TELECOM ACCESS STANDARDS NEWSLETTER NO. 134 SEPTEMBER 2002 CONTENTS 1. VOICE OVER INTERNET PROTOCOL (VOIP) 2. CONVERSATION QUALITY 3. DATA VERSUS VOICE 4. TRADITIONAL TRANSMISSION PLANNING AND VOIP 5. WHAT NEEDS TO BE DONE? RETURN TO MAIN INDEX 1. VOICE OVER INTERNET PROTOCOL (VOIP) Setting reasonable Quality-of-Service objectives for Voice-over-IP private networks connected to the PSTN has been a major issue for Access Standards in the past few months. This Newsletter gives an overview of the issues that are accentuated by the use of VoIP technology. It is clear that: * VoIP offers cost savings and increased functionality, so it will increasingly be used in both private and public networks. * There is the usual trade-off between performance and cost. Voice quality approaching that of the existing PSTN is achievable, but only if there is not too much emphasis on cost cutting. * If too many customers accept lower quality at lower cost, this will over time lower the voice quality of all connections. * There is a collective responsibility on providers/operators of private and public networks to ensure that calls involving multiple public and private networks are satisfactory. 2. CONVERSATION QUALITY Telephone conversation is an interactive process that is affected by both the fidelity/level of the received speech and the delay between what is said and what is heard. Voice quality is all about ease of conversation. Most readers will remember how hard it was to carry out a normal conversation on an international call carried via satellite because of the unnatural delays that were involved. VoIP technology can also add both significant delay (and distortion) when compared with older telephony technologies. Mouth-to-ear delays exceeding about 150ms make interactive conversation quite difficult. In previous newsletters we have concentrated on CPE loudness ratings. Ensuring that overall loudness ratings of end-to-end connections are near optimal is the first step in achieving VoIP voice quality. Getting Loudness right costs no more than getting it wrong, and it allows other impairments to be tolerated before ease of conversation is significantly affected. Digital encoding, decoding, and transcoding processes all degrade speech quality to some extent, particularly with low bit rate encoding algorithms. These are often used to reduce the cost of linking circuits, despite the increased latency (delay) they introduce. Significantly more delay can be added in VoIP due to packetisation of the speech signal, packet switching times, buffering to accommodate jitter (variation in delay) and avoid packet loss. Limiting packet loss is essential if further degradation of the voice signal is to be avoided. Delay and packet loss can be reduced to acceptable levels by giving voice packets priority and ensuring that sufficient bandwidth is provided in the first place. However, this all costs more. 3. DATA VERSUS VOICE The usual claim is that savings can be made by integrating the voice services with an existing data LAN. However, it is important to note some key differences between data and voice. Data is not particularly sensitive to delays of even a few seconds, let alone milliseconds. On the other hand, it is essential that data is error-free, so processes are in place to check for lost or degraded packets and re-transmit them where necessary. Voice is different in that anything more than 150 milliseconds end-to-end have a significant impact on conversation quality. Spoken language is highly redundant and can tolerate at least some distortion or packet loss without loss of basic communication. Because of these differences, it is not simply a case of adding voice capability into a data LAN. It is essential that the voice facilities are properly designed if conversation quality is to be maintained. 4. TRADITIONAL TRANSMISSION PLANNING AND VOIP Traditional transmission planning has been based on establishing nominal (preferred) and limiting values for loudness and other parameters. Budgets for end-to-end connections are allocated for each component; the CPE, the access lines, and the core network as far as the point of interconnection to other national networks and the international network. Similar design practices are used by other networks and countries in line with the ITU-T Recommendations for end-to-end call quality. This approach has resulted in most connections having near optimal voice quality with only a very small proportion of connections experiencing worst case conditions. The probability of telephones and access lines at both ends of any given connection all having worst case transmission conditions at the same time is relatively low. R-value is a quality rating as determined using the ITU-T Rec.G.107 E-model. For an all digital network the maximum possible R-value is 93.2. An R-value of 70 indicates some users would be dissatisfied and an R-value of 50 that nearly all users would be dissatisfied. This is illustrated by the curve on the right-hand side of the figure below, which indicates the quality of circuit-switched PSTN local calls, with very few calls having an "R-score" of less than 80.. It is expected that growing use of VoIP in both private and public networks will mean that more and more calls traversing the PSTN will experience at least some additional impairments. If there is too much cost-cutting and these networks are not properly designed to maintain good voice quality, the trend will be for higher percentages of calls to experience poorer transmission, leading to increased user dissatisfaction. This trend is illustrated by the left-hand curve of the figure. WARNING: Telecom acknowledges and supports the move to VoIP, but if too many VoIP private network implementations elect to trade voice quality for lower cost then the end result may be far worse than indicated above, with a lot more calls falling into the "below 70" class. Should this occur, EVERYONE will be affected to at least some extent. However, poor service quality will be far more apparent to those customers connected to a "cut cost" private network, as they will have higher probability of connecting to another similar network than will the average caller. 5. WHAT HAS TO BE DONE? Newsletter No. 135 and subsequent Newsletters will provide a lot more detail on private network design matters. Unfortunately, many of the various design parameters and impairments are interactive and it is no longer a simple case of setting limits on the impairments or characteristics of each network or CPE item in isolation. This is where there is a collective responsibility for not only public network operators, but also for each private network operator to ensure that a more moderate approach is taken to cost cutting. DOUG BURRUS Manager Access Standards