1 Advanced Multidisciplinary System Engineering or “How I learned to think outside of MY box!” Dr. Joseph R. Guerci Director DARPA/SPO [email protected] All material cleared for Public Release
Oct 06, 2020
1
Advanced Multidisciplinary System Engineeringor
“How I learned to think outside of MY box!”
Dr. Joseph R. GuerciDirector
DARPA/[email protected]
All material cleared for Public Release
2
Outline
• Breakthrough systems/technologies are almost always multidisciplinary– Arise from cross-fertilization– “Cross-fertilization” occurs in someone’s mind
• “Thinking outside the box” = “Thinking outside your box”– Examples:
• KASSPER• HISS
• New Trend in Multidisciplinary Systems Engineering– Level 1: System = Interconnected set of single-purpose subsystems– Level 2: System = Interconnected set of multi-purpose subsystems– Level 3: System = Embedded multi-purpose subsystems w/o clear
boundaries
• Example: ISIS• Summary
Sample SPO Projects(A Multidisciplinary Systems Technology Office)
PRODUCTIAR
IRSGPRODUCT
IDAPathogen DNA DNA Polymerase
5’3’
5’3’
5’3’
5’3’
PRODUCT
Pathogen RNA
Nicking Enzyme
RNAPolymerase
RNAPolymerase
ToxinPRODUCTIAR
IRSGPRODUCT
IDAPathogen DNA DNA Polymerase
5’3’
5’3’ 5’3’
5’3’
5’3’
PRODUCT
Pathogen RNA
Nicking Enzyme
RNAPolymerase
RNAPolymerase
Toxin
Next Generation Chem/Bio Sensors & Protection
Advanced Intelligent Signal Processing &Embedded Systems
Revolutionary Space and Near-Space Technologies
3
mD Adaptive Signal Processing
• Example: Space-Time Adaptive Processing (STAP)
Space-Time Adaptive Beamformer
“Ideal” Adapted Pattern
Optimum Solution
• Weiner-Hopf
1R−=w s(Optimum space-time beamformer weights)(Desired signal “steering vector”)
(Inverse of total interference covariance matrix)
, NM
NM NM
CR C ×
∈
∈
w s
~ 10' 100'NM s s−
4
Covariance Estimation Problem
• Practical implementation example and real data example (White Sands DARPA Mountain Top Radar)
Sample Covariance Estimation Measured Data
ˆi i
iR
∈Ω
′= ∑x x
Ideal (Stationary) Data
• Heterogeneous Clutter– Rapidly varying terrain
• Mountainous (rapid elevation/reflectivity variation)• Rapid land cover variations (e.g., littoral)
• Dense “Target” Backgrounds– “Moving Clutter”
• Military/civilian vehicles
• Large Discretes and “Spiky” Clutter– Urban clutter– Power lines, towers, steep mountainous terrain
• Range-Varying (Nonstationary) Clutter Loci– Bi/Multistatics– Nonlinear array geometries (e.g., circular arrays)
Welcome to the Real-World!
Extremely suboptimal radar performance can occur if one or more of the following occurs: (High false alarm rates and/or low Pd)
One or More of the Above is Almost Always Present in Real-World Ops!
5
Serious Performance Impacts!!(KASSPER ’02 Data Cube & APTI Data Set)
x distance (km) −> (longitude)
y d
ista
nc
e (
km
) −
> (
lati
tud
e) Rx
(35.73°,118.5°)0 20 40 60 80 100
0
10
20
30
40
50
Doppler (Fraction PRF)
Rang
e Bi
n #
GMTI Range−Doppler Data (dB−thermal)
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
10
20
30
40
50
60
−50 0 50 100 150−70
−60
−50
−40
−30
−20
−10
0
SIN
R/S
NR
o (
dB
)
Doppler (m/s)
rang bin 240 (38.6km)
optimalPCI−40MWF−40post−Doppler (3 bin)
SINR Loss High False Alarm Rates
−10 −5 0 5 10 15 20 25 3010
−5
10−4
10−3
10−2
10−1
100
Pixel SINR (dB)
Frac
tion
Exc
eedi
ng V
alue
AMF Exceedance
STAP Only STAP w/ Pre−Whitening
Radar Environmental Knowledge Bases(DTED/DFAD/LCLU, SAR, etc.)
Knowledge-Aided Sensor Signal Processing & Expert Reasoning (KASSPER)
Clutter Knowledge Base
ICCR iiiKA22
CellsClutter Over Sum
σγ +∑ ′=
iγClutter Cell Returns
GPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
iCClutter Steering Vectors
Clutter Knowledge Base
ICCR iiiKA22
CellsClutter Over Sum
σγ +∑ ′=
iγClutter Cell Returns
GPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
GPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
GPS/INSGPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
iCClutter Steering Vectors
-60
-50
-40
-30
-20
-10
0
10
20
-0.5 0 0.5
-0.5
0
0.5-60
-50
-40
-30
-20
-10
0
10
20
-0.5 0 0.5
-0.5
0
0.5
XNonstationary Clutter
(plus Signal)
XRY KA21−=
21−
KAR 21ˆ −
SMIR
Reduced-RankConventional FilterKA Pre-Filter
YRZ SMI21ˆ −=
Detector
1st StageKnowledge-Aided
Pre-Filter Response
2nd StageConventional
Filter
KASSPERClutter Knowledge Base
ICCR iiiKA22
CellsClutter Over Sum
σγ +∑ ′=
iγClutter Cell Returns
GPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
iCClutter Steering Vectors
Clutter Knowledge Base
ICCR iiiKA22
CellsClutter Over Sum
σγ +∑ ′=
iγClutter Cell Returns
GPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
GPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
GPS/INSGPS/INS
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
N Elements
[ x1 x2 x3 • • • xM ]
.
.
.
T TM Pulses
. . . .
.
.
T T
M Pulses
Array Snapshots
Sensor Characteristics
iCClutter Steering Vectors
-60
-50
-40
-30
-20
-10
0
10
20
-0.5 0 0.5
-0.5
0
0.5-60
-50
-40
-30
-20
-10
0
10
20
-0.5 0 0.5
-0.5
0
0.5
XNonstationary Clutter
(plus Signal)
XRY KA21−=
21−
KAR 21ˆ −
SMIR
Reduced-RankConventional FilterKA Pre-Filter
YRZ SMI21ˆ −=
Detector
1st StageKnowledge-Aided
Pre-Filter Response
2nd StageConventional
Filter
KASSPER
Measured(DARPA Mtn Top)
Predicted(DTED Level-1)
Ran
ge
Doppler
Bald Earth
1980
Physical
20001980
HPEC
Real-TimeDatabase
EM ModelingTools
PhysicalDatabases
6
CACFARAGC, etc.
IF SidelobeCanceler
Fully AdaptiveArray
Space-TimeAdaptive (STAP)
Radar
Advanced and Real-Time
STAP50’s 60’s 70’s 80’s 90’s
Reinventing Adaptive Radar
First Gen Statistical Signal Processing
KASSPER
Intelligent Adaptive Radars“Real-world nonstationarity does NOT
support conventional adaptivity”
00’s 10’s
Real-time
knowledge-aided
KASSPER
Classic
Savant
FLOPS/Throughput
Knowledge
Data type/MBytes
High-speed, single function
Multi-function, slow access
speeds
True “Intelligent” Processing
+
SAR Roads VMAP Discrete
Radar returns
Old
New
∑ ′=Ω
kkiR xxˆ
Space-TimeSnapshot
Vector
RangeCells
Test Cell
“Guard”Cells
2
1
1
2
−
−
+
+
i
i
i
i
i
xxxxx
...
...
Ω
Ω
ConventionalSpace-Time Filtering
sw 1−= R
QR Factorization w/ Back substitution(from Antenna-Based Signal Processing Techniques for Radar, A. Farina, Artech House)
Highly Parallel Systolic Array Implementation(Achieves 100’s to 1000’s of GFLOPS)
KASSPER HPEC Challenge:Optimizing adaptation by injecting environmental knowledge “intelligently” into the front-end signal flow
First Gen Real-Time KASSPER HPEC
ClutterKnowledge
Base
IntelligentSignal
Processing
• KASSPER requires memory access interrupts
• Optimal interrupt scheduling
• Optimized ISP• “Look-Ahead” scheduling
• KASSPER requires memory access interrupts
• Optimal interrupt scheduling
• Optimized ISP• “Look-Ahead” scheduling
Conventional vs. KASSPER HPEC Processing
7
“Look-Ahead” Scheduling Addresses Memory Latency Issues
CPU Registers
Registers
FasterSpeedHigherCost
StagingTransfer Unit
100s Bytes<1s ns
Cache10s-100s K Bytes1-10 ns
Source: Dave Patterson, Graduate Computer Architecture Course, University of California, Berkeley, Spring, 2001
Disk10s G Bytes10 ms
TapeInfinitesec-min
Main MemoryM Bytes100-300 ns
Cache
Memory
Disk
Tape
Instr. Operands
Blocks
Pages
Files
Prog./Compiler 1-8 Bytes
Cache Controller 8-128 Bytes
OS512-4K Bytes
User/Operator MBytes Larger
SizeLowerCost
CapacityAccess Time
CPU Registers
Registers
FasterSpeedHigherCost
StagingTransfer Unit
100s Bytes<1s ns
Cache10s-100s K Bytes1-10 ns
Source: Dave Patterson, Graduate Computer Architecture Course, University of California, Berkeley, Spring, 2001
Disk10s G Bytes10 ms
TapeInfinitesec-min
Main MemoryM Bytes100-300 ns
Cache
Memory
Disk
Tape
Instr. Operands
Blocks
Pages
Files
Prog./Compiler 1-8 Bytes
Cache Controller 8-128 Bytes
OS512-4K Bytes
User/Operator MBytes Larger
SizeLowerCost
CapacityAccess Time
Problem:KASSPER
“Look-Ahead”Interrupt Scheduling
ttt ∆+
ClutterKnowledge
Base
Predictor
Solution:
Next-Gen KASSPER HPEC Testbed
• Architecture:– Base computer and I/O cards
purchase order completed– Lab computer configuration complete– Various processing concepts in review– PDR planned for late June 03– Demonstration at DARPATech 04
• Parallel Vector Library (PVL) chosen for open standards programming language
– LL reviewing initial KASSPER algorithms for library impacts
– Coding started on basic radar signal processing components (pulse compression, data retrieval, etc.)
– Algorithm developers will program the hardware Vendor
Hardware
PortableLibrary
Maps
ApplicationCode
VendorSoftware
Open standards for real-time processing
MP-510 mercury processing
Multiple high-speed RAID drives
ASIC high-speed cache memory devices
VendorHardware
ApplicationCode
VendorSoftware
• Upgrades restricted to hardware remapping & new features
8
Pre-filtering Followed by Conventional STAP
0
5
10
15
20
25
30
Doppler (Fraction PRF)
Ran
ge
Bin
#
GMTI AMF Output (dB−thermal)
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
10
20
30
40
50
60
0
5
10
15
20
25
30
Doppler (Fraction PRF)
Ran
ge
Bin
#
GMTI AMF Output after Whitening (dB−thermal)
−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4
10
20
30
40
50
60
−10 −5 0 5 10 15 20 25 3010
−5
10−4
10−3
10−2
10−1
100
Pixel SINR (dB)
Fra
ctio
n E
xcee
din
g V
alu
e
AMF Exceedance
STAP Only STAP w/ Pre−Whitening
Adaptive Matched Filter
With Prefiltering
Without Prefiltering
Better Behaved “Tail”
Pre-Filtering Reduces The “Tail” of the Exceedance Function
13 dB!
KASSPER:“It’s an Architecture, NOT an Algorithm”
KASSPER is an architecture for real-time adaptation of multidimensional sensor systems in real-world environments
KASSPER is an architecture for real-time adaptation of multidimensional sensor systems in real-world environments
• KASSPER Architecture– Environmental context is key to efficient adaptation
• Sensors, like humans, benefit from context!– Key enablers: “look-ahead” scheduling and resource allocation – Multiresolution philosophy: blurring the boundaries between SAR
and GMTI– KASSPER as a modern manifestation of the “Bayesian” method!
• KA-STAP Bayesian STAP• The DARPA KASSPER Challenge: Creatively explore the
possibilities– Re-examine entire adaptive signal processing paradigm with an eye
towards maximizing knowledge-aided “robust” methods– Robust STAP algorithms AND KASSPER architecture
• Environmental knowledge base as “read/write” scratch memory– What is “implementable”? 2010? 2020?– Environmentally aware sensors have a future!
• KASSPER Architecture– Environmental context is key to efficient adaptation
• Sensors, like humans, benefit from context!– Key enablers: “look-ahead” scheduling and resource allocation – Multiresolution philosophy: blurring the boundaries between SAR
and GMTI– KASSPER as a modern manifestation of the “Bayesian” method!
• KA-STAP Bayesian STAP• The DARPA KASSPER Challenge: Creatively explore the
possibilities– Re-examine entire adaptive signal processing paradigm with an eye
towards maximizing knowledge-aided “robust” methods– Robust STAP algorithms AND KASSPER architecture
• Environmental knowledge base as “read/write” scratch memory– What is “implementable”? 2010? 2020?– Environmentally aware sensors have a future!
9
Emerging Field
• Special Issue of IEEE Signal Processing Magazine
Handheld Isothermal Silver Standard Sensor(HISSS)
The goal of the HISSS program is to develop a handheld sensor that is capable of identifying biological threats including bacteria, viruses and toxins.
Polymerase Chain Reaction (PCR) Machine Notional Sensor
DNA detection
RNA detection
Protein detection
Fluid handling
DNA readout
RNA readoutProtein readout
System check
Notional Sample Cartridge
How to shrink into a handheld?• Order-of-mag faster!• At least as accurate!
10
PCR vs. Isothermal
∆t ~ 60 sec
Anneal at 55ºC
Starting the process:Primers PolymerasePathogen DNA 5’
3’
Products: copies of Pathogen DNA5’3’
Extendat 72ºC
Denature at 95ºC
5’
5’3’
3’
3’
5’5’
3’5’5’
3’
5’
5’3’
Polymerase Chain Reaction
3’ 5’
3’ 5’
3’
Cleave5’3’
Product falls off5’3’
Polymerase re-binds
5’3’
5’
Extend
3’
Products: copies of reporter
Nicking enzyme
Starting the process:
Polymerase
Trigger template
Pathogen DNAIsothermal
∆t ~ 3 sec
HISS DNA Amplification
11
HISSS Progress
• Progress:– Demonstrated false alarm rates, using ROC curve analysis for HISSS assays that
are equal to or better than current DNA, RNA, and protein assays– Successfully developed and utilized a flow-through testbed to test all assays
0
0.2
0.4
0.6
0.8
1
0.01 0.1 10.10.01 1Pfa
DNA ROC Curves
0.8
0.6
0.4
0.2
0.0
1.0
Pd
PI (1:99)PII (1:99)PCR (1:99)
0
0.2
0.4
0.6
0.8
1
0.01 0.1 10.10.01 1Pfa
DNA ROC Curves
0.8
0.6
0.4
0.2
0.0
1.0
Pd
PI (1:99)PII (1:99)PCR (1:99)
PI (1:99)PII (1:99)PCR (1:99)
0
0.2
0.4
0.6
0.8
1
0.001 0.01 0.1 10.001 0.10.01 1
0.8
0.6
0.4
0.2
0.0
1.0
Pd
Pfa
RNA ROC Curves
PI (1:82)PII (1:82)RT-PCR (1:82)
0
0.2
0.4
0.6
0.8
1
0.001 0.01 0.1 10.001 0.10.01 1
0.8
0.6
0.4
0.2
0.0
1.0
Pd
Pfa
RNA ROC Curves
PI (1:82)PII (1:82)RT-PCR (1:82)
PI (1:82)PII (1:82)RT-PCR (1:82)
0
0.2
0.4
0.6
0.8
1
0.001 0.01 0.1 1
Pd
0.01 10.10.001Pfa
0.8
0.6
0.4
0.2
0.0
1.0Protein Toxin ROC Curves
PI (1:3000)PII (1:3000)ELISA (1:3000)
0
0.2
0.4
0.6
0.8
1
0.001 0.01 0.1 1
Pd
0.01 10.10.001Pfa
0.8
0.6
0.4
0.2
0.0
1.0Protein Toxin ROC Curves
PI (1:3000)PII (1:3000)ELISA (1:3000)
PI (1:3000)PII (1:3000)ELISA (1:3000)
PI Static (1:99)PII Flow (1:99) PCR (1:99)
PI Static (1:82)PII Flow (1:82) RT-PCR (1:82)
PI Static (1:3000)PII Flow (1:3000) ELISA (1:3000)
New Airship Design Philosophy
MDA Airship
Payload bay
Conventional Airship
Capability cannot be added to airship after development
Payload: ~2% of system mass
ISIS requires integration of sensor and airship
Payload: 30-40% of system mass
Turn a disadvantage (large size) into an advantage (large antenna)!
12
The “First” ISIS?
Echo 1
Simultaneous AMTI/GMTI Operation via Dual Band (UHF/X-Band) Aperture
Most Powerful Airborne GMTI/AMTI Radar & Comms Ever Conceived
Long-range AMTI/GMTI/COMM
FOPEN GMTI
Cruise Missile Defense
Steep Grazing AnglesDetect/Track Dismounts
Extremely High Capacity CommsNear Zero Platform Speed
No In-Theater Ground Support – 99% on station availability for 1+ years600km radar horizon at 70kft operational altitude
13
ISIS
Joint STARSJoint STARS
AWACSAWACS
Global HawkGlobal HawkGlobal Hawk
109108107106105104103102100 101 109108107106105104103102100 101
160,000300,000,000
1.0 Relative Search Capability (PA/R2)1.0 Relative Track Capability (PA2/λ2/R4)
2405,100
3,30015,000
ISAT 140610
VHFX
SS
XX
XX
Unprecedented Radar Performance
Platform Carries the Antenna
Antenna Is the Platform
Sustained Operations Logistics
• Aircraft-based ISR Requires– Local air base– Multiple aircraft to keep 1 flying – Air crews– Ground crews– Fuel supplies– Maintenance facilities
• ISIS – Unmanned– Deploys worldwide from U.S.
base– Regenerative Fuel Sources– One-year continuous ISR
capability
14
Wind Conditions DrivePropulsion Power Needs
ηρ
2
33/2 vVCP d⋅
=Where ρ = air density at altitude
V = volume of airshipv = relative velocity of airη = efficiency of propellers
Propulsion Power for V = 106 m3
(Cd=0.022)
0500
1,0001,5002,0002,500
0 10 20 30 40 50
Wind Speed (m/sec)
Pow
er R
equi
red
(kw
)
Propulsion Power for V = 106 m3
(Cd=0.022)
0500
1,0001,5002,0002,500
0 10 20 30 40 50
Wind Speed (m/sec)
Pow
er R
equi
red
(kw
)
44.95 m/s
Max winds drive powersystem requirements
Station Keeping
• ISIS Objective: 99% on-station availability for 1 year– Function of airship speed (sustained and sprint) and available
energy (regenerative and stored fuel)
• Need operational algorithms for maximizing availability– Managing airship energy ala satellite delta-v
-90-80-70-60-50-40-30-20-10
0102030405060708090
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
Wind Speed (m/s)
Latit
ude
(deg
ress
)
Mean Wind Speed Average 99 PercentileMean Wind Speed Average 99 Percentile
Maxim
um S
print Speed
15
Requires Large Mass Reductions
Mass
VolumePo
wer
Mass
VolumePo
wer
MV ∝
3/2VP ∝
PM ∝ • ISIS designs are mass-centric– Lifting gas has reached the maximum limit: – 0.061kg per 1m3 of He @ 21km– 0.066kg per 1m3 of H2 @ 21km
• ISIS focusing on:– Removing mass from largest contributors– Integration, INTEGRATION, INTEGRATION!
avionicspropulsionpropulsion
dairradar
power
poweraperturehullhgasISIS
avionicspropulsionpowerradarstructuregasliftingairdisplaced
MMvVCPAVcVM
MMMMMMM
++⎟⎟⎠
⎞⎜⎜⎝
⎛++++=
+++++=
ηρ
ηρ
ρρρ2
33/23/2
Integration
Components
Summary
• Breakthrough systems/technologies are almost always multidisciplinary
– System engineers need to be continually learning about new technologies and methods across ALL disciplines
• “Be an annoying know-it-all!”– Tactic: “Can the thermal engineer give the flight control engineer’s
briefing?”– Often “cross fertilization” can occur even if with only a 1st or 2nd
order understanding of multiple disciplines• Balance of depth and breadth
– How should engineering programs be structured in light of above?• Undergraduate programs typically have the breadth, but don’t seem to
“close the deal”– Example: Senior class semester devoted to dissecting a complex system
• Emergence of a “Level 3” systems integration– Multidisciplinary from its inception!