Benesty J., Paleologu C., Gänsler T., Ciochina S. A Perspective on Stereophonic Acoustic Echo Cancellation

Springer, 2011. — 146 p. — ISBN10: 364222573X, ISBN13: 978-3642225734.The objectives of this book are to recast the stereophonic echo cancellation problem using the widely linear (WL) model, as well as in this framework present and analyze some of the typical algorithms applied to the stereophonic case. Chapter 2 describes the stereophonic echo cancellation problem as a WL model and redefines some of the evaluation criteria commonly used in echo cancellation. General identification of the stereophonic echo paths using the Wiener formulation in the WL stereo framework is discussed in Chapter
3. This chapter also analyzes the nonuniqueness problem and presents a new approach to preprocessing the loudspeaker signals. Three chapters are devoted to classical as well as improved variants of adaptive filters for the SAEC problem. Stochastic gradient methods, of which the normalized least-meansquare (NLMS) algorithm belongs, is the topic of Chapter
4. This chapter also discusses in detail how to appropriately regularize the algorithms. Regularization is extremely important for practical implementations of echo cancelers. Moreover, variable step-size control for NLMS based algorithms are presented. For the stereophonic problem, the ability of the adaptive algorithm to exploit the spatial correlation between the channels is important. A family of algorithms with this ability is based on affine projections (APs). Chapter 5 goes into details of these algorithms. AP algorithms (APAs) have less degrees of freedom for spatial decorrelation compared to RLS based algorithms. However, the APA is less computationally complex compared to the RLS and is therefore an interesting alternative for realtime implementations. RLS adaptive filters are the most flexible algorithms when it comes to handling the problems occurring in stereophonic echo cancellation systems. Hence, a full derivation of the WL model-based RLS as well as a fast version are described in Chapter
6. The problems of double-talk and residual echo and noise handling are treated in Chapters 7 and 8, respectively. Chapter 9 presents extensive simulation results from most of the algorithms described in previous chapters.Problem Formulation.
System Identification with the Wiener Filter.
A Class of Stochastic Adaptive Filters.
A Class of Affine Projection Algorithms.
Recursive Least-Squares Algorithms.
Double-Talk Detection.
Echo and Noise Suppression as a Binaural Noise Reduction Problem.
Experimental Study.