Integration of Sensing & Processing Doug Cochran, Fulton School - - PowerPoint PPT Presentation

▶
integration of sensing processing
SMART_READER_LITE
LIVE PREVIEW

Integration of Sensing & Processing Doug Cochran, Fulton School - - PowerPoint PPT Presentation

Aug 28, 2023 348 likes •709 views

Integration of Sensing & Processing Doug Cochran, Fulton School of Engineering 30 January 2006 ASU, 30 January 2006 Outline 1. Introduction Introduction 1. Traditional sensing system design and operation Traditional sensing system

slide-1
SLIDE 1

ASU, 30 January 2006

Integration of Sensing & Processing

Doug Cochran, Fulton School of Engineering 30 January 2006

slide-2
SLIDE 2

ASU, 30 January 2006

Outline

1.

  • 1. Introduction

Introduction

  • Traditional sensing system design and operation

Traditional sensing system design and operation

  • The integrated sensing & processing vision

The integrated sensing & processing vision 2.

  • 2. Closing the sensing loop

Closing the sensing loop

  • Issues and approaches

Issues and approaches

  • Experiment sequence for target classification

Experiment sequence for target classification

  • Waveform scheduling

Waveform scheduling

  • Coded-aperture sensors

Coded-aperture sensors 3.

  • 3. Processing on the physical layer

Processing on the physical layer

  • Analogue-to-Information conversion

Analogue-to-Information conversion

  • Optical reference structure devices

Optical reference structure devices

  • Combined analog-digital signal processing

Combined analog-digital signal processing

slide-3
SLIDE 3

ASU, 30 January 2006

Outline

1.

  • 1. Introduction

Introduction

  • Traditional sensing system design and operation

Traditional sensing system design and operation

  • The integrated sensing & processing vision

The integrated sensing & processing vision 2.

  • 2. Closing the sensing loop

Closing the sensing loop

  • Issues and approaches

Issues and approaches

  • Experiment sequence for target classification

Experiment sequence for target classification

  • Waveform scheduling

Waveform scheduling

  • Coded-aperture sensors

Coded-aperture sensors 3.

  • 3. Processing on the physical layer

Processing on the physical layer

  • Analogue-to-Information conversion

Analogue-to-Information conversion

  • Optical reference structure devices

Optical reference structure devices

  • Combined analog-digital signal processing

Combined analog-digital signal processing

slide-4
SLIDE 4

ASU, 30 January 2006

Traditional Sensing

Physical Physical Phenomenology Phenomenology

slide-5
SLIDE 5

ASU, 30 January 2006

PHYSICAL LAYER

Digital Representation

finite-precision finite-dimensional

Traditional Sensing

… …101100010111010100… 101100010111010100… 1011000101110101010001010111101010010100101001101010100 1011000101110101010001010111101010010100101001101010100 1010101110000101100100010101010111010100101001001000101 1010101110000101100100010101010111010100101001001000101 0010101010010100101010100101010010101010101010101110101 0010101010010100101010100101010010101010101010101110101 0000110100100010010111110010100001001010101010010110010 0000110100100010010111110010100001001010101010010110010 1011000101110101010001010111101010010100101001101010100 1011000101110101010001010111101010010100101001101010100 1010101110000101100100010101010111010100101001001000101 1010101110000101100100010101010111010100101001001000101 0010101010010100101010100101010010101010101010101110101 0010101010010100101010100101010010101010101010101110101 0000110100100010010111110010100001001010101010010110010 0000110100100010010111110010100001001010101010010110010 1011000101110101010001010111101010010100101001101010100 1011000101110101010001010111101010010100101001101010100 1010101110000101100100010101010111010100101001001000101 1010101110000101100100010101010111010100101001001000101 0010101010010100101010100101010010101010101010101110101 0010101010010100101010100101010010101010101010101110101 0000110100100010010111110010100001001010101010010110010 0000110100100010010111110010100001001010101010010110010 1011000101110101010001010111101010010100101001101010100 1011000101110101010001010111101010010100101001101010100 1010101110000101100100010101010111010100101001001000101 1010101110000101100100010101010111010100101001001000101 0010101010010100101010100101010010101010101010101110101 0010101010010100101010100101010010101010101010101110101 0000110100100010010111110010100001001010101010010110010 0000110100100010010111110010100001001010101010010110010 1011000101110101010001010111101010010100101001101010100 1011000101110101010001010111101010010100101001101010100 1010101110000101100100010101010111010100101001001000101 1010101110000101100100010101010111010100101001001000101 0010101010010100101010100101010010101010101010101110101 0010101010010100101010100101010010101010101010101110101 0000110100100010010111110010100001001010101010010110010 0000110100100010010111110010100001001010101010010110010

slide-6
SLIDE 6

ASU, 30 January 2006

Traditional Sensing

PHYSICAL LAYER

PROCESSING LAYER

Transformed Digital Representation 10110001011101010100010 10110001011101010100010 10101011100001011001000 10101011100001011001000 00101010100101001010101 00101010100101001010101 00001101001000100101111 00001101001000100101111 0111 0111 0101 0101

slide-7
SLIDE 7

ASU, 30 January 2006

Traditional Sensing

PHYSICAL LAYER PROCESSING LAYER

EXPLOITATION LAYER

Symbolic Output 0111 0111 0101 0101   H

H1

1

slide-8
SLIDE 8

ASU, 30 January 2006

Outline

1.

  • 1. Introduction

Introduction

  • Traditional sensing system design and operation

Traditional sensing system design and operation

  • The integrated sensing & processing vision

The integrated sensing & processing vision 2.

  • 2. Closing the sensing loop

Closing the sensing loop

  • Issues and approaches

Issues and approaches

  • Experiment sequence for target classification

Experiment sequence for target classification

  • Waveform scheduling

Waveform scheduling

  • Coded-aperture sensors

Coded-aperture sensors 3.

  • 3. Processing on the physical layer

Processing on the physical layer

  • Analogue-to-Information conversion

Analogue-to-Information conversion

  • Optical reference structure devices

Optical reference structure devices

  • Combined analog-digital signal processing

Combined analog-digital signal processing

slide-9
SLIDE 9

ASU, 30 January 2006

Agile sensing opportunities Agile sensing opportunities

  • Optical:

Optical: e.g., high-speed spatial light e.g., high-speed spatial light modulators, femtosecond pulse-shaped lasers modulators, femtosecond pulse-shaped lasers

  • RF:

RF: e.g., software-driven transmitters & e.g., software-driven transmitters & receivers receivers

  • Acoustic:

Acoustic: e.g., steerable & waveform-agile e.g., steerable & waveform-agile sources sources

  • Configurable networks:

Configurable networks: e.g., deployable motes, e.g., deployable motes, unmanned vehicles unmanned vehicles

  • Tunable materials:

Tunable materials: e.g., e.g., electrically tunable electrically tunable materials, photonic materials, photonic band-gap materials band-gap materials

  • Chemical:

Chemical: e.g., e.g., artificial dogs’ noses artificial dogs’ noses

Integrated Sensing & Processing Vision

slide-10
SLIDE 10

ASU, 30 January 2006

Integrated Sensing & Processing Vision

Hsiao Zhua-Zi Hsiao Zhua-Zi “ “Beast” Beast” 1984-2004 1984-2004

slide-11
SLIDE 11

ASU, 30 January 2006

PHYSICAL LAYER PROCESSING LAYER EXPLOITATION LAYER

Integrated Sensing & Processing Vision

Develop holistic approaches to sensor system design and operation to enable optimal end- to-end performance

1) Supplant currently prevalent feed-forward operational concepts with feedback ideas  Allow back-end exploitation requirements (e.g., target ID) to task front-end sensor elements!!

slide-12
SLIDE 12

ASU, 30 January 2006

Integrated Sensing & Processing Vision

PHYSICAL LAYER PROCESSING LAYER EXPLOITATION LAYER

INTEGRATED INTEGRATED SENSOR/PROCESSOR SENSOR/PROCESSOR

Develop holistic approaches to sensor system design and operation to enable optimal end- to-end performance

1) Replace independent optimization of sensor system components with end-to-end system optimization  Integrate processing and sensing functionality; e.g., “processing on the physical layer”

slide-13
SLIDE 13

ASU, 30 January 2006

DARPA ISP Program

  • Complexity and

volume of raw measurements

  • Increased
  • perational tempo
  • Concepts of
  • peration with

immediate information sharing

  • More flexible

(tunable, mode/waveform selectable, configurable, etc.) sensor elements

Defense and national security sensing systems supporting next- generation reconnaissance, surveillance, and weapon capabilities face dramatically increased demands:

ISP is developing critical enabling methodology for the next generation of sensor/exploitation networks

slide-14
SLIDE 14

ASU, 30 January 2006

Outline

1.

  • 1. Introduction

Introduction

  • Traditional sensing system design and operation

Traditional sensing system design and operation

  • The integrated sensing & processing vision

The integrated sensing & processing vision 2.

  • 2. Closing the sensing loop

Closing the sensing loop

  • Issues and approaches

Issues and approaches

  • Experiment sequence for target classification

Experiment sequence for target classification

  • Waveform scheduling

Waveform scheduling

  • Coded-aperture sensors

Coded-aperture sensors 3.

  • 3. Processing on the physical layer

Processing on the physical layer

  • Analogue-to-Information conversion

Analogue-to-Information conversion

  • Optical reference structure devices

Optical reference structure devices

  • Combined analog-digital signal processing

Combined analog-digital signal processing

slide-15
SLIDE 15

ASU, 30 January 2006

Closing the Loop: Issues

Myopic Perspective:

Get the most out of the next measurement

  • The most what?
  • Requires quantification of exploitation objectives
  • Even single-step propagation of conditional densities is problematic
  • Non-linear
  • Non-Gaussian

Finite Horizon Perspectives

Know as much as possible at some fixed future time Reach a desired confidence level as quickly as possible

… …

  • Myopic issues still apply
  • Combinatorics quickly get out of hand
slide-16
SLIDE 16

ASU, 30 January 2006

Closing the Loop: Approaches

Bayesian Analysis / Embedded Simulation

  • Particle filtering propagates arbitrary quantized conditional

densities through nonlinear systems

  • But it can be slow – particularly in multi-stage problems

Testbed applications:

  • Radar beam dwell

management (MIT Lincoln Lab, General Dynamics)

  • Waveform scheduling

(Melbourne, DSTO)

  • UXO and mine search (Duke)
  • Chemical sensing (JHU)
slide-17
SLIDE 17

ASU, 30 January 2006

Closing the Loop: Approaches

Multi-armed Bandit Formulation

Testbed applications

  • Multi-stage waveform scheduling (Melbourne University, DSTO, CSU)
  • Radar dwell and mode management (Alphatech, Boston University, NRL)
  • NMR probing of macromolecules (Harvard)

Optimal Control Perspective

(Stochastic Dynamic Programming)

Look-ahead Over Short-term Window

Current State

+

Approximate Reward-to-Go

J J ~ ~

  • Gittens index provides multi-stage solution
  • Computing the index has traditionally been

intractable except for small problems

  • Rich theory provides exact optimal solutions to broad

classes of sensor scheduling problems

  • Optimal solutions typically require extensive

memory and communication

slide-18
SLIDE 18

ASU, 30 January 2006

Closing the Loop: Examples

Select Optimal Select Optimal Measurement Measurement Configure & Take Configure & Take Measurement Measurement Resolved Resolved ? ?

Y Y

Done Done

N N Measurement 1 Measurement 1 Measurement 2 Measurement 2 Measurement N Measurement N

Design of experiment Design of experiment sequence for target sequence for target classification classification

slide-19
SLIDE 19

ASU, 30 January 2006

Closing the Loop: Examples

Classification Movie Classification Movie

slide-20
SLIDE 20

ASU, 30 January 2006

Closing the Loop: Examples

Myopic Waveform Scheduling: Performance value University of Melbourne, DSTO

slide-21
SLIDE 21

ASU, 30 January 2006

Closing the Loop: Examples

Non-myopic Waveform Scheduling: Performance value Melbourne, DSTO

Two steps ahead vs. one step ahead Two steps ahead vs. one step ahead waveform scheduling in target tracking waveform scheduling in target tracking example example 1.

  • 1. Position RMSE as a function of

Position RMSE as a function of time time 2.

  • 2. Re-visit count as a function of time

Re-visit count as a function of time

Can we develop rigorously- based heuristics for when multi-stage processing is worth the cost?

slide-22
SLIDE 22

ASU, 30 January 2006

CMOS Base

“on” +10°

  • 10°

“off”

Coded-Aperture Sensing Devices Yale University & FMAH Inc.

  • Digitally controlled light source used as a spectrometer or direct

chemometric analysis system

  • Algorithmically optimizing the illumination spectrum allows

discrimination of materials

Broadband Source Scene Illumination

Closing the Loop: Examples

slide-23
SLIDE 23

ASU, 30 January 2006

Mirror “on” Mirror “off”

Spatial Dimension (Slit Height) Spectral Dimension

Filter A

(larger λ’s, top of slit)

1600 1650 1700 1750 1800 1850 1900

Wavelength (nm) Energy

1600 1650 1700 1750 1800 1850 1900

Wavelength (nm) Energy

Filter B

(medium λ’s, mid-slit)

Filter C

(small λ’s, bottom slit)

Filters A + B + C

1600 1650 1700 1750 1800 1850 1900

Wavelength (nm) Energy

1600 1650 1700 1750 1800 1850 1900

Wavelength (nm) Energy

Spatio-spectral filtering with coded-aperture sensor

Closing the Loop: Examples

slide-24
SLIDE 24

ASU, 30 January 2006

Detection of particular signatures via coded-aperture spectroscopy Detection of particular signatures via coded-aperture spectroscopy White White Light Light Prism Prism Micromirror Micromirror Array Array Analyte Analyte Photodetector Photodetector 0/1 0/1 Decision Decision Rapid digitally- Rapid digitally- controlled “experiment” controlled “experiment” sequence sequence

Closing the Loop: Examples

slide-25
SLIDE 25

ASU, 30 January 2006

Fake vegetation

Coded-aperture sensor: application

Under broadband Under broadband illumination illumination With optimized With optimized spectral discrimination spectral discrimination

Closing the Loop: Examples

slide-26
SLIDE 26

ASU, 30 January 2006

Outline

1.

  • 1. Introduction

Introduction

  • Traditional sensing system design and operation

Traditional sensing system design and operation

  • The integrated sensing & processing vision

The integrated sensing & processing vision 2.

  • 2. Closing the sensing loop

Closing the sensing loop

  • Issues and approaches

Issues and approaches

  • Experiment sequence for target classification

Experiment sequence for target classification

  • Waveform scheduling

Waveform scheduling

  • Coded-aperture sensors

Coded-aperture sensors 3.

  • 3. Processing on the physical layer

Processing on the physical layer

  • Analogue-to-Information conversion

Analogue-to-Information conversion

  • Optical reference structure devices

Optical reference structure devices

  • Combined analog-digital signal processing

Combined analog-digital signal processing

slide-27
SLIDE 27

ASU, 30 January 2006

Analog-to-Information (A/I) Conversion

Analog Analog Front End Front End Digital Processing Digital Processing and Exploitation and Exploitation

Uniformly quantized & Uniformly quantized & spaced digital samples spaced digital samples

Tunable Tunable Analog Analog Transduction Transduction Digital Processing Digital Processing and Exploitation and Exploitation

Highly compact Highly compact digital output digital output Control Control

Physical-Layer Sensing & Processing

slide-28
SLIDE 28

ASU, 30 January 2006

A/I Conversion: Frequency–Hopping Receiver

Physical-Layer Sensing & Processing

Time Frequency

BPF Bandwidth Ω

Ω samples/second Matched Filter A/D Time Frequency

Tunable BPF Bandwidth Ω/Ν

Ω/Ν samples/second Hop Scheduler A/I

slide-29
SLIDE 29

ASU, 30 January 2006

Physical-Layer Sensing & Processing Optical Reference Structure Devices Duke University

Introduction of known structure in optical path can…

  • compute certain linear

transforms optically before A/D

  • regularize otherwise ill-posed

inverse problems

Mapping desired transform description (optimally) to feasible structure design is an engineering challenge

slide-30
SLIDE 30

ASU, 30 January 2006

Physical-Layer Sensing & Processing

slide-31
SLIDE 31

ASU, 30 January 2006

Physical-Layer Sensing & Processing

Combined Analog-Digital Signal Processing – CADSP (GA Tech)

  • Advances in floating gate device technologies being applied to develop

highly-flexible signal processing elements

  • Potential five-year payoffs: 6 ICs  1 IC; 2-3 W  2-3 mW; $100 

$10

slide-32
SLIDE 32

ASU, 30 January 2006

Waveforms for Active Sensing Program (WASP)

  • Transmitter (& receiver) technologies have enjoyed great

advances in the past two decades

  • Development of mathematical techniques to capitalize on these

advances has not kept pace!

  • Optical (micromirror arrays; SLMs; ultra-

short (shaped) pulse and high-power lasers

  • RF (Agile, software-driven devices; tunable materials)
  • Acoustic (Air-coupled & liquid-

coupled microsensors; agile software-controllable coherent array sources)

slide-33
SLIDE 33

ASU, 30 January 2006

WASP Vision

Develop a unified, rigorous methodology for waveform design and scheduling in active sensing systems

Possible Objectives

  • Adaptive spatio-temporal-spectral optical sensing
  • Libraries of signal classes for diversity sensing
  • Real-time, closed-loop adaptive waveform design
  • Coordinated irregular pulsing
  • Bio-inspired pulse scheduling

Status

  • Program proposal under development
  • Anticipated start in FY 2005

Another upcoming MTO/DSO program: A/I Conversion

slide-34
SLIDE 34

ASU, 30 January 2006

End