Traditional circuit-switched networks have been optimised over many years to inherently deliver time-sensitive voice traffic with high quality, essentially by allocating dedicated bandwidth and using analogue-to-digital coding techniques without compression. As a result, the voice quality delivered by the fixed PSTN has set a standard that end users have become accustomed to and will not tolerate at a sub-standard level.

With the deployment of converged next generation networks, voice quality is neither inherently guaranteed nor predictable but must be closely monitored and managed. The debate about how best to do this has smouldered for some considerable time. Typically, some operators have relied on gauging it indirectly by measuring network parameters like packet loss, jitter and latency believing that quality is not an issue so long as the network is properly engineered. Others have sought to supplement this by using measures more closely related to actual user experience. Now there is a new speech assessment standard – known as ITU-T P.862 – which for the first time accurately measures voice quality on packet networks as perceived by the end user. Will this change the debate and will it have an impact in competitive markets?

How well can you hear me . . . ?

The first significant technique used to measure speech quality was developed in the late 1980s and literally asked large numbers of human listeners to rate the clarity of either two-way conversations or one-way listening of samples transmitted over a network. Because of the large numbers involved, statistically valid subjective clarity scores could then be produced. Called Mean Opinion Scoring, or MOS, the ITU published recommendations P.800 and P.830 to describe the methods employed in detail.

Of course, this approach has several drawbacks. Not only is it expensive to recruit the large numbers of people required and set up the tests, it is essentially not repeatable, not necessarily consistent and quite impractical to use for frequent testing – as is needed both for network design and for routine network monitoring.

To cater for the need for an objective, automated and repeatable testing method, a technique was developed in the early 1990s at KPN Research in the Netherlands called Perceptual Speech Quality Measurement (PSQM). Subsequently, from 1993 to 1996, several methods for objectively measuring speech quality where investigated by the ITU. It compared the results of these to determine which correlated best with subjective test results (such as those produced by MOS), chose PSQM and published recommendation P.861 in 1996. It has since gained wide acceptance as a consistent and accurate measurement of speech quality based on human perception factors.

 . . . not so good on an IP network

Unfortunately, one of PSQM’s drawbacks is that it does not accurately report the impact of distortion when caused by packet loss or other types of time clipping. Under such conditions, it reports better clarity than a human listener would.

To cater for this, PSQM+ was devised which provides some adjustment in this type of situation and PSQM99 provided further improvements in 1999. Meanwhile, an alternative approach called Perceptual Analysis Measurement System (or PAMS) was also devised at British Telecom in the UK and first published in 1998.

To cut a long story short, the ITU conducted a review of all new methods during 1998 to 2000 and concluded that both PSQM99 and PAMS matched subjective testing best. As a result, it was determined that each had significant merits and that it would be beneficial to the industry to combine the merits of both in a new standard. This was duly christened Perceptual Evaluation of Speech Quality (or PESQ), with recommendation P.862 published in 2001.

What now?

As IP Telephony and related added-value services are increasingly implemented in next generation networks and as multi-service traffic across such networks grows, ensuring that voice quality is comparable to traditional PSTN-based voice services will gain considerably greater significance – for both fixed and mobile use.

Perhaps of even greater importance will be the ability to prove it to end users. For business users, Service Level Agreements will inevitably be required to incorporate more relevant user quality measures. In domestic markets, users will be presented with more alternatives if they are unhappy with the quality offered by their current provider. Constant monitoring by service providers is therefore likely to become a feature of the market. 

It seems inevitable, therefore, that PESQ and its successors will play an increasingly crucial role in defining and measuring competitive service offerings in the future. 

 © e-principles 2001

Robin Duke-Woolley

Any comments on this article? Please send them to : Editor@e-principles.com