Techniekboeken: Advances in Network and Acoustic Echo Cancellation J Benesty

Aangeboden: Advances in Network and Acoustic Echo Cancellation J Benesty Prijs: T.e.a.b. ADVANCES IN NETWORK AND ACOUSTIC ECHO CANCELLATION Digital Signal Processing J. Benesty - T. Gänsler - D.R. Morgan - M.M. Sondhi - S.L. Gay NIEUW boek met harde grijs/rode kaft met zwart/witte letteropdruk; 222 pagina's met 70 tabellen, schema's, illustraties enz. For many decades, hybrid devices have been used to connect 2-wire local circuits and 4-wire long distance circuits in telephone lines. This leads to a well known problem, whereby echoes are generated. The delay introduced by telecommunication satellites exacerbated this problem and the need for new methods of echo control soon became obvious. The best solution to date for solving this problem was invented in the 1960s at Bell Labs by Kelly, Logan, and Sondhi, and consists of identifying the echo path generated by the hybrid by means of an adaptive filter, a technique that became known as an echo canceler. The echo canceler allowed full-duplex communication which was not possible with older echo suppression techniques. Later, with the development of hands-free teleconferencing systems, another echo problem appeared; but this time the echo was due to the coupling between the loudspeaker and microphone. It is not surprising that the same solution was proposed to solve this problem, and most of today's teleconferencing systems have an acoustic echo canceler. More recently, attention has been given to the very interesting problem of multichannel acoustic echo cancellation, which leads to more exciting applications that take advantage of our binaural auditory system. Advances in Network and Acoustic Echo Cancellation is a book on echo cancellation in general and adaptive filtering in particular. Although the topic may sound a little bit old, most of the ideas presented here are new and several of them were never published before. There is no necessary order for reading the chap ters of this book. All the chapters were written in such a way that can be read independently. Chapter 1 introduces the general subject of echo cancellation, giving a historical perspective and overall background to the problem. This serves to set the stage for the various contemporary topics addressed in the subsequent chapters of this book. In Chap. 2, we talk about a whole class of normalized least mean square (NLMS) algorithms, called proportionate NLMS, which converge very fast when identifying the sparse impulse responses typically encountered in network echo cancellation. In Chap. 3, we propose to model the error signal with a robust distribution and deduce from it a robust fast recursive least-squares adaptive algorithm. We then show how to successfully apply this new algorithm to the problem of network echo cancellation when combined with a double-talk detector. Chapter 4 presents some ideas on how to efficiently implement network echo cance l ers designed to simultaneously handle a large number of channels. We show how the computational requirement can be reduced by a very large factor — perhaps as large as thirty. Chapter 5 explains why and for what kind of applications multichannel sound is important for telecommunication. We explain the fundamental lifference from the single-channel case and study a so-called nonuniqueness problem. We also present conventional multichannel time-domain adaptive algorithms. Chapter 6 develops a fast normalized cross-correlation (FNCC) method for double-talk detection. This method is independent of echo path gain but relies on some of the computations performed in the fast recursive leastsquares algorithm (FRLS) used in the acoustic echo canceler (AEC). We also generalize the algorithms to the multichannel case. The combination of the FNCC detector and a robust FRLS algorithm results in a system with extremely low sensitivity to double-talk, few detection errors, a n d fast convergence after echo path changes. Chapter 7 describes a stereo teleconferencing system and the implementation of a stereo echo canceler in subbands. The emphasis here is on some of the more practical issues that must be dealt with in an actual implementation. In Chap. 8, we introduce a new theory on how to develop a whole class of adaptive filtering algorithms in the frequency domain from a recursive leastsquares criterion, with a block size independent of the length of the adaptive filter. Then, we deduce an exact adaptive algorithm in the frequency domain and study the convergence of this generalized algorithm. We suggest a very useful approximation, deduce several well-known algorithms, and give hints on how to choose values for some very important parameters. Finally, we show a rigorous link between the multi-delay filter (MDF) and the affine projection algorithm (APA) and generalize some of these ideas to the multichannel case. In Chap. 9, we de scribe a system for echo cancellation and double-talk control based on frequency-domain algorithms. The advantages of this approach are, among others, algorithm stability, fast convergence and tracking, low computational complexity, and simplicity of implementation. In detail, we derive a new way of calculating a statistic for double-talk detection, based on a robust adaptive multichannel frequency-domain algorithm. In Chap. 10, we show that many well-known variables or equations such as the Kalman gain, the Wiener-Hopf equation, the input signal covariance matrix, and the Schur complements can be explicitly formulated in terms of linear interpolation. Also, the so-called principle of orthogonality is generalized. From this theory, we then give a generalized least mean square algorithm and a generalized affine projection algorithm. We hope this book will serve as a guide for researchers and developers, as well as students who desire to have a new and fresh lo ok a t echo cancellation. including some new ideas on adaptive filtering in general. We also hope it vill inspire many of the readers and will be the source of new ideas to come. As Prof Hansler once wrote about echo cancellation [63]: "From algorithms to systems—It's a rocky road." So let's make it a safe highway! Acknowledgments We would like to thank Eric Diethorn, Gary Elko, and Peter Kroon for carefully reading a draft of this book and offering many useful comments for its improvement. Bell Laboratories, Murray Hill, January 2001 Jacob Benesty Tomas Gänsler Dennis R. Morgan M. Mohan Sondhi Steven L. Gay CONTENTS 1. An Introduction to the Problem of Echo in Speech Communication 1.1 Introduction 1.2 Line/Network Echoes 1.2.1 The Echo Suppressor 1.2.2 The Line/Network Echo Canceler 1.3 Adaptive Cancellation 1.3.1 The Stochastic Gradient Algorithm 1.3 .2 Other Algorithms 1.4 Single-Channel Acoustic Echo Cancellation 1.4.1 Subband Approach 1.4.2 Frequency-Domain Algorithms 1.5 Multichannel Acoustic Echo Cancellation 1.5.1 Misalignment: the Nonuniqueness Problem 1.5.2 Search for Solutions 1.6 Concluding Remarks 2. A Family of Robust PNLMS-Like Algorithms for Network Echo Cancellation 2.1 Introduction 2.2 Adaptive Proportionate Step-size Algorithms 2.2.1 The PNLMS Algorithm 2.2.2 A Signed Regressor PNLMS Algorithm 2.2.3 A Proportionate Affine Projection Algorithm (PAPA) 2.3 The Geigel DTD 2.4 The Robust Algorithms 2.4.1 Estimating the Scale Factor 2.4.2 Detailed Derivation of the Scale Estimate 2.5 Complexity Comparison 2.6 Simulations 2.6.1 Results with Speech as Excitation Signal 2.6.2 Results with the Composite Source Signal as Excitation Signal 2.7 Conclusions 3. A Robu st Fast Recursive Least-Squares Adaptive Algorithm 3.1 Introduction 3.2 A Robust Fast Recursive Least-Squares Adaptive Algorithm 3.3 Application to Network Echo Cancellation and Simulations 3.4 Conclusions 4. Dynamic Resource Allocation for Network Echo Cancellation 4.1 Introduction 4.2 The PNLMS and PNLMS++ Algorithms 4.3 DRA Algorithm 4.3.1 A Simple Algorithm to Update only Active Channels and Coefficients 4.3.2 Stopping Adaptation when Small Residual Error is Detected 4.3.3 Theoretical Reduction in Complexity 4.3.4 Overload Management 4.4 Simulations 4.5 Discussion 5. Multichannel Acoustic Echo Cancellation 5.1 Introduction 5.2 Multichannel Identification and the Nonuniqueness Problem 5.3 The Impulse Response Tail Effect 5.4 Link Between the Coherence Function and the Covariance Matrix 5.5 Some Different Solutions for Decorrelation 5.6 Con venti onal Multichannel Time-Domain Adaptive Filters 5.6.1 The Multichannel RLS Algorithm 5.6.2 The Multichannel LMS Algorithm 5.6.3 The Multichannel APA 5.7 Application: Synthesized Stereo and AEC for Desktop Conferencing 5.7.1 Interchannel Differences for Synthesizing Stereo Sound 5.7.2 Choice of Interchannel Differences for Stereo AEC 5.7.3 Simulations 5.8 Conclusions 6. A Fast Normalized Cross-Correlation DTD Combined with a Robust Multichannel Fast Recursive Least-Squares Algorithm 6.1 Introduction 6.2 The Generic Double-Talk Detector Scheme 6.2.1 Properties of DTDs and Adaptive Algorithms 6.2.2 The Geigel DTD 6.3 Robust Recursive Least-Squares Algorithms 6.3.1 Derivation of the Robust Multichannel RLS 6.3.2 Examples of Robust PDFs 6.3.3 Scale Factor Estimation 6.3.4 The Robust Multichannel Fast RLS 6.4 The Normalized Cross-Correlation (NCC) DTD 6.4.1 Fast NCC (FNCC) 6.4.2 The NCC DTD as a Neyman-Pearson Detector 6.4.3 The Link Between FNCC and the Two-Path DTD Scheme 6.5 Practical Modifications of the FNCC DTD and FRLS Echo Canceler 6.6 Evaluation and Simulations 6.6.1 Receiver Operating Characteristic (ROC) 6.6.2 AEC/DTD Performance During Double-Talk 6.6.3 AEC/DTD Sensitivity to a Time-Varying Echo Path 6.6.4 Two-Channel AEC/DTD Performance During Double-Talk 6.7 Conclusions 7. Some Practical Aspects of Stereo Teleconferencing System Implementation 7.1 Introduction 7.2 Hard Limiter 7.3 Subband Filter Banks 7.3.1 Subband Prototype Filters 7.3.2 Subband Analysis Filter Banks 7.3.3 Subband Synthesis Filter Bank 7.4 Non-Causal Taps 7.5 Stereo Acoustic Echo Canceler 7.6 Dynamic Suppression 7.7 Nonlinear Processing 7.8 Comfort Noise Fill 7.9 Conclusions 8. General Der ivat ion of Frequency-Domain Adaptive Filtering 8.1 Introduction 8.2 General Derivation of FD Adaptive Algorithms 8.2.1 Criterion 8.2.2 Normal Equation 8.2.3 Adaptive Algorithm 8.2.4 Convergence Analysis 8.3 Approximation and Particular Cases 8.3.1 Approximation 8.3.2 Particular Cases 8.4 An FD Affine Projection Algorithm and a Link with the MDF 8.5 Generalization to the Multichannel Case 8.6 Conclusions 9. Multichannel Acoustic Echo and Double-Talk Handling: A Frequency-Domain Approach 9.1 Introduction 9.2 Definitions and Notation for the Multichannel FD Algorithm 9.3 An Outlier-Robust Multichannel FD Algorithm 9.3.1 Scale Factor Estimation 9.3.2 A Regularized Two-Channel Version 9.4 Double-Talk Detection Based on a Pseudo-Coherence Measure 9.4.1 Choosing an Appropriate Detection Threshold 9.5 A Tracking Improvement Technique 9.6 Evaluati on an d Simu lations 9.6.1 Receiver Operating Characteristic 9.6.2 Data for the Two-Channel Case 9.6.3 AEC Tracking and DTD Sensitivity to Transmission Room Echo Path Changes 9.6.4 AEC Tracking and DTD Sensitivity to Receiving Room Echo Path Changes 9.6.5 AEC/DTD Performance During Double-Talk 9.7 Conclusions 10. Linear Interpolation Applied to Adaptive Filtering 10.1 Introduction 10.2 A Generalized Least Mean Square (GLMS) Algorithm 10.3 A Generalized Affine Projection Algorithm (GAPA) 10.4 Conclusions References Index This book brings together many advanced topics in network and acoustic echo cancellation which are aimed towards enhancing the echo cancellation performance of next-generation telecommunication systems. The general subject nature relates to algorithms with increased convergence speed, improved detection of double-talk from near-end speech, robust immunity to undete cted do uble-talk, increased computational efficiency, and multi-channel capability. The resulting compendium provides a coherent treatment of such topics not found otherwise in journals or other books. The chapters are related with a common terminology, but still can be read independently. "Adaptive filtering is now part and parcel of wide range of research, development and production systems. … here is one of the more useful and usable publications the reviewer has read. All-in-all, reading the book would be a worthwhile activity for anyone actively working with adaptive signal processing … ." (Scott D Snyder, The Physicist, Vol. 38 (6), 2001) "This book is clearly and well written and will be of great interest to those interested in adaptive filtering and echo cancellation. It is a fine piece of engineering showing the impact of practical considerations and numerical aspects of optimization algorithms and their implementation." (A. Akutowicz, Zent ralblat t MATH, Vol. 965, 2001)

meer info