ALGORITHMS FOR ESTIMATION AND CONTROL OF PACKET VOICE PLAYOUT DELAY SO AS TO MINIMISE THE END-TO-END VOICE DELAY IN PACKET TELEPHONY

摘要

A technique using algorithms called EXD-TLP on excess play out delay estimation, with a target loss probability in packet telephony for maximum reduction of end-to-end mouth to ear (M to E) delay. The invention' directly estimates the excess delay required at the beginning of the talk spurt thereby facilitating the straggling packets to catch up. The algorithms use the target loss probability (TLP) as a parameter and produce an excess delay that will achieve this TLP. The first algorithm is based on the stochastic approximation (SA) approach, which involves the loss probability function Ploss(h,b), where 'h' is the voice activity detector (VAD) hangover and 'b' is the play out delay. The approach being to improve the estimation of 'b', the adjustments being driven by the errors between the observed loss probability and the TLP (Algorithm being called EXD-TLP-SA). The second algorithm gives an estimate of the "optimal" play out delay for each talk spurt, using the exponentially weighted moving average (EWMA) approach to obtain a running estimate of the play out delay from the "observed" play out delays. This algorithm is called EXD-TLP-EWMA. The method is to dynamically adjust 'h', so that a small value of 'h' results in reducing play out delay and a larger 'h' helps to make the received speech less sensitive to silence period fitter, and thus in the equation Ploss(h,b)=P*, both the hang over 'h' and the excess play out delay V can be chosen, so as to minimize V while keeping 'h' above some minimum desirable value. In optimizing the play out delay, the receiver therefore needs to periodically feed back new 'h' , values to the feeder.