7.4 Network Quality Requirements

The measurement of voice quality is rather difficult. A subjective rating scale of 1 to 5, called mean opinion score (MOS), ^[16] is used to state voice quality. Wireline voice quality is normally within 4 to 4.5 MOS, the current wireless voice quality lies between 3.5 and 4 MOS. To improve the quality of voice signal for wireless networks, G.729 adopted 10-millisecond frame times. The computation time is 10 milliseconds and look-ahead delay is 5 milliseconds. This results in a total one-way codec delay of 25 milliseconds. In addition to the delay criteria, speech performance depends also on the bit error rate. The objective of performance under random bit error rate < 10^-3 is recommended not to be worse than that of G.726 under similar conditions. ^[17] To introduce VoIP, the appropriate selection of coding technology is necessary to meet the criteria of delay and bit error rate of the network.

In network applications of speech coding, coded voice signals are transmitted through multiple nodes and links, as shown in Figure 7.2. All these network links and nodes cause impediments to the coded voice signals. The ITU Recommendations G.113 and G.114 ^[18] specify several system requirements, including:

End-to-end noise accumulation is limited to 14 QDU (quantization distortion unit), where each QDU is equal to the noise of a single 64-kbps PCM device
End-to-end transmission delay budget is 300 milliseconds
G.114 limits the processing delay for codec at each end to 10 milliseconds

Among these network requirements, the most important one in the design of VoIP using wireless is the end-to-end delay budget of 300 milliseconds. In digital networks, because of the synchronous nature of transmission, this delay budget is mostly used for the switching and transmission delay. There is very little variation of this delay, called jitter, within the synchronous digital network for voice signals. In the current wireless voice links, the air interface introduces additional delay because of the air link multiple access standards. Most of the wireless link designs attempt to meet the delay requirements of 300 milliseconds for single-link voice calls in a national network. So single-link wireless calls without intermediate satellite links perform with a reasonable MOS rating today, however, if two ends of the connection are wireless links, the speech quality deteriorates below MOS 3.0.

There are three important network performance parameters for wireless VoIP service:

Performance to set up and tear down the call
Quality of voice payload packets during conversation
Performance of the voice session handoff

These parameters will depend on the selection of VoIP service protocols to set up the voice session, the voice signal coding and transporting scheme, and the micro/macro mobility protocols used for VoIP services.

The call setup delay depends on the successful transfer of the current Q.931 and ISDN-type messages and the additional message sets for capability check of the terminals and media packet synchronization in the IP network. Both H.323 and SIP signaling protocol implement this function. The average call setup delay of the present wireless voice network is about 3 seconds, H.323 and SIP implementations will have to meet this delay requirement.

Network voice quality will depend on the contribution of the different components of a hypothetical connection of VoIP in wireless network. A hypothetical connection, as shown in Figure 7.2, includes network components such as a wireless terminal, a wireless link, an access point, a wireless gateway, the Internet, a media gateway, and a landline terminal. The landline phone is assumed connected through an analog line. The wireless terminal includes the codec and the function to map the coded frames into the wireless data channel for communication with the wireless access point. The common wireless channel is shared between multiple users, so to get access to the capacity of the channel is a delay process. The fading and propagation loss of the wireless channel causes the frame error that introduces additional impediments to the transfer of data. Once the packet is received at the wireless access point, it is transferred to the terminating media gateway through the Internet. Internet performance depends on the delay at the different routers and the propagation delay through the transport links. Due to the use of fiber transmission, the transmission related bit error rate or packet loss is almost nonexistent in the links and propagation delay is very small. The delay in the router depends on the long-range dependency of the traffic and link congestion. Using proper engineering techniques, this Internet delay can be maintained within strict limits. The DiffServ and MPLS protocols will be able to support the core Internet with minimum delay and jitter. At the media gateway, the coded voice packets are reconverted to analog voice signals and transferred to the terminating analog voice terminals using copper wire connection. The end-to-end delay of the voice packet for this hypothetical connection can be represented by the following equation:

D_end-to-end = d_jitter + d_wt + d_wc + d_w₁ + d_wap + d_internet + d_mgw

where

d_jitter	=	Delay introduced by the jitter buffer. To compensate for the fluctuating network conditions, it is necessary to implement a jitter buffer in voice gateways or terminals. This is a packet buffer that holds incoming packets for a specified amount of time before forwarding them to decoding. This has the effect of smoothing the packet flow, increasing the resiliency of the codec to packet loss, delayed packets, and other transmission effects. The downside of the jitter buffer is that it can add significant additional delay in the path. It is not uncommon to see jitter buffer settings approaching 80 milliseconds for each direction.
d_wt	=	Delay at the wireless terminal for coding and decoding voice packets and creating voice frames conforming to the Internet frame packet format (TCP or UDP). Each coding algorithm has certain built-in delay. For example, G.723 adds fixed 30-millisecond delay. To reduce the IP overhead, multiple voice packets may be mapped to one Internet frame and thus introduce bundling delay.
d_wc	=	Delay at the wireless terminal to get a wireless data channel and map the Internet voice packet to it. This includes delay for buffer allocation such as GSM TBF (temporary buffer flow)² allocation. In uplink transmission: Before sending the data to the base station, the mobile station must access the common channel in the uplink direction to send the request. Getting permission to send data in the uplink direction takes time, which increases the end-to-end delay.
d_w1	=	Delay to transfer the voice packets to the wireless access point, including the retransmission delay to protect the frame error during propagation.
d_wap	=	Delay at the wireless access point to assemble and reassemble the voice frame from wireless frame formats to the Internet format.
d_internet	=	Delay to transfer the packet through the Internet to the media gateway. This is the delay incurred in traversing the VoIP backbone. In general, reducing the number of router hops minimizes this delay. Alternatively, it is possible to negotiate a higher priority to voice traffic than for delay-insensitive data.
d_mgw	=	Delay at the media gateway to convert Internet voice packets to analog voice signals and transfer to analog voice lines.

The TCP retransmission delay impacts the delay parameters of d_wc and d_wap. The end-to-end delay, D_end-to-end, of voice packets for conversation should be less than 300 milliseconds, as specified by ITU G.114; one-way delay greater than 300 milliseconds has an adverse impact on conversation, and the conversation seems like half duplex or push-to-talk.

The wireless VoIP session handoff between the two wireless access points is determined by the handoff mechanism supported by the wireless mobility protocol. There are two mobility functions for the wireless IP network: (1) micro-mobility and (2) macro-mobility. The consensus between the different standards bodies is that current Mobile IP may be suitable for macro-mobility, but a new technique is necessary for micro-mobility. The potential micro-mobility protocols are GPRS GMM, Cellular-IP, Hawaii, TIMIP, ^[19] IDMP, ^[20], ^[21] etc. The challenge of these mobility protocols is to ensure that the voice packets can be routed to the new access point without any packet loss or significant additional delay on the voice path. Most of the protocols in their current form have difficulty meeting the micro-mobility requirements of the voice packets. The current soft handoff of CDMA, the make-before-break mechanism on the GSM, and hard handoff in TDMA maintain the continuity of voice packet flow during handoff. The method of GPRS and 3G packet handoff do not allow similar mechanisms at this time.

^[16]Lakaniemi, A. and Parantainen, J., On Voice Quality of IP Voice over GPTS, IEEE International Conference on Multimedia, ICME, 3, 751–754, 2000.

^[17]Cox, R.V., Three new speech coders from the ITU cover a range of applications, IEEE Communications Magazine, 39(9), 40–47, Sep. 1997.

^[18]International Telecommunications Union, G.114: Mean One-Way Propagation Time, Recommendation G.114, Telecom Standardization Sector, Geneva, Switzerland, Nov. 1988.

^[19]Grilo, A., Estrela, P., and Nunes, M., Terminal independent mobility for IP (TIMIP), IEEE Communications Magazine, 39(12), 34–46, Dec. 2001.

^[20]Das, S. et al., IDMP: An intra-domain mobility management protocol for next generation wireless networks, IEEE Wireless Communications, Special issue on Mobile and Wireless Internet: Architectures and Protocols, Agrawal, P., Omidyar, G., and Wolisz, A., Guest Eds., 9 (3), 38–45, 2002.

^[21]Misra, A. et al., IDMP-based fast handoffs and paging in IP-based 4G mobile networks, IEEE Communications, Special issue on 4G Mobile Technologies, Lu, W., Guest Ed., 40 (3), 138–145, 2002.