`
`g'vnazwzfiv‘m
`
`...
`u...
`H- un-.«-.. \.-
`
`.a-n-ow...-~....V....,
`..——.u-a
`
`.--..»-....
`,. ~oo.~”o~-.--.~-‘d ‘--4.. --.--...
`....~..-.. -
`.v...---cm.7- ..-.....
`-_,_‘_..-..~
`-- .r......
`
`A
`J... --....-..- --....
`...
`.
`.
`-e-.....,--......-.-..
`..
`.4 .-,-... .-.. . .-...-- -0...-.~._..... .
`.4
`.
`-un...-....--.-— "v-..
`..y._.."4...-- .w—b’..o~00 -‘--<-..-
`..-... ~a..---.-.. .. 0.“... ..
`..-......-..--....-..
`n .--.
`-----n....-..t---
`.
`-....:.
`.-.-......~...---....-.
`nu...
`.-..A........-....-‘.-
`"07..-.
`.
`...-.-r...---.--.--..- -.....-....-
`_. -...._......-.....,...-..-.. .-...
`~--,..._.
`.--.....-.....--..--.«. ---J-.._.._.,
`
`.
`
`..- o-—-«~~--
`.........q -.‘>
`..--
`.. .- .....-
`.u a....
`"—-
`’
`.......v.....
`
`.. ,-,....
`-...p
`
`_
`
`4.....-M. u-.—
`,.--.~- -W-.~~.. ~.u,
`, g. .«n «v... luv-mu- Ann... ou—ocn
`,... Mann... m..«»~..~—4 -.... .u-N ..
`,...u~.w.‘ .nvunvuoovv-n. 1..
`.
`..... .~-un%........-.m.-...
`u‘
`.
`..-.---aha"..-~-.-...¢--~—~uu..
`__... -.~u a-...-. ~19. baa ... V.
`,
`.umm-h..—....n.a .v.
`...‘.
`.-u.-u. Muwa .wrvtw.c.-m..
`...c.—- “we-n-gou...._.-......-~. .
`u... .--..———...-.,._--..~.~.. ...
`.
`.adv.»-...-.._......—..-“- ....-.,,.
`~..—<—.-mv—u..--a..
`.-...— ofl-ut3~'*¢>.* ...--.
`.-..~—--~-—--.~.
`.a-. c...V..
`......>¢-ow-u.
`.qauvnn-.. ~ ......-,.,.~
`cv’-~vc~~uuvn---v§.-r—v--Lua
`‘...-~-b.llnvhu.yv
`V -
`_..o—.‘ J—-4-o~ ~.---uo.~.- s.. —-
`.-.<.~-—vc‘.a.....
`...‘ .‘..
`-.VV 1.”
`.uo-~».u.-...._4J.
`. .-..
`—-.a.~.-.«-
`.u......-..-.
`
`
`
`
`
`
`
`Ex.1005/Page1 of16
`Apple v. Saint Lawrence
`
`Ex. 1005 / Page 1 of 16
`Apple v. Saint Lawrence
`
`
`
`
`
`EX. 1005 / Page 2 0f 16
`
`Ex. 1005 / Page 2 of 16
`
`
`
`<®> IEEE TRANSACTI C) N S C) N
`(Contents Continuedfrom Front Cover) 0\\ AC I;
`
`Recursive Wiener Filtering for Image Restoration .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`.
`.
`M S
`ihmed and K K
`I
`.
`. T
`.
`.
`.
`.
`.
`.
`Some Results on MatrIx Symmetries and a Pattern Recognition Application
`fiDaIta and S. D. 11:31:05:
`990 A
`C—R Bounds for AR Parameter Estimation as a Function of the Data Length'
`.
`.
`.
`.
`.
`.
`.
`.
`. .. F. Giannella
`.
`.
`.
`A Note on “Realization of Digital Filters Using Input-Scaled Floating-Point Arithmetic” .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. J. R. Heath
`.
`.
`995
`.
`Index Transforms of Multidimensional Cyclic Convolutions and Discrete Fourier Transforms
`J I
`I
`.
`g
`i
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`I k
`.
`High-Performance Signal-Modeling by PenCII-of-Functions Method: Band-Pass Case .
`.
`.
`.
`.
`.
`.
`. .. V. K. Jain and T. K. Surf/(:11:
`.
`Corrections to “Finding the Poles of the Lattice Filter" .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. W.
`. J
`5
`'
`Structured Fast Hartley Transform Algorithm
`3 Cofkfv’g‘ifizg if??? 1000
`Efficient Calculation of Signal and Noise Variances
`H.-C Lue and M
`luv
`1000
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`Effect of Word-Length Truncation on Quantized Gaussian Random Variables
`. D. R. Margan and W Cassatt?
`133i
`.
`.
`Measurement of Time Delay Using the Time Shift Property of the Discrete Fourier Transform (DFT)
`.
`.
`.
`.
`.
`.
`.
`.
`.
`I
`.K Na ai'
`1006
`.
`.
`.
`.
`.
`.
`.
`.
`Pruning the Decimation—in-Time FFT Algorithm with Frequency Shift
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`I
`I
`I
`i
`i
`i
`i
`i K A’Vaga'
`1008
`The Stepdown Procedure for Complex Predictor Coefficients .
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. J. Pieone, D. P. Prezas, W. T. Hartwell.
`icfrti
`1010
`i
`Realizauon of First—Order Two-Dimensional All—Pass Digital Filters
`M. S. Reddy S. C. Dutta Roy andiS JV Ha r
`1011
`Error Feedback in~a Class of Orthogonal Polynomial Digital Filter Structures
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`D. W
`I
`z a
`i
`I
`illiamson and S. Sridharan
`1013
`
`,
`
`I
`
`.
`
`'
`OCT
`N U M B E R 5
`V O LU M E A S S P 3 4
`OBER 1 986
`r. PUBLICATION OF THE IEEE ACOUSTICS, SPEECH, AND SIGNAL PROCESSING SOCIETY
`
`_
`('8 S N O 0 9 6 3 5 18)
`
`CORRESPONDENCE
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`. ..
`
`.
`
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`,
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`.
`.
`.
`.
`
`BOOK REVIEWS
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`Adaptive Signal Processing~B. Widrow and S D Stearns
`I
`I
`I
`r
`.
`.
`.
`.
`.
`.
`OptImum Signal Processing: An IntroductIon—S. J. Orfanidis .
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`,
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`.
`.
`
`'
`. Reviewed [7 D. R. M
`.
`.
`. Reviewed by
`
`Abstracts of Manuscripts in Review .
`
`.
`
`r
`
`‘
`
`_
`
`~
`
`V
`
`_
`
`h
`
`_
`
`_
`
`‘
`
`V
`
`t
`
`.
`
`_
`
`‘
`
`.
`
`'
`
`'
`
`'
`
`'
`
`'
`
`*
`
`‘
`
`'
`
`'
`
`'
`
`-
`
`-
`
`' *-
`
`1020
`
`(Contents Continued on Back Cover)
`
`H. Wakgii‘ig\PAPERS
`1.2 Underwater Signal Processing
`. .. B. M. Bell and T. E. Ewart
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`SepartItIng Multipttths by Global Optimization ofa Multidimensional Matched Filter .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. R. Sr‘hreiber
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`Implementation of Adaptive Array Algorithms .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`Simple, Effective Computation of Principal Eigenvectors and Their Eigenvalues and Application to High-Resolution Estimation of
`Frequencies .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. .. D. W. Tufts and C. D. Melissinos
`2./ Speech Transmission and Coding
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`RCgular~Pulse Excitation~A Novel Approach to Effective and Efficient Multipulsc Coding of Speech .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. P. Kroon. E. F. Depretrere. and R. J. Sluyter
`2.3 Speech Analysis and Reconstruction
`-\ High—Quality Digital Filterbank for Speech Recognition Which Runs in Real Time on a Standard Microprocessor .
`2.4 Speech Synthesis
`Synthesis of Natural Sounding Pitch Contours in Isolated Utterances Using Hidden Markov Models .
`3.! One-Dimensional DSP
`. Y. Bresler and A. Maeot‘xki
`.
`.
`.
`.
`.
`.
`.
`.
`EXuCt Maximum Likelihood Parameter Estimation of Superimposed Exponential Signals in Noise .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. R. V. Cox
`.
`.
`.
`,
`.
`.
`.
`.
`The Design of Uniformly and Nonuniformly Spaced Pseudoquadrature Mirror Filters .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. E. Eleftheriou and D. D. Falconer
`.
`.
`.
`Tt‘deing Properties and Steady-State Performance of RLS Adaptive Filter Algorithms .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. W. A. Gardner
`.
`.
`.
`Measurement of Spectral Correlation .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. A. Nehorai and B. Porat
`.
`.
`.
`Ad'dplive Comb Filtering for Harmonic Signal Enhancement .
`.
`.
`.‘
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`T. Ning and C. L. Nikias
`Multichannel AR Spectrum Estimation: The Optimum Approach in the Reflection Coefficient Domain
`. .. J. P. Princen and A. 8. Bradley
`.
`.
`.
`.
`.
`A"a|.V~\‘iS/Synthesis Filter Bank Design Based on Time Domain Aliasing Cancellation .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. .G. R. L. Sohie and L. H. Sibul
`,
`.
`.
`.
`.
`.
`.
`Slochastic Operator Norms for Two-Parameter Adaptive Lattice Filters .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. F. J. Taylor
`.
`.
`.
`.
`.
`An Analysis of the Distributed Arithmetic Digital Filter .
`.
`.
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`.
`.
`.
`. .J. K. Tugnail
`.
`.
`.
`.
`.
`Modeling and Identification of Symmetric Noncausal ImpuISe Responses .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. G. Waekersreuther
`.
`.
`.
`.
`.
`Some New Aspects of Filters for Filter Banks .
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. H. Wang and M. K1108}!
`.
`.
`.
`.
`.
`0111M Performance of Signal-Subspace Processing——Part l: Narrow-Band Systems .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`Roundoff Noisc Minimization and Pole-Zero Sensitivity in Fixed—Point Digital Filters Using Residue Feedback .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. .D. Williamson
`3.2 Multidimensional DSP
`'
`.
`,
`.
`,
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`,
`.
`.
`Edge Detection Using the Linear Model .
`. .. S. M. Kay and G. J. Lemay
`.
`Morphological Skeleton Representation and Coding of Binary Images
`P. A. Maragos and R. W. Sehafer
`Delcction of Wave Reflectors and Wave Sources in the Earth Subsurface .
`. l. Pita: and A. N. Venetsanopoulos
`N“Mince” Space-Variant Postprocessing of Block Coded Images .
`.
`.
`.
`.
`. B. Ramamurthi and A. Gersho
`.
`.
`.
`.
`.
`.
`Image Reconstruction from Zero Crossings .
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`. D. Rotem and Y. Y. Zeevt
`Subband Coding or Images .
`.
`.
`_
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`,
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. J. W. Wood: and s. D. O‘Neil
`VLSI and Hardware Implementations
`.
`-
`1289
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`0I’llmtzation of One-Bit Full Adders Embedded in Regular Structures .
`, K. [wane and K. Sleight:
`A MUllimiCroprocessor Architecture for Rea1~Time Computation of a Class of DFT Algorithms
`1301
`~
`-
`-
`~
`~
`A
`~
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`.
`. V. Milutinovit“. J. A. B. Fortes. and L. H. Jamieson
`.
`.
`1310
`A Ring Array PrOCessor Architecture for Highly Parallel Dynamic Time Warping .
`.
`.
`. .
`.
`. J.-l. Takahashi. S. Haltori. T. Kimura, and A. lwata
`R“—
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`,
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`,
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`,
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`.
`.
`.
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`.
`
`. A. Ljolje and F. Fol/side
`
`1074
`
`1081
`1090
`1097
`1
`1 l l
`1124
`1139
`1153
`1162
`1165
`1171
`1182
`1201
`1210
`
`1221
`1228
`1245
`1258
`1269
`1278
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`,
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`.
`.
`,
`.
`.
`
`.
`.
`.
`.
`,
`
`.
`.
`.
`.
`.
`
`.
`.
`.
`.
`.
`
`i
`
`i
`
`'
`
`'
`
`I
`
`'
`
`I
`
`i
`
`'
`
`i
`
`i
`
`'
`
`'
`
`'
`
`i
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`‘
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`'
`
`‘
`
`'
`
`‘
`
`'
`
`'
`
`'
`
`'
`
`‘
`
`'
`
`‘
`
`‘
`
`'
`
`‘
`
`‘
`
`‘
`
`'
`
`‘
`
`‘
`
`‘
`
`1029
`1038
`
`1046
`
`1054
`
`. H. F. Silvermnn
`
`1064
`
`Ex. 1005 / Page 3 of 16
`
`
`
`'®
`
`IEEE ACOUSTICS. SPEECH. AND SIGNAL PROCESSING SOCIETY
`
`The Acoustics, Speech. and Signal Processing Society is an organization. within the framework of the IEEE. of members with
`in thelteehnology of transmissmn. recording. reproduction. processing. and measurement of s
`acoustic. mechanical. and optical means. th
`
`TRANSACTIONS ON ACOUSTICS. SPEECH, AND SIGNAL PROCESSING, VOL. ASSP-34, NO. 5, OCTOBER l986
`
`l029
`
`“‘0 Separating Multipaths by Global Optimization of a
`Multidimensional Matched Filtn
`
`ADMINISTRATIVE COMMITTEE
`Officers
`
`D. VI. ETTER. Vice President
`Univ New Mexico
`Albuquerque. NM 87l3|
`
`J. S. LI M. Secretary-Treasurer
`MIT.
`Cambridge. MA 02 I39
`
`T. H. CRYSTAL. President
`Institute for Defense Analyses
`Thanet Rd.
`Princeton. NJ 08540
`
`R. E. CROCHIERE. Past President.
`AT&T Bell Laboratories
`Murray Hill. NJ 07974
`
`'84 & '85
`
`Transactions
`PIERCE WHEELER. Managing Editor
`AT&T Bell Laboratories
`Short Hills. NJ 07078
`
`Term Ending December 31. [985
`M. G. BELLANGER
`T H. CRYSTAL
`A. E. ROSENBERG
`
`Term Ending December 3]. I986
`D. M. ETTER
`S. L. MARPLE
`L. L. SCHARF
`SOCIETY PUBLICATIONS EDITORIAL BOARD
`SOCIETY PUBLICATIONS
`M. KAVEH, Chairman
`Univ. of Minnesota
`Minneapolis. MN 55455
`
`S. T. ALEXANDER (2/88)
`N. Carolina State Univ
`Raleigh. NC 27695
`
`S. BOLL (2/87)
`lTT-DCD
`San Diego. CA 92l3l
`
`TRANSACTIONS ASSOCIATE EDITORS
`J. M. CIOFFt (4/87)
`G. W. ELKO (IO/87b
`Stanford Univ
`AT&T Bell Labs
`Stanford. CA 94305
`Murray Hill. NJ 07974
`
`L. J. GRIFFITHS (2/88)
`Univ. Southern Calif.
`Los Angeles. CA 90089
`
`C. L. NIKIAS (6/87)
`Northeastern Univ.
`Boston. MA 02l15
`
`L, H. JAMtESON 11/88)
`Purdue Univ.
`West Lafayette. IN 47907
`
`T. RAMSTAD (6/88)
`Univ. Trondheim
`Trondheim. Norway
`
`B. H. JL‘ANG (2/88)
`AT&T Bell Labs
`Murray Hill. NJ 07974
`
`M. M. SONDHI (IO/85)
`AT&T Bell Labs
`Murray Hill. NJ 07974
`
`T, L MARZETTA (6/86)
`Schlumberger-Doll Research
`Ridgefield. CT 06877
`
`H. J. TRL'SSELL (6/88)
`N. Carolina State Univ
`Raleigh, NC 27695
`
`THE INSTITUTE OF ELECTRICAL AND ELECTRONICS ENGINEERS, INC.
`Officers
`BRUNO O. WEINSCHEL. President
`HENRY L. BACHMAN. President-Elect
`EMERSON W. PUGH. Executive Vice President
`MICHIYUKI UENOHARA. Secretary
`EDWARD J. DOYLE. Treasurer
`
`THOMAS W. BARTLETT. Controller
`DONALD CHRISTIANSEN. Edilor of Spectrum
`IRVING ENGELSON. Staff Director. Technical Activities
`LEO FANNING. Staff Director, Professional Activities
`SAVA SHERR. Staff Director. Standards
`
`0096-3518!86/ 1000-1029$01 .00 © 1986 IEEE
`
`CYRIL J. TUNis, Vice President. Educational Activities
`CARLETON A. BAYLESS. Vice President. Professional Activitiet
`CHARLES H. HOUSE. Vice President. Publication Activities
`DENNIS BODSON. Vice President, Regional Activities
`I
`V
`I
`_
`MERLIN G. SMITH. Vice President. Technical Activities
`N. REX DIXON. Du‘lSlOn IX Director—Signals and Applications Division
`Headquarters Staff
`ERIC HERZ. Executive Director and General Manager
`ELWOOD K. GANNETT, Deputy General Manager
`DAVID L. STAIGER. Staff Director, Publishing Services
`CHARLES F. STEWART, JR.. Staff Director, Administration Services
`DONALD L. SUPPERS. Staff Director, Field Services
`THOMAS C. WHITE, Staff Director, Public Information
`ANTHONY L. RlGAS. SiaffDirector. Educational Services
`Publications Department
`Publication Managers: ANN H. BURGMEYER. GAIL S. FERENC. CAROLYNE TAMNEY
`Associate Editor.- SUSAN HART ANDERSON
`
`Term Ending December 31, 1987
`B. FRIEDLAVDER
`.I. S. LlM
`J. H. MCCLELLAN
`
`Magazine
`D. M. ETTER, Editor
`Univ. New Mexico
`Albuquerque. NM 87l3|
`
`C. GAt AND (9/87)
`IBM France
`La Gaude. Alpes~Maritimcs
`France
`
`DC. MIJNSON (9/87)
`Univ. of Illinois
`Urbana. IL 61801
`
`G. H. WAKt-HFI I)
`Book Reviews
`Univ. Michigan
`Ann Arbor. MI 48109
`
`Abstract—A transmitted signal can arrive at a receiver via several
`[refracted Fermat paths. If the paths are independent in the Fresnel
`use, then the received signal can be modeled as the sum of amplitude
`scaled and time shifted copies of a predetermined replica plus white
`noise. We present an algorithm that uses the replica to determine the
`time shifts and amplitudes for each path. It is referred to as an n-
`tlimensional matched filter algorithm by analogy with the well-known
`nltched filter algorithm. The cross correlation between the received
`ignnl and the replica oscillates near the center frequency of the trans-
`mitted signal. This causes the n-dimensional matched filter output to
`lure many local maxima that are not globally optimal. The time shifts
`and amplitude scalings for the Fermat paths are determined by maxi-
`mizing the output of the n-dimensional matched filter. The algorithm
`lr more robust and efficient than others currently available. Simulated
`realizations of received signals were generated with multipath and noise
`characteristics similar to an ocean acoustic transmission case. These
`realizations were then separated into arrival times and corresponding
`amplitudes by the algorithm. The results of these tests and the general
`limitations of the algorithm are discussed.
`
`1.
`
`INTRODUCTION
`
`Asound pulse transmitted in the ocean almost always
`arrives at a receiver via several paths. Theoretically,
`separating the pulse energy in the individual paths can be
`‘ impOSSIble. Path separation is possible if their arrival
`i times differ by at least one-half cycle of the pulse center
`‘ frequency. This condition is closely related to the Fresnel
`zones of the paths. (Fresnel zones ofray paths are defined
`and discussed in Section 7.1 of [1].) The signal process-
`.ng literature has many references to techniques such as
`lmatched filters, inverse filters, and many modifications of
`these and other methods which purport to determine the
`arrival time and amplitude of each path. These estimates
`are biased when the pulses from two or more paths over-
`lap in time. Nilsson3-[2] introduced a method that elimi-
`nates these biases. Ehrenberg et al. [3] and Ewart et al.
`i4] present a technique to evaluate Nilsson’s method,
`demonstrate the improved accuracies in the arrival time
`find amplitude estimates, and apply the technique to some
`limoustic observations. Ewart and Reynolds [5] discuss
`
`-.
`‘
`
`I Manuscript received July 17, 1985; revised February 28, 1986. The work
`"f3. M. Bell was supported by the Office of Naval Research under Code
`5AR. The work of T. E. Ewart was supported by the Office of Naval
`h under Code 4250A.
`r
`. B. M. Bell is with the Applied Physics Laboratory, College of Ocean
`Fishery Sciences, University of Washington, Seattle, WA 98105.
`E. Ewart is with the Applied Physics Laboratory and School of
`'gr-aphy, College of Ocean and Fishery Sciences, University of
`hington, :Seattle, WA 98105. '
`EB Log Number 8609044.
`
`
`
`
`
`
`
`
`
`
`
`AMPLITUDE
`
`t (i0'3st
`
`Fig. 1. Replica s(t) and realization r(t).
`
`an application of the technique to acoustic data taken dur-
`ing the Mid—Ocean Acoustic Transmission Experiment,
`MATE. They also discuss the limitations of the method
`due to volume scattering. It was in their paper that the
`terminology “n-dimensional matched filter” was intro- 1
`duced. In its present implementation, the algorithm dc— '-
`scribed by Ehrenberg et al. would require 1—2 years on a
`VAX 750 to analyze the MATE data. It also would re-‘
`quire operator intervention about once every #0 received, ,
`pulses. The algorithm presented here can process the
`MATE data set in 20 days without operator intervention.
`The MATE data set consists of 2.9 X 105 received.
`
`,
`
`pulse can consist of up to three paths. The received signal
`r(t) is modeled as the sum of amplitude scaled and time-
`shifted copies of a replica so) plus white noise e(t); i.e.,
`r(t) = Efil ajs(t — r) + e(t). The term replica refers to
`the idealized received signal for the single path case with—
`out noise or scattering effects.
`A
`Throughout this paper, examples and results willbe
`presented. These have been computed with the replicas
`that were used to analyze the MATE data. In Fig.
`l‘ the
`replica s(t) for a 2083 Hz pulsed tone (four cycles input“
`to the transducer) is plotted together with sevétplreal
`izations of r(t). Bach realization has three pathsihwh
`overlap well within the pulse duration-:The signal-toeho
`ratio is 100; we define the signal-to-noise ratio as the
`
`pyright law if”
`page, provided the per-copy fee indicated in the coders pald
`articles Without fee. Instructors are permitted to photocopy
`reprint. or republication
`rmission, write to D'
`cto . Publishin Services
`_
`I
`.
`_
`at IEEE Headquarters. All rights reserved. Copyright © 1986 by The Institute of Electrical and Electronig: Engineers. Inc. Pririizd iiiU .S.A. Secgond~class
`POS‘fige Paid at New York. NY, and at additional mailing offices. Postmaster: Send address changes to IEEE, 445 Hoes Lane. Piscataway. N} 0885441I 50.
`
`,"Student Servicesl: 445 Hoes Lane.
`~7910. Controller—7748. Educational
`
`Ex. 1005 / Page 4 of 16
`
`
`
`IEEE TRANSACTIONS ON ACOUSTICS. SPEECH. AND SIGNAL PROCESSING. VOL. ASSP~34. NO. 5, OCTOBER 1985
`Regular-Pulse Excitation—A Novel Approach to
`Effective and Eflicient Multipulse Coding of Speech
`IEEE, ED F. DEPRETTERE, MEMBER,
`iEEE, AND ROB J. SLUYTER
`It
`
`PETER KROON, STUDENT MEMBER,
`
`I
`
`Abstract—This paper describes an effective and efficient time do-
`main speech encoding technique that has an appealing low complexity, 4f
`and produces toll quality speech at rates below 16 kbits/s. The pro—
`posed coder uses linear predictive techniques to remove the short-time
`correlation in the speech signal. The remaining (residual) information
`is then modeled by a low bit rate reduced excitation sequence that,
`when applied to the time-varying model filter, produces a signal that
`is “close” to the reference speech signal. The procedure for finding
`the optimal constrained excitation signal incorporates the solution of a
`few strongly coupled sets of linear equations and is of moderate com-
`plexity compared to competing coding systems such as adaptive trans-
`form coding and multipulse excitation coding. The paper describes the
`novel coding idea and the procedure for finding the excitation se-
`quence. We then show that the coding procedure can be considered as
`an “optimized” baseband coder with spectral folding as high-fre—
`quency regeneration technique. The effect of various analysis param-
`eters on the quality of the reconstructed speech is investigated using
`both objective and subjective tests. Further, modifications of the basic
`algorithm, and their impact on both the quality of the reconstructed
`speech signal and the complexity of the encoding algorithm, are dis-
`cussed. Using the generalized baseband coder formulation, we dem-
`onstrate that under reasonable assumptions concerning the weighting
`filter, an attractive low-complexitylhigh-quality coder can he obtained.
`
`I.
`INTRODUCTION
`N interesting application area for digital speech cod-
`ing can be found in mobile telephony systems and
`computer networks. For these
`
`i
`
`called delayed decision coders (DDC) [1, ch. 9], seems
`to be promising for these applications. Coders that belong
`to this class utilize an encoding delay to find the “best”
`quantized version of the input speech signal or a trans-
`formed version of it. Quite effective algorithms can be
`designed by combining predictive and DDC techniques to
`yield low bit rate waveform matching encoding schemes.
`A powerful and common approach is to use a slowly time-
`
`revised March 5, 1986. This work
`ch Laboratories, Eindhoven, The
`Netherlands, and by the Dutch National Applied Science Foundation under
`Grant STW DEL 44.0643.
`P. Kroon was with the Department of Electrical Engineering, Delft Uni-
`» versity of Technology, Delft, The Netherlands. He is now with the Acous-
`tics Research Department, AT&T Bell Laboratories, Murray Hill, NJ
`07974.
`E. F. Deprettere is with the Department of Electrical Engineering, Delft
`University of Technology, Mekelweg 4, 2628 CD Delft, The Netherlands.
`R. J. Sluyter is with the Philips Research Laboratories, 5600 MD Eind-
`hoven, The Netherlands.
`IEEE Log Number 8609633.
`
`H. BASIC CODER STRUCTURE
`
`The basic coder structure can be viewed as a residual
`modeling process, as depicted in Fig. 1. In this figure. the
`residual r(n) is obtained by filtering the speech signal SO!)
`0096-3518/86/1000—1054$01.00 © 1986 IEEE
`
`-.. s.
`
`varying linear predictive (LP) filter to model the short—
`time spectral envelope of the quasi—stationary speech sig-
`nal. The problem that remains is how to describe the re-
`sulting prediction residual that contains the necessary in-
`formation to describe the fine structure of the underlying
`spectrum. In other words, what is the “best” low-capac-
`ity model for the speech prediction residual subjected to
`one or more judgment criteria. These may include objec-
`tive and subjective quality measures (such as rate distor-
`tions and listening scores, respectively), but coder com-
`plexity can also be taken into account. Although certain
`models have been shown to behave very satisfactorily [2]—
`[4], the question of optimality remains difficult to answer.
`In this paper we address the problem of finding an ex-
`citation signal for an LP speech coder that not only en-
`sures a comparable quality with existing approaches, but
`is also structurally powerful. By the latter we mean that a
`fast
`realization algorithm and a corresponding high
`throughput (VLSI) implementation can be obtained. We
`propose a method in which the prediction residual is mod-
`eled by a signal that resembles an upsampled sequence
`and has, therefore, a regular (in time) structure. Because
`of this regularity, we refer to this coder as the regular-
`'
`‘
`PE) coder [5]. The values of the non-
`zero samples in this signal are optimally determined by a
`least-squares analysis-by-synthesis fitting procedure that
`‘ can be expressed in terms of matrix arithmetic.
`In Section II we describe in more detail the regular—
`pulse excitation coding procedure and the algorithm for
`finding the excitation sequence. In Section III we show
`that the proposed encoding procedure can be interpreted
`in terms of optimized baseband coding. In Section IV, the
`influence of the various analysis parameters on the quality
`of the reconstructed speech is investigated.7 Further, to ex-
`ploit the long-term correlation in the speech signal, the
`use of a pitch predictor is discussed. Modifications to the
`basic procedure, to attain a further reduction in complex-
`ity without noticeable quality loss, are described in Sec-
`tion V. Finally, in Section VI, we describe the effect of
`quantization on the quality of the reconstructed speeCh
`signal.
`
`KROON el al.: REGULAR-PULSE EXCITATION
`
`5(0)
`
`MINIMIZATION
`
`EXCITATION
`- GENERATOR
`
`_'
`NZ)
`
`V‘")
`
`('3)
`
`A
`Sm)
`
`In this figure, the locations of the pulses are marked by a
`vertical dash and the zero samples by dots. If k (k =. l,
`2,
`'
`-
`-
`, N) denotes the phase of the upsampled versron
`of the vector b“), i.e., the position of the first nonzero
`sample in a particular segment, then we have to compute
`for every value of k the amplitudes b“"(-) that minimize
`the accumulated squared error. The vector that yields the
`minimum error is selected and transmitted. The decoding
`procedure is then straightforward, as is shown in Fig.
`1(b).
`
`Fig. 1. Block diagram of the regular—pulse excitation coder: (a) encoder.
`(b) decoder.
`
`A. Encoding Algorithm
`
`Denoting by Mk the Q by L position matrix with entries
`
`m,j=1ifj=i*N+k—1
`
`k = 4= .
`
`.
`
`Fig. 2. Possible excitation patterns with L = 40 and N = 4.
`
`through a pth-order time—varying filter A(z),
`p
`
`Ac) = 1 + £11 akz‘k.
`
`(1)
`
`which can be determined with the use of linear prediction
`(LP) techniques as described in, e.g., [6]. The difference
`between the LP-residual r(n) and a certain model‘resrdual
`v(n) (to be defined below) is fed through the shaping filter
`l/A(z/y),
`
`1
`
`=
`
`A(z/v)
`
`1
`
`P
`1 + [(1:31 ak'ykz_k
`
`,
`
`0
`
`S 'y S l.
`
`(2)
`
`This filter, which serves as an‘ error weighting function,
`plays the same role as the feedback filter in adaptive pre-
`dictive coding with noise shaping (APC-NS) [7]/ and the
`weighting filter in multipulse excitation (MPE) coders [2].
`The resulting weighted difference e(n) is squared and ac-
`cumulated, and is used as a measure for determining the
`effectiveness cf the presumed model v(n) of the resrdual
`r(n).
`.
`The excitation sequence 1) (n) is determined for adjacent
`frames consisting of L samples each, and is constrained
`as follows. Within a frame, it is required to correspond to
`an upsampled version of a certain “optimal” vector b =
`(17(1),
`'
`'
`'
`, b(Q)) of length Q(Q < L). Thus, each seg-
`ment of the excitation signal contains Q equidistant sam-
`ples of nonzero amplitude, while the remaining samples
`are equal to zero. The spacing between nonzero samples
`is N = L/Q. For a particular coder, the parameters
`and
`N are optimally chosen but are otherwise fixed quantities.
`The duration of a frame of size L is typically 5 ms. Each
`excitation frame can support N sets of Q equidistant non-
`zero samples,
`resulting in N candidate excrtation se-
`cIuences. Fig. 2 shows the possible excitation patterns for
`a‘frame containing 40 samples and a spacing of N = 4.
`
`..
`
`O S.
`i
`0 S j
`my = 0 otherwise
`the segmental excitation row vector 00‘), corresponding to
`the kth excitation pattern, can be written as
`
`Q — l
`
`s L — l,
`
`(3)
`
`v‘” = b‘kIMk.
`
`(4)
`
`.1;
`Let H be an uppertriangular L by L matrix whose jth row
`(j = 0,
`-
`'
`-
`, L — 1) contains the (truncated) response
`h(n) of the error weighting filter 1/A(z/y) caused by a ntrit
`impulse 6(n — j). That is,
`.2
`h(0) h(1)
`
`ML — 1)
`
`O
`
`O
`
`O
`
`h(0)
`
`0
`
`h(L - 2)
`
`h(L — 3)
`
`(5)
`
`0 _
`
`0
`
`h(0)
`
`If e0 denotes the output of the weighting filter due to'tlie
`memory hangover (i.e. , the output as a result of the initial
`filter state) of previous intervals, then the signal e(n) pro-
`duced by the input vector H“ can be described as
`
`em = etO) _ btkin’
`
`k = 1,
`
`-
`
`-
`
`-
`
`, N,
`
`where
`
`e<°> = e0 + rH,
`
`Hk = MkHa
`
`(6)
`
`.(7)
`
`and the vector r represents the residual r(n) for the current 1
`frame. The objective is to minimize the squared error
`
`Ear) = emetic»,
`
`(9)
`
`where t denotes transpose. For a given phase the optimal
`amplitudes b(k)(-) can be computed from (6) and (9), by
`requiring eme: to be equal to zero. Hence,
`
`1,00 = e‘°)H;,[H,,H[]_‘.
`
`(10)
`
`By substituting (10) in (6) and thereafter the resulting. ..
`expression in (9), we obtain the following expression for V
`the'error:
`
`(k) = (0) _ H: HHI —lHIJ e(0)t-
`e [1
`kl
`k
`k]
`_
`
`E
`
`,
`\MA
`
`Ex. 1005 / Page 5 of 16
`
`
`
`IEEE TRANSACTIONS ON ACOUSTICS. SPEECH, AND SIGNAL PROCESSING. VOL. ASSP~34. N0. 5, OCTOBER 1986
`1056 ;
`The vector b“) that yields the minimum value of E 0‘) over
`all k is then selected. The resulting optimal excitation
`vector v“) is entirely characterized by its phase k and the
`corresponding amplitude vector b“). The whole procedure
`'
`’
`f linear equations as
`y (10). A fast algorithm to compute the N vectors
`b“) simultaneously has been presented in [8] and [9]. We
`shall show in Section V that a further reduction in com-
`plexity can be obtained by exploiting the nature of the
`matrix product HkHL in (10).
`
`7(n)
`
`(k)
`
`”
`
`(n)
`
`(k)
`
`b
`
`I")
`
`_
`
`I
`I
`I
`1
`l—
`,_____
`a
`i'
`MINIMI A I
`1.3.3110
`L ____ -3191;
`Fig. 3. Block diagram of a BBC coder (solid lines). and an RPE coder
`(solid and dashed lines).
`
`_
`
`I'II
`
`
`
`, N), an FIR filter Fk(z) such that the weighted
`'
`‘
`'
`— 1,
`Define ka as
`least-squares error 2,, e2(n) over the interval L is minimal.
`L—I
`
`Fk(z) = E) fi")z“’,
`
`(12)
`
`and
`
`A
`
`
`
`
`
`RELATIVEPOWER(d8)
`
`KROON et al.: REGULAR-PULSE EXCITATION
`
`
`r
`hr \
`3M1
`p
`(\/\
`i,
`/”
`I
`//\/V/WV \\/VV
`.
`AN. VJ (it
`-16
`-|6
`NA
`i
`-24
`uN
`('uN
`k=2
` sa
`1.5
`2'0
`30
`4.
`00
`to
`5.0
`$0
`4.0
`FREQUENCY (kHz)
`FREQUENCY (kHz)
`
`_|
`I
`A
`F
`/
`'
`IA
`[\fnrh ,’ I
`(I/IlI/“I
`v
`
`
`
`
`
`0.RELATIVEPOWER(d8)
`
`\rV/‘x
`
`0 a
`
`I
`I
`
`0
`
`-8 1r V \ (AW
`-|6
`I
`(yII
`24
`.32
`4O
`' 00
`
`
`('no
`
`k=3
`
`-
`
`k=4
`
`30
`2.0
`1.0
`FREQUENCY (kHz)
`
`4.0 400.0
`
`10
`2.0
`[0
`FREQUENCY (kHz)
`
`4.0
`
`I6
`
`32
`TIME (ms)
`
`4e
`
`64
`
`Fig.'4. Power spectra le(e ")I‘ for different values of k, obtained from a
`5 ms speech segment.
`
`I
`0
`.
`.
`.
`-
`Hence, with the matrix H and the Initial error e‘ ) as de—
`fined in the previous section,
`
`ear) = em) _ f(k)RkMkH
`
`'
`FI
`
`’
`'
`l for).
`(C)
`(a) Speech sIgnal s(n), (b) reconstructed speech srgna
`. 5.
`Excitation signal u(n), and (d) difference srgnal s(n) — £(n) In the RPE
`coding procedure.
`
`TABLE I
`DEFAULT PARAMETERS RPE ANALYSIS
`
`.
`
`Value;
`
`III. GENERALIZED BASEBAND CODING
`It may be observed that the regular—pulse excitation se-
`quence bears some resemblance to the excitation signal of
`excited baseband coder (BBC) using spectral folding as
`high~frequency regeneration technique [4], [10]. In this
`section we show that the RPE coder can be interpreted as
`a generalized version of this baseband coder. For this pur-
`pose We use the block diagram of Fig. 3. The blocks
`_ drawn with solid lines represent the conceptual structure
`of a residual excited BBC coder with
`this coder, the index k has no signi
`zero. In this scheme,
`the LP-resid
`tained by filtering the speech signal through the filter A(z),
`is band-limited by an (almost) ideal low-pass filter F0(z),
`downsampled to b(°)(n) and transmitted. At the receiver,
`this signal is upsampled to v(°’(n) to recover the original
`bandwidth, and is fed through the synthesis filter to re-
`trieve the speech signal fin). When the dashed blocks are
`included in Fig. 3, one provides a possibility to optimize
`the filter Fk(z), i.e.,
`to replace the ideal low—pass filter
`Fo(z) by another filter, which is more tailored to “opti—
`mal” waveform matching, where the optimality criterion
`is to minimize the (weighted) mean-squared error between
`the original and the reconstructed signal.
`We shall now show that for this “optimized” BBC ver—
`sion, the output of the filter Fk(z), after down- and upsam-
`pling, is exactly the excitation signal v(k)(n) as computed
`by the RPE algorithm, Thus, let there exist for each k, (k
`
`r ;I
`
`= et") — f<’°R,.H,,.
`Minimizing 9009002 we obtain as solution
`
`(18)
`
`Parameter
`
`f“" = e‘°)(RiHI)‘IRka(RkHI>‘I“.
`(19)
`Substituting this result in (15), we obtain the \kIector‘bU‘),
`which is equal to the pulse amplitude vector b‘ ) obtained
`via the procedure described in Section II (see the proof 1n
`the A endix .
`.
`Figfn; give)s an example of the spectra le(e "9)l2 ob—
`tained from real speech data. From this figure we see that
`the filters Fk(z) are rather different from the one (Fo(z))
`used in the classical baseband coder, and have a more all—
`pass character.
`'
`Although the RPE algorithm and the optimal BBC al-
`gorithm are conceptually equivalent, the optimized BBC
`variant will in general not offer any com