VoIP Voice and Fax Signal Processing

4.8 SUMMARY ON VAD/CNG

For waveform-based narrowband codecs such as G.711 and G.726, two popular VAD/CNG algorithms are used. The simplest VAD/CNG works on power levels. In this scheme, VAD detection is made when power level falls below a certain threshold, usually of −42 to −44 dBm represented in dBov units. After sending one VAD packet, another VAD packet will be sent after holding for two frames and power-level changes by more than 2 dB. In the receiver, VAD packets will go through AJB, and this can create a jitter buffer under run. It is suggested to keep memory of the AJB statistics in the transitions of silence and speech. On getting speech packets, established jitter buffer characteristics can be reimposed. In some designs, VAD packets are sent at regular intervals between 100 and 1000 ms to ensure that VAD/CNG works even under packet losses and avoids RTP session disconnects in long silences.

In payload, all negative values of dBov will appear as positive numbers in the range of 0 to 127 dBov. It will correspond to power levels of 6.17 to −120.83 dBm. Ideal channel noise of −70 dBm is the minimum signal in the system. Hence, catering or clipping at −70 dBm or 76 dBov will be sufficient in deciding the power levels. In case of a sudden power change within the VAD power limits, it is required to limit the rate of a VAD packet update. As per G.169, a signal gain change of 10 dB per second is suggested. In the case of VAD-I, the interpretation is to send packets once in 200 ...