Acoustic Echo and Noise Control: A Practical Approach -- Contents -- List of Figures -- List of Tables -- Preface -- Acknowledgments -- Abbreviations and Acronyms -- Part I Basics -- 1 Introduction -- 1.1 Some History -- 1.2 Overview of the Book -- 2 Acoustic Echo and Noise Control Systems -- 2.1 Notation -- 2.2 Applications -- 2.2.1 Hands-Free Telephone Systems -- 2.2.2 Car Interior Communication Systems -- 2.2.3 Public Address Systems -- 2.2.4 Hearing Aids -- 3 Fundamentals -- 3.1 Signals -- 3.1.1 Speech -- 3.1.2 Noise -- 3.1.3 Probability Density of Spectral Amplitudes of Car Noise -- 3.2 Acoustic Echoes -- 3.2.1 Origin of Acoustic Echoes -- 3.2.2 Electronic Replica of LEM Systems -- 3.3 Standards -- 3.3.1 Standards by ITU and ETSI -- Part II Algorithms -- 4 Error Criteria and Cost Functions -- 4.1 Error Criteria for Adaptive Filters -- 4.2 Error Criteria for Filter Design -- 4.3 Error Criteria for Speech Processing and Control Purposes -- 5 Wiener Filter -- 5.1 Time-Domain Solution -- 5.2 Frequency-Domain Solution -- 6 Linear Prediction -- 6.1 Normal Equations -- 6.2 Levinson-Durbin Recursion -- 7 Algorithms for Adaptive Filters -- 7.1 The Normalized Least Mean Square Algorithm -- 7.1.1 Control Structures -- 7.1.2 Stability Analysis -- 7.2 The Affine Projection Algorithm -- 7.2.1 Derivation of the Algorithm -- 7.2.2 Fast Versions of the AP Algorithm -- 7.3 The Recursive Least Squares Algorithm -- 7.3.1 Derivation of the Algorithm -- 7.3.2 Fast Versions of the RLS Algorithm -- 7.4 The Kalman Algorithm -- 7.4.1 Initialization -- 7.4.2 Prediction -- 7.4.3 Correction -- 7.4.4 Colored Noise -- Part III Acoustic Echo and Noise Control -- 8 Traditional Methods for Stabilization of Electroacoustic Loops -- 8.1 Adaptive Line Enhancement -- 8.1.1 FIR Structure -- 8.1.2 IIR Structure -- 8.1.3 Comparison of Both Structures -- 8.2 Frequency Shift.

8.2.1 Basic Idea -- 8.2.2 Hilbert Filter -- 8.2.3 Results -- 8.3 Controlled Attenuation -- 8.3.1 Automatic Gain Control -- 8.3.2 Loss Control -- 8.3.3 Limiter -- 9 Echo Cancellation -- 9.1 Processing Structures -- 9.1.1 Fullband Cancellation -- 9.1.2 Block Processing Algorithms -- 9.1.3 Subband Cancellation -- 9.2 Stereophonic and Multichannel Echo Cancellation -- 9.2.1 Stereophonic Acoustic Echo Cancellation -- 9.2.2 Multichannel Systems -- 10 Residual Echo and Noise Suppression -- 10.1 Basics -- 10.1.1 Echo and Noise Suppression in the Frequency Domain -- 10.1.2 Filter Characteristics -- 10.2 Suppression of Residual Echoes -- 10.2.1 Comfort Noise -- 10.3 Suppression of Background Noise -- 10.3.1 Postprocessing -- 10.4 Combining Background Noise and Residual Echo Suppression -- 11 Beamforming -- 11.1 Basics -- 11.1.1 Spatial Sampling of a Sound Field -- 11.2 Characteristics of Microphone Arrays -- 11.2.1 Directional Pattern -- 11.2.2 Array Gain -- 11.2.3 Further Performance Criteria -- 11.3 Fixed Beamforming -- 11.3.1 Delay-and-Sum Beamforming -- 11.3.2 Filter-and-Sum Beamforming -- 11.4 Adaptive Beamforming -- 11.4.1 Generalized Sidelobe Canceler -- Part IV Control and Implementation Issues -- 12 System Control-Basic Aspects -- 12.1 Convergence versus Divergence Speed -- 12.2 System Levels for Control Design -- 13 Control of Echo Cancellation Systems -- 13.1 Pseudooptimal Control Parameters for the NLMS Algorithm -- 13.1.1 Pseudooptimal Stepsize -- 13.1.2 Pseudooptimal Regularization -- 13.1.3 "Relationship" between Both Control Methods -- 13.2 Pseudooptimal Control Parameters for the Affine Projection Algorithm -- 13.2.1 Pseudooptimal Regularization -- 13.2.2 Pseudooptimal Stepsize -- 13.3 Summary of Pseudooptimal Control Parameters -- 13.4 Detection and Estimation Methods -- 13.4.1 Short-Term Power Estimation.

13.4.2 Estimating the System Distance -- 13.4.3 Detection of Remote Single Talk -- 13.4.4 Rescue Detectors -- 13.4.5 Concluding Remarks -- 13.5 Detector Overview and Combined Control Methods -- 13.5.1 Stepsize Control for the NLMS Algorithm -- 13.5.2 Combined Stepsize and Regularization Control for the NLMS Algorithm -- 13.5.3 Regularization Control for AP Algorithms -- 14 Control of Noise and Echo Suppression Systems -- 14.1 Estimation of Spectral Power Density of Background Noise -- 14.1.1 Schemes with Speech Pause Detection -- 14.1.2 Minimum Statistics -- 14.1.3 Scheme with Fast and Slow Estimators -- 14.1.4 Extensions to Simple Estimation Schemes -- 14.1.5 Concluding Remarks -- 14.2 Musical Noise -- 14.3 Control of Filter Characteristics -- 14.3.1 Overestimation Factor -- 14.3.2 Spectral Floor -- 15 Control for Beamforming -- 15.1 Practical Problems -- 15.1.1 Far-Field Assumptions -- 15.1.2 Sensor Imperfections -- 15.2 Stepsize Control -- 16 Implementation Issues -- 16.1 Quantization Errors -- 16.2 Number Representation Errors -- 16.3 Arithmetical Errors -- 16.4 Fixed Point versus Floating Point -- 16.5 Quantization of Filter Taps -- Part V Outlook and Appendixes -- 17 Outlook -- Appendix A Subband Impulse Responses -- A.1 Consequences for Subband Echo Cancellation -- A.2 Transformation -- A.3 Concluding Remarks -- Appendix B Filterbank Design -- B.1 Conditions for Approximately Perfect Reconstruction -- B.2 Filter Design Using a Product Approach -- B.3 Design of Prototype Lowpass Filters -- B.4 Analysis of Prototype Filters and the Filterbank System -- References -- Index -- List of Figures -- 1 Introduction -- 1.1 Bell's new telephone -- 1.2 The hands-busy telephone -- 1.3 Wallmounted telephone from 1881 -- 1.4 Desk telephone from 1897 -- 2 Acoustic Echo and Noise Control Systems -- 2.1 Structure of a monophonic hands-free telephone system.

2.2 Frequency responses of different communication directions within a car -- 2.3 Structure of a car interior communication system -- 2.4 Behind-the-ear hearing aid -- 2.5 Architecture of a hearing aid -- 3 Fundamentals -- 3.1 Example of a speech sequence -- 3.2 Voiced and unvoiced speech segments -- 3.3 Normalized Gaussian probability density function and normalized approximation for speech signals -- 3.4 Approximations of the probability density function -- 3.5 Probability density function of the logarithmic normalized power spectral density of speech and histograms -- 3.6 Estimates of the power spectral densities of noises measured in a car and in offices -- 3.7 Noise produced by a PC ventilator -- 3.8 Examples of engine noise -- 3.9 Time-frequency analyses and power spectral densities of wind noise -- 3.10 Noise during a change in road surface -- 3.11 Noise caused by a ventilator -- 3.12 Probability density functions of the logarithmic normalized power spectral density of car noise and histograms -- 3.13 Model of the loudspeaker-enclosure-microphone system -- 3.14 Floorplan of an office -- 3.15 Impulse response and absolute value of the frequency response of an office -- 3.16 Impulse responses measured in different environments -- 3.17 Relationship between echo delay and required attenuation -- 3.18 Composite source signal -- 4 Error Criteria and Cost Functions -- 4.1 Half-band filter designed with the Remez exchange algorithm -- 4.2 Cepstral transformation as a preprocessor for cost functions -- 4.3 Structure of a single-channel hands-free telephone system -- 4.4 Double-talk detection in hands-free telephone systems -- 5 Wiener Filter -- 5.1 Wiener filter -- 5.2 Example of a Wiener filter -- 5.3 Error surfaces -- 6 Linear Prediction -- 6.1 Prediction error filter.

6.2 Power spectral densities of a speech signal and the related decorrelated signal -- 6.3 Original speech signal and prediction error signals -- 6.4 Predictor of order k -- 6.5 Inverse spectral power density of the predictor input and absolute values of the predictor transfer function -- 7 Algorithms for Adaptive Filters -- 7.1 Adaptive filter for system identification -- 7.2 Convergence with colored noise -- 7.3 Adaptive filter for system equalization -- 7.4 Example of a vector stepsize control -- 7.5 Prewhitening structure -- 7.6 Contraction-expansion parameter -- 7.7 Simulation examples and theoretic convergence -- 7.8 Convergence plane in the absence of local noise -- 7.9 Convergences with different filter orders -- 7.10 Time-variant prewhitening structure -- 7.11 Convergence examples-only stepsize control -- 7.12 Convergence examples-only regularization control -- 7.13 Control strategies -- 7.14 NLMS versus AP -- 7.15 Convergences for different algorithms -- 7.16 Kalman filter: state space description of coefficient update -- 7.17 Statespace model of the echo canceling filter and the Kalman filter -- 7.18 Sequence of steps of the Kalman filter algorithm -- 8 Traditional Methods for Stabilization of Electroacoustic Loops -- 8.1 Basic scheme of adaptive noise cancellation -- 8.2 Structure of an FIR filter for adaptive line enhancement -- 8.3 Simulation example for FIR adaptive line enhancement. -- 8.4 Structure of an IIR filter for adaptive line enhancement -- 8.5 Simulation example for IIR adaptive line enhancement. -- 8.6 Time-frequency analysis of the estimated distortion signals -- 8.7 Structure of a public address system -- 8.8 Frequency response measured in a small lecture room -- 8.9 Spectra of original and shifted signals -- 8.10 Structure of the frequency shift -- 8.11 Frequency responses of Hilbert filters.

8.12 Additional gains due to the frequency shift.