Customer Service and Booking Hotline: +44 (0) 845 672 0175

Search

Voice Quality Measurement

VoIP is vulnerable to poor quality through mechanisms that did not affect traditional circuit-switched, PSTN telephony. Excessive delay, delay variation (jitter) and packet loss can all make the quality of calls to end users unacceptable. Fortunately the means to define and measure voice quality have matured in the last several years....


The original measure for speech quality based on listening tests outlined in the ITU-T P.800 specification has test subjects (listeners) in a special design closed room. They are then asked to score the sounds they hear on a scale of 1-5. This is the basis of the so called Mean Opinion Score (MOS).

This rudimentary scoring mechanism is somewhat out of date for packet based voice and video communications as many different factors all contribute to the quality a customer experiences

  • latency
  • jitter
  • packet loss
  • transcoding
  • echo

There is now a large industry dedicated to providing carriers with passive and proactive tools to monitor and report on VoIP and Video over IP quality.

The ITU-T improved on the original MOS score model with Perceptual Speech Quality Measure (PSQM) in recommendation P.861, which was subsequently replaced by Perceptual Evaluation of Speech Quality (PESQ) in standard P.862 for narrow-band codecs (3.1KHz).

Interestingly, the PSQM measurement did not provide a comprehensive evaluation of two-way speech quality. PESQ is the combination of a number of efforts to be able to evaluate better the factors that affect voice quality in IP networks; it incorporates the PSQM measures and work done in BT Group called Perceptual Analysis/Measurement System (PAMS)

ITU-T recommendation G.107 describes a model for transmission planning called the E-model. The E-model is combines a number of different impairments to calculate an overall quality measure, R.

R = Ro - Is - Id - Ie + A

The R factor has a value between 0 and 100, with 100 representing the best quality and 0 the worst quality.

  • Ro is the signal to noise ratio including background noise and transmission noise
  • Is represents the simultaneous impairment factor
  • Id represents impairment caused by delay
  • Ie is the equipment impairment factor representing impairments from low-bit codecs and packet loss
  • A is the advantage factor, which is a compensator to off-set the others. For example using a DECT handset or a mobile phone gives user mobility advantages which off-set the impairment caused by (for example) errors in the radio interface

The other important factor when measurements of voice (and video) quality are undertaken is to understand the need for both active and passive testing. Passive testing involves probes being present both in various points in the network and potentially in the devices which are generating and carrying the voice traffic. Active testing involves the generation of test traffic between points in the network to continually assess impairments. The vendors of QoS monitoring equipment have products which do both of these. It is also important to ensure whole calls are monitored as packet network behaviour can be highly variable over durations of several minutes.

The IETF has recommended a more up to date approach for measuring and reporting voice quality, defined in RFC3611. This describes an extension to the Real-Time Control Protocol (RTCP) called RTCP-XR.

The values measured are based on an enhancement to the E-Model which improves on the calculation of the equipment impairment factor.

RTCP defines six types of packet used to report performance, as follows:

Sender Reports include

  • Number of RTP packets sent
  • RTP timestamp
  • NTP timestamp
  • senderís packet count
  • Sender Identity

Receiver Reports include

  • Number of lost packets
  • Inter arrival jitter
  • sender identity
  • last sender report
  • delay since last sender report

Source Description Packet

Goodbye Packet

Application Specific Packet

Extended Report Packet (XR)

The original specification for RTP and RTCP had RTP occupying even UDP port numbers and RTCP occupying the odd-numbered port immediately above the RTP port; however this restriction was later relaxed, although a number of devices may well still conform to this.

Bookmark this article

Share this article using the following sites:

Courses by category...

Glossary Search

Newsletter Sign-up

Our RSS Feeds...