QoS evaluation of VoIP end-points

We evaluate the QoS of a number of VoIP endpoints, in terms of mouth-to-ear (M2E) delay, clock skew, silence suppression behavior and robustness to packet loss. Our results show that the M2E delay depends mainly on the receiving endpoint. Hardware IP phones, when acting as receivers, usually achieve a low average M2E delay (45-90 ms) under low jitter conditions. Software clients achieve an average M2E delay from 65 ms to over 400 ms, depending on the actual implementation. All tested end-points can compensate for clock skew, although some suffer from occasional playout buffer underflow. Only a few of the tested end-points support silence suppression. We find that their silence detectors have a fairly long hangover time (> 1 sec), and they may falsely detect music as silence. All the hardware IP phones we tested support some form of packet loss concealment better than silence substitution. The concealment generally works well for two to three consecutive losses at 20 ms packet intervals, but voice will quickly deteriorate beyond that.