Adaptive Quantization for Multichannel Wiener Filter-Based Speech Enhancement in Wireless Acoustic Sensor Networks

Algorithm 1

Greedy adaptive quantization for MWF in WASN.

(1) Choose a metric from , , or .

(2) Initialize to the dynamic range of each sensor.

(3) Initialize the bit depth assignment to the maximum bit depth allowed by the hardware.

(4) Initialize using equation (12).

(5) while    do

(6)  Each signal is quantized in time domain with bits using (6).

(7)  Receive signal frames from .

(8)  Apply STFT to the received frames.

(9)  Compute based on the quantized microphone signals using equation (15).

(10) Update using and equation (12) (The update is done with in order for the metric to predict

     what would happen if the bit depth of the thsignal is reduced by 1 bit. However only one signal gets its actually reduced

     in step (14)).

(11)  Compute the selected metric according to equation (29), (30), (31) or (27) respectively.

(12)  Combine using equation (32).

(13)  Find the index of the signal with minimal .

(14)  Reduce by bit.

(15)  If equals 0 after the reduction, remove the th signal for subsequent iterations.

(16) end while