Memory Proportionate APA with Individual Activation Factors for Acoustic Echo Cancellation

Affine projection M-estimate subband adaptive filters for robust adaptive filtering in impulsive noise

“…In order to compare the tracking capability of the algorithms, the impulse response of the echo path is multiplied by À1 at iteration 40,000. The input signal is either an autoregressive (AR) signal or a speech signal [6]. The output of the echo path is corrupted by an independent white Gaussian noise with a signal-tonoise ratio (SNR) of 30 dB [6].…”

“…The input signal is either an autoregressive (AR) signal or a speech signal [6]. The output of the echo path is corrupted by an independent white Gaussian noise with a signal-tonoise ratio (SNR) of 30 dB [6]. In addition, an impulsive interference signal is added to the output with a signal-tointerference ratio (SIR) of À10 dB [17].…”

“…The difference between them is that the input signal matrix of the APA consists of the fullband signal vectors, whereas those of the IMSAF and APM-SAF are composed of the subband signal vectors. Hence, similar to the weight update equation of the proportionate-type APA in [5][6][7][8], the proportionatetype IMSAF and APM-SAF update their coefficients in a straightforward way wðk þ1Þ ¼ wðkÞþμGðkÞUðkÞðU T ðkÞGðkÞUðkÞÞ À 1 eðkÞ ð 23Þ…”

“…Thus, the larger coefficients receive larger increments and so does the convergence rate. Since the different choice of g l ðkÞ will generate the different proportionate-type adaptive filtering algorithm [5][6][7][8], a class of proportionate-type IMSAF and APM-SAF will be obtained. Among these, the following choice is one of the most attractive choices because of its robustness to impulse responses with different sparseness degrees [5][6][7][8] …”

“…One of the useful approaches to accelerate convergence is to use affine projection algorithm (APA) [4] and its variants [5][6][7][8]. Since the APA updates the weight coefficients with multiple, most recent input vectors, its convergence rate increases and so does the computational complexity.…”