in 

Interdepartmental Sciences from the American International College, Springfield, MA. 

Abraham L. Sonenshein is a Professor of Molecular Biology and Microbiology at the Tufts University School of Medicine, 

and has studied the regulation of gene expression and sporulation in Bacillus subtilis for more than 30 years. He was the 

Editor of the Journal of Bacteriology from 1990 to 1996, and served as the Editor-in-Chief of “Bacillus subtilis and Other 

Gram-Positive Bacteria: Biochemistry, Physiology and Molecular Genetics and Bacillus subtilis” and “Its Closest Relatives: 

from Genes to Cells,” both published by ASM Press. Dr. Sonenshein was elected to fellowship in the American Academy of 

Microbiology in 1993 and has served on its Board of Governors since 1999. He received a BS in Biochemical Sciences from 

Princeton University (1965) and a PhD in Biology from MIT (1970). 

Cristina E. Davis is a Principal Member of the Technical Staff and Group Leader of Bioengineering at Draper Laboratory. 

Her work focuses on biological detection and analysis of biomarkers using the DMS sensor technology. Previous research 

includes high-throughput electrophysiology systems, membrane biophysics, signal transduction pathways in cell volume 

regulation, voltage-sensitive fluorescent dyes, and MEMS biosensor research. Dr. Davis received a BS from Duke University, 

Durham, NC, with a double major in Mathematics and Biology (1994), and MS and PhD degrees in Biomedical Engineering 

from the University of Virginia (1996 and 1999, respectively). 

41

Autonomous Mission Management for Spacecraft Rendezvous 

Using an Agent Hierarchy 

Mark Jackson, Christopher D'Souza, Hobson Lane 

Copyright © 2005 by The Charles Stark Draper Laboratory, Inc. Presented at Infotech@Aerospace. Arlington, VA, 

September 26-29, 2005 

INTRODUCTION 

An agent-based autonomous mission manager for spacecraft is developed and applied to a spacecraft 

rendezvous problem. The mission manager monitors, plans, and executes rendezvous activities 

for a chaser satellite with low-thrust and high-thrust effectors. The software is based on a hierarchy of 

“activity agents,” each of which plans, monitors, executes, and replans an activity. These activity agents 

are instantiated and scheduled by a software framework capable of reconfiguring the agent hierarchy 

when replanning occurs. Targeting and planning algorithms are incorporated into the agent hierarchy 

to orchestrate a rendezvous with an orbiting spacecraft. These algorithms make use of low-thrust guidance 

to arrive in the neighborhood of the target, followed by a series of high-thrust burns to arrive at 

a specified offset point at a specified time. During both the low-thrust and high-thrust activities, the 

mission manager monitors subsystem status as well as the relative navigation state of the chaser. When 

low-thrust failures occur, the manager completes the rendezvous by scheduling additional high-thrust 

burns. The mission manager, along with low- and high-thrust guidance algorithms, is implemented in 

simulation. A demonstration of several replanning events – both failure and error driven – is provided 

with 3-dimensional (3D) graphics as well as an “agent activity” display that depicts the active agents 

and shows the hierarchy reconfiguration process when replans occur. 

The next 10 to 20 years are likely to see an increasing need 

for spacecraft capable of autonomous rendezvous with 

other spacecraft or space objects. The National 

Aeronautics & Space Administration (NASA) is currently 

developing a set of exploration vehicles, both manned and 

unmanned, to accomplish several missions, including 

International Space Station (ISS) crew exchange, ISS 

resupply, lunar transit and landing, lunar return, and 

ultimately, Mars exploration. These missions will all 

require spacecraft rendezvous and docking in various 

42 

orbital environments, including low Earth orbit, lunar 

orbit, Martian orbit, and possibly, rendezvous and docking 

at or near Lagrange points. Additionally, there is significant 

military and commercial interest in satellite rendezvous for 

resupply, intelligence, and other operations. 

Many of these operations will be conducted by unmanned 

spacecraft in arenas for which close ground contact and 

supervision are neither practical nor desirable. Spacecraft 

will be required to autonomously plan rendezvous maneuvers, 

monitor navigation and subsystem status, diagnose

Autonomous Mission Management for Spacecraft Rendezvous Using an Agent Hierarchy 

problems with the existing maneuver plans, and replan 

when required. To accomplish this, software will be 

required with capabilities above and beyond classic guidance 

algorithms. 

Toward this end, Northrop Grumman Corporation in collaboration 

with Draper Laboratory has applied an 

agent-hierarchy approach developed by Draper to a satellite 

rendezvous problem in simulation. The satellite incorporates 

Northrop Grumman’s low-thrust rendezvous guidance. 

In addition, it is equipped with high-thrust actuators and a 

sensor suite consisting of Global Positioning System (GPS) 

for absolute navigation, relative GPS, and light detection and 

ranging (LIDAR) for close-in rendezvous. 

Three sections follow. First, some background is provided 

on the general principles of mission management as they 

apply to spacecraft. This section discusses the role of the 

mission manager in spacecraft and discusses mission management 

principles and processes. The second section 

applies these principles to the rendezvous problem mentioned 

above. Finally, some simulation results are presented 

that demonstrate the capability of the mission manager to 

react to subsystem failures and navigation state updates. 

Background on Mission Management from Spacecraft 

Role of the Mission Manager 

For space vehicles, the role of autonomous mission management 

software is to configure guidance, navigation, and 

control (GN&C) and subsystem components to meet mission 

objectives (Figure 1). The inputs to the mission 

manager are the states and status of the vehicle and its subsystems 

from failure detection, isolation, and recovery 

(FDIR) and from navigation, as well as the current mission 

objectives. The outputs are commands to guidance and 

control (G&C), and modes and configurations for subsystems. 

The mission manager, therefore, multiplexes a great 

deal of state and status data to create the required commands. 

The principal challenge to the practical 

implementation of autonomous systems in spacecraft is 

not in determining the optimal path or best course of 

action for a given situation (or set of inputs), but rather, it 

is the complexity that arises from the range of possible failure 

situations and vehicle configurations that the software 

must handle. A structured approach to the design and 

implementation of an autonomous mission manager is 

absolutely critical. An ad-hoc design will quickly become 

brittle and may be virtually untestable. 

Crew/Mission Control 

Objectives 

Flight Computer Subsystems 

FDIR 

Nav 

Mission 

Manager G&C 

Sensors 

Effectors 

Other 

Figure 1. Mission manager interfaces. 

Much of the modern literature acknowledges four basic 

functions that the mission manager must perform to 

accomplish this multiplexing task. We shall refer to these 

as: monitoring, diagnosis, planning, and execution. These 

correspond to the observe, orient, decide, and act functions 

that are prevalent in the military literature. [1] We 

review these here since they figure prominently in our discussion 

of hierarchical architectures and their application 

to spacecraft rendezvous. 

Monitoring is the process of collecting state and status 

inputs and checking for failures or problems with the current 

plan. The primary job of the monitoring function is to 

answer the question: Given the current state and status, 

will the current plan and vehicle configuration meet my 

objectives within constraints? Often, for space applications, 

answering this question involves propagating the 

current state of the vehicle forward in time to the end of a 

current activity or mission phase. 

Diagnosis is the process of identifying problems with the 

current plan or configuration and deciding whether 

replanning or reconfiguration is necessary. Note that this 

diagnosis function is focused on identifying problems with 

plans or configurations, not on finding mechanical problems 

within the subsystems. Although these diagnosis jobs 

may overlap, we generally consider the diagnosis of subsystem 

faults to be part of the FDIR software. 

Planning is the process of creating a series of commands or 

configurations that achieve the mission objectives within 

constraints. When primary mission objectives cannot be 

met, planners must be capable of selecting alternate objectives. 

In the worst case, alternate objectives may include 

mission abort or spacecraft safing. Planning may occur at 

the beginning of a mission phase or activity or whenever 

diagnosis determines that replanning is necessary. 

Finally, the execution software enacts the plan by sequencing 

commands to G&C algorithms and vehicle 

subsystems. This sequencing may be time or event-driven 

depending on the nature of the plan. 

Note that these four functions form a continuous loop with 

monitoring and execution running continuously, and diagnosis 

and planning happening when failures occur or 

modeling errors accrue. The next section starts with planning 

and shows how a hierarchical planning process can be 

used to make decisions or optimize a path or configuration. 

Hierarchical Planning 

Consider the problem of moving a vehicle from point A to 

point C in Figure 2. We assume that subplanner 1 is capable 

of finding an optimal path from A to B, while 

subplanner 2 is capable of finding the optimal path from B 

to C, and that the coordinates of the intermediate point, B 

are B1 and B2. These coordinates are known as interface 

variables. This problem may be solved by a simple hierarchy 

of three planning entities as shown. The supervisor 

planner invokes subplanner 1 by requesting it to solve a 

problem consisting of the start state, A, the goal state, B, 

43


and any applicable constraints. Subplanner 1 returns the 

cost of the optimal trajectory from A to B, call it J1. 

Similarly, the supervisor planner may request subplanner 

2 to solve the B to C problem and receives the B to C 

cost, J2. The total cost, J1 + J2, is plotted topographically 

on the right as a function of the interface states 

(coordinates of B). The supervisor optimizes the A to C 

trajectory by searching for the optimal coordinates of B. 

Depending on the attributes of the problem, the supervisor 

may find the optimal solution by using one of any 

number of optimization techniques, gradient descent for 

example. 

Figure 2. Optimization or decision-making via hierarchical 

planners. 

On the other hand, the supervisor may not be required to 

find the optimal solution at all. It may instead search 

through a predefined set of interface states (red Xs in the 

figure) choosing the option with the lowest cost. When 

the supervisor evaluates a set of options, we shall call the 

process decision-making. When an iterative search is 

used, the process may be an optimization process. The 

multilevel optimization technique described here is 

referred to as interaction coordination. Other techniques, 

such as price coordination are also available. [2] 

Of course, this iterative process is not new to those familiar 

with optimization. Many optimization algorithms 

work by iterating on a set of control variables to minimize 

a cost function. But diagramming the problem this 

way highlights some of the advantages of thinking about 

planning in a hierarchical sense. Each subplanner may be 

an expert at its piece of the problem. Subplanner 2 need 

not have any knowledge of subplanner 1’s problem and 

vice versa. Further, the supervisor need not have all the 

details of the subproblems to solve the overall problem. 

Second, once the vehicle has passed point B, there is no 

longer any need to consider the A to B problem, and subplanner 

1 (and all its associated data and memory) may 

be deactivated. Finally, note that the time horizon for 

each planner gets longer as we move up the hierarchy 

with the supervisor having the lowest level of detail while 

planning over the longest time horizon. 

It is easy to imagine an extension of the two-level hierarchy 

described above in which each subplanner in turn 

solves its problem by invoking subplanners of its own. 

44 

A 

Subplanner 

1 

Cost = J1 

Supervisor 

Planner 

B = interface 

state 

B 

Subplanner 

2 

Cost = J2 

Total Cost, J = J1 + J2 

(out of page) 

B2 

J=50 J=40 

B1 

C 

Discrete decision points 

J=30 

J=50 

Optimum 

This sort of hierarchical planning breakdown has one 

other very important advantage. The same process that 

creates the planning tree results in a natural breakdown 

for all the functions required of a mission manager. In 

fact, the process of invoking subplanners and sub-subplanners 

can be used to create a hierarchy of agents, each 

responsible for monitoring, diagnosing, planning, and 

executing a section of the mission. We shall refer to these 

agents as activity agents. Activity agents that are connected 

to a superior agent in the hierarchy are referred to as 

children of the superior, while the superior agent is the 

parent of each child. Agents without children (bottom 

level nodes) are referred to as leaf agents, or leaf nodes. 

Figure 3 shows an example for a spacecraft rendezvous 

and docking mission. Each of the activity agents has four 

sub-boxes, corresponding to monitoring, diagnosis, 

planning, and execution. Just as for planning, the time 

horizons for each activity increases with its level in the 

hierarchy, while the detail of its functions decreases. 

This process of breaking down a complex problem or 

mission into activities or phases is referred to as temporal 

decomposition. The process of defining the functionality of 

each activity agent is referred to here as functional decomposition. 

In this context, the functional decomposition 

consists of defining the monitoring, diagnosis, planning, 

and execution functions required by each 

activity. 

Rendez 

D P 

M E 

Burn 

D P 

M E 

D P 

M E 

Survey 

D P 

M E 

Acquire Insert 

D P 

M E 

Mission 

D P 

M E 

Maneuver 

D P 

M E 

Approach 

D P 

M E 

Maintain Init Final 

D P D P D P 

M E M E M E 

Figure 3. Example hierarchy of activity agents. 

The interfaces between parent and child activities in 

Figure 3 consist of problem specifications from parent to 

child and solution information from child to parent. For 

vehicle applications, the problem specification usually 

consists of a start state for the child activity, a goal state, 

and a set of constraints. Sometimes, the goal state is only 

partially specified or constrained, so the child must 

return the final state in the solution information. 

Solution information may also contain sensitivity data, 

resource data, and cost data, etc. At the lowest level, the 

interfaces are commands to the GN&C and subsystems. 

These commands may be thought of as problem specifications 

to the vehicle. 

The implementation of these hierarchical mission management 

architectures requires a software framework

capable of instantiating the agent hierarchy during the 

planning process, sequencing the monitoring and diagnosis 

functions, and rebuilding the hierarchy from any 

level when replanning is required. The framework provides 

the designer with built-in activity interfaces, 

activity code templates, and locations for user-defined 

monitoring, diagnosis, planning and execution functions. 

The framework used in the example application 

below is the All Domain Execution and Planning 

Technology (ADEPT) developed by Draper Laboratory. [2] 

More details on the ADEPT framework will be provided 

with the examples. 

Design Process 

In the development of complex mission management 

applications, the design process is as important as the 

software architecture. Figure 4 illustrates the recommended 

design process. The process starts with an 

analysis of vehicle requirements and operations concepts. 

Once these are understood, a temporal 

decomposition is performed to create the activity hierarchy. 

The rendezvous application described below 

provides an example of a temporal decomposition. Rules 

of thumb are provided with the example to assist designers 

in creating a manageable breakdown without 

overdividing the problem. 

REQUIREMENTS 

Concept of 

Operations 

Operational 

Requirements 


DECOMPOSITION 

Temporal 

Functional 

Algorithm 

Selections 

Mathematical 

Problems 

Algorithm 

Knowledge Base 

PEM 

Schematic 

Selected 

Algorithms 

Populate ADEPT 

Architecture 

Figure 4. ADEPT design process. 

Next, a functional decomposition of each activity is performed 

to define requirements for each of the four 

activity functions. These requirements are posed as 

mathematical problems that may be solved by appropriate 

decision-making or optimization algorithms. These 

algorithms are then used to populate the framework of 

activity agents. Once the temporal and functional 

decompositions are complete, designers should step 

through the agent responses to hypothetical scenarios or 

“use cases” to identify any gaps in the activity hierarchy 

and to further identify algorithm requirements. 

The algorithms used by the mission manager may be 

divided into two broad categories: domain-specific algorithms 

and decision-making/optimization algorithms. 

The domain-specific algorithms act on vehicle state and 

status data to provide information about future consequences 

of actions. These may include state propagation 

routines, environmental models, and vehicle or subsystem 

models. The decision-making algorithms use this 

information about consequences to make decisions, to 

optimize, or to plan. These techniques may range from 

simple logical comparisons to sophisticated optimization 

or artificial intelligence algorithms. 

The next sections apply this design process to a spacecraft 

rendezvous problem. 

Application: Prototype Autonomous Mission 

Manager 

The hierarchy of activity agents described above was 

applied to an autonomous spacecraft rendezvous mission. 

This Prototype Autonomous Mission Manager 

(PAMM) was implemented in simulation aboard a chaser 

spacecraft whose mission was to rendezvous with a 

dynamically passive target. For this application, the target 

spacecraft was assumed to have GPS capability. This 

allowed the chaser navigation system to update the target 

location via ground uplinks and to utilize relative GPS 

techniques for mid-field rendezvous. The chaser effector 

models included a low-thrust propulsion system as well 

as high-thrust engines capable of actuating nearly impulsive 

burns. Chaser sensor models included a GPS system 

for absolute inertial position, a relative GPS model to 

provide relative state information within about 25 km, 

and a LIDAR rendezvous sensor capable of returning 

range, azimuth, and elevation information for close-in 

operations (

Y eci (me) 


Figure 5. Low-thrust portion of rendezvous. 

The low-thrust guidance algorithm works by optimizing 

the attitude profile during long periods of constant thrust 

to accomplish rendezvous, minimizing time and fuel consumption 

for the rendezvous. 

The low-thrust portion of the rendezvous may take several 

days, each consisting of dozens of orbits. This means 

that the mission manager must be capable of planning 

high-thrust burns to complete the rendezvous from a variety 

of failure conditions, resulting in a range of phase 

angles, altitude differentials, and lateral errors. 

If all goes nominally, the low-thrust guidance delivers the 

satellite to a point approximately 150 km behind and 30 

km below the RSO. Figure 6 shows the final 2500 km of 

the rendezvous. In the figure, the RSO is at the origin and 

is moving to the left. The solid green trace is the relative trajectory 

of the chaser in the orbit plane. As the figure shows, 

in the nominal case, two final high-thrust burns precisely 

complete the trajectory to the desired offset point. 

Altitude Difference (km) 

Figure 6. Chaser downrange and altitude errors from RSO 

(+Vbar left). 

46 

x107 1 

0.8 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

-0.6 

-0.8 

X eci (me) x106 -1 

-8 -6 -4 -2 0 2 4 6 8 

10 

0 

-10 

-20 

-30 

-40 

-50 

-60 

-70 

-80 

Target offset point 

Nominal Mission (No Failures) 

Low-thrust cutoff 

Downrange (km) 

Green: Chaser 

Red: RSO 

RSO Orbit 

Chaser lowthrust 

spiral 

trajectory 

High-thrust burns 

Chaser lowthrust 

trajectory 

-90 0 500 1000 1500 2000 2500 

When failures occur, the high-thrust system has enough 

impulse to accomplish a rendezvous from approximately 

100 to 2500 km behind the RSO. Should the failure occur 

outside this region, the mission manager waits until the 

chaser has “lapped” the RSO and arrived back in the region 

“behind” the RSO prior to scheduling high-thrust burns. 

In testing an autonomous rendezvous concept, it is important 

to include navigation errors, particularly in the 

knowledge of the target vehicle (RSO) state. Filter convergence 

and target state updates test the mission manager 

response at several levels. Smaller corrections cause 

replanning at lower levels in the hierarchy since they tend 

to affect only the current or upcoming activities. Large corrections 

can cause a complete replan of a trajectory or 

burn sequence. Also, target state estimation updates are 

very similar, from the mission manager point of view, to 

target maneuvers. 

The chaser navigation software includes a dual inertial 

state extended Kalman filter. This means that the filter 

tracks and propagates the states of both the chaser and 

RSO. During far field rendezvous, the chaser inertial state 

is updated by the onboard GPS sensor, while the RSO state 

is updated by ground uplinks. Once inside communication 

range, relative GPS provides continuous relative RSO 

location information, which is used to update the filter’s 

inertial RSO state estimate. Navigation state updates and 

sensor acquisitions can cause the mission manager to execute 

replanning. 

The above discussion of the CONOPS has prepared us to 

list a set of objectives explicitly for the rendezvous portion 

of the PAMM mission. These will be enacted by the agent 

hierarchy as described in the sections on temporal and 

functional decomposition below. 

These mission objectives are simple and clear and may be 

stated as: 

1. Achieve the target offset point at the specified time 

using low thrust to 150 km and high-thrust burns 

thereafter. 

2. If #1 cannot be achieved (possibly due to low-thrust 

system failures outside 150 km, target maneuvers, or 

large target state estimation errors), achieve the target 

offset point at the specified time using the high-thrust 

system from the point of failure. 

3. If the target offset point cannot be achieved at the specified 

time, adjust the arrival time to an achievable value 

and continue the mission. 

4. If the target offset point cannot be achieved with 

onboard resources, abort the rendezvous and enter a 

circular orbit. 

5. If failures or resource limitations prevent any of the 

above, enter safe mode.


PAMM Mission Temporal Decomposition 

The temporal decomposition process starts with the 

CONOPS and the mission objectives. Once these are 

understood, a nominal mission timeline can be developed 

with the goal of identifying mission activities and 

their interface states. As a rule of thumb, activity boundaries 

are set whenever a new subsystem or guidance 

mode must be invoked to continue the mission. From 

the CONOPS then, the PAMM temporal decomposition 

will include a low-thrust activity and a high-thrust activity 

(Figure 7). During the low-thrust rendezvous, the 

mission manager invokes the low-thrust agent and 

assigns it a target state relative to the RSO. The lowthrust 

agent continuously uses the optimal guidance 

scheme described above to assess the vehicle’s capability 

to reach the low-thrust target state given current 

resources. The agent also monitors the low-thrust 

propulsion system for failure indications and resource 

usage. The low-thrust agent can correct for small errors 

that may accrue in the trajectory due to navigation errors 

or thrust control errors. If for some reason the low-thrust 

agent determines that the target state cannot be achieved 

from the current vehicle state, given the current 

resources (propellant) and health status, then it reports 

this failure to the mission activity agent. The mission 

activity agent may change the low-thrust target state to a 

state that is both reachable by the low-thrust system and 

from which a high-thrust rendezvous may be completed 

to the target point. The low-thrust target then, is the 

interface state between the two agents as described previously 

in the Hierarchical Planning section. 

The high-thrust rendezvous nominally consists of two 

burn events with coasting flight between the burns as 

described in the CONOPS section. The first burn raises 

the orbital apogee to the altitude of the desired RSO offset 

Low-Thrust 

Rendezvous 

High-Thrust 

Rendezvous 

Mission 

Figure 7. Temporal hierarchy for PAMM mission. 

point, while the second burn places the chaser in an orbit 

that is co-elliptic and co-altitude with the RSO. The second 

burn is timed to cause the chaser to arrive at the 

offset point at the desired time. As we shall see later, 

additional burns may be required to achieve the offset 

point if the high-thrust agent is invoked for any reason at 

a starting state other than close to the nominal low-thrust 

termination state. 

The high-thrust maneuvers include periods of thrusting 

and periods of nonthrusting attitude hold or “pointing.” 

In contrast, the segments of low-thrust maneuvering 

involve continuous thrusting while pointing of the space 

vehicle (and thrust vector) to effect the solution to the 

minimum-time rendezvous problem according to an 

adaption of the Kechechian solutions. Because the lowthrust 

trajectory employs a constant thrust approach, the 

minimum time solution is also the minimum fuel solution. 

To employ this low-thrust algorithm effectively, the 

mission manager must account for these periods of constrained 

attitude requirements in order to ensure the 

satisfaction of other mission constraints such as power 

generation, target sensing, and navigation. During the 

high-thrust rendezvous, the burns will be short and long 

periods will be spent in the pointing activity. The mission 

planner may schedule pointing to optimize solar power 

generation, improve thermal conditions, acquire the RSO 

with sensors, etc. Pointing activities could be planned 

onboard, scheduled on the ground, and tasked as mission 

objectives, or ground-planned activities could be 

treated as constraints to the mission manager. Since the 

focus for this activity was on the rendezvous problem, 

pointing and attitude hold activities were not implemented, 

and the mission manager was implemented in a 

3-degree-of-freedom (DOF) simulation. The implemented 

activity agents are shaded in Figure 7. Reference [3] 

Burn 

Burn Burn 

Thrust Point Height Point CoCo- 

Point Safe 

Raising 

ellipticelliptic Abort 

Proximity 

Operations 

Docking 

47

details many of the considerations for planning the attitude 

profile of a satellite during rendezvous. 

Once the activities have been identified for the nominal 

mission, failure and off-nominal conditions are considered 

to complete the problem decomposition. We 

already have the capability to handle many failure types 

since the mission activity agent can request the highthrust 

system to take over should low-thrust failures 

occur, and the low-thrust agent can replan the lowthrust 

profile should burn errors or navigation state 

updates require it. However, when low-thrust failures 

occur very early in the rendezvous, or if high-thrust failures 

prevent completion of the rendezvous, then it may 

not be possible to achieve the desired offset point. When 

this occurs, the mission agent may invoke an “abort” 

agent to place the spacecraft in a safe orbit co-elliptic 

with the RSO, or if this is not possible, to safe the spacecraft 

subsystems. Note that the abort agent utilizes the 

same burn co-elliptic agent used by the high-thrust 

agent to accomplish its safe orbit objective. Only the 

agents required to accomplish the current mission 

objectives are actually invoked by the mission agent and 

instantiated by the ADEPT framework, so the abort 

agent does not appear in the successful cases shown in 

the results section below. 

Once the desired offset point has been achieved, the 

PAMM spacecraft may conduct proximity operations, 

docking operations, or other RSO-relative maneuvers. 

The proximity operations and docking agents are not 

implemented here, but are displayed in the results section 

as placeholders for future development. Reference 

[3] describes some considerations and techniques for 

autonomous proximity operations and docking. 

PAMM Functional Decomposition 

Once the hierarchy of agents has been determined 

through the temporal decomposition process, we can 

begin the process of defining the required functionality 

of each agent by defining its monitoring, diagnosis, 

planning, and execution functions. For the sake of 

brevity, this section will describe only the more interesting 

functions associated with the shaded elements in 

Figure 7. 

Mission Agent Planning 

Beginning with the mission activity agent, we start by 

discussing planning, since the planning process actually 

creates and reconfigures the agent hierarchy. The mission 

agent planning function looks at the current relative navigation 

state of the chaser with respect to the RSO, along 

with subsystem state and status information. If subsystems 

are nominal and the altitude differential between 

the vehicles is large, then the mission agent will spawn a 

low-thrust child and assign it the objective of reaching a 

nominal target state below and behind the RSO (refer 

48 


again to Figure 6). The low-thrust child agent in turn 

invokes the optimal guidance algorithm to plan a trajectory 

to the assigned relative state. If low-thrust planning 

is successful, the mission agent creates a high-thrust 

child, assigns it the low-thrust termination state as an 

initial condition and the desired offset point as an objective. 

The high-thrust agent in turn calculates a series of 

burns to achieve the desired state by invoking the corresponding 

burn agents. If high-thrust planning is 

successful, then the mission agent terminates planning 

and begins the monitoring cycle. 

If low-thrust planning is unsuccessful, the mission agent 

attempts to accomplish the remaining rendezvous by 

assigning the high-thrust agent the current state as initial 

condition, rather than the predicted low-thrust 

termination state. As part of the high-thrust problem 

specification, the mission agent may also relax the 

arrival time requirement. If the high-thrust agent is successful, 

the mission is continued with high thrust only; 

if not, an abort is commanded by invoking the abort 

agent. 

Note that no assumptions were made about the initial 

vehicle conditions for the planning process. This means 

that the replanning process is exactly the same as the initial 

planning process, although the resulting plan – and 

corresponding agent hierarchy – may be different. 

Mission Agent Monitoring 

Once the rendezvous plan is complete, the mission agent 

begins to monitor subsystem status, as well as the status 

from all its executing children. The job of the monitoring 

function is to determine whether the mission objectives 

may be accomplished given the current plan and the current 

vehicle state and status. Two types of conditions may 

cause the mission agent to trigger a replan by executing its 

diagnosis function. First, a child agent (e.g., low thrust) 

may report that it can no longer achieve the objective 

assigned by the mission agent, and second, a subsystem 

failure may be detected that directly triggers the mission 

diagnosis function. The distinction between these two 

types of failure conditions becomes important in the diagnosis 

process. 

Mission Agent Diagnosis 

Diagnosis is executed whenever monitoring detects a 

condition that prevents achieving the mission objective 

with the current plan (set of activities). The job of diagnosis 

is to determine the reason for the failure of the 

plan in order to select a replanning strategy. If, for example, 

diagnosis is triggered because the low-thrust agent 

has determined (through its own monitoring and diagnosis 

process) that it can no longer reach its commanded 

state, then the mission agent may attempt to assign a 

new goal to the low-thrust child agent. This could occur 

for example if low-thrust propellant usage was larger


than expected, or if a navigation state update placed the 

RSO further out of the orbit plane than previously 

thought. 

If, on the other hand, the reason that diagnosis was triggered 

was due to a low-thrust engine failure, then the 

replanning strategy would be to attempt to use only the 

high-thrust system. 

Mission Agent Execution 

The execution portion of the mission agent, as for all nonleaf 

nodes, is simply to command the next activity when a 

child activity is complete. For the leaf nodes, the execution 

functions send configuration and desired state commands 

to GN&C and other vehicle subsystems. 

Low-Thrust Agent Planning 

The low-thrust planning algorithm uses a trajectory optimization 

technique initially developed by J.A. 

Kechechian. [4],[5] One of the keys to the simplifications 

accomplished in Kechechian’s low-thrust transfer algorithm 

is the formulation of the orbital differential 

equations in a polar coordinate frame that is nonsingular. 

The singularities in conventional orbital element 

approaches tend to frustrate optimization attempts. 

Likewise, Cartesian position-velocity formulations lead 

to highly nonlinear cost functions and inaccurate numerical 

integration due to the cyclical and rapidly varying 

characteristics of these state variables. Kechechian avoids 

these pitfalls by using nonsingular equinoctial orbital 

elements. The Northrop Grumman adaptation of this 

approach corrects several errors in the published 

Kechechian formulation and extends the technique to 

include optimization of mean anomaly (for rendezvous) 

along with the other five orbital elements optimized by 

Kechechian (for orbit transfer). The resulting nonlinear 

optimization solution outputs a time sequence of spacecraft 

attitude that accomplishes the desired rendezvous 

in minimum time given the thrust constraints of the EP 

system. This profile is then executed and monitored 

within the PAMM heirarchy. 

High-Thrust Agent Planning 

The high-thrust agent planning algorithm is responsible 

for generating a set of burns that causes the spacecraft 

to arrive at the desired RSO offset point at a specified 

time, given an initial condition at the termination or 

failure of low thrust. The approach to solving this problem 

is to generate a series of co-elliptic orbits that 

stair-step the chaser to the offset point (Figure 8). A coelliptic 

orbit keeps the chaser at an approximately 

constant altitude differential with the RSO. Since the 

altitude difference determines the rate of closure with 

the RSO, the time of arrival at the offset point may be 

controlled by varying the height of each co-elliptic, as 

well as the amount of time spent on each. Although a 

single co-elliptic could be used to control the arrival 

time, multiple co-elliptics are used to prevent long burn 

durations, as well as to cause the closure rate to diminish 

with relative range. The algorithm adds co-elliptic 

orbits until arrival time objectives and burn constraints 

are met. 


10 

0 

Low-thrust failure 15000 s prior to transition 

-10 


-20 

-30 

-40 

Co-elliptic orbits 

-50 

-60 

-70 

-80 

Low-thrust failure 

-90 

0 500 1000 1500 2000 2500 


Figure 8. Multiple co-elliptic trajectory. 

The chaser and RSO states are propagated using an 

Encke-Nystom integration scheme with J2, J3, and J4 

zonal gravitational coefficients. The Encke-Nystrom integration 

scheme takes into account the second-order 

nature of the equations of motion, requiring fewer function 

evaluations (only three for a fourth-order scheme). 

The advantage of the Encke-Nystrom integrators is that it 

allows large step sizes because it evaluates the trajectory 

along the Keplerian orbit and computes the perturbations 

from the Keplerian orbit. Thus, for low Earth orbit, 

step sizes on the order of 200 s can be taken while maintaining 

numerical precision. 

Burn Agent Planning and Monitoring 

In the process of generating the burn plan, the highthrust 

agent creates child burn agents of two types: 

height raising and co-elliptic. The height-raising agents 

are responsible for raising apogee to the height of the 

next co-elliptic, while the co-elliptic agents cause burns 

at a desired co-elliptic altitude to place the chaser at a 

constant altitude differential with the RSO. The final 

transfer to the Vbar is a 135-deg transfer to correct for 

out-of-plane errors that have built up during the final coelliptic 

orbit and to take advantage of improved 

navigation toward the end of the rendezvous sequence. 

The interface states between the burn agents are the 

states just prior to the next burn. The height-raising 

agent, for example plans to place the chaser at the 

desired altitude at the ignition time of the co-elliptic 

burn. Similarly, the co-elliptic agent plans to keep the 

chaser at the current altitude until the next height-raising 

burn. 

49


The type of the first burn in the sequence depends on the 

initial state. The high-thrust agent checks to see if the initial 

state is already close to a co-elliptic and, if so, 

schedules the first burn as a height-raising burn at some 

later time (Figure 8 is an example of this case). In this, the 

burn plan will include an even number of burns, since a 

pair of burns is required to transition from one co-elliptic 

to another. On the other hand, if the chaser is not initially 

in a co-elliptic orbit, an immediate burn will be scheduled 

to place it in one – resulting in an odd number of burns. 

When small navigation state updates occur, it is desirable 

to avoid replanning the entire burn sequence, so the active 

burn agent continuously monitors the error between the 

desired and predicted states at the next burn’s ignition 

time. If this error exceeds a threshold, the burn is 

replanned to achieve the desired altitude differential. Only 

if the burn magnitude exceeds the engine on-time constraint 

does the burn agent diagnoser deliver a failed status 

to the high-thrust agent, who initiates a replan of the 

entire burn schedule. 

PAMM Simulation 

The PAMM was implemented in a 2-body, 3-DOF 

Simulink-based simulation. Figure 9 shows the block 

diagram of the chaser vehicle in the simulation with 

ADEPT-based mission manager accepting state and status 

data from GN&C and subsystems and returning modes 

and commands. The ADEPT framework software is Clanguage 

code written in an object-oriented style to ease 

integration into real-time systems. The agent hierarchy 

and agent functions described above were implemented 

within the framework and interfaced to the simulation 

via a Simulink S-function. The ADEPT framework also 

supplies a developer interface (Figure 10) that displays 

current activities in green, future activities in blue, activities 

that are replanning in red, and deactivated, or 

“pruned” activities in grey. 

50 

Mission Objectives 

State 

and 

Status 

Telemetry_Chaser 

Telemetry_RSO 

Planet_Data 

Relative_State 

Telemetry_RSO 

PAMM 

GN&C 

Vehicle 

Models & 

Dynamics 

Mode 

and 

Commands 

Figure 9. PAMM chaser simulation. 

Telemetry_Chaser 

Figure 10. Agent hierarchy display. 

The GN&C block in the PAMM chaser vehicle included 

a dual-inertial state, nonlinear filter to provide realistic 

estimates of RSO and chaser inertial positions and velocities. 

The filter interfaced with the sensor models 

described in the CONOPS section. An RSO ground 

update model was provided to emulate the process 

whereby a ground station tracking the RSO would send 

the chaser an updated RSO state in order to better plan 

the rendezvous. The accuracy (covariance) of the RSO 

ground update was also uplinked to the chaser in order 

to allow the navigation filter to accurately process the 

measurement (Figure 11). 

4000 

2000 

0 

Chaser Nav Target Position Error (m) 

Target GPS Uplink 

-2000 

Relative GPS Acq 

-4000 

0 0.5 1 1.5 2 2.5 3 3.5 

4 

2 

0 

-2 

MissionAct 

MissionAct 

LowThrRend HighThrRend ProxOps Dock 

Burn Burn Burn Burn 

Chaser Nav Target Velocity Error (m/s) 

x10 4 

x10 4 

0 0.5 1 1.5 2 2.5 3 3.5 

x10 

Figure 11. Effects of target state uplink and relative GPS 

acquisition on target state error. 

The PAMM control laws were very simple as this was a 3- 

DOF implementation. Steering logic converted velocity 

change commands to thrust commands and sequencing 

logic timed the high-thrust burns. Low- and high-acceleration 

engines with representative thrust were included in 

the effector modeling. The dynamics included J2 gravity 

for both chaser and RSO bodies. 

The simulation also included graphical capability for visualization 

of the rendezvous geometry. 

4 

-4

Error (m) 

RESULTS 


The objective of this application was to demonstrate the 

mission manager response to two types of events: subsystem 

failures and navigation state updates. This section 

shows results from two example runs with low-thrust failures. 

Both cases also include replanning events from 

navigation state updates. For reference to the nominal 

case, see the CONOPS section. Some comments on the 

results of all the PAMM testing are included at the end of 

this section. 

Example 1: Early Failure of Low-Thrust System 

The intent of this example was to demonstrate the mission 

manager response to a low-thrust failure as well as a 

change in estimated target position. The plot in Figure 12 

is the in-plane relative position plot (orbital velocity points 

left) that summarizes the vehicle trajectory. The figure also 

shows the planner response to the failure. The top diagram 


10000 

Chaser Nav Target Position Error (m) 

0 0.5 1 1.5 2 2.5 3 3.5 4 

Chaser Nav Target Velocity Error (m/s) x104 -5000 

5 

Error (m/s) 

5000 

0 

0 

10 

0 

-10 

-20 

-30 

-40 

-50 

-60 

-70 

-80 


-90 0 500 1000 1500 2000 2500 


Figure 12. Low-thrust failure triggers replan event. 

x10 4 

Relative 

GPS Acq 

-5 

0 0.5 1 1.5 2 

Time (s) 

2.5 3 3.5 4 


Chaser position 


10 

0 

-10 

-20 

-30 

-40 

-50 

-60 

-70 

-80 

Relative 

GPS Acq 

Figure 13. GPS acquisition triggers replan event. 

shows the agent hierarchy at the moment that the mission 

agent’s diagnoser has triggered a replan event due to 

the low-thrust failure. The bottom diagram shows the 

state of the hierarchy immediately after the replan is 

complete. The blocks on the left (labeled MissionAct and 

LowThrRend) represent the agents that were executing 

just prior to the replan. These will be deactivated on the 

next pass of the PAMM. Since this is a complete failure of 

the low-thrust system, the mission agent spawns only a 

high-thrust child. Note that the high-thrust agent’s planner 

perceives the relative state at failure to be close to a 

co-elliptic trajectory. For this reason, no immediate burn 

is required. The final result is that four burns are planned 

to effect a double co-elliptic path to the desired offset 

point. 

Figure 13 (left plot) shows the chaser navigation estimate 

errors for the RSO inertial state. In this example, no 

MissionAct 

MissionAct 


Burn Burn 

-90 

0 500 1000 1500 2000 2500 


MissionAct 





MissionAct 

MissionAct 

HighThrRend ProxOps Dock 

Burn Burn Burn 

MissionAct 



51

ground updates are received prior to relative GPS acquisition, 

so the errors have grown to over 5 km when relative 

GPS is acquired. 

Throughout this period, the monitor function within the 

height-raising burn agent is continuously predicting the 

relative state just prior to the final burn at the target offset 

point. The navigation state update causes this prediction 

to exceed an allowable threshold, so the height-raising 

agent replans the burn to achieve the offset point objective 

(Figure 13, center plot). Since the objective is achieved 

successfully by the new burn parameters, no replanning is 

required at any higher levels. 

The right side of Figure 13 shows the planner state after 

the replan. All prior activities are pruned and the current 

activities are displayed in green. The formerly 

active burn agent is displayed in gray. This run completes 

with the chaser arriving at the offset point at the 

desired time. 

Example 2: Late Failure of Low-Thrust System 

Figure 14 summarizes this case, which shows the effect 

of a low-thrust failure close to the targeted low-thrust 

termination point. The failure occurs with significant 

vertical relative velocity so an odd number of burns were 

planned – in this case three. The first burn placed the 

chaser in a co-elliptic orbit (see the left plot of the figure). 

The burn was timed to control the relative height of 

the co-elliptic in order to control the arrival time. The 

right side of Figure 14 shows the mission manager’s 

response to the replan. Note that the original two-burn 

solution was replaced by three burn agents. 

52 


10 

0 

-10 

-20 

-30 

-40 

-50 

-60 

-70 

-80 







-90 0 500 1000 1500 2000 2500 

Relative GPS acquisition occurred just prior to the final 

height-raising burn (as in Figure 13) and, as in the previous 

example, the height-raising burn was replanned to 

arrive at the offset point at the required time. 

Additional Testing Results Comments 

The mission manager performance was tested for several 

dozen cases with low-thrust failures between 100 and 

2500 km behind the target on the final portion of the 

low-thrust rendezvous. In all these cases, the agent hierarchy 

was able to achieve a high-thrust burn solution to 

arrive at the offset point, although sometimes the time of 

arrival was not achievable. 

As shown in Figure 5, the nominal low-thrust trajectory 

“spiraled” outward toward the RSO, lapping it several 

times as the altitude difference was reduced. When lowthrust 

failures occurred outside the tested downrange 

window, the mission agent waited until the chaser was 

inside this window, and then activated the high-thrust 

agent. The success of the high-thrust replan depended on 

the phase angle at the time of failure and the total 

remaining altitude differential. When the high-thrust 

replan was not successful, the mission would have been 

aborted. No attempt was made to quantify the region of 

successful high-thrust replanning; this is left for future 

work. 

CONCLUSIONS 

Figure 14. Late low-thrust failure triggers replan event. 

A hierarchy of activity agents has been applied to solve a 

satellite rendezvous problem. The nominal rendezvous 

profile included a low-thrust and high-thrust phase. The 

MissionAct 

MissionAct 


Burn Burn 

MissionAct 


Burn Burn Burn


agent hierarchy was developed to handle failures of the 

low-thrust system, as well as updates to the chaser 

knowledge of the RSO state. 

The agent hierarchy was implemented within a software 

framework that instantiates the appropriate number and 

types of agents when planning or replanning occurred. 

The framework also provides handles to user-definable 

functions for the monitoring, diagnosis, planning, and 

execution the functions required of each agent. 

The agent hierarchy included low-thrust and high-thrust 

agents. The low-thrust agent utilized trajectory planning 

algorithms provided by Northrop Grumman Corp., 

which generated an optimal trajectory to the vicinity of 

the RSO. From there, a high-thrust agent planned a series 

of near-impulsive burns to arrive at a desired RSO offset 

point at a desired time. 

The PAMM was implemented in simulation (Figure 15) 

and tested for low-thrust failures and navigation state 

updates. The agent hierarchy was able to predict the consequences 

of failures and/or changes in navigation states, 

diagnose problems with the current plan, replan a new 

set of activities when mission objectives were not met, 

and execute the new set of activities to accomplish the 

rendezvous. 

Figure 15. Visualization of PAMM satellite arriving at offset 

point. 

When failures or state updates occurred within the region 

described above, rendezvous replanning was successful. 

Outside this region, success depended on the phase angle 

and altitude differential at the time of the failure. If planning 

was not successful, a mission abort was commanded. 

Future work should include more testing to determine the 

region of a successful rendezvous capability for a set of 

spacecraft parameters. Additionally, the capability to handle 

other system failures (such as sensor failures) should 

be added and tested within the PAMM simulation. 


Portions of this work were supported by the Northrop 

Grumman Agile Space Vehicle Research and Development 

program and by the Draper ADEPT development team led 

by Richard Hildebrant. The authors also wish to acknowledge 

Jennifer Hamelin, who led the PAMM project through 

the concept development phase. 

REFERENCES 

[1] Boyd, J.R., “The Essence of Winning and Losing,” Excerpts in 

presentation format at http://www.defense-and-society.org/fcs/ 

ppt/boyds_ooda_loop.ppt, January 1996. 

[2] Ricard, M. and S. Kolitz, “The ADEPT Framework for 

Intelligent Autonomy,” VKI Lecture Series on Intelligent Systems 

for Aeronautics, von Karman Institute, Brussels, Belgium, May 

2002. 

[3] Zimpfer, D., P. Spehar, F. Clark, C. D’Souza, M. Jackson, 

“Autonomous Rendezvous and Capture Guidance, Navigation, 

and Control,” 2005 Flight Mechanics Symposium, NASA 

Goddard Space Flight Center, October 2005. 

[4] Kechechian, J.A., “Trajectory Optimization Using Nonsingular 

Orbital Elements and True Longitude,” Journal of Guidance, 

Control, and Dynamics, Vol. 20, No. 5, September-October 

1997. 

[5] Kechechian, J.A., “Minimum-Time Constant Acceleration Orbit 

Transfer with First-Order Oblateness Effect,” Journal of 

Guidance, Control, and Dynamics, Vol. 23, No. 4, July-August 

2000. 

[6] Hu, H.C, T. Straube, J. Madsen, M. Ricard, “Shuttle Abort Flight 

Management (SAFM) – An Onboard Real-time Abort 

Determination and Assessment Application for Crew Safety and 

Awareness,” 2002 Core Technologies for Space Systems 

Conference, November 2002. 

[7] Jackson, M.C., T. Straube, T.J. Fill, S. Nemeth, “Onboard 

Determination of Shuttle Glide Capability for Shuttle Abort 

Flight Manager (SAFM),” 2002 Core Technologies for Space 

Systems Conference, November 2002. 

[8] Hu, H., R. Proud, J. Hart, “Spacecraft Mission Assessment and 

Replanning Tool (SMART) – a Real-Time, Intelligent 

Autonomous Flight Management (AFM) System for Increased 

Safety and Performance of Human Spaceflight Vehicles,” AIAA- 

2004-946, 42nd AIAA Aerospace Sciences Meeting and 

Exhibition, Reno, NV, January 5-8, 2004. 

[9] Abramson, M., M. Cleary, S. Kolitz, “Steps Toward Achieving 

Robust Autonomy,” Proceedings of the AAAI Spring Symposium, 

Stanford, CA, March 2001. 

53


[10] Adams, M.B., O.L. Deutsch, J.V. Harrison, “A Hierarchical 

Planner for Intelligent Systems,” Proceedings of SPIE - The 

International Society for Optical Engineering, Vol. 548, 

Arlington, VA, April 9-11, 1985. 

[11] Albus, J.S., R. Lumia, H. McCain, “Hierarchical Control of 

Intelligent Machines Applied to Space Station Telerobots,” IEEE 

Transactions on Aerospace and Electronic Systems, Vol. 24, No. 

5, September 1988. 

[12] Wade, M., “Kistler K-1,” article on the Web at http://www.astronautix.com/lvs/kislerk1.htm, 

2003. 

[13] Jackson, M. and R. McDonald, “Draper Simulation Analysis 

Tool (DSAT): Graphical Object Simulation Techniques and 

Tools for Simulink,” 2004 AIAA Conference on Simulation 

Tools and Techniques, August 2004. 

[14] Jackson, M. “Prototype Autonomous Mission Manager and 

Simulation,” The Charles Stark Draper Laboratory, Inc., July 25, 

2004. 

54


(l-r) Christopher D’Souza and Mark Jackson 

Mark Jackson is a Principal Member of the Technical Staff in the Mission Design and Analysis Group of 

the Algorithms and Software Directorate. He is currently located at the Draper field site office at the 

Johnson Space Center in Houston, TX. He holds a BS in Systems Engineering from the U.S. Naval 

Academy (1982) and an MS in Aerospace Engineering from MIT (1994). 

Christopher D’Souza is currently employed in the Autonomous Flight Systems Office of the Aeroscience 

and Flight Mechanics Division of the Johnson Space Center. He is responsible for orbital analysis (rendezvous, 

cislunar, and lunar operations) for the CEV. Prior to that, he was employed at Draper Laboratory 

from 1996 to 2005. His research areas and specialties are guidance and navigation systems for spacecraft 

and missiles as well as hybrid optimal control systems. Dr. D’Souza received BS and MS degrees from the 

University of Illinois (Urbana) and a PhD from the University of Texas in Austin, all in Aerospace 

Engineering. 

Hobson Lane is a Section Head within the Control Systems Department at Northrop Grumman Space 

Technology. At the time of the research described in this paper, he was the Principal Investigator of the 

Agile Space Vehicle Technology Development program at Northrop Grumman. More recently, he managed 

the Autonomous Walking Inspection and Maintenance Robot program under NASA’s exploration initiative, 

and he is currently the System Engineering Team Lead on a satellite laser communication system technology 

demonstration program. Mr. Lane received a double BS from Vanderbilt University in Physics and 

English (1993) and an MS degree from Georgia Tech (1996). 

55


Peter Y. Kwok, Marc S. Weinberg, Kenneth S. Breuer 

Copyright © 2005 by The Charles Stark Draper Laboratory, Inc. Printed with permission in Journal of Microelectromechanical Systems, 

Vol. 14, No. 4, August 2005, pp. 770 – 781 

INTRODUCTION 

Squeeze-film damping and hydrodynamic lift for a micromechanical perforated proof mass are 

calculated and measured. This work has resulted in closed-form expressions that can be used to design 

accelerometers, tuning-fork gyroscopes, and other micromechanical devices. The fluid damping and 

lift are determined using finite-element analyses of the normalized and linearized governing equations 

where the boundary condition of the pressure relief holes is derived using pipe flow analysis. The 

rarefaction of gas is incorporated in the governing equations based on slip flow condition. As a further 

check, a one-dimensional network model is developed to account for the boundary condition of the 

holes on a tilted proof mass. Both closed-form and numerical solutions are compared against experimental 

data over a range of pressures. 

Fluid damping can be an important consideration in various 

microelectromechanical system (MEMS) designs. 

Typical motions in a MEMS device involve oscillations of a 

proof mass in the horizontal or vertical directions, and the 

fluid (gas) flow between the proof mass and the substrate 

can be described by classic Couette flow or Poiseuille 

flow, [1] respectively. When the mean free path of the fluid is 

comparable to the characteristic length of the MEMS 

device, the gas does not behave entirely as a continuum, 

but rather exhibits some of the characteristics associated 

56 

with discrete particles. This phenomenon is known as rarefaction, 

and it often occurs when the MEMS unit is placed 

in a sealed package at very low pressure in order to minimize 

the fluid damping. As the mean free path of the fluid 

increases, the continuum assumptions of fluid begin to 

break down, and the flow transitions into a slip flow regime 

and ultimately, a free molecular flow regime. The structural-fluid 

interaction is further complicated when pressure 

relief holes are present in the proof mass, resulting in flow 

both beneath and through the moving structure.

A micromechanical tuning-fork gyroscope (TFG) is used 

to study the effect of rarefied gas on micromachined structures. 

It consists of two proof masses with evenly 

distributed square holes, as shown in Figure 1. The proof 

masses are attached to anchors through elastic beams, and 

separated from the substrate by a small gap. The design 

parameters are depicted in Figure 2. The proof masses 

oscillate in the x direction when voltage is supplied at the 

outer combs. Any angular rate applied in the y-axis will 

cause the proof masses to also oscillate in the z direction 

due to Coriolis acceleration. By measuring the differential 

capacitance of the proof masses based on the vertical displacements, 

the angular rate can be determined. 

Figure 1. Photograph of a micromechanical TFG with the 

coordinate system defined. 

y 

b 

z 

y 

x 

Elastic Beam 

L 

x gc 

Lc 

wc 

Ltc 

Inner Combs 

Proof Mass Combs 

Base Beam 

Figure 2. Illustration of the TFG design parameters. 

Proof mass oscillation is damped by the fluid surrounding 

the structure. This is known as squeeze-film damping for 

the vertical oscillation since the gas film is squeezed by the 

vibrating plate. When the proof mass is oscillating horizontally, 

shear forces act on its surfaces. If the proof mass 


Lp 

Lp 

Lh 

Anchor 

Outer Combs 

is unintentionally tilted during fabrication, hydrodynamic 

lift will be generated from the pressure built up underneath 

the proof mass as it moves horizontally. This is 

known as “surfboarding.” 

TFG performance is related to squeeze-film damping and 

possible hydrodynamic lift from surfboarding. High 

squeeze-film damping degrades gyro bias stability (lowfrequency 

variations of output with no actual input 

angular rate), and it also leads to high hydrodynamic lift, 

which is discussed in the “Lift from Surfboarding Effect” 

section. For some gyro mechanizations, good performance 

requires that the TFG response not lag the proof mass 

oscillation because of phase shifts induced by squeeze-film 

damping. In addition, the holes are important in determining 

bias stability since they typically reduce damping 

and hydrodynamic lift by about 100X (compared with no 

holes). Therefore, in order to improve the TFG performance, 

it is imperative to understand the detailed flow 

characteristics within the gyroscope unit. 

Squeeze-film motion and the surfboarding effect are common 

in many MEMS designs. Previous work [2]-[4] has been 

performed based on the Reynolds equation to analyze 

squeeze-film damping on a solid flat plate. The equation 

had been modified [5]-[7] to account for rarefaction using 

slip flow conditions. Similar approaches were also performed 

on flow analysis in small channels. [8],[9] References 

[10] and [11] are recent analyses of squeeze-film damping 

in perforated plates with finite thickness. The approach in 

Reference [10] is heavily numerically based and does not 

give a simple closed-form equation suitable for trade-off 

analyses for the squeeze damping. Reference [11] offers a 

closed-form approximation to the damping of a finite 

thickness, perforated plate. The derivation given in the 

“Closed-Form Approximation for Squeeze-Film Damping 

and Surfboarding Lift” section is a more direct, simpler 

alternative that is valid for most cases of practical interest. 

The derivation of Reference [11] is also discussed further 

in that section. Neither References [10] nor [11] consider 

rarefaction or hydrodynamic lift (surfboarding) effects. 

The hydrodynamic lift for a tilted solid flat plate has been 

evaluated based on Reynolds lubrication theory. [12] Since 

the pressure built up underneath the proof masses from 

squeeze-film motion and surfboarding is vented through 

the pressure relief holes in the proof masses, each hole 

serves as a “chimney” and generates additional damping 

resistance. 

The goal of this work is to develop easy-to-use, closedform 

solutions for squeeze-film damping and surfboarding 

lift for a MEMS design with holes on the proof mass. The 

numerical models of squeeze-film damping and surfboarding 

and the experimental results are used to validate the 

solutions. The derivation of closed-form solutions for 

squeeze-film damping and hydrodynamic lift with numerical 

and experimental verification at various pressure levels 

provides very simple and useful tools for trade-off studies 

57

for gyroscopes, oscillators, and other high mechanical 

quality factor devices. Our experimental data also offer 

valuable information that is not readily available in literature 

for comparison with various analytical models. 

This paper’s unique contributions include: 

1. A closed-form solution for the squeeze-film damping in 

a finite thickness perforated plate. This offers a simple 

alternative to the results of Reference [11]. 

2. The previous work on thin plates assumed a roughly 

hexagonal configuration of circular holes. This paper 

uses a pattern of square holes and spacing. 

3. The model is extended to rarefied gases and to hydrodynamic 

lift, situations of practical importance to 

gyroscope design. 

4. Test results. 

5. Lumped parameter and finite-element verifications of 

closed-form models. The lumped parameter model 

offers an alternative to the finite-element method 

(FEM) and does not include the assumption that all 

flow in a cell goes through the nearest hole. This 

assumption is embedded in the Reynolds equation 

approach. [11],[13] 

This paper begins with the derivation of the governing 

equation for squeeze-film damping with rarefaction 

included, followed by the development of the chimney 

boundary condition in the following section. The governing 

equation is normalized, linearized, and solved using 

FEM. The significance of the various design parameters on 

squeeze-film damping is also illustrated. 

The third section begins with evaluating the shear damping 

based on a flat proof mass that is parallel to the 

substrate. Next, the surfboarding analysis is carried out 

based on the slider bearing equation [12] with rarefaction 

included. The equation is then solved numerically to estimate 

the upper and lower bounds of the hydrodynamic 

lift. A one-dimensional network model is developed to 

include the chimney boundary condition. 

In fourth section, the closed-form solution for squeezefilm 

damping is derived with comb teeth considered. The 

closed-form estimation of the hydrodynamic lift from the 

squeeze-film damping for small proof mass tilts is also 

illustrated. 

Lastly, an experimental validation is presented and the test 

data are compared with the analytical and computational 

results. 

Fluid Under the Effect of Squeeze-Film Damping 

The pressure gradient of gas between the proof mass and 

the substrate is related to the velocity profiles in the x and 

y directions when the proof mass is undergoing squeezefilm 

motion. The flow in the x direction can be described 

by Navier-Stokes equation [1] 

58 


where 

p = pressure 

u = fluid velocity in the x direction 

µ = gas viscosity 

x,y,z = Cartesian coordinates defined in Figure 1. 

Equation (1) is valid under the assumption that (a) the 

gap height is always much smaller than the lateral extent 

of the proof mass plate; (b) the motion is sufficiently slow 

that the unsteadiness can be neglected; (c) the gas obeys 

the ideal gas law; and (d) the system is isothermal, with 

the density proportional to the pressure. Using a slip 

boundary condition [14] u = λ((∂u)/(∂z)) at z = 0 and 

u = –λ((∂u)/(∂z)) at z = h, where h is the gap height and 

λ is the mean free path of gas, the governing equation 

becomes 

or 

considering the flow in both x and y directions. [15] 

Equation (3) is the Reynolds equation with the rarefaction 

term, 6h 2λ. It is then normalized [16] according to the 

parameters listed in Table 1. 

(1) 

(2) 

(3) 

Table 1. Normalized Parameters for Squeeze-Film 

Equation. 

Parameter Symbol Normalization 

Normalized Pressure Ψ p/Pa 

Normalized Gap Height H h/ho 

Normalized Coordinate X, Y x/L, y/b 

Normalized Time T ωzτ 

where 

Pa = ambient pressure 

ho = initial gap height 

ωz = frequency of oscillation in z direction 

τ = time 

L,b = length and width of a proof mass 

In addition, Knudsen number (Kn) is defined as the ratio 

between λ and the characteristic length, h; and Kno is 

defined as the Knudsen number at the initial condition, 

i.e., λo/ho. The normalized governing equation becomes

where σ is the squeeze number, [3] (12µωzL2)/(Pah2 o). 

Equation (4) can be linearized using perturbation expansion. 

[2] Assuming the variation of gap height is small 

compared with its mean value and the variation in pressure 

is also small compared with the ambient pressure 

level, the normalized pressure and gap height can be 

expressed as 

(4) 

Ψ = 1 + εΨ^ (5) 

H = 1 + ε cos T (6) 

where ε = δ/ho and δ is the maximum vertical displacement 

of the proof mass. Note that Eq. (6) sets the proof 

mass at its highest position at T = 0, thereby begins 

squeezing initially. Substituting Eqs. (5) and (6) into Eq. 

(4), the equation of order ε becomes 

Then, assuming the solution is in the form [3] 

(7) 

Ψ^ = Ψ0 sin T + Ψ1 cos T (8) 

the time dependence can be eliminated 

(9) 

(10) 

One can consider the term, (σ/(1 + 6Kn)), to be an effective 

squeeze number with rarefaction included. 

Blech [3] has solved the coupled linearized squeeze film 

Eqs. (9) and (10) with Kn = 0 for a rectangular plate. The 

nondimensional damping (f0) and spring forces (f1) can be 

obtained through 

(11) 

(12) 

where A is the total area acted on by pressure. For a viscously-damped 

free vibration, [17] the damping 

coefficient, c, is determined by (f0 . Pa . A)/W, where W 

is the vertical velocity of the proof mass. The critical 

damping coefficient, Cc, is defined as 2mωn, where m is 

the mass and ωn is the natural frequency of the system. 

The quality factor, Qsqueeze, defined as Cc/2c, can be 

expressed as 


(13) 

As the plate moves down, the fluid underneath the proof 

mass is forced out through the pressure relief holes. This 

“chimney” flow can be modeled as pipe flow, [18] with the 

pipe length being the thickness of the proof mass. The 

pressure gradient (∇Ψ0 and ∇Ψ1) is zero at the line of 

symmetry between the holes. The schematic of the chimney 

flow is shown in Figure 3. A circular cross section is 

assumed for simplicity. The entrance length of the flow is 

about 25% of the total chimney length (using a Reynolds 

number of order of 10 -1), and thus, the entrance length 

effect can be neglected as a first approximation. The governing 

equation of the fluid through the chimney is 

therefore 

(14) 

where r is the radial coordinate and R is the radius of the 

chimney. 

z 

Symmetry 

of Flow 

x 

t 

Ambient 

Pressure, Pa 

Proof Mass 

Figure 3. Schematic of “chimney” flow. 

Solving for the velocity profile with slip boundary condition 

u = –λ((∂u)/(∂r)) at r = R and (∂u)/(∂r) = 0 at r = 0, 

and integrating over the cross section to obtain the flow 

rate, the average flow velocity is then determined by dividing 

the flow rate by the cross sectional area, πR2 (15) 

From energy conservation, [18] the pressure drop along the 

chimney is balanced by the friction loss from the flow; also 

substituting Lh as the hydraulic diameter of a square cross 

section, [19] the pressure drop is given by 

(16) 

Again, after normalizing the pressure based on Table 1 and 

using the perturbation analysis, with Ψ = 1 + εΨ^, the 

equation for the pressure becomes 

Lp 

Lh 

h P 

w = dh 

dτ 

59

(17) 

Since the squeeze number is low, the flow can be assumed 

to be incompressible, and Vchimney can be represented 

based on conservation of mass (volume) by 

(18) 

where ωzδ is the proof mass velocity W. Substituting Eq. 

(18) into Eq. (17) and rearranging the terms, the normalized 

pressure at the chimney boundary condition can be 

derived as 

or 

(19) 

(20) 

Equation (20) shows that the boundary condition of the 

hole can be represented by the product of the squeeze 

number, a geometric term, and a rarefaction term. The 

ratios h/Lp and Lh/Lp are particularly important due to the 

exponential dependences on their value. A contour plot of 

log(Kgeometry) with various h/Lp and Lh/Lp ratios is shown 

in Figure 4. 

Figure 4. Contour plot of log(Kgeometry) with various h/Lp 

and Lh/Lp ratios. 

60 

Lh/Lp 

Log10(Kgeometry) 

0.9 

0.8 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 

h/Lp 


4 

3 

2 

1 

0 

-1 

-2 

-3 

With the chimney boundary condition defined, a numerical 

simulation is performed using PDEase, [20] a 

finite-element solver capable of solving field equations in 

an arbitrary two-dimensional domain. As mentioned previously, 

the geometric symmetry allows the problem to 

be defined only on a square “cell” domain with a pressure 

relief hole at the center. The total pressure can be 

obtained by multiplying the solution to the number of 

cells in the proof mass. Figure 5 illustrates the problem 

definition with the appropriate boundary conditions. 

Based on the dimensions of the design parameters for the 

TFG tabulated in Table 2, Kgeometry is 1.4 with h/Lp = 0.3 

and Lh/Lp = 0.45, and Krarefaction is 5.5e-5 with Pa = 5 

mTorr. 

∇Ψ^ = 0 

Kgeometry × Krarefaction × σ 

∇Ψ^ = 0 

∇Ψ^ = 0 

∇Ψ^ = 0 

Figure 5. Problem definition for squeeze-film damping 

with rarefaction effect. 

Table 2. Dimensions of the Design Parameters of the 

Selected TFG. 

Parameter Dimension (µm) 

Proof Mass Length (L) 450 

Proof Mass Width (b) 400 

Proof Mass Thickness (t) 10 

Gap Height (h) 3 

Hole Width (Lh) 4.5 

Hole Spacing (Lp) 10 

Combs Teeth Length (Ltc) 50 

Combs Overlap Length (Lc) 25 

Combs Teeth Width (wc) 2 

Combs Gap (gc) 3 

A solution contour of Ψ0 from the numerical simulation is 

shown in Figure 6. Since the simulation is only carried out 

on a single cell, the length in the squeeze number calculation 

would be using Lp (10 µm). The effective squeeze 

number with rarefaction included is 0.0304, based on Pa = 

5 mTorr, µ = 1.85e-5 N-s/m 2 at 300 K, ωz = 27 kHz × 2π, 

and Kn = 3000.

y 

1 

0.8 

0.6 

0.4 

0.2 

0 

0 0.2 0.4 0.6 0.8 1 

x 

Figure 6. Solution contour of nondimensional damping 

pressure (Ψ0) with σ = 0.0304. 

Lift from Surfboarding Effect 

The proof masses will oscillate in the x direction when 

excited at their horizontal vibration mode frequency. Shear 

will be exerted by the fluid between the stationary substrate 

and the moving proof mass. The fluid flow between 

the proof mass and the substrate can be described by 

Couette flow, [1],[21] where the pressure is constant along 

the x direction. The shear damping force, fshear, acting on 

the proof mass under horizontal oscillation is 

(21) 

where U is the horizontal velocity of the proof mass (1.57 

m/s for the selected TFG). 

In addition, the oscillation motion also induces Couette 

flow in between the comb teeth. The average shearing area 

Acomb = Lct is used for the calculation. Neglecting the 

squeezing of the fluid by the tip of each comb at the end, 

the resulting shear is similar to Eq. (21). Hence, the total 

quality factor becomes 

(22) 

where nc is the number of comb teeth (nc = 80). 

It can be seen from Eq. (22) that as pressure becomes really 

low, in which λ >> h and λ >> gc, the quality factor can 

be approximated [9] by 

(23) 

The quality factor becomes independent of the gap 

heights; hence, the shear stress on the top surface should 

also be included. 

The pressure is constant along the x and y directions when 

the gap height is uniform across the proof mass. However, 

the pressure underneath the proof mass will build up and 


generates lift if the proof mass is tilted at an angle, and 

such flow can be modeled by a stationary tilted plate with 

the substrate moving horizontally at velocity U, as depicted 

in Figure 7. 

z 

x 

Figure 7. Fluid flow model for surfboarding analysis. 

The governing equations for the fluid flow are the same as 

Eq. (1), in both x and y directions. Proceeding with slip 

boundary conditions: [5] u = U + λ((∂u)/(∂z)) and v = 

λ((∂v)/(∂z)) at z = 0 and u = –λ((∂u)/(∂z)) and v = 

–λ((∂v)/(∂z)) at z = h, and assuming the density is proportional 

to the pressure for an isothermal flow, the equation 

of motion becomes 

(24) 

This is simply the Reynolds equation for slider bearing 

with rarefaction effect. [22] The equation is then normalized 

using the same nondimensional parameters from 

Table 1 

X 

h1 

H 3 

Y H3 

X + 6H2 Kn X + 

Y + 6H2 Kn Y = 

(25) 

where Λ is the bearing number, (6µLU)/(Pah 2). The gap 

height is defined as 

and, therefore, the normalized H is 

( H) 

X 

(26) 

(27) 

where h1 and h2 are the gap heights at higher and lower 

end, respectively, and ε = (h1 – h2)/(h1). 

After substituting in the perturbation expansion using Ψ = 

1 + εΨ^ and neglecting higher-order terms, the normalized 

equation of order ε becomes 

L 

U 

h2 

(28) 

61

Equation (28) can be solved over the proof mass, however, 

the boundary conditions of the pressure relief holes 

are not uniform due to the varying gap heights, hence, 

the “cell simulation” approach may not provide the most 

accurate result. At the same time, evaluating the boundary 

condition on each hole and simulating the entire 

proof mass surface (1800 holes) may be impractical. As a 

result, FEM is used to provide the upper and lower 

bounds for the hydrodynamic lift, based on the results 

from proof mass with no holes and perfectly vented 

holes, respectively. The problem domain and definition 

are similar to those in Figure 5, except with governing 

Eq. (28). The upper bound estimate assumes a flat proof 

mass with no hole for pressure relief; and the lower 

bound estimate of hydrodynamic lift is determined with 

ambient pressure at the hole boundaries. The results are 

shown later in Figure 14. 

In addition, a one-dimensional network model is developed 

using Matlab [23] to account for the chimney effect 

and the varying gap heights. The network model consists 

of pressure nodes underneath the proof mass and the 

holes. Each pressure node is connected by “pressure resistances,” 

as shown in Figure 8. 

Figure 8. One-dimensional flow network for surfboarding 

analysis. 

For n holes, there are 2n+1 pressure nodes (Pi), n chimney 

resistances (Rc), and 2n+2 proof mass resistances 

62 

Pa 

Pa 

Lp 

Lh 

Rc Rc 

Pa 

P1 P 2 P3 P 4 

P5 P6 Pi-1 Pi Pa 

R1 R2 R3 R4 R5 R6 Ri Ri+1 

l 


(Ri). l is defined as Lp – Lh. For tutorial purposes, the 

resistances are derived using a single “block” from the first 

column (width = Lp). From Kirchoff’s law [24] 

(29) 

with Pa set to zero. The driving flow rates, Q1 and Q2, are 

Q1 = (Lph1U)/(2) and Q2 = (Lph2U)/(2), respectively. 

Solving for Ri 

(30) 

with the modified viscosity to account for the rarefaction 

effect, [4] i.e. 

(31) 

The pressure resistance of chimney can be derived similarly 

as Eq. (16) 

(32) 

A set of equations can therefore be written based on conservation 

of mass 

(33) 

for n is odd (pressure node under the proof mass), and 

(34) 

for n is even (pressure node under the hole). Arranging the 

2n+1 equations into a matrix form, we have Eq. (35). 

(35)

Solving for the pressure nodes simultaneously, the hydrodynamic 

lift is obtained by integrating the pressure 

distribution over the area of the proof mass and multiplying 

the number of columns. 

To verify the network model, Lh is set to 0.1 µm and t is 

set to 1020 m, the effect of pressure relief holes is therefore 

minimized by the large chimney resistance. The network 

solution is then compared against the one-dimensional flat 

plate solution [25] 

(36) 

where h = h(x) as defined in Eq. (26). 

The comparison of the pressure distributions of a 450-µm 

flat plate is shown in Figure 9, with Kn = 17.5, and ε = 

0.0645 (h1 – h2 = 0.2 µm). The network solution with infinitely 

long (1020 m) and small (0.1 µm) chimney agrees 

well with the flat plate solution. 

Pressure (Pa) 

0.9 

0.8 

0.7 

0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

0 

0 100 200 300 400 500 

X (µm) 

Figure 9. Comparison of pressure distributions between 

analytical solution and network model with “infinite” 

chimney resistance. 

Closed-Form Approximation for Squeeze-Film 

Damping and Surfboarding Lift 

In this section, a closed-form solution for estimating the 

squeeze-film damping is derived. The rarefaction effect is 

embedded in the air viscosity term, µ. Consider the shaded 

trapezoid in Figure 10, which represents the fluid 

flow in a single cell. The flow passing through line x is 

given by 


Analytical Solution for Flat Plate 

Network Model Solution 

(37) 

Symmetry Lines 

Figure 10. Illustration of fluid flow in a single cell under 

squeeze-film motion. 

With a cross section of h by x, the pressure gradient along 

the flow direction is 

Hence, the pressure at position x is 

(38) 

(39) 

The pressure drop from chimney can be approximated by 

fully developed circular pipe flow with Lh as the hydraulic 

diameter, [18] i.e. 

(40) 

The vertical damping force for the entire cell (8 trapezoids) 

is obtained by integrating the pressure under the proof mass 

(41) 

and the squeeze-film damping coefficient from the proof 

mass is therefore 

where η = Lp/Lh, nh = number of holes (nh = 1800). 

y 

Hole 

x 

Lh/2 

Lp/2 

x 

(42) 

63

However, when the damping calculated with chimneys 

exceeds that calculated from flat plate, the flat plate solution 

[3] should be used. In this limit, fluid is squeezing out 

from the sides rather than through the chimneys. 

The damping contributed by the combs is also estimated. 

Each comb tooth is assumed to be infinitely long so that 

the flow is in the short direction across the comb tooth. 

Using the overlap comb bottom surface area, Ltcwc/2, the 

flow in one gap is given by 

(43) 

The last term in Eq. (42) is the increase in damping 

caused by the finite length of the holes. Using the formula 

Eq. (38) for fully developed flow in a wide rectangular 

channel – the gap between the fixed and moving comb 

teeth – the pressure at the corner of the comb tooth is 

determined by 

(44) 

Assuming that Ltc >> wc, the damping coefficient from the 

comb teeth is 

(45) 

The first term in Eq. (45) represents the squeeze film 

beneath the comb teeth, and the second term represents 

squeezing the flow through the gaps between moving and 

fixed teeth. This gap occurs along only one half the tooth 

length. The total damping coefficient, cz, is therefore the 

sum of Eqs. (42) and (45). 

Next, the hydrodynamic lift is estimated based on the similarity 

between the squeeze film and surfboarding 

equations. To begin, incompressible gas is assumed so that 

density drops from the formulation, thus, the Reynolds 

equations are simplified and the ambient pressure for the 

boundary conditions can be set to zero. Since typical 

solved pressures are within 0.1% of the ambient pressure, 

the constant density assumption is acceptable. 

When the constant density assumption is applied, the 

pressure equation with both vertical motion and surfboarding 

for low Reynolds number laminar flow is 

64 


(46) 

where U, V, and W are the proof mass velocities in x, y, and 

z directions, respectively. 

For a flat plate, the derivatives with respect to x and y are constants. 

For this study, ∂h/∂y = 0. When expanding to 

first-order terms, the parentheses on the left side become zero 

so that Eq. (24) results with the rarefaction condition added. 

The right-hand side represents constant driving terms. 

For the linear analysis, the pressures from squeeze-film 

damping, the response to W, and from surfboarding, 

response to U, are proportional. This result is important 

since one can rather easily envision the squeeze-damping 

process and the assumption that for a large plate, the 

flow from each cell (hole plus surrounding land) can be 

calculated individually. 

Because the equation for pressure, Eq. (46), is linear, the 

hydrodynamic lift can be calculated directly from the 

fluid-caused sense axis damping by 

(47) 

Applying this equation on the case from Table 2 and 

Figure 6 where P = 5 mTorr, µ = 1.85e-5 N-s/m 2 at 300 

K, ωz = 27 kHz × 2π, Kn = 3000, U = 1.57 m/s, and 

assuming the tilt, h1 – h2, is 0.2 µm, the hydrodynamic 

lift per hole determined from the network model is 

2.56e-12 N, and the closed-form solution is 2.31e-12 

N. 

In Reference [11], the Reynolds equation for flat plates is 

modified by including a term for the vertical flow per unit 

area. This term is developed by considering the flow in a 

closed, round cell with a vertical cylinder. Where 

Reference [11] continues onto the modified Reynolds 

equations, the closed-form approach herein uses the 

results of a square cell directly; thus, the mathematical 

derivation is simplified considerably. 

Experimentation and Result Comparison 

Experiments are performed on TFG samples with the 

dimensions listed in Table 2. The fluid damping from 

squeeze-film motion is obtained and compared with the 

analytical results in terms of quality factor. The lift force 

from surfboarding is also determined and compared 

with the numerical and closed-form predictions. 

A TFG sample unit is placed inside a bell jar with controlled 

pressure. The temperature of the TFG unit is 

monitored by a thermistor. The proof masses begin to 

oscillate vertically with excitation at their vertical resonant 

frequency. The decay response of the TFG is 

recorded in an oscilloscope after switching off the 

power supplied. A sample decaying signal is shown in 

Figure 11.

Output Voltage (V) 

1 

0.8 

0.6 

0.4 

0.2 

0 

-0.2 

-0.4 

-0.6 

Figure 11. A decaying signal recorded at pressure = 200 

mTorr and temperature = 305 K. 

Each response has approximately 450 data points, and its 

envelope exhibits exponential decay in accordance to free 

vibration with viscous damping. The upper envelope of 

the decaying signal is extrapolated and the damping ratio 

is determined by curve fitting. The noise floor of the signal 

is eliminated to ensure more accurate curve fitting. The 

measurement and curve fitting procedure is carried out for 

a range of pressures at room temperature. 

Five sample units are tested at pressures ranging from 1 

mTorr to 1 Torr. The Knudsen number is determined 

based on the pressure and temperature recorded. The data 

show that as the pressure decreases, the fluid damping 

decreases and structural damping becomes dominant, and 

the quality factor eventually becomes the structural quality 

factor. The structural damping can be determined by 

curve-fitting the data. [15] -0.8 

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 

Time (s) 

The fluid damping can be isolated 

from the total damping using 

(48) 

The quality factor is isolated from structural damping 

using Eq. (48), and a selected sample is plotted against the 

numerical results from finite-element analysis for a range 

of Knudsen number in Figure 12. It can be seen that the 

quality factor from fluid damping follows a linear trend 

with pressure. The quality factor measurements show reasonable 

agreement with the prediction from the 

finite-element and closed-form solutions. The average discrepancy 

between the measured data and the 

finite-element solution is 59%, and 39% for the closedform 

results. This is considered to be good agreement, 

since the solution spans several orders of magnitude. In 

fact, many calculations [26],[27] have shown a factor of four 

higher in predicting squeeze-film damping in other MEMS 

devices. These calculations neglected the pressure buildup 

at the holes, which further illustrates the significance of 

chimney resistance. The continuum flow model is also 


1.00E+05 

1.00E+04 

1.00E+03 

Qsqueeze 1.00E+06 

1.00E+02 

1.00E+01 

Test Data 

Closed Form 

FEM 

Continuum 

1.00E+00 

0.01 0.1 1 10 100 1000 10000 100000 

Knudsen Number 

Figure 12. Comparison of measured quality factors with 

continuum and slip flow predictions for 

squeeze-film damping. 

plotted to illustrate the difference due to rarefaction. Note 

that the solutions merge as the Knudsen number 

approaches continuum regime. In addition, the comb 

teeth only contribute about 0.03% to the overall quality 

factor calculation, hence, not a significant source for 

squeeze-film damping. 

The experimental process is also carried out for horizontal 

oscillation. The decay response of the TFG is again 

recorded in an oscilloscope when the power supplied is 

switched off. The same post-processing method is used 

to evaluate the quality factor in the x-axis. Four sample 

TFG units are tested at pressures ranging from 1 mTorr 

to 1 Torr. The fluidic quality factor is isolated from the 

structural quality factor using Eq. (48). 

For the Knudsen number range considered, the shear 

areas A and Acombs should be doubled for the quality factor 

calculations as discussed previously. For illustration 

purposes, the quality factor curves from single-surface 

and double-surface calculations are shown in the log 

Qshear vs. log Kn plot in Figure 13. In single-surface calculation, 

only the bottom surface of the proof mass is 

used to calculate quality factor, whereas in double-surface 

calculation, both top and bottom surfaces of the 

proof mass are used as the effective area. As mentioned 

in Eq. (23), the shear on the top surface of the proof mass 

should also be included when the mean free path is 

much larger than the gap height, and the quality factor 

becomes independent of the gap heights. 

The quality factor measurements show reasonable agreement 

with the prediction from slip flow model. The 

average discrepancies between the data and analytical 

results are 58% and 29%, for single and double area calculations, 

respectively. As with the squeeze film results, 

we consider the agreement here to be excellent, considering 

the fact that the solution spans several orders of 

magnitude. Based on the design parameters of the selected 

TFG, the contribution to the total quality factor from the 

65

combs is about 39%. Thus, the combs shearing can be an 

important source of damping in horizontal oscillation. 

Again, the continuum flow model is also plotted to illustrate 

the difference due to rarefaction. 

For surfboarding, the experimental setup combines the 

horizontal and vertical oscillations. The TFG unit is 

placed inside the bell jar and oscillated at resonant horizontal 

frequency; at the same time, the vertical output 

signal is monitored. With no angular rate input, the lift 

force is determined from the sensor output, which 

includes the motor and sense coupling forces that are 

also in-phase with the bias readout. These other in-phase 

biases included in the output can cause the measured lift 

to be 2 to 4 times higher than the actual lift generated 

from the surfboarding effect. Unlike squeeze film and 

shear damping measurements, each TFG unit may have 

different tilt orientation, and thus, only a sample TFG 

unit with prominent bias output is tested and the measurement 

is plotted against the typical bias and tilt range 

observed. This TFG has the same dimensions as those 

listed in Table 2, except for the proof mass thickness, 

which is 8 µm instead of 10 µm. 

Figure 14 shows the comparison of surfboarding lift evaluated 

from measurements and the various solutions over 

a range of pressures (Kn). The typical bias of this TFG 

design measured at the sealed pressures of the production 

units ranges between 0.01 to 0.05 deg/s/mTorr. The 

typical hydrodynamic lift over the range of pressures 

(Kn) is then estimated by inverse pressure scaling, represented 

by the shaded region. As seen in the figure, 

measurements at very low pressure can be difficult as the 

instruments are approaching their resolution limits, thus, 

the data do not remain a constant slope at high Kn. 

Knowing that the typical differential tilt of the proof 

masses for this TFG design is about 0.2 to 0.3 µm, the 

measured data that are mostly above the shaded region 

indicate that this particular tested sample should have a 

tilt about 0.4 µm. 

66 

1.00E+08 

1.00E+07 

1.00E+06 

1.00E+05 

1.00E+04 

Quality Factor in X-Axis 1.00E+09 

1.00E+03 

Test Data 

Slip Flow (bottom surface) 

Slip Flow (both surfaces) 

Continuum Flow 

1.00E+02 

0.01 0.1 1 10 100 1000 10000 100000 1000000 


Figure 13. Comparison of measured quality factors with 

continuum and slip flow predictions for horizontal 

shear damping. 


Hydrodynamic Lift (N) 

1.00E+06 

1.00E+07 

1.00E+08 

1.00E+09 

1.00E+10 

1.00E+11 

1.00E+12 

1.00E+13 

Test Data 

Upper Bound 

Lower Bound 

Closed Form 

Network Model 

1.00E+14 

10 100 1000 


10000 100000 

Figure 14. Comparison of measured surfboarding lift with 

various solutions over a range of pressures with 

h1 – h2 = 0.4 µm. 

To justify the approximation of the tilt, the upper and 

lower bounds are established based on the PDEase solutions 

with ε = 0.1333, or a gap height difference of 0.4 µm. 

The one-dimensional network model solution and the 

closed-form solution are also plotted based on the same 

tilt. The results show that the lift calculated from upper and 

lower bounds differs by 3 orders of magnitude, illustrating 

the significance of the chimney resistance. As mentioned, 

the lift determined from the bias readout also contains 

other in-phase bias elements, such as motor coupling force. 

The closed-form and network solutions show nice agreement, 

and they are within a factor of 2 after considering 

the influence of the motor coupling force. With the uncertainty 

in the measurement and in build parameters, the 

models describe the phenomena. 

Comparison with Molecular Flow Model 

It is important to note that differences in the velocity profiles 

can be found between our first-order slip model and 

other molecular flow models in the transitional and free 

molecular flow regimes, even for simple two parallel 

plates. [28] One way to obtain a better understanding is by 

performing a molecular dynamic simulation on a simple 

case with two parallel plates and comparing the results. In 

fact, for the two parallel-plate case, a continuum equation 

similar to Navier-Stokes can be formed that is applicable 

for the whole Knudsen regime, however, with modified 

viscosity and boundary models. 

Another way is to compare the shear stress per unit velocity 

for two parallel plates. This allows our model to be 

compared directly to a molecular flow model without a 

large effort. The comparison also addresses the issue that, 

although the velocity distributions may differ between the 

slip flow and molecular models, the agreement of the 

shear stresses in the molecular flow region is suitable for 

many engineering calculations. 

Recall Eq. (21), the shear stress per unit velocity for parallel 

plates is given by

(49) 

where µ is 1.9e-5 N-s/m2 for air at 310 K, λ is 5.3e-3 m at 

310 K and 10 mTorr, and h is 3e-6 m and omitted for large 

mean free path. 

Taken from Eq. (7) of Reference [28], which referred to 

Reference [29], the shear stress per relative plate velocity 

in the free molecular flow regime is given by 

(50) 

where ρ is 1.45 × 10-5 kg/m3 for air at 310 K and 10 

mTorr, M is 28 gm/gm-mole, and Ru is 8.31 N-m/gmmole-K. 

The various models for the shear stress cited in 

Eqs. (8)-(11) of Reference [28] all converge to Eq. (50) for 

high Knudsen numbers. 

In the limit of large mean free path, both Eqs. (49) and 

(50) are proportional to pressure. If one assumes that viscosity 

is proportional to the square root of temperature, 

both Eqs. (49) and (50) are inversely proportional to the 

square root of temperature. For simple slip flow, Eq. (49), 

the stress per relative velocity is 0.0018 N-s/m3; for the 

idealized molecular flow, Eq. (50), the stress per relative 

velocity is 0.0035 N-s/m3, based on the parameters of air 

at T = 310 and P = mTorr. Calculated by greatly different 

techniques and noting that the velocity distributions 

differ, [28] the two shear stress results differ by less than a 

factor of two. This calculation corroborates our experimental 

result that the slip flow model gives usable 

engineering estimates for damping over a wide pressure 

range, even into the molecular flow region. 

SUMMARY AND DISCUSSION 

Closed-form solutions for squeeze-film damping and surfboarding 

effects on a micromechanical TFG have been 

developed with rarefaction and perforation considerations. 

The solutions are in good agreement with numerical solutions 

and experimental data. The methodology can be 

applied to other micromachined structures such as 

accelerometers. 

Sample TFG units are tested for the squeeze-film damping 

and horizontal shear damping. The solutions show an 

average discrepancy within 40%, even though the flow 

belongs to a free molecular flow regime based on the 

Knudsen number. This is a very reasonable approximation 

considering that the range of quality factors covers few 

orders of magnitude. 

It is found that for the pressure range considered (1 Torr 

to 1 mTorr), the rarefaction effect could contribute up to 

four orders of magnitude increase in quality factor over the 

solution obtained with the continuum assumption. This is 

the same for shear damping from horizontal oscillation. In 

the case of TFG with an operating pressure ranging from 1 


to 50 mTorr and a gap height of 3 µm, the Knudsen number 

is on the order of 10 3. Although the Knudsen number 

value indicates free molecular flow – far beyond the 

regime where this first-order slip flow model is valid, the 

squeeze-film damping and surfboarding agrees quite well 

with experiment. The agreement between theory and 

measurement over such a wide range of Kn is surprising 

and far better than one would expect. One must be cautious, 

and we emphasize that this does not imply that 

first-order rarefaction theory is uniformly valid at such low 

pressures. However, for this particular flow geometry, the 

agreement can be explained to some extent. If one 

expands the equations of motion to include higher-order 

corrections for rarefaction (e.g., the Burnett equations), 

one finds that the additional terms drop out for simple 

geometries such as parallel (or nearly parallel) plates. For 

this reason, the approximation used here is formally accurate 

at least to second order and perhaps even further. We 

also note that the better-than-expected agreement between 

theory and experiment at high Knudsen numbers has also 

been found in other related investigations. [13],[30],[31] 

Finally, a TFG unit with prominent bias was tested for 

surfboarding lift. The test results represent a bias roughly 

four times the contribution from the real surfboarding lift 

because of other in-phase effects such as motor coupling, 

and show that the amount of tilt is between 3 to 5 µm over 

the 450-µm proof mass length. Hydrodynamic lift from 

the numerical models and the closed-form solution are 

evaluated based on a 4-µm tilt. The closed-form and 

lumped parameter network solutions are able to predict 

the real surfboarding within a factor of 2. 

In summary, the closed-form solutions for squeeze-film 

damping and surfboarding lift agree with more complicated 

FEM and lumped parameter models and with measured data. 

ACKNOWLEDGMENT 

The authors would like to thank Mr. Greg Kirkos and Mr. 

Eli Weinberg for their contribution to the finite-element 

model and network flow model, and Mr. Richard Elliot for 

assisting with the experimental setup. 

REFERENCES 

[1] Fay, J.A., Introduction to Fluid Mechanics, MIT Press, 1994. 

[2] Griffin, W.S., H.H. Richardson, S. Yamanami, “A Study of Fluid 

Squeeze-Film Damping,” Journal of Basic Engineering, June 

1966, pp. 451-456. 

[3] Blech, J.J., “On Isothermal Squeeze Films,” Journal of 

Lubrication Technology, Vol. 105, October 1983, pp. 615-620. 

[4] Veijola, T., H. Kuisma, et al., “Equivalent-Circuit Model of the 

Squeeze Gas Film in a Silicon Accelerometer,” Sensors and 

Actuators A, Vol. A48, 1995, pp. 239-248. 

[5] Burgdorfer, A., “The Influence of the Molecular Mean Free Path 

on the Performance of Hydrodynamics Gas Lubricated Bearing,” 

Journal of Basic Engineering, Vol. 81, 1959, pp. 94-99. 

[6] Schaaf, S.A. and F.S. Sherman, “Skin Friction in Slip Flow,” 

Journal of Aeronautical Sciences, Vol. 21, No. 2, 1953, pp. 85-90. 

67

[7] Rohsenow, W. and H. Choi, Heat, Mass, and Momentum 

Transfer, Prentice-Hall, 1961. 

[8] Arkillic, E. and K.S. Breuer, “Gaseous Slip Flow in Long 

Microchannels,” J. Microelectromechanical Systems, Vol. 6, No. 

2, 1997, pp. 167-178. 

[9] Cho, Y-H., B.M. Kwak, A.P. Pisano, R.T. Howe, “Slide Film 

Damping in Laterally Driven Microstructures,” Presented at 

MEMS’99, Orlando, Florida, January 1999. 

[10] Yang, Y-J and C-J Yu, “Macromodel Extraction of Gas Damping 

Effects for Perforated Surfaces with Arbitrarily-Shaped 

Geometries,” Proc. MSM 2002, pp. 178-181. 

[11] Bao, M., H. Yang, et al., “Modified Reynolds Equation and 

Analytical Analysis of Squeeze-Film Air Damping of Perforated 

Structures,” Journal of Micromech. Microeng., Vol. 13, No. 6, 2003. 

[12] Gross, W.A., Gas Film Lubrication, John Wiley and Sons, 1962. 

[13] Veijola, T. and T. Mattila, “Compact Squeezed-Film Damping 

Model for Perforated Surface,” Proceedings of Transducers ‘01, 

Munchen, Germany, June 2001, pp. 1506-1509. 

[14] Arkilic, E.B., M.A. Schmidt, K.S. Breuer, “Gaseous Slip Flow in 

Long Microchannels,” Journal of Microelectromechanical 

Systems, Vol. 6, No. 2, June 1997. 

[15] Kwok, P., Fluid Effects in Vibrating Micromachined Structure, 

MS Thesis, Mechanical Engineering, MIT, Cambridge, MA, 

1999. 

[16] Langlois, W.E., “Isothermal Squeeze Films,” Quarterly Applied 

Mathematics, Vol. XX, No. 2, 1962, pp. 131-150. 

[17] Nashif, A.D., D.I. Jones, J.P. Henderson, Vibration Damping, 

John Wiley & Sons, 1985. 

[18] Munson, B.R., D.F. Young, and T.H. Okiishi, Fundamentals of 

Fluid Mechanics, John Wiley & Sons, 1994. 

[19] Potter, M.C. and D.C. Wiggert, Mechanics of Fluids, Prentice- 

Hall, Inc., 1991. 

[20] Macsyma Inc., PDEase2D Reference Manual, 3rd Edition, 

November 1996. 

68 


[21] Ye, W., X. Wang, et al., “Air Damping in Laterally Oscillating 

Microresonators: A Numerical and Experimental Study,” 

JMEMS, Vol.12, No. 5, 2003, pp. 557-566. 

[22] Cercignani, C. and A. Daneri, “Flow of a Rarefied Gas Between 

Two Parallel Plates,” Journal of Applied Physics, Vol. 34, No. 12, 

December 1963, pp. 3509-3513. 

[23] MathWorks Inc., Matlab® Function Reference, Vol. 1-3, Version 

5, October 1997. 

[24] Standard Handbook for Mechanical Engineers, 7th Edition, 

Baumeister, T. and L.S. Marks, Editors, McGraw-Hill, New 

York, NY, 1967. 

[25] Sherman, F.S., Viscous Flow, McGraw-Hill Publishing 

Company, 1990. 

[26] Kim, E., Y. Cho, M. Kim, “Effect of Holes and Edges on the 

Squeeze-Film Damping of Perforated Micromechanical 

Structures,” MEMS’99, Orlando, FL, January, 1999. 

[27] Bhave, S.A., J.I. Seeger, et al., “An Integrated, Vertical-Drive, In- 

Plane-Sense Microgyroscope,” Transducers ‘03, Boston, MA, 

June 9-12. 

[28] Bahukudumbi, P., J.H. Park, A. Besok, “A Unified Engineering 

Model for Steady and Quasi-Steady Shear-Driven Gas 

Microflows,” J. Microscale Thermophysical Engineering, Vol. 7, 

2003, pp. 291-315. 

[29] Kogan, M.N., Rarefied Gas Dynamics, Plenum Press, New York, 

1969. 

[30] Arkilic, E., M.A. Schmidt, K.S. Breuer “Slip Flows and 

Tangential Momentum Accommodation in Micromachined 

Channels” Journal of Fluid Mechanics, Vol. 437, 2001, pp. 29- 

44. 

[31] Veijola, T., H. Kuisma, J. Lahdenpera, “The Influence of Gas- 

Surface Interaction on Gas-Film Damping in a Silicon 

Accelerometer,” Sensors and Actuators, A, Vol. A66, 1998, pp. 

83-92.


Marc S. Weinberg 

Peter Y. Kwok was a Draper Fellow from 1998-99, after which he joined the Volpe National 

Transportation Systems Center. In 2001, he joined Foster-Miller, Waltham, MA as a Project Engineer, 

working on structural analysis. He is a Member of the American Society of Mechanical Engineers (ASME). 

He received an MS in Mechanical Engineering from MIT (1999). 

Marc S. Weinberg served in the United States Air Force Aeronautical System Division, Wright-Patterson 

AFB during 1974 and 1975, where he applied modern and classical control theory to design turbine 

engine controls, and at the Air Force Institute of Technology, where he taught gas dynamics and feedback 

control. At Draper Laboratory, he has been responsible for the design and testing of a wide range of micromechanical 

gyroscopes, accelerometers, hydrophones, microphones, angular displacement sensors, chemical 

sensors, and biomedical devices. He holds 24 patents with 12 additional in application. Dr. Weinberg 

has been a Member of ASME since 1971. He received BS, MS, and PhD degrees in Mechanical Engineering 

from MIT (1971, 1971, and 1974, respectively), where he also held a National Science Foundation 

Fellowship. 

Kenneth S. Breuer was in the Department of Aeronautics and Astronautics, MIT, from 1990 to 1999, 

before moving to Brown University where he is currently a Professor of Engineering. His research interests 

are in the fluid mechanics of micron and nanometer-scale systems. He is a member of ASME. Dr. 

Breuer received a PhD from MIT (1988). 

69

A Reusable Design for Precision Lunar Landing Systems 

Proceeding of the 2005 International Lunar Conference 

Toronto, Canada, September 2005 

Abramson, M.; Bryant, C.; Carr, F.; Kolitz, S.; Kahn, A.; 

Key, J.; Sakamoto, P. 

Robust Planning Under Uncertainty 

American Institute of Aeronautics and Astronautics 

(AIAA) Infotech@Aerospace, September 26-29, 2005, 

Crystal City, VA. Sponsored by: AIAA 

Abramson, M.; Carter, D.; Collins, B.; Kolitz, S.; Scheidler, 

P.; Strauss, C. 

Improving the Value of EO-1 Observations 

Proceedings of the 2005 Earth Science Technology 

Conference, College Park, MD, June 2005 

Adams, M.; Manobianco, J.; Bickford, J.; Manobianco, D.M. 

In Situ Atmospheric Profiling Using Mobile Ad Hoc 

Sensor Networks 

Nano Science and Technology Institute (NSTI) 

Nanotechnology Conference and Trade Show, Anaheim, 

CA, May 8-12, 2005. Sponsored by: NSTI 

70 


Artist’s Concept of the Lunar Surface Access Module (LSAM) 

Image courtesy of NASA/John Frassanito & Associates 

Adams, M.; Lepanto, J.; Hanson, M. 

Mixed Initiative Command and Control of 

Autonomous Air Vehicles 

AIAA Journal of Computing, Information, and 

Communication, Vol. 2, February 2005 

Ardini, J. 

A High-Performance Radix-2 FFT in ANSI-C for RTL 

Generation 

8 th Military and Aerospace Programmable Logic Devices 

(MAPLD), Washington, D.C., September 7-9, 2005. 

Sponsored by: MAPLD 

Ardini, J. 

Demand and Penalty-Based Resource Allocation for 

Reconfigurable Systems with Runtime Partitioning 

8 th Military and Aerospace Programmable Logic Devices 

Conference, Washington, D.C., September 7-9, 2005. 

Sponsored by: MAPLD

Barton, G.; Marinis, T.; Pryputniewicz, D. 

A Multichip Module Implementation of a Radiation- 

Tolerant Microprocessor for Space Applications 

32 nd International Microelectronics and Packaging 

Society (IMAPS) New England Annual Symposium and 

Exhibition, Boxborough, MA, May 17, 2005. Sponsored 

by: IMAPS 

Barton, G.; Marinis, T.; Pryputniewicz, D. 

A Multichip Module Implementation of a Radiation- 

Tolerant Microprocessor for Space Applications 

38 th International Symposium on Microelectronics, 

Boxboro, MA, September 25-29, 2005. Sponsored by: 

IMAPS 

Bedrossian, N.; Jang, J-W; Alaniz, A.; Johnson, M.; 

Sebelius, K.; Mesfin, Y. 

International Space Station U.S. GN&C Attitude 

Hold Controller Design for Orbiter Repair Maneuver 

Guidance, Navigation, and Control Conference and 

Exhibit, San Francisco, CA, August 15-18, 2005. 

Sponsored by: AIAA 

Bickford, J.; Adams, M.; Manobianco, J. 

GEMS: A Mobile Wireless Network for Atmospheric 

Sampling 

Defense and Security Symposium, Orlando, FL, March 

28-April 1, 2005. Sponsored by: International Society 

for Optical Engineering (SPIE) 

Bickford, J.; George, S.; Manobiance, J.; Adams, M.; 

Manobianco, D. 

Large-Scale Deployment and Operation of Distributed 

Sensor Assets Optimized for Robust Mars Exploration 

Evolvable Hardware Conference, Washington, D.C., 

June 29-July 1, 2005. Sponsored by: National 

Aeronautics and Space Administration/Department of 

Defense (NASA/DoD) 

Bloom, I.K. 

Aliquots of Adhesives Presentation 

15 th Alternatives to Toxic Materials in Industrial 

Processes, Phoenix, AZ, January 10-13, 2005 


Borenstein, J.; Fiering, J.; Mescher, M.; Kujawa, S.; Sewell, 

W.; McKenna, M. 

Microsystems Technology for Long-Term Drug 

Delivery to the Inner Ear 

5 th International Symposium Meniere’s Disease and 

Inner Ear Homeostasis Disorders, Los Angeles, CA, 

April 2-5, 2005 

Brazzel, J.; Clark, F.; Spehar, P. 

RPOP Enhancements to Support the Space Shuttle R- 

Bar Pitch Maneuver for Tile Inspection 

Proceedings of the AIAA Guidance, Navigation, and 

Control Conference, August 2005 

Brown, R. 

Automating Space Operations Using Timeliner and 

Adept 

Enhancing Space Operations Workshop, Houston, TX, 

May 4-6, 2005. Sponsored by: AIAA 

Campbell, D.; Fill, T.; Hattis, P.; Tavan, S. 

An On-Board Mission Planning System to Facilitate 

Precision Airdrop 

Infotech@Aerospoace, Arlington, VA, AIAA-2005-7071, 

September 2005, pp. 26-29 

Carlen, E.; Heng, K-H; Bakshi, S.; Pareek, A.; 

Mastrangelo, C. 

High-Aspect Ratio Vertical Comb-Drive Actuator 

with Small Self-Aligned Finger Gaps 

Journal of Microelectromechanical Systems, Vol. 14, No. 

5, October 2005, pp. 1144-1155 

Carr, F.; Kolitz, S.; Lepanto, J.; Scheidler, P.; Wilde, J.; 

Smith, J. 

An Air Traffic System Simulation of the NAS with 

CNS and CD&R Models 

Modeling and Simulation Technologies, Conference and 



Carter, D.; George, S.; Hattis, P.; Singh, L.; Tavan, S. 

Autonomous Guidance, Navigation, and Control of 

Large Parafoils 

Aerodynamic Decelerator Systems Conference (ADSC), 

Munich, Germany, May 23-26, 2005 

71

Chen, Z.; Kujawa, S.; McKenna, M.; Fiering, J.; Mescher, 

M.; Borenstein, J.; Leary Swan, E.; Sewell, W., 

Inner Ear Drug Delivery Via a Reciprocating 

Perfusion System in the Guinea Pig, 

J. Controlled Release, Vol. 110, No. 1, 2005, pp. 1-19. 

Cleary, M.; Dai, L. 

Situation Awareness Improvements for UUVs 

Unmanned Systems, North America, Baltimore, MD, 

June 28-30, 2005. Sponsored by: Association for 

Unmanned Vehicle Systems International (AUVSI) 

Colwell, W. 

Implementing Collocation Groups 

Facing the Future, Tivioli Storage Manager (TSM) 

Symposium, September 26-29, 2005. Sponsored by: 

TSM 

Coskren, D.; Easterly, T.; Polutchko, R. 

More Bang, Less Buck. Low-Cost GPS/INS Guidance 

for Navy Munitions Launches 

GPS World, September 2005, pp. 22-28 

Davis, C. 

Microfabricated Sensors for Clinical Diagnostic 

Applications 

Gordon Research Conference on Chemical Sensors and 

Interfacial Design, Oxford, England, August 28- 

September 2, 2005 

Diel, D.; Debitetto, P. 

Epipolar Constraints for Vision-Aided Inertial 

Navigation 

Computer Society Workshop on Motion and Video 

Computing, Breckenridge, CO, January 5-6, 2005. 

Sponsored by: IEEE 

Fidkowski, C.; Kaazempur-Mofrad, M.; Borenstein, J.; 

Vacanti, J.; Langer, R.; Wang, Y. 

Endothelialized Microvasculature Based on a 

Biodegradable Elastomer 

Tissue Engineering, Vol. 11, No. 1-2, January 2005, pp. 

302-309 

72 


Flueckiger, K.; Gustafson, D.; Dowdle, J. 

INS/GPS Deep Integration Navigation Hardware 

Testbed 

61 st Institute of Navigation (ION) Annual Meeting, 

Cambridge, MA, June 27-29, 2005. Sponsored by: ION 

Friend, S.; Monopoli, D.; Natoli, L.; Haas, D. 

SH-60B HUMS Experience Using a Satellite Data 

Link 

Aerospace Conference, Big Sky, MT, March 5-12, 2005. 


Fuhrman, L. 

The NASA Human Exploration Program 

Aerospace Control and Guidance Systems Committee 

Meeting, Salt Lake City, UT March 2-4, 2005. Sponsored 

by: Society of Automotive Engineers (SAE). 

Fuhrman, L.; Fill, T.; Forest, L.; Norris, L.; Paschall II, S.; 

Tao,Y.C. 

A Reusable Design for Precision Lunar Landing 

Systems 

Proceeding of the 2005 International Lunar Conference, 

Toronto, Canada, September 2005 

Furis, M.; Cartwright, A.; Waldron, E.; Schubert, E. 

Spectral and Temporal Resolution of Recombination 

from Multiple Excitation States in Modulation- 

Doped AlGaN/GaN Multiple Quantum Well 

Heterostructures 

Applied Physics Letters, Vol. 86, No. 16, April 18, 2005 

George, S.; Carter, D.; Berland, J-C.; Dunker, S.; Tavan, S.; 

Barber, J. 

The Dragonfly 4,500-kg Class Guided Airdrop 

System 

Infotech@Aerospace, Arlington, VA, September 26-29, 

2005. Sponsored by: AIAA 

Guerra, C.; Page, L. 

Autonomous Planning for Spacecraft Rendezvous 

and Proximity Operations 

Infotech@Aerospace, Arlington, VA, September 26-29, 

2005. Sponsored by: AIAA

Gustafson, D.; Dowdle, J.; Monopoli, D.; Anszperger, J. 

King Air Demonstration of Deep Integration 

(KADDI) 

2005 Joint Navigation Conference, Orlando, FL, April 

2005 

Hall, R.; Barrington, R.; Kirchwey, K.; Alaniz, A.; 

Grigoriadis, K. 

Shuttle Stability and Control During the Orbiter 

Repair Maneuver 

3 rd International Energy Conversion, Engineering 

Conference, San Francisco, CA, August 15-18, 2005. 


Hopkins, R. 

Inertial Navigation System Technology: A Short 

Course 

30 th Joint Navigation Conference, Orlando, FL, April 

11, 2005. Sponsored by: Joint Services Data Exchange 

(JSDE) 

Hopkins, R.; Miola, J.; Sawyer, W.; Setterlund, R.; Dow, B. 

The Silicon Oscillating Accelerometer: A High- 

Performance MEMS Accelerometer for Precision 

Navigation and Strategic Guidance Applications 

10 th ION National Technical Meeting, San Diego, CA, 

January 24-26, 2005. Sponsored by: ION 

Hopkins, R.; Miola, J.; Setterlund, R.; Dow, B.; Sawyer, W. 

The Silicon Oscillating Accelerometer: A High- 

Performance MEMS Accelerometer for Precision 

Navigation and Strategic Guidance Applications 

61 th ION Annual Meeting, Cambridge, MA, June 27-29, 

2005. Sponsored by: ION 

Huntington, G.; Rao, A. 

Optimal Reconfiguration of a Tetrahedral Formation 

via a Gauss Pseudospectral Method 

Proceedings of the 2005 Astrodynamics Specialist 

Conference, AAS Paper 05-338, Lake Tahoe, California, 

August 7-10, 2005 

Huntington, G.; Rao, A. 

Optimal Spacecraft Formation Configuration Using 

a Gauss Pseudospectral Method 

15 th Space Flight Mechanics Meeting, Copper 

Mountain, CO, January 23-27, 2005. Sponsored by: 

American Astronautic Society (AAS)/AIAA 


Huntington, G.; Benson, D.; Rao, A. 

Post-Optimality Evaluation and Analysis of a 

Formation Flying Problem via a Gauss Pseudospectral 

Method 

Proceedings of the 2005 Astrodynamics Specialist 

Conference, AAS Paper 05-339, Lake Tahoe, California, 

August 7-10, 2005 

Jackson, M.; D’Souza, C.; Lane, H. 

Autonomous Mission Management for Spacecraft 

Rendezvous Using an Agent Hierarchy 

Infotech@Aerospace. Arlington, VA, September 26-29, 

2005. Sponsored by: AIAA 

Kaya, T.; Lin, P.; Noubir, G.; Qian, W. 

Efficient Multicast Trees with Local Knowledge on 

Wireless ad hoc Networks 

Wireless Communications and Networking Conference 

(WCNC), New Orleans, LA, March 13-17, 2005. 


Kaya, T.; Lin, P.; Noubir, G.; Qian, W. 

Efficient Multicast Trees with Local Knowledge on 

Wireless Ad Hoc Networks 

3 rd Wired/Wireless Internet Communications, International 

Conference, Xanthi, Greece, May 11-13, 2005 

Key, R. 

Routing in Stochastic Networks 

International Conference on Networking, Sensing, and 

Control, Tucson, AZ, March 19-22, 2005. Sponsored by: 

IEEE 

Khademhosseini, A.; Yeh, J.; Eng, G.; Karp, J.; Kaji, H.; 

Borenstein, J.; Farokhzad, O.; Langer, R. 

Cell Docking Inside Microwells Within Reversibly 

Sealed Microfluidic Channels for Fabricating 

Multiphenotype Cell Arrays 

Lab on a Chip, Vol. 5, No. 12, 2005, Royal Soc. 

Chemistry, Cambridge, England, pp. 1380-1386 

Kostas, T.; Lin, P. 

Flexible Satellite Communication Design 

Intelligence Community Chief Information Officer (IC 

CIO), Digital Convergence Conference, Chantilly, VA, 

May 16-17, 2005. Sponsored by: IC CIO 

73

Krebs M.; Tingley R.; Zeskind J.; Kang J.; Holmboe M.; 

Davis C. 

Autoregressive Modeling of Analytical Sensor Data 

Can Yield Classifiers in the Predictor Coefficient 

Parameter Space 

Bioinformatics, Vol. 21, No. 8, April 15, 2005, pp. 

1325-1331 

Krebs, M.; Zapata, A.; Nazarov, E.; Miller, R.; Costa, I.; 

Stonenshein, A.; Davis, C. 

Detection of Biological and Chemical Agents Using 

Differential Mobility Spectrometry (DMS) 

Technology 

IEEE Sensors Journal, Vol. 5, No. 4, August 2005, pp. 

696-703 

Kwok, P.; Weinberg, M.; Breuer, K. 


Journal of Microelectromechanical Systems, Vol. 14, No. 

4, August 2005, pp. 770-781 

Malasky, J.; Forest, L; Kahn, A; Key, J. 

Experimental Evaluation of Human-Machine 

Collaboration in Resource Allocation and Planning 

for Multiple UAVs 

International Conference on Systems, Man, and 

Cybernetics, Hawaii, October 10-12, 2005. Sponsored 

by: IEEE 

Marinis, T.; Soucy, J.; Lawrence, J.; Owens, M. 

Wafer-Level Vacuum Packaging of MEMS Sensors 

55 th Electronic Components and Technology 

Conference, Lake Buena Vista, FL, May 31-June 3, 

2005. Sponsored by: IEEE/Components, Packaging, and 

Manufacturing Technology (CPMT) Society 

Mescher, M.; Lutwak, R.; Varghese, M. 

An Ultra-Low-Power Physics Package for a Chip- 

Scale Atomic Clock 

13 th International Conference on Solid-State Sensors, 

Actuators, and Microsystems (Transducers), Seoul, 

Korea, June 5-9, 2005. Sponsored by: IEEE 

Monopoli, D.; Natoli, L.; Haas, D.J.; Baker, T. 

JAHUMS ACTD Operational Experience 

Annual Forum 61, New Frontiers in Vertical Flight, 

Grapevine, TX, June1-3, 2005. Sponsored by: American 

Helicopter Society (AHS) 

74 


Parry, J.; Ricard, M. 

Vectors of Autonomy 

14 th Unmanned Untethered Submersible Technology, 

Durham, NH, August 21-24, 2005. Sponsored by: 

Autonomous Undersea Systems Institute (AUSI) 

Plump, J.; Ricard, M.; Keegan, M.; Jarriel, M. 

ONR’s Maritime Reconnaissance Demonstration – 

Simulation-Based Test and Integration 

Unmanned Systems North America, Baltimore, MD, 

June 28-30, 2005. Sponsored by: AUVSI 

Postma, B.; Jang, J.; Bedrossian; N.; Spanos, P. 

Robust Constrained Optimization Approach for 

International Space Station Centrifuge Rotor Auto- 

Balancing Controller 




Proulx, R.; Carter, D.; Cefola, P.; Draim, J.; Inciari, R.; 

The Orbital Perturbation Environment for the 

COBRA and COBRA Teardrop Elliptical 

Constellations 

Journal of the Astronautical Sciences, Vol. 53, No. 2, 

April-June 2005, pp. 111-129 

Ricard, M.; Nervegna, M.; Keegan, M.; Jarriel, M. 

Autonomy and Control Architectures 

Unmanned Systems North America, Baltimore, MD, 

June 28-30, 2005. Sponsored by: AUVSI 

Rosch, G.; Hall, R.A. 

Stability of Angular Rate Damping for the NASA 

Space Shuttle 




Ross, I.; D’Souza, C. 

Hybrid Optimal Control Framework for Mission 

Planning 

AIAA Journal of Guidance Control and Dynamics, Vol. 

28, No. 4, July-August 2005, pp. 686-697

Sawyer, W.; Prince, M.; Brown, G. 

SOI-Bonded Wafer Process for High-Precision MEMS 

Inertial Sensors 

Journal of Micromechanics and Microengineering, Vol. 

15, No. 8, pp. 1588-93 

Sherman, P.; Kourepenis, A.; Holmes, S.; Girolamo, H.; 

Zimmer, G.; Sokolowski, S. 

Personal Navigation for the Warfighter 

Joint Navigation Conference, Orlando, FL, April 11-15, 

2005. Sponsored by: DoD 

Shnayderman, M.; Mansfield, B.; Yip, P.; Clark, H.; Krebs, 

M.; Cohen, S.; Zeskind, J.; Ryan, E.; Dorkin, H.; 

Callahan, M.; Stair, T.; Gelfand, J.; Gill, C.; Hitt, B.; 

Davis, C. 

Species-Specific Bacteria Identification Using 

Differential Mobility Spectrometry and 

Bioinformatics Pattern Recognition 

Analytical Chemistry, Vol. 77, No. 18, August 12, 2005, 

pp. 5930-5937 

Singh, L.; Yang, L.; McConley, M.; Appleby, B. 

Autonomous Guidance and Control for Agile UAV 

Maneuvering 

2005 American Helicopter Society Forum 61, 

Grapevine, Texas June 1-3, 2005 

Singh, L.; Lapp, T. 

Model Predictive Control in Nap-of-Earth Flight 

Using Polynomial Control Basis Functions 

Proceedings of the 2005 American Control Conference, 

Vol. 2, pp. 840-5 

Stocker, D.; Waldron, E.; Boyle, J.; Kisler, Y. 

Neutron-Induced Degradation of CCD and CMOS 

Imager Optical Responsivity 

Hardened Electronics and Radiation Technology 

(HEART) Conference, Monterey, CA, March 1-5, 2005. 

Sponsored by: DoD/Department of Energy (DoE) 

Stolfi, M.; Negro, L.; Michel, J.; Duan, X.; LeBlanc, J.; 

Haavisto, J.; Kimerling, L.C. 

CMOS Compatible Erbium Coupled Si Nanocrystal 

Thin Films for Microphotonics 

Materials Research Society Symposium Proceedings, Vol. 

832, 2005, pp. F11.8 


Sullivan, M.; Bedrossian, N.; Nagarajaiah, S. 

State Estimation for International Space Station 

Centrifuge Rotor 

Guidance, Navigation, and Control Conference, San 

Francisco, CA, August 15-18, 2005. Sponsored by: 

AIAA 

Tetewsky, A.; Anszperger, J. 

Exactly What Does the GPS Satellite Transmit? 

61 st Institute of Navigation Annual Meeting, Cambridge, 

MA, June 27-29, 2005. Sponsored by: ION 

Tudryn, C.; Schweizer, S.; Hopkins, R.; Hobbs, L.; Garratt- 

Reed, A. 

Characterization of Si and CVD SiC to Glass Anodic 

Bonding Using TEM and STEM Analysis 

Journal of the Electrochemical Society, Vol. 152, No. 4, 

March 7, 2005, pp. E131-E134 

Underwood, J.; Poppe, D.; Weck de, O. 

Distributed Satellite Communications Systems: 

First-Order Interactions Between System and 

Network Architectures 

International Communications Satellite Systems 

Conference (ICSSC), Rome, Italy, September 25-28, 

2005. Sponsored by: ICSSC 

Van Beusekom, C.; Miotto, P.; Shepperd, S. 

Guidance and Control for Highly Constrained 

Rendezvous 

15 th Flight Mechanics Conference, Copper Mountain, 

CO, January 23-27, 2005. Sponsored by: AAS/AIAA 

Weinberg, E.J.; Kaazempur-Mofrad, M.R., 

A Large-Strain Finite-Element Formulation for 

Biological Tissues with Application to Mitral Valve 

Leaflet Tissue Mechanics 

Journal of Biomechanics, July 20, 2005 

Weinberg, E.J.; Kaazempur-Mofrad, M.R. 

On the Constitutive Models for Heart Valve Leaflet 

Mechanics 

Cardiovascular Engineering, Vol. 5, No. 1, 2005, pp. 37- 

43 

75

Weinberg, M.; Wall, C., III 

Vestibular Prostheses for the Balance Impaired 

5 th International Symposium Meniere’s Disease and 

Inner Ear Homeostasis Disorders, Los Angeles, CA, 

April 2-5, 2005 

Weiss, L.; White, T. 

Autonomy Standards for Unmanned Undersea 

Vehicles 

ASTM Standardization News, November 2005, pp. 30- 

33. 

Wholey, L.; Singh, L. 

Automatic Low-Visibility Trajectory Optimization 

for Visually Identifying a Suspected Aircraft 

Guidance, Navigation, and Control, Conference and 



Willig, R. 

Solid-Core Laminar Photonic-Crystal Waveguide 

Optics East. Sensors and Applications, Boston, MA 

October 23-26, 2005, Sponsored by: SPIE 

76 


Zhu, M.; Perreault, D.; Caliskan, V.; Neugebauer, T.; 

Guttowski, S.; Kassakian, J. 

Design and Evaluation of Feedforward Active Ripple 

Filters 

IEEE Transactions on Power Electronics, Vol. 20, No. 2, 

March 2005, pp. 276-285 

Zimpfer, D.; Kachmar, P.; Tuohy, S. 

Autonomous Rendezvous, Capture, and In-Space 

Assembly: Past, Present, and Future 

1 st Space Exploration Conference: Continuing the 

Voyage of Discovery, Orlando, FL, January 30-February 

1, 2005. Sponsored by: AIAA 

Zimpfer, D.; Spehar, P.; Clark, F.; D’Souza, C.; Jackson, M. 

Autonomous Rendezvous and Capture Guidance, 

Navigation, and Control 

Flight Mechanics Symposium, Greenbelt, MD, October 

18-20, 2005. Sponsored by: NASA

Low-Cost, Low-Power Geolocation System: 

Patent Number 6,934,626 

Draper Laboratory’s enduring trademark is its ability to 

integrate widely diverse technical capabilities and technologies 

into innovative and creative solutions for 

problems of national importance. Draper scientists and 

engineers are encouraged in their quest to advance the 

application of science and technology, expand the functions 

of existing technologies, and to create new ones. 

The disclosure of inventions is an important step in documenting 

these creative efforts and is required under 

Laboratory contracts (and by an agreement with Draper 

that all employees sign). Draper has an established patent 

policy and understands the value of patents in directing 

attention to individual accomplishments. Pursuing patent 

protection enables the Laboratory to pursue its strategic 

mission and to recognize its employees’ valuable contributions 

to advancing the state-of-the-art in their technical 

areas. An issued patent is also recognition by a critical 

third party (the U.S. Patent Office) of novel and creative 

work for which the inventor should be justly proud. 

Achromatic Fiber-Optic Power Splitter and Related Methods: 

Patent Number 6,959,131 B2 

Patents Introduction 

On average, Draper’s Patent Committee typically recommends 

seeking patent protection for 50 percent of the 

disclosures received. Millions of U.S. patents have been 

issued since the first patent in 1836. Through December 

31, 2005, 1252 Draper patent disclosures have been submitted 

to the Patent Committee since 1973; 638 of which 

were approved by Draper's Patent Committee for further 

patent action. As of December 31, a total of 479 patents 

have been granted for inventions made by Draper personnel. 

Sixteen patents were issued for calendar year 2005. 

This year’s featured patent is: 

Optically Rebalanced Accelerometer 

Tuning-Fork Gyro: 


The following pages contain an overview of the technology 

covered in the patent, followed by the official patent 

abstract issued by the U.S. Patent Office. 

77

78 


Patent No. 6,867,411 B2 Issued: March 15, 2005 

William Kelleher, Stephen Smith, Richard Stoner 

This invention is an all-optical accelerometer that uses radiation pressure to stabilize the position 

of a proof mass and a rebalance mechanism to measure acceleration. Instruments that can sense departures 

of their own reference frame from an inertial reference frame are used in many areas, including 

inertial navigation and guidance. Such departures include accelerations, which are commonly sensed 

by measuring either the displacement of a proof mass in response to an inertial force or by the restoring 

force necessary to restore the displacement of a proof mass. 

In the past, such sensors have been constructed from relatively large and expensive electromagnetic 

components. More recently, MEMS sensors have been fabricated from silicon wafers using techniques 

such as photolithography. Advantages of microfabricated sensors include small size and weight and 

the possibility of lower-cost large-scale production. There is now a growing manufacturing base and 

body of skilled workers in the fiber-optic communications industry as well as a large and growing 

infrastructure. An inertial sensor that uses only electro-optical components would thus share many 

subsystems and components with the fiber-optics industry and can be built economically. 

This innovative all-optical accelerometer could provide many advantages over prior art electromechanical 

inertial sensors. It has no moving wear surfaces: its projected lifetime is much greater than 

that of electromechanical accelerometers, since the lifetime of an all-optical accelerometer is limited 

only by the optical source lifetime. It may also be built as a flexureless and very linear instrument, 

eliminating the need for building flexural support structures into the device. Further, unlike prior-art 

MEMS sensors, it would be possible to recalibrate an all-optical inertial sensor during the operation 

of the device. An all-optical inertial sensor can be built as a closed-loop instrument with a high 

dynamic range, and using integrated optics and fiber-optic components, the space and energy requirements 

of the accelerometer can be minimized. 

Optically Rebalanced Accelerometer: 


340 

310 

320 

330 

325

The Optically Rebalanced Accelerometer: U.S. Patent No. 6,867,411 B2 

79

80 

The Optically Rebalanced Accelerometer: U.S. Patent No. 6,867,411 B2 

(l-r) Stephen P. Smith, William P. 

Kelleher, and Richard E. Stoner 

William P. Kelleher joined the Electromagnetics Group at Draper Laboratory after working at Los Alamos 

National Laboratory. He has since served as group leader in electro-optics, SP23 IFOG Task Leader, and 

SP23 Sensor Development Technical Director. He is currently an Associate Division Leader in Guidance 

Hardware. Dr. Kelleher received a PhD in Nuclear Engineering from Rensselaer Polytechnic Institute. 

Stephen P. Smith developed a novel fiber-optic-based quench diagnostic for massive superconducting 

magnet systems to be used in Tokamak plasma fusion. He then worked as a post-doctoral researcher in 

the laboratory of Prof. Mara Prentiss at Harvard, where he was involved in a diverse array of research projects 

entailing biomedical applications of optical tweezers, laser cooling and trapping, and others. Except 

for a brief interlude, he has been employed at Draper since 1999, where he serves as Task Leader for the 

SP23 IFOG program as well as GBE-1 Group Leader. Mr. Smith received BS and ScD degrees in Electrical 

Engineering from MIT. 

Richard E. Stoner worked as a post-doctoral researcher and then a staff Scientist at the Harvard- 

Smithsonian Center for Astrophysics until June 2000. His work there comprised precision tests of fundamental 

symmetries using noble gas spin masers that he developed and constructed. He then joined 

Draper’s technical staff, where he has focused on the development of radiation-hard interferometric fiberoptical 

gyro (IFOG) technology, primarily in support of the SP23 program. He received the Draper 

Distinguished Performance Award in 2002 for his work on precision radiation-hard scale-factor control for 

IFOGs. Dr. Stoner received a PhD in Physics from MIT.

Anderson, R.; Connelly, J.; Hanson, D.; Soucy, J.; Marinis, T. 

Integrated Sensor and Electronics Package 

Patent Number 6,891,239 B2, May 10, 2005 

Anderson, R.; Hanson, D.; Kasparian, F.; Marinis, T.; 

Soucy, J. 

Stress Isolation System 

Patent Number 6,936,479 B2, August 30, 2005 

Ash, M.; Martorana, R. 

Borehole Navigation System 

Patent Number 6,895,678 B2, May 24, 2005 

Ash, M.; DeBitetto, P.; Kourepenis, A.; Thorvaldsen, T. 

Compact Navigation System and Method 

Patent Number 6,918,186, July 19, 2005 

Bickford, J.; Petrovich, A.; Weinberg, M. 

Dual Microwave Cavity Accelerometer 

Patent Number 6,928,875, August 16, 2005 

Borenstein, J.; Sawyer, W. 

Method for Microfabricating Structures Using 

Silicon-on-Insulator Material 

Patent Number 6,946,314 B2, September 20, 2005 

Borenstein, J.; Connelly, J.; Cousens, J.; Duwel, A.; 

Kourepenis, A.; Lento, C.; Sawyer, W.; Weinberg, M. 

Tuning-Fork Gyro 

Patent Number 6,862,934 B2, March 8, 2005 

Bousquet, R.; Haley, H.; Hildebrant, E.; Martinez, S.; 

Ward, P. 

Charge Amplifier Device Having Fully Integrated 

DC Stabilization 

Patent Number 6,873,206, March 29, 2005 

Dual Microwave Cavity Accelerometer: 


2005 Patents Issued 

Cunningham, B.; Williams, J. 

Flexural Plate Wave Sensor and Array 

Patent Number 6,837,097 B2, January 4, 2005 

Cunningham, B.; Williams, J. 

Flexural Plate Wave Sensor and Array 

Patent Number 6,851,297, February 8, 2005 

Dubé, C.; Petrovich, A.; Williams, J. 

Sensor Readout Circuit 

Patent Number 6,972,553, December 6, 2005 

Kelleher, W.; Smith, S.; Stoner, R. 



Lane, P.; Tapalian, C. 

Thermo-Optical Switch Using Coated Microsphere 

Resonators 

Patent Number 6,934,436 B2, August 23, 2005 

Miller, R.; Nazarov, E.; Eiceman, G.; Krylov, E. 

Method and Apparatus for Electrospray Augmented 

High Field Asymmetric Ion Mobility Spectrometry 

Patent Number 6,972,407, December 6, 2005 

Tingley, R. 

Low-Cost, Low-Power Geolocation System 

Patent Number 6,934,626, August 23, 2005 

Willig, R. 

Achromatic Fiber-Optic Power Splitter and Related 

Methods 


Stress Isolation System (green section): 


81

Photo credit: National Academy of Engineering 

The NAE presented the 2006 Charles Stark Draper Prize to the inventors 

of charge-coupled devices, Willard S. Boyle and George E. Smith, on 

February 21 in Washington, D.C. The NAE’s citation reads, “for the invention 

of the Charge-Coupled Device (CCD), a light-sensitive component at 

the heart of digital cameras and other widely used imaging technologies.” 

Boyle and Smith share the $500,000 honorarium for their invention. 

CCDs, the first practical solid-state imaging devices, were invented in 1969 

by Boyle and Smith at Bell Laboratories. Since CCDs are silicon-based 

devices, they are relatively inexpensive to produce and are compact and 

fairly rugged, thus eminently suitable for commercial use. High sensitivity, 

excellent stability, and lack of distortion make them attractive for use in 

scientific research imaging systems. CCDs can image a variety of sources, 

such as optical, x-ray, ultraviolet, and infrared emissions. Currently, CCDs 

are used extensively in consumer products, including camcorders and 

cell phone cameras, and in such advanced electronic imaging tools as 

telescopes and imaging satellites. 

82 

Willard S. Boyle 

The 2006 Charles Stark Draper Prize 

The Charles Stark Draper Prize was established in 1988 to honor the 

memory of Dr. Charles Stark Draper, “the father of inertial navigation.” 

Awarded annually, the Prize was instituted by the National 

Academy of Engineering (NAE) and endowed by Draper Laboratory. 

It is recognized as one of the world’s preeminent awards for 

engineering achievement, honoring individuals who, like Dr. Draper, 

developed a unique concept that has contributed significantly to the 

advancement of science and technology and the welfare and freedom 

of society. 

For information on the prize, contact the Public Affairs Office 

at the National Academy of Engineering at (202) 334-1237 or 

http://www.nae.edu. 

Willard S. Boyle received a PhD in Physics at McGill University and worked for a year as a postdoctoral 

fellow in the Radiation Laboratory. After 2 years as an Assistant Professor at the Royal 

Military College, he joined the research staff of Bell Laboratories in 1953, and held positions in 

New Jersey, Washington, and Pennsylvania. He became Executive Director of Device 

Development and then Executive Director of the Communication Science Division until his 

retirement in 1979. His early work was on the electronic properties of semiconductors. He was 

the first to use infrared spectroscopy to measure donor energy levels and cyclotron energy levels. 

With Gene Kunzler, he discovered large, low-temperature magnetothermal oscillations, a powerful 

new tool for mapping out Fermi surfaces. With Donald Nelson, he developed the first 

continuously pumped ruby laser. He co-authored a major review paper with George E. Smith 

summarizing the unique properties of bismuth. Dr. Boyle is best known as the co-inventor, with 

George E. Smith, of the CCD. The first paper on the CCD was published in the Bell System 

Technical Journal on January 29, 1970. Dr. Boyle is a fellow of the American Physical Society and 

IEEE and a member of the National Academy of Engineering. He was awarded an Honorary 

Doctor of Laws from Dalhousie University. Since his retirement, he has served on the Research 

Council of the Canadian Institute of Advanced Research and the Science Council of the Province 

of Nova Scotia. 

Photo credit: National Academy of Engineering

The 2006 Charles Stark Draper Prize (cont.) 

George E. Smith received a BA in Physics from the University of Pennsylvania and MS and 

PhD degrees in Physics from the University of Chicago. He joined Bell Laboratories in 

1959,where he studied the electrical properties and band structures of semimetals, mostly 

bismuth and bismuth-antimony alloys, conducted microwave-resonance experiments 

and investigated magnetothermoelectric and galvanomagnetic effects. Heading the Device 

Concepts Department, he conducted investigations on junction lasers, semiconducting 

ferroelectrics, electroluminescence, transition-metal oxides, the silicon-diode-array 

camera tube, and CCDs. His major technical accomplishment was the development of the 

CCD with Willard S. Boyle, for which they have received numerous prestigious awards. 

They hold the basic patent (U.S. 3,858,232) and published the first paper disclosing the 

device concept, accompanied by a paper on its experimental verification in 1970. In April 

1986, he retired from Bell Laboratories as head of the VLSI Device Department, where he 

oversaw work on the physics of devices made with submicron lithography and their use 

in high-performance digital and analog circuits. Dr. Smith is a member of Pi Mu Epsilon, 

Phi Beta Kappa, Sigma Xi, and the National Academy of Engineering and is a fellow of 

the IEEE and American Physical Society. He holds 31 U.S. patents and has authored more 

than 40 papers. He received the IEEE Electron Devices Society Distinguished Service 

Award in 1997. 

Recipients of the Charles Stark Draper Prize 

George E. Smith 

2005: Minoru S. Araki, Francis J. Madden, Don H. Schoessler, Edward A. Miller, James W. 

Plummer for their invention of the Corona reconnaissance satellite technology. 

2004: Alan C. Kay, Butler W. Lampson, Robert W. Taylor, and Charles P. Thacker for the development 

of the Alto computer at Xerox’s Palo Alto Research Center (PARC). 

2003: Ivan A. Getting and Bradford W. Parkinson for their technological achievements in the 

development of the Global Positioning System. 

2002: Robert Langer for bioengineering revolutionary medical drug delivery systems. 

2001: Vinton Cerf, Robert Kahn, Leonard Kleinrock, and Lawrence Roberts for their individual 

contributions to the development of the Internet. 

1999: Charles Kao, Robert Maurer, and John MacChesney for spearheading advances in fiberoptic 

technology. 

1997: Vladimir Haensel for the development of the chemical engineering process of 

“Platforming” (short for Platinum Reforming), which was a platinum-based catalyst to 

efficiently convert petroleum into high-performance, cleaner-burning fuel. 

1995: John R. Pierce and Harold A. Rosen for their development of communication satellite 

technology. 

1993: John Backus for his development of FORTRAN, the first widely used, general-purpose, 

high-level computer language. 

1991: Sir Frank Whittle and Hans J.P. von Ohain for their independent development of the 

turbojet engine. 

1989: Jack S. Kilby and Robert N. Noyce for their independent development of the monolithic 

integrated circuit. 

83 

Photo credit: National Academy of Engineering

The CSAC program team developed the critical component 

for the world’s smallest and lowest power atomic clock – 

the physics package. This package, which uses MEMS 

technology, contains a VCSEL/photodiode, a mirror, a 

heater/sensor, and a cesium cell. The CSAC physics package 

achieves temperature control with ultra-low power by 

means of a novel thermal isolation mechanism (suspension 

tether system). The team demonstrated that the package 

consumes an order of magnitude lower power than the 

closest competing approach and two orders of magnitude 

lower power than commercially available atomic clock 

physics packages. This innovative design allowed Draper 

to meet DARPA Phase III goals during Phase II. 

The development of signal processing algorithms for 

GPS in a weak signal environment addresses the unique 

challenges in evaluating the time and Doppler search 

space. Determining location using GPS depends on the 

ability to receive signals from GPS satellites and to measure 

the amount of time between the signal’s broadcast and its 

reception. Poor signal reception can compromise the accuracy 

of position determination. The team considered signal 

integration methods to extract the highest signal-to-noise 

ratio within the constraints imposed by the GPS signal 

structure. The algorithms they recently developed, implemented, 

and refined clearly enhance the effectiveness 

of the processing required for many real-world signal 

environments. 

84 

The Draper Distinguished Performance Awards 

Chairman of the Board Dr. John Kreick and President Vince Vitto presented the 2005 Draper Distinguished Performance 

Awards (DPAs) to two teams at the Annual Dinner of the Corporation on October 5. These awards recognize the development 

of a MEMS Physics Package for a Chip-Scale Atomic Clock (CSAC), performed by team members John Le Blanc, Mark 

Mescher, Gary Tepolt, and Mathew Varghese; and Signal Processing Algorithms for GPS in a Weak Signal Environment, 

performed by James Donna and James Scholten. 

(l-r) Mark Mescher, Mathew Varghese, John Le Blanc, 

and Gary Tepolt 

(l-r) James Scholten and James Donna 

Established in 1989, the Annual DPA is the most prestigious award that Draper can bestow to recognize extraordinary 

achievements by individuals or teams. These achievements must represent a high standard of excellence, 

provide significant benefit to the Laboratory, and be considered a major advance by the outside community. This 

year’s DPA Screening Committee was chaired by Heidi Perry and included James Bickford, Christopher Gibson, 

Roger Medeiros, David Owen, Donald Schwartz, Andrew Staugler, Heather Tahan, and Scott Uhland.

The Howard Musoff Student Mentoring Award 

The 2005 Howard Musoff Student Mentor Award was presented 

to Dr. Michael J. Ricard. In addition to his many 

responsibilities at Draper, he has served as thesis supervisor 

and mentor for nine Draper Fellows at both the 

master’s and PhD level. He has been a member of Draper’s 

technical staff since 1995. His work has focused on the 

design and development of software for unmanned, 

autonomous vehicles, and his specific research interests 

include path planning, mission planning, and other topics 

in applied combinatorial optimization. Dr. Ricard has 

broad experience in undersea programs. He is currently 

the Technical Director of the Maritime Reconnaissance 

Demonstration (MRD) and the Risk-Aware Mixed- 

Initiative Dynamic Replanning (RMDR) programs funded 

by the Office of Naval Research. He is also the Technical 

Director of Draper’s involvement with the US-UK 

Partnership demonstrating antisubmarine warfare (ASW) 

concepts with UUVs. 

Dr. Ricard has also been the Task Leader for the mission 

planner used on the Autonomous Minehunting and 

Mapping Technologies (AMMT) UUV sponsored by 

DARPA. He designed and implemented the real-time software 

that served as the high-level mission controller for 

the vehicle and was involved in the testing of that software 

Michael J. Ricard 

in both the simulation lab and in the field. He has also 

been the Principal Investigator on many joint research 

efforts with the Naval Undersea Warfare Center, Division 

Newport that extended the autonomous capabilities of 

UUVs. He holds a BS in Computer Science from Boston 

College (1988) and SM and PhD degrees in Operations 

Research from MIT (1992 and 1995, respectively). 

The Howard Musoff Mentor Award was established in 

2005. Musoff, who died in 2004, was a Draper employee 

for more than 40 years and advised and mentored 

many Draper Fellows. This award, given each February 

during National Engineers week, recognizes staff members 

who, like Musoff, share their expertise and 

supervise the professional development and research 

activities of Draper Fellows. The award is endowed by 

the Howard Musoff Charitable Foundation and includes 

a $1,000 honorarium and a plaque. Each Engineering 

Division Leader may submit one nomination of a staff 

person from his Division. The Education Office assists in 

the process by soliciting comments from students who 

were residents during that time period. The Selection 

Committee consists of the Vice President of Engineering, 

the Principal Director of Engineering, and the Director 

of Education. 

85

Agrawal, P.; Supervisors: Davis, C.; Belcher, A. 

Characterization of AP-MALDI and ESI for a 

Differential Mobility Spectrometer 

Master of Science Thesis, MIT, May 2005 

Akin, J.; Supervisors: Davis, C.; Irvine, D. 

Characterization of Human Skin Emanations by 

Solid Phase Microextraction (SPME) Extraction of 

Volatiles and Subsequent Analysis by Gas 

Chromatography-Mass Spectrometry (GC-MS) 


Becker, T.; Supervisors: Bottkol, M.; Schmidt, G. 

Approaches to Optimal Inertial Instrument 

Calibration Using Slewing 

Master of Science Thesis, MIT, June 2005 

Calhoun, P.; Supervisors: Hall, S.; White, D. 

Frequency Synthesis Using MEMS Piezoelectric 

Resonators 

Master of Science Thesis, MIT, February 2005 

Carr, K.; Supervisors: Keshava, N.; Greenberg, J. 

Detection of Contaminants Using a MEMS FAIMS 

Sensor 


Chen, D.; Supervisors: Lin, P.; Modiano, E. 

Minimum Energy Path Planning for Ad Hoc 

Networks 

Master of Science Thesis, MIT, September 2005 

Cooper, A.; Supervisors: Roy, N.; Gustafson, D.; 

McConley, M. 

A Comparison of Data Association Techniques for 

Simultaneous Localization and Mapping 


Diel, D.; Supervisors: DeBitetto, P.; Rowell, D. 

Stochastic Constraints for Vision-Aided Inertial 

Navigation 

Master of Science Thesis, MIT, January 2005 

Dobuzhskaya, M.; Supervisors: Brown, R.; Miller, R. 

Timeliner Integrated Development Environment 

Master of Engineering Thesis, MIT, June 2005 

86 


During 2005, the Draper Fellow Program consisted of approximately 60 students from MIT and several other universities. 

Abstracts of theses completed this year are available on the Laboratory’s web site at www.draper.com. The list of completed 

theses follows. 

Earnest, C.; Supervisors: Page, L.; Ricard, M. 

Dynamic Action Spaces for Autonomous Search 

Operations 


Gandhi, R.; Supervisors: Yang, L.; Roy, N. 

Examination of Planning Under Uncertainty 

Algorithms for Cooperative Unmanned Aerial 

Vehicles 


Hawkins, A.M.; Supervisors: Proulx, R.; Fill, T.; Feron, E. 

Constrained Trajectory Optimization of a Soft Lunar 

Landing from a Parking Orbit 


Hickie, M.; Supervisors: Homer, M.; Barnhart, C.; Appleby, B. 

Behavioral Representation of Military Tactics for 

Single-Vehicle Autonomous Rotorcraft via 

Statecharts 


Hickman, R.; Supervisors: Nervegna, M.; Tsitsiklis, J. 

Interception Algorithm for Autonomous Vehicles 

with Imperfect Information 


Hohreiter, L.; Supervisors: Duwel, A.; Chen, G. 

The Effects of Mechanical Coupling on the Electrical 

Impedance of MEMS Resonators for UHF Filter 

Applications 


Hyler, J.; Supervisors: Yu, C.; Madden, S. 

An Augmentation Algorithm for Improving 

Longevity in Ad Hoc Wireless Networks 


Liang, J.; Supervisor: Thordarson, P. 

The Application of the Value-Added Activity Model 

for the Mark-6 LE Integration Project 

Master of Engineering Thesis, MIT, June 2005 

Lommel, P.; Supervisors: Roy, N.; McConley, M.; Peraire, J. 

An Extended Kalman Filter Extension of the 

Augmented Markov Decision Process 

Master of Science Thesis, MIT, May 2005

MacLellan, S.; Supervisor: Staugler, A. 

Orbital Rendezvous Using an Augmented Lambert 

Guidance Scheme 


Malasky, J.; Supervisors: Barnhart, C.; Adams, M. 

Human Machine Collaborative Decision-Making in a 

Complex Optimization System 


McNurlen, A.; Supervisors: Magee, R.; Liskov, B. 

Petscope: A Standardized System for Ballistic 

Missile Guidance Data Analysis 


Merrick, W.; Supervisors: Davis, C.; Greenberg, J. 

Characterization of Human Expired Breath by Solid 

Phase Microextraction and Analysis Using Gas 

Chromatography-Mass Spectrometry and Differential 

Mobility Spectrometry 

Master of Engineering Thesis, MIT, September 2005 

Naresh, P.; Supervisors: Weinstein, W.; Shrobe, H. 

Distributing Network Identity to Mitigate Denial-of- 

Service Attacks 

Master of Engineering Thesis, MIT, August 2005 

Noonan, J.; Supervisors: White, D.; Dawson, J. 

Design of a High-Efficiency RF Power Amplifier for 

an MCM Process 


Oliver, M.; Supervisors: Harris, W.; Barrows, T. 

A Parametric Analysis of the Start-Up Procedure and 

Flight Characteristics of a Gliding Autogyro 


Postma, B.; Supervisor: Jang, J.W. 

Robust Constrained Optimization Approach to 

Control Design for International Space Station 

Centrifuge Rotor Auto Balancing Control System 

Master of Science Thesis, Rice University, April 2005 

Sakai, M.; Supervisors: Carter, D.; Tuller, H. 

Fabrication Process Changes for Performance 

Improvement of an RF MEMS Resonator: 

Conformable Contact Lithography, Moire Alignment, 

and Chlorine Dry Etching 



Stern, S.; Supervisors: Brown, R.; Berwick, R. 

An Extensible Object-Oriented Executor for the 

Timeliner User Interface Language 


Sullivan, M.; Supervisor: Bedrossian, N. 

State Estimation of International Space Station 

Centrifuge Rotor with Incomplete Knowledge of 

Disturbance Inputs 

Master of Science Thesis, Rice University, May 2005 

Thessin, R.N.; Supervisors: Bogner, A.; Herring, T.; 

Peraire, J. 

Atmospheric Signal Delay Affecting GPS 

Measurements Made by Space Vehicles During 

Launch, Orbit, and Reentry 


Thompson, G.; Supervisors: Hall, S.; Soltz, J. 

Inertial Measurement Unit Calibration Using Full 

Information Maximum Likelihood Optimal Filtering 


Torgerson, J.; Supervisors: Deyst, J.; George, S. 

Simulation and Control Design of a Gliding 

Autogyro for Precision Airdrop 


Underwood, J.; Supervisors: Poppe, D.; de Weck, O. 

Distributed Satellite Communication System 

Design: First-Order Interactions Between System 

and Network Architectures 


Van Beusekom, C.; Supervisors: Miotto, P.; Battin, R. 

A New Guidance Method for a DeltaV and Reentry 

Constrained Orbit Transfer Problem 


Warren, J., III; Supervisors: Perreault, D.; Neugebauer, T. 

Cell Modulated dc/dc Converter 

Master of Engineering Thesis, MIT, September 2005 

Weinberg, E.; Supervisors: Borenstein, J.; Mofrad, M. 

Dynamic Simulation of Heart Mitral Valve with 

Transversely Isotropic Material Model 


Wholey, L.; Supervisors: Singh, L.; Appleby, B. 

Trajectory Optimization with Detection Avoidance 

for Visually Identifying an Aircraft 


87

Each year, Draper Laboratory hosts a Technology 

Exposition (Tech Expo) to showcase recent projects and 

highlight Draper’s core competencies. The 2005 Tech 

Expo was held on October 5 and 6, which coincided with 

the fall meeting of Draper’s Board of Directors and the 

Annual Meeting of the Corporation. In addition to 

employees and Corporation members, guests included 

students from local universities and Cambridge public 

schools, as well as journalists and sponsors. 

The exhibits featured technologies under development in 

the Laboratory’s program areas: strategic, tactical, space 

systems, special operations, biomedical engineering, and 

independent research and development. The exhibits also 

reflected the Laboratory’s core competencies: guidance, 

navigation, and control; embedded, real-time software; 

microelectronics and packaging; autonomous systems; 

distributed systems; microelectromechanical systems; biomedical 

engineering; and prototyping system solutions. In 

coordination with Draper’s Education Office, a number of 

the projects also included graduate or undergraduate students 

on their teams. 

Tech Expo also featured Draper’s subsidiary venture capital 

firm, Navigator Technology Ventures, LLC (NTV). NTV 

displayed information about a number of its portfolio 

companies, which currently includes Actuality Systems, 

Aircuity, Assertive Design, Food Quality Sensor (FQS) 

International, HistoRx, Polnox Corp., Polychromix, 

Renalworks Medical Corp., Sionex Corp., and Tizor 

Systems. 

Dr. Angela Zapata describes technology under development in the 

Biomedical Engineering Group to local students. 

88 

2005 Technology Exposition 

Draper’s developing technology and systems for low-intensity conflict, 

surveillance, counterterrorism, and homeland defense are outlined by 

Dr. Paul Rosenstrach, Director of Special Operations (left). 

The Space Systems exhibit highlights Draper’s current focus on 

autonomous and highly reliable flight systems to meet the nation’s 

need for advanced space-based systems. 

David J. Carter describes advances in nanolithography and nanotechnology 

to Draper employee Andrea Prudente.

Inside Back Cover

Headquarters: 

The Charles Stark Draper Laboratory, Inc. 

555 Technology Square 

Cambridge, MA 02139-3563 

Phone: (617) 258-1000 

Fax: (617) 258-1131 

E-mail: techdigest@draper.com 

Business Development 

Phone: (617) 258-2124 

E-mail: busdev@draper.com 

Washington Area Offices: 

Suite 501 

1555 Wilson Boulevard 

Arlington, VA 22209 

Phone: (703) 243-2600 

Fax: (703) 528-5918 

www.draper.com 

#*

TECHNOLOGY DIGEST - Draper Laboratory

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?