evolvable systems for space applications · 2003. 11. 24. · comparison: pattern analysis for...

Jason D. Lohn

Evolvable Systems GroupComputational Sciences Division

NASA Ames Research CenterMountain View, CA USA

Jason D. Jason D. LohnLohn

Evolvable Systems GroupEvolvable Systems GroupComputational Sciences DivisionComputational Sciences Division

NASA Ames Research CenterNASA Ames Research CenterMountain View, CA USAMountain View, CA USA

Evolvable Systems for

Space Applications

Evolvable Systems Evolvable Systems forfor

Space ApplicationsSpace Applications

Stanford, CS426/BMI226November 24, 2003

Ames Research CenterAmes Research Center

Outline

• NASA Interest in Evolvable Systems

• Example Applications• Self-repairing FPGAs• Circuit Design• Antenna Design

• Future & Conclusion


The NASA Mission

To understand and protect our home planet

To explore the Universe and search for life

To inspire the next generation of explorers


What Are Evolvable Systems?

A system that changes over time to become better…

Changing its shape, its function, being automatically designed…

All under the guidance of a process derived from biological evolution.


Examples of Evolvable Systems

PlanetaryExploration

CoordinationAmong Modular

RoboticStructures

CoordinationCoordinationAmong ModularAmong Modular

RoboticRoboticStructuresStructures

Evolved ControllerCircuits EnableMobile Drilling

Evolved ControllerCircuits EnableMobile Drilling

Defect TolerantNanosystems

Defect TolerantDefect TolerantNanosystems Nanosystems

Automated Design with New Devices

DistributedRoboticControl

Evolutionary and Adaptive Algorithms

ElectronicsSurvive Extreme

Radiation, Temperature

ElectronicsElectronicsSurvive Survive ExtremeExtreme

Radiation, Radiation, Temperature Temperature

AutomaticFault Recovery


NASA Interest in Evolvable Systems

Abundance of:• Optimization problems• Design problems• Applications requiring automation, fault recovery,

control

Agency-wide:• Many projects using EAs• Ames & JPL each have groups dedicated to evolvable

systems


Human-Competitive Performance

• Evolutionary Algorithms are beginning to design complex engineering structures at the level of human experts

• John Koza’s list of human-competitive performance (many infringe on patents)

Automated Antenna Design

CoevolutionaryAlgorithms

Evolutionary Scheduling for Satellite

Fleets

Fault Recovery for Reconfigurable

Systems

Automated Circuit Design


Evolutionary Design Engine

ParallelSearch

Algorithm

Best Solution Found

outin simulator

GA • GP • Hybrid • Coevolution • AnnealingMemetic • Heuristic • Traditional • Others

• Evolutionary Design Engine:– C/C++– Easy to “drop-in” new

application areas– Parallel, any Unix – Integrated native code for

NEC4 antenna simulator– Over 10,000 lines of code

• Beowulf cluster:– simulation farmed out to

diskless nodes (currently 120 cpus)

– low bandwidth communication


Fault Recovery ForReconfigurable Systems

(1 Digital,1 Analog)


Fault Recovery

EvolutionaryAlgorithm

• Description:– Fault tolerance / self-repair in

extreme environments • High temperature• High radiation

• Output: adaptive algorithms for autonomous self-repair of re-programmable logic chips

radiation

damaged logicboard

damaged FPGA

damaged regiondetected by

parity/checksum

evolved solutionmakes use of

healthy resources

evolutionaryalgorithm

Dynamic Evolution for Fault Tolerance

chip reprogrammed by algorithm


Quadrature Decoder

• Applications requiring determination of angular translation (or speed)• Example: DC-motor to drive system for a mobile robot we may wish to

move forward (or reverse) by a fixed distance• Decoder determines rotation direction


Quadrature Decoder• Finite state machine

Input and State traces

State transition table


Genetic Representation

• Logic bits in the LUTs• Routing bits specify how to connect LUT outputs to

LUT inputs

LUT 0 BITS

R-CLB = REMOTE CLB

R-LUTR-CLB

R-LUT = REMOTE LUT

R-LUTR-CLB...LUT 0 INPUTS

... R-LUTR-CLB R-LUTR-CLB...LUT 3 INPUTS

LUT 3 BITS ...CLB 0 CLB 1

LUT0

LUT1

LUT2

LUT3

• • •LUT

0

LUT1

LUT2

LUT3

LUT0

LUT1

LUT2

LUT3

CLB 0 CLB 1 CLB n


Results

Evolved Quad Decoder Configuration


Hardware Evolution

Hardware Evolution

• 20 random latchup faults are simulated• Quad decoder is repaired automatically in minutes• FPGA chip is reprogrammed about 5 times per second – we

will soon go much faster

Celoxica RC1000 boardXilinx Virtex 1000e


Hardware Setup

Virtex FPGA


Programmable Transistor Array Cell – FPTA2

Interconnection amongst cells

Input(96)

Output(64)

N

Data bus D0-15(Configuring bitstring)

E W

S

Control Logic

Address bus(9 bits)

Cell SchematicChip Architecture

• TSMC 0.18micron 1.8v; Area 5mmX 7mm; 64 reconfigurable cells; 96 analog in, 64 analog out• 256 pins; 16bits data bus/9bits address bus; 5000 programming bits• Each cell OpAmp level


Extreme Environment integrating FPTA-2 and SABLES

AirTorch

Ceramic

High Temperature BoardCeramic

Chip Die

0.5”

0.875”

0.5”

0.625 “

Align AirTorch and ceramic hole with laser

Air flow of 4cfm

ControlsDSP

GA runs here

DAC

ADC

ConnectorTo

Testbed

Analog Array

Digital Interface

FPTA-2

High Temperature Board

SABLESstimulation

response

digital

FPTA Under test

ConnectorTo Testbed

DSP


Recovery of Half-wave rectifier at 280C

• Reconfigurable Cell (FPTA-2)– 2 cells: 0,1– Input: 1 Analog Input (0.9 Volt amplitude – 2 kHz)

• Evolution Mechanism:– Goal: Synthesize Half-way rectifier at 280oC. – Constrained evolution to Cell0 and Cell1– Fitness Function: Absolute error to the target;

• Genetic Algorithm– 400 individuals– Chromosome: 160 bits– less than 50 generations – 1 min

0 3

1 2

12 15

13 14

4 7 8 11

5 6 9 10

IN1

Output

Input

O1

IN1

FPTA-2Input

Output


Recovery of Halfwave Rectifierat 280C

Input

Output

Temperature above 50C:Degradation of response

Output

Input

Output

Input

Temperature 27C:Obtained by evolution

Recovery by Evolution of Halfway rectifier:• 2 Cells of FPTA2 used by evolution• Input Freq.: 2 kHz• Input Amplitude: 0.9 Volt Temperature 280C:

Recovery by Evolution


Automated Circuit Design


Circuit Construction

• Template circuit

• Set of selectable components:


Circuit Representation

• a small circuit construction language is used• instructions involve placing components end-to-end

(laying a trail)• examples:

– place_resistor(144.92, new_node)– place_transistor(ground, new_node, previous_node)– place_capacitor(183e-6, ground)


From Representation toFitness Calculation

circuitdesign

geneticrepresentation

SPICEcircuit

simulator

circuit-constructingprogram SPICE

netlist FitnessCalculation


Circuit Design Application

Design and optimization

evolved circuit


Evolved 85 dB Amplifier

Evolved Circuit-Constructing Programtransistor(N, ACTIVE_NODE, NEW_NODE, INPUT_NODE);transistor(N, BASE, ACTIVE_NODE, PREVIOUS_NODE);resistor_cast_to_ps(4.618467e+04);capacitor_cast_to_input(1.628423e-04);transistor(N, NEW_NODE, ACTIVE_NODE, GROUND_NODE);resistor_cast_to_ps(9.396477e+04);transistor(N, NEW_NODE, ACTIVE_NODE, GROUND_NODE);transistor(P, NEW_NODE, ACTIVE_NODE, PS_NODE);transistor(N, NEW_NODE, ACTIVE_NODE, PS_NODE);transistor(N, PS_NODE, ACTIVE_NODE, NEW_NODE);resistor_move_to_output(1e-06);


Automated Antenna Design

and Optimization


ST5 Mission

• New Millennium Program mission

• Three nanosats• Measure effect of solar

activity on the Earth's magnetosphere


ST5 Antenna Requirements• Transmit Frequency: 8470 MHz• Receive Frequency: 7209.125 MHz• Antenna RF Input: 1.5W = 1.76 dBW = 31.76 dBm• VSWR: < 1.2 : 1 at the antenna input port at Transmit Freq, < 1.5 : 1 at the antenna input port

at Receive Freq• Antenna Gain Pattern: Shall be 0 dBic or greater for angles 40 <= theta <=80; 0 <= phi <= 360• Antenna pattern gain (this shall be obtained with the antenna mounted on the ST5 mock-up) shall

be 0.0 dBic (relative to anisotropic circularly polarized reference) for angles 40 <= theta <=80; 0 <= phi <= 360, where theta and phi are the standard spherical coordinate angles as defined in the IEEE Standard Test Procedures for Antennas, with theta=0 to direction perpendicular to the spacecraft top deck. The antenna gain shall be measured in reference to a right hand circular polarized sense (TBR).

• Desired: 0 dBic for theta = 40, 2 dBic for theta = 80, 4 dBic for theta = 90, for 0 <= phi <= 360• Antenna Input Impedance: 50 ohms at the antenna input port• Magnetic dipole moment: < 60 mA-cm^2• Grounding: Cable shields of all coaxial inputs and outputs shall be returned to RF ground at the

transponder system chasis. The cases of all comm units will be electrically isolated from the mounting surface to prohibit current flow to the spacecraft baseplate.

• Antenna Size: diameter: < 15.24 cm (6 inches), height: < 15.24 cm (6 inches)• Antenna Mass: < 165 g


ST5 Quadrifilar Helical Antenna


Antenna Genotype

• Genotype specifies design of 1 arm in 3-space• Genotype is tree-structured computer program that

builds a wire form

• Commands:• forward(length radius)• rotate_x(angle)• rotate_y(angle)• rotate_z(angle)

• Branching in genotype branching in wire form

rx f

f

f f

rz rx

f

2.5cm

5.0cm

Feed Wire


Antenna Genotype

• Forward commands can have 0,1,2, or 3 children• Rotate commands have exactly 1 child (always non-

terminal)

rx f

f

f f

rz rx

f


EA Parameters

• Generational• Rank-based selection• Elitism of 2• Population=200 (seeded w/individuals from previous

run)• Crossover: 1pt, 50%• Mutation: 50%:

– Mutate a constant,– Mutate a function (terminals remain terminals),or– Add/delete node (for adds: terminals remain terminals)

• Crossover, mutation mutually exclusive


Fitness Function• Antenna designs are evaluated by NEC4 running on Linux

Beowulf supercomputer• 3 copies w/added noise are evaluated for each design• Fitness function (to be minimized):

F = VSWR_Score * Gain_Score * Penalty_Score• VSWRs from both freqs are

scaled and multiplied

• Gain is sampled at 5 degree increments between theta=40 and theta=90

f = 0 if gain > 0.5 dBf = 0.5 – gain if gain < 0.5 dB

• Penalty: proportional to # gain samples less than 0.01 dB

VSWR


Fitness Function: VSWR• voltage standing wave ratio: quantifies reflected-wave interference • the ratio between the highest voltage and the lowest voltage inthe signal envelope along a transmission line


Fitness Function: Gain


Evolved Antennas

ST5-3-10 ST5-4W-03


Antenna for ST5 Mission


VSWR for Antenna ST5-3-10

Data measured at GSFC on prototype ST5-3-10, Build LIR-2003-02-08


Gain for Antenna ST5-3-10



Prediction vs Measurement forAntenna ST5-3-10

Phi = 0 degreesNEC4 Simulator


Prediction vs Measurement for deck 3.10 at 7.2 GHz

-50.00

-45.00

-40.00

-35.00

-30.00

-25.00

-20.00

-15.00

-10.00

-5.00

0.00

5.00

10.00

-180

-160

-140

-120

-100

-80

-60

-40

-20

0 20 40 60 80 100

120

140

160

180

Elevation (Degrees)

Gai

n (d

Bic

)

Prediction

Measurement

Prediction vs Measurement for deck 3.10 at 8.47 GHz

-50.00

-45.00

-40.00

-35.00

-30.00

-25.00

-20.00

-15.00

-10.00

-5.00

0.00

5.00

10.00

-180

-160

-140

-120

-100

-80

-60

-40

-20

0 20 40 60 80 100

120

140

160

180

Elevation (Degrees)G

ain

(dB

ic)

PredictionMeasurement


Comparison: Pattern Analysisfor Antennas @ 7.2 GHz

Evolved Antenna Conventional Antenna

Max and Min Gain vs Theta for 7.2 GHz

-40

-35

-30

-25

-20

-15

-10

-5

0

5

10

0 10 20 30 40 50 60 70 80 90

Theta (deg)

Gai

n Max

Min

Shaded Yellow Box Denotes Area In-Spec, According to Original Mission Requirements


Comparison: Pattern Analysisfor Antennas @ 8.47 GHz

Evolved Antenna Conventional Antenna

Max and Min Gain vs Theta for 8.47 GHz

-35

-30

-25

-20

-15

-10

-5

0

5

10

0 10 20 30 40 50 60 70 80 90

Theta (deg)

Gai

n Max

Min

Shaded Yellow Box Denotes Area In-Spec, According to Original Mission Requirements


-30

00.511.522.533.544.555.566.577.588.599.51010.51111.51212.5 1313.5 1414.5 15 15.5 16 16.5 17 17.5 18 18.5 19 19.5 20 20.5 21 21.5 22 22.5 23 23.5 24 24.525 25.526 26.527 27.52828.52929.53030.53131.53232.53333.53434.53535.53636.53737.53838.53939.54040.54141.54242.54343.54444.54545.54646.54747.54848.54949.55050.55151.55252.55353.55454.55555.55656.55757.55858.55959.56060.56161.56262.56363.56464.56565.56666.56767.56868.56969.57070.57171.57272.57373.57474.57575.57676.57777.57878.57979.58080.58181.58282.58383.58484.58585.58686.58787.58888.58989.59090.59191.59292.59393.59494.59595.59696.59797.59898.59999.5100100.5101101.5102102.5103103.5104104.5105105.5106106.5107107.5108108.5109109.5110110.5111111.5112112.5113113.5114114.5115115.5116116.5117117.5118118.5119119.5120120.5121121.5122122.5123123.5124124.5125125.5126126.5127127.5128128.5129129.5130130.5131131.5132132.5133133.5134134.5135135.5136136.5137137.5138138.5139139.5140140.5141141.5142142.5143143.5144144.5145145.5146146.5147147.5148148.5149149.5150150.5151151.5152152.5153153.5154154.5155 155.5156 156.5 157 157.5 158 158.5 159 159.5 160 160.5 161 161.5 162 162.5 163 163.5 164 164.5 165 165.5 166 166.5 167167.5 168168.5169169.5170170.5171171.5172172.5173173.5174174.5175175.5176176.5177177.5178178.5179

179 5180-179.5-179-178.5-178-177.5-177-176.5-176-175.5-175-174.5-174-173.5-173-172.5-172-171.5-171-170.5-170-169.5-169-168.5-168-167.5-167-166.5-166-165.5-165-164.5-164-163.5-163-162.5-162-161.5-161-160.5-160-159.5-159-158.5-158-157.5-157-156.5-156-155.5-155-154.5-154-153.5-153-152.5-152-151.5-151-150.5-150-149.5-149-148.5-148-147.5-147-146.5-146-145.5-145-144.5-144-143.5-143-142.5-142-141.5-141-140.5-140-139.5-139-138.5-138-137.5-137-136.5-136-135.5-135-134.5-134-133.5-133-132.5-132-131.5-131-130.5-130-129.5-129-128.5-128-127.5-127-126.5-126-125.5-125-124.5-124-123.5-123-122.5-122-121.5-121-120.5-120-119.5-119-118.5-118-117.5-117-116.5-116-115.5-115-114.5-114-113.5-113-112.5 -112 -111.5 -111 -110.5 -110 -109.5 -109 -108.5 -108 -107.5 -107 -106.5 -106 -105.5 -105 -104.5 -104 -103.5 -103 -102.5 -102 -101.5 -101 -100.5 -100 -99.5 -99 -98.5 -98 -97.5 -97 -96.5 -96 -95.5 -95 -94.5 -94 -93.5 -93 -92.5 -92 -91.5 -91 -90.5 -90 -89.5 -89 -88.5 -88 -87.5 -87 -86.5 -86 -85.5 -85 -84.5 -84 -83.5 -83 -82.5 -82 -81.5 -81 -80.5 -80 -79.5 -79 -78.5 -78 -77.5 -77 -76.5 -76 -75.5 -75 -74.5 -74 -73.5 -73 -72.5 -72 -71.5 -71 -70.5 -70 -69.5 -69 -68.5 -68 -67.5-67-66.5-66-65.5-65-64.5-64-63.5-63-62.5-62-61.5-61-60.5-60-59.5-59-58.5-58-57.5-57-56.5-56-55.5-55-54.5-54-53.5-53-52.5-52-51.5-51-50.5-50-49.5-49-48.5-48-47.5-47-46.5-46-45.5-45-44.5-44-43.5-43-42.5-42-41.5-41-40.5-40-39.5-39-38.5-38-37.5-37-36.5-36-35.5-35-34.5-34-33.5-33-32.5-32-31.5-31-30.5-30-29.5-29-28.5-28-27.5-27-26.5-26-25.5-25-24.5-24-23.5-23-22.5-22-21.5-21-20.5-20-19.5-19-18.5-18-17.5-17-16.5-16-15.5-15-14.5-14-13.5-13-12.5-12-11.5-11-10.5-10-9.5-9-8.5-8-7.5-7-6.5-6-5.5-5-4.5-4-3.5-3-2.5-2-1.5-1

-0.50

100

20

40

80

60

120

140

160 180 -160

-140

-120

-100

-80

-60

-40

-20 -3

-6

-9-12-15

-18

-21-24

-24

-27-30

Directionality Comparison

Evolved Antenna (red) vs QHA (green)Data is normalized to 0 dBic (RHCP) for each antenna -3 dB for each division


Evolved Antenna Animation

• genetic representation• genotype phenotype mapping• position on satellite


Benefits

• Potential Lower Power: evolved antenna achieves high gain (2-4dB) across a wider range of elevation angles– allows a broader range of angles over which maximum data throughput

can be achieved– may require less power from the solar array and batteries

• More Uniform Coverage: Very uniform pattern with small ripples in the elevations of greatest interest (40 – 80 degrees)– allows for reliable performance as elevation angle relative to the ground

changes• Inexpensive Design Cycle: 3 person-months to run algorithms and

fabricate the first prototype as compared to 5 person-months for contractor team

• On Schedule to be First Evolved Hardware in Space when mission launches in 2004


Future & Conclusion


Future

• Automated design across multiple levels: macro, micro, nano

• New antenna design problems• Evolved nano-devices and architectures• FPGA Self-Repair (real-time/subsecond repair)• MEMS design & manufacturing problems• Ornithopters (macro/micro)• Modular robotic control


Conclusion

• Evolutionary design/optimization:– Fast design cycles save time/money– Fast design cycles allow iterative “what-if”– May discover designs that challenge expert

designers• Which NASA applications area a good fit for EAs?

• Evolved antenna for ST5:– Meets mission requirements– Unusual organic topology– Currently undergoing space qualification

evolvable systems for space applications · 2003. 11. 24. · comparison: pattern analysis for...

Documents