processing and analysis: statistical …signal processing and analysis: statistical techniques,...
TRANSCRIPT
0D-8193 244 SIGNAL PROCESSING AND ANALYSIS: STATISTICAL TECHNIQUES 1/1EFFICIENT ALGORITH..(U) MOORE SCHOOL OF ELECTRICALENGINEERING PHILADELPHIA PA DEPT 0. S A KASSAM
SSFE 7AUG 66 RFOSR-TR-97-S943 AFOS-94-S255 F/G 12/5 N
UNL flflflflflm flflIlIloi
W 8I8 L W5II I .I~
1~.~25 1 '' ,I.6
MICROCOPY RESOLUTION TEST CHARTNATIONAL BUREAU OF STANOAROS-I963-A
%P -w % -
A .A=
(J CLASSI PCATI'*jotToTilPAGE
REPORT DOCUMENTATION PAGEAT SECURITY CLASSIFICATION 1b. RESTRICTIVE MARKINGS
AiTY CL.ASSIFICATIONd AUINORITY 3. DISTRlBUTION/A VAI LAM LD! 9A SapO~r"
I ASSIFICATION/OOWt4GRAOING CEUE tt
M MING ORGANIZATION RE1PORT NUMBER(MS) S. MONITORI nOUnA:O T..iI
of PERFORMING ORGANIZATION 0. OFFICE SYMBOL 7&. NAME OP MONITORING ORGANIZATION
University of Pennsylvania ________ AFOSR/NiMIe ADDRESS (City. State and ZIP Code) 7b. ADORES (City. Stan and ZIP Code)
Dept. of Electrical Eng. BI31F 410Philadelphia, PA 19104-6390 Boll :Ai-'32 D Z0Z32..AAA
b. NAME or FUNOING/SPONSORING 8111. OFFICE SYMBOL. B. PROCUREMENT INSTRUMENT IDENTIFICATION4 NUMBERORGANIZATION it applicable
AFOSR j NAFOSR -84-025B. A0OP IR4(i4. Stale and ZIP Code) 10. SOURCE OF FUNDING NOS. _____
PROGRAM PROjaCT TASK WORK UNIT
1-. Iid D C3.64 E LEME NTNO. No. jA NO. No.
fnr~~112 2304~t A5rgin~Aiz~
12. PERSONAL AUTHOR(S)
TOf~~n .1 A __K ~r13& TYPE OF REPORT i3b. TIM% COVERED 14. OA TE0FPR E POAT (Yr...o...Dy) Is.PAGE COUNT
FialFROM 7/13/84 TO 7/14/81 3/27/8614. SUPPLEMENTARY NOTATION
1?. COSATi CODES 1B. SUBJECT TERMS (Continue on rr if nectizary and identy by block numbers
FIELD0 GROUP suB. GR.
1B. ABSTRACT (Contianue on revers 'if necesary and identify by' blocle numb.',
A "AX 11/750 minicomputer and an image display and graphics controller andadditional pieces of hardware were acquired. This equipment has enabled the signalProcessing Research Laboratory to be established at the University of Pennsylvania.The laboratory supports the needs of four faculty members and ten graduate studentsworking on AFOSR and ONR sponsored research.
DTICELECTEpAUG 19 9 87
26 DISTAl BUTION/A VAI LABILITY OP ABSTRACT 21. ABSTRACT SECURITY CLASSIFICATION 0UNCLASSIFIEOIUNLIMITEO 0 SAME AS APT. E: OTIC USERS C
lie. NAME OP RESPONSIBLE INDIVIDUAL 22b. TELEPHONE NUMBER 22c. OFFICE SYMBOLIinetud. A e Code)
Maj. Woodruff 1(202)767-5027
00 FORM 1473,083 APR EDITION OF I JAN 73 IS OBSOLETE.
8 v ' 2 8 4 2Z SECURITY CLABSIFICATION OP TmIS PAGE
AFOSR.Tht. 870943
FINAL TECHNICAL REPORT
AFOSR EQUIPMENT GRANT 84-0255
SIGNAL PROCESSING AND ANALYSIS: STATISTICAL TECHNIQUES, EFFICIENTALGORITHMS, NONLINEAR PROCESSING, EM ANALYSIS
Grant Amount: $141,339
Grant Year. July 15, 1984 - July 14, 1985
Report Date: August 27, 1986
Saleem A. KassamPrincipal Investigator
Accesiorn ForJINTIS CRAWiDTIC TAB L)
Justificijt~IO
By..........TDii ib ktto i o
66
This equipment grant was awarded for the purpose of upgrading our researchcapability in the areas of signal processing and analysis, on which we have beenactive for over ten years primarily under AFOSR sponsorship. At the beginning of theequipment grant period we had no laboratory associated with our research; we nowpossess a very useful laboratory facility, the SIGNAL PROCESSING RESEARCHLABORATORY, which has been established in two rooms made available to us in theelectrical engineering department. An industrial cash grant was obtained to meet thepower supply and air conditioning costs for the laboratory. The total area of thefacility is approximately 800 square feet and houses the hardware which has beenacquired under the equipment grant.
The two main hardware systems that have been possible through this grant are:
1) A VAX 11/750 minicomputer system with 4Mb of main memory, afloating point accelerator, communication ports, two 456 Mb disk drives,a tape drive, four terminals including two graphics terminals, two printersand two plotters;
2) An image display and graphics controller (Raster Tech) with two colormonitors, which is interfaced to the VAX and to a DEC Rainbowmicrocomputer, and a video camera with a frame grabber resident on anIBM PC XT microcomputer for image capture.
Additional pieces of hardware acquired under this grant include a digital signalprocessing development system installed in an IBM PC XT microcomputer, an IBMPC AT microcomputer, communication modems and printers.
A primary need that the new laboratory has been serving very well is that of areasonably powerful computational and simulation facility which can be coupleddirectly to signal processing hardware. Another major function of this laboratory is toprovide image processing facilities which have turned out to greatly enhance ourresearch activity in the applications of new signal processing ideas in this importantarea. The signal processing research laboratory is currently supporting the needs offour faculty members (Profs. Kassam, Ansari, Berkowitz and Kritikos) and tengraduate students, working primarily on AFOSR sponsored research and on otherrelated research programs supported by ONR. In the area of statistical and digitalsignal processing this laboratory is supporting research in nonlinear and adaptivefiltering techniques, nonparametric and robust detection and filtering,multidimensional digital filters, spectrum estimation, and computationally efficientalgorithms in signal processing. In the area of electromagnetic analysis theequipment is supporting work in three-dimensional EM imaging, design of smartsignal processing antenna arrays, and other related concepts.
2
11J !', l l * * * * 1* *1:,1pll ll
The computer facility is operated entirely by the users themselves. Maintenanceand software costs are being met by laboratory fees charged to grants and contracts,as well as from industrial sources of gifts and grants. In fact, one very positive resultof the establishment of this laboratory is that industrial equipment and cash gifts havebeen easier to get.
This laboratory has proved to be a tremendous resource to us. We are hopingthat we will be able to continue to keep it efficient and current through the addition ofnew hardware and software, for which we are always seeking new funds and giftsfrom various sources.
The attached set of copies of viewgraphs are from a recent talk that we haveprepared describing our primary research areas at the laboratory.
3
at the
SIGNAL PROCESSING RESEARCH LABORATORY
'C
. ~ %C~*C~S*~SCCSCCJ I .... . *---.. S.
STATISTICAL SIGNAL DIGITAL SIGNALPROCESSING PRGCESSNG
Detection, Estimation, and Digital Filter StructuresFiltering Two-Dimensional Filters
Robust and Nonparametric Multi-Rate ProcessingTechniques Signal ReconstructionQuantization and Coding Spectrum Estimation
IMPLEMENTATIONCONSIDERATIONS
Applications-SpecificVLSI Processors
Analog/Digital VLSI
p.
APPLICATION AREAS 1Image Processing and CodingSpatial Array ProcessingRadarSpeech Processing
UI
.3
* ~a *. ~*.~*.p *S*-*~~*ar*.* -
-x %. ~,A
i ,I _ _t _1 _I _ _ _ _ , . . . .. m ; = . -
SIGNAL PROCESSING RESEARCH LABORATORY
SOME SPECIFIC RESEARCH AREAS
" SIGNAL AND IMAGE RESTORATION
" SPATIAL ARRAY SIGNAL PROCESSING
" DIGITAL FILTER STRUCTURES
* VLSI IMPLEMENTATIONS
4 ~ ,~.. %**S*~ - - -f
* SIGNAL AND IMAGE RESTORATION
1. APPLICATION OF ROBUST ESTIMATION TOSMOOTHING, FILTERING, AND ENHANCE-MENT OF SIGNALS AND IMAGES
" Non-Gaussian (Impulsive) Noise Suppression
" Preservation of Edges in Signals
* Frequency Selectivity
NEW NONLINEAR DATA-ADAPTIVE FINITE-WINDOWFILTERS (VIEWED AS GENERALIZATIONS OF MEDIANFILTERS) PERFORM VERY WELL
" Extension to Image Sequence Processing* Applications in Edge Detection, Motion Estimation,
Recognition, Tracking
* Application in CFAR Radar
* Fast Hardware Implementations
'I-
F I I *~ F ',.
202
0..
-211mdh~ian101 W cxn W (2, 7DW 120 t) noisy signal 10202
a.0.
O.rigirnaJ Image "'OLSY Image Median Fiit-er
Cc)npAq is I ON OF
FILTER~S (L.7DWF)
FOR SICKNA~LfRe1rRAIO
&*W (2,2 LZII II UI I. I' (3,
I JI -a.@ fOrgnlSignal
OriginalCorrupted SignalaE~ 7. w
CegsOgcRunn*enOtu
Meia utu
0 n ia- . Ii m i ?
cm We
UnserFIR Otput -2-rMedan Output
U.0 A.0
.
to n~ 314 a
SIGNAL AND IMAGE RESTORATION
2. RESTORATION OF SIGNALS AND IMAGESFROM PARTIAL INFORMATION
Restoration from Fourier Transform Magnitude Only
NEW HILBERT SPACE FRAMEWORK ALLOWSREPRESENTATION OF CONSTRAINTS AS CONVEX SETS.GUARANTEED CONVERGENCE TO GOOD SOLUTIONS
• Extension to Image Sequence Processing
* Consideration of Other Constraint Combinations
" Study of Tolerance to Noise
V.
S
, !r, , ,, . , € ' '.. ,' -' '',,', ,; °,, ' ;,' . %,,; , .- .. ,,;.', ".....,-'o.-. ",-:....',-'. "..'% ..'-.;%,-.;,-I
SPATIAL ARRAY SIGNAL PROCESSING
1. LOW-SIDELOBE BEAM CHARACTERISTICSFROM NARROWBAND HIGH-RESOLUTIONBOUNDARY ARRAYS
• Array Elements Spaced Along Boundary of ConvexPlanar Aperture
NEW SYNTHESIS THEOREM ALLOWS EFFECTIVEINTERPOLATION INSIDE BOUNDARY. ACHIEVED BYPROCESSING TRANSMIT/RECEIVE DATA TO SYNTHESIZEINDIVIDUAL BEAMS, LINEAR COMBINATION OF THESECONVERGES TO IDEAL SOLUTION
• Study Effects of Noise and Element PositionUncertainities
• Combination of Super-Resolution and Low-SidelobeProcessing
* Implementation Considerations
"F
.4l
U,
TWO-WAY PATTERNS,. 20A SQUARE APERTURE
01 3
CONTINUOUS 1BOUNDARY ONLY-1
4 ***.*,,.r................
_~A' A - .~-0
CONTINUOUSLY . .3FILLED APERTURE b
-10 -
i1:>.. ~-20 -
- xz
.~t4 * .. .. ~ .* . 40 -
BOUNDARY______________0,5A~ SPACING. 0 1320 TERM SYNTHESISSIMULATION -0 f
--- -40
-50-
It
SPATIAL ARRAY SIGNAL PROCESSING
2. DIGITAL INTERPOLATION BEAMFORMING
High-Speed Beamforming for Linear Arrays
USE OF MULTI-RATE PROCESSING AND NEW RECURSIVEDIGITAL FILTER STRUCTURES FOR COMPUTATIONALLYEFFICIENT BEAMFORMING AND SCANNING
* Extensions to Non-Uniformly Spaced and TwoDimensional Arrays
I,
SVl
I," ' " ' ' " ... " " ' "', " I.-
.c. ISM
Zerospie
SCEM FO CIATEPLTO' EMOMN
r1
SCNTERPOLATO UIG7T- SECTIONSI~ INAPARALLEL
I ' *
r .' 'i I OUTPUT ATLOW
''. i- SAMPLING RATE
* *- I,
I
": _______.___ __ __ _
' " ;-".
* . ; , *
: , OUTPUT AT HIGH
* " SAMPLING RATE
-- }
- ' X
2.. / ,,
. ,4.
S 'I OUTPUT AFTER
i . INTERPOLATION
' I '
V.A ' ,V'
. ... el--
A&
~, -~ %~ %~~V %N
- DIGITAL FILTER STRUCTURES
INFINITE IMPULSE RESPONSE STRUCTURESFOR FILTERING AND SAMPLING-RATECONVERSION
" Multi-band, Multi-level Infinite Impulse Response(IIR) Filters
* Linear-Phase IIR Filters
* Interpolators and Decimators
CONFIGURATIONS OF ALLE A BUILDING BLOCKS RESULTIN SCHEMES WITH ATTRIBUTES OF LOW COMPLEXITY,LOW QUANTIZATION EFFECTS, AND MODULARITY
* Incorporation of coefficient constraints in allpass building
blocks
* Allpass and generalized separable two-dimensional filters
* Applications in image processing and coding and adaptivefiltering
ib
CI
Ideal response of M-level filter (levels P' ..Pm,
bo1
M-level filter configuration using alipass sections
r-4
_ _ _ _ _ _ _ _ _ _ _ _ 0i-
>~S~Z)~
C-1
tm -06
A c,.) 2
1.0
7r, I
to -IDA MANTD EPNEO ARWADFpiFLE
P. (-)S
7rU
APROI1AT 11-EA IEDSPRBErL
* VLSI IMPLEMENTATIONS
EFFICIENT IMPLEMENTATION OF HIGH-RESOLUTION SPECTRUM ESTIMATIONALGORITHM
Multiple Signal Classification (MUSIC) AlgorithmImplementation
SYSTOLIC ARRAY SCHEME BASED ON THE USEOF CORDIC BUILDING BLOCKS
* Use of CORDIC blocks for filter implementations (e.g. allpass
blocks)
" Interpolation beamforming using CORDIC blocks
* New general-purpose signal processor for filterimplementations with structural constraints
-!V -- d'E' ,,*
PROCE55ING ELEMENT5 (PE's):
X, -- ' IC039 --A,,SinOCOMPLEX - CS9jc sn
y -CORDIC -x aineg +:Kos CO DI - 'ii9~c
Pe 0 e PEe
SYSTOLIC ARRAY FOR EIGENVALLE-VECTQR.COMPU TAT ION
was
U32. U23
U3, U1u 13031 UZI U13
0311 0
amaal,
I.,,I
I I /~
U), LLa UVU's1 U1
I.
I