Monitor voice quality

Quality-of-service parameters can be measured in real time.

Enterprises providing voice services over IP (VoIP) on their corporate networks need to measure call quality in the same way that they measure availability, response time or utilization. They need a way to measure the quality parameter in quality of service.

To date, tools available to enterprises and service providers have been predominately lab and planning tools, taking such input parameters as delay and packet-loss rate, or doing time-consuming comparative signal analysis between audio files sent and received. They are not getting, however, a true picture of the quality of service their users are experiencing, or that the network is providing.

Enterprises and third-party providers often overprovision the network to reduce the rate of quality problems. This leads to high network capital, maintenance, manpower and operating costs.

In order to understand the end-user experience and to mitigate the need to overprovision, proactively monitoring the quality of every call made in a network is essential for network operations staff. This is not fundamentally different than using remote monitoring probes to measure traffic rates and volumes. With voice traffic, however, the issue is not how much but at what quality.

There are several shortcomings with existing voice-quality measurement tools. The major problems with current tools are:

• They may have an inability to incorporate the impact of time-varying impairments, such as packet loss, which result from the bursty nature of the traffic, and cause voice quality to vary during a call.
• They do not consider the effects of human perception and memory. Network managers and service providers need to know the end-user’s opinion of the call-quality or service, especially when the network’s experience is different from the end-user experience.
• They are intrusive, which means that test calls must be made through the network, and run repeatedly in order to produce sufficient samples to make informed decisions. This uses additional network bandwidth, possibly reduces the quality of other calls in progress and may steal bandwidth from pending calls.
• They are compute intensive, requiring time and high-performance, PC-based hardware. Only a small number of such test calls can be made, resulting in only occasional snapshots of performance.

Generally speaking, three issues impact the quality of voice over IP. They are delay, jitter and packet loss. Delay mostly affects conversational quality, rather than received voice quality. For example, expecting a reply, a speaker may repeat utterances, which collide with the delayed response from the other end, causing double talk. Jitter is largely compensated for by adaptive jitter buffers, which, in general, turn the jitter problem into a delay and packet-loss problem, where excessively delayed packets are dropped (i.e., lost).

Packet loss typically occurs due to congestion or rerouting in the network and is likely to occur in bursts. The degree of burstiness has a major impact on voice quality. Packet-loss-concealment algorithms can disguise isolated lost packets but are much less effective when many packets are lost in bursts.

For example, a three-minute VoIP call may contain two seconds of 50% packet loss. Current voice-quality monitors would average this over the entire call, assuming a uniform 0.5% of packet loss, and thus predict no degradation in quality. That prediction would be different if the actual packet loss distribution and end-user’s perception of the call were considered. Existing real-time voice-quality monitoring technology is available that can recognize that a significant burst of packet loss was perceived by the listener and predict its subjective impact.

At the other end of the packet-loss spectrum—at very low packet-loss rates—burst packet-loss can actually sound better than random loss as the gaps between packet-loss events become large. At packet-loss rates in excess of 3%, burst packet loss has a more significant effect on voice quality than random packet loss. This, too, is unnoticed by current tools.

Enterprises and service providers alike need a voice-quality monitor that tracks all calls in real time and properly models the effects of time-varying impairments on end-user-perceived quality. Only with such visibility can network managers control the various operational, planning and support aspects of their networks.

Massad is VP of Marketing for Telchemy Inc., Atlanta. Comments for publication should be sent to guest@comnews.com.