In VoIP, audio signal from the emitting source is encoded to numerical data and encapsulated into UDP packets before being sent to the receiving end. Since voice is a real time phenomena, audio data must be processed and played-out in the receiver side in a timely manner. This article discusses the effect of wireless handoff delays on the QoS (Quality of Service) of VoIP sessions.

In digital audio coding and transmission, the original speech is divided into portions S1, S2,.. at regular time intervals. The main reason is that if the whole speech were to be encoded and sent altogether, the delay will be too long before the receiving end obtains the signal. The “pipeline” approach is thus used in digital audio transmission by dividing the original signal into smaller portions that can be sent in regular intervals. The receiving end regroups the chunks of data and can start playing the audio signal before the speaker in the other end finishes the whole speech.

Assume E1, is the encoded form (sample) of an audio signal (or portion of speech) S1. Similarly, E2 is the encoding of audio signal portion S2 that follows S1. It is necessary that the receiving end processes the samples E1 and E2 and play them back sequentially, with minimum delay between them in order to reproduce the continuous audio signal composed of S1 and S2 recorded at the emitting end.

Jitter in computer sciences is a term used to refer to the variation in network delays between two end-points. For example, imagine you are downloading a file using ftp. The packets from the ftp server to your machine may take a delay D with a deviance (+/-)J. Meaning that the delay the packet takes is not exactly D but it is somewhere between D-J and D+J. The jitter is the parameter J in this case.

If the packet that contains E1 arrives at instant T1, then the packet that contains E2 must arrive on time, so that, right after playing E1, the receiving end can start playing E2. Let T2 be the instant when the packet that contains E2 arrives. The following condition must be met :

T2 - T1 > T_Play

Where T_Play is the time required to decode and play audio samples. Basically, the portion E2 must arrive before the receiver finishes playing the previous portion E1.

If there is a jitter, packets that take longer than D to travel the network may not be in the receiving side when expected and will be dropped or the audio signal will be interrupted.

In order to resolve the problem of “too late” packets, VoIP applications use a buffer that compensates for network delay variations. The receiving end delays the beginning of the processing for a delay DJ longer that D+Jitter (DJ corresponds to the longest tolerable packet transit time). This allows the receiving end to buffer more packets and give chance for packets with extra delay to arrive on time.

This has been working fine in current deployments because jitters in today's Internet are quiet stable.

Wireless handoffs are not the Internet, and is not yet a tamed phenomena. when a wireless station changes of access point. Link layer procedures take place to build a new association with the new wireless access point. This procedure referred to as handoff may take large delays (couple of seconds in some cases). During the handoff process, packets are buffered by the infrastructure until the wireless station is ready to receive them. This delay propagates to higher layers and cause sudden increases in packet delays (positive jitter). Assuming that the wireless station is the receiving end in a VoIP session. All buffered packets in the de-jitter buffer can be processed before the wireless station receives more audio packets. This situation occurs when the delay introduced by the handoff is larger than the tolerated delay DJ.

One cam think of increasing the tolerated delay of the de-jitter when wireless mobile communications are involved. However the drawback in this approach is that the interactivity will not be optimal. The ITU (International Telecommunication Union), recommends mouth-to-ear delay less than 150ms. The size of the de-jitter is thus limited. The encoding of the audio signal + The transmission of the packet over the Internet + The waiting time in the de-jitter + the decoding and play-out time needs to be less than 150ms. This leaves only a dozens of milliseconds to allocate for the de-jitter.

The only remaining approach is to reduce wireless handoff delays. The handoff must not introduce delays larger than the allowed de-jitter length, this is typically about 50ms.

T-REC-G.114: International telephone connections and circuits – General Recommendations on the transmission quality for an entire international telephone connection