Pruebas de tecnologia

1022

IEEE JOURNAL ON SELECTEDAREAS IN COMMUNICATIONS, VOL. SAC-1, NO.

6, DECEMBER 1983

Techniques for Packet Voice Synchronization

Abstruct -Packet switching has been proposed as an effective technology for integrating voice and data in a single network. An important aspect of packet-switched voice is the reconstruction of a continuous stream of speech from the set of packets thatarrive at the destination terminal, each of which may encounter a different amount of buffering delay in the packet few network. The magnitude of the variation in delay may range from a milliseconds in alocalarea network to hundreds of millisecondsina long-haul packet voice and data network. This paper discusses several aspects of the packet voice synchronization problem, and techniques that can beused to address it. These techniques estimate in some way the delay encountered by each packet and use the delay estimate to determine how speech is reconstructed. The delay estimates produced by these techniques can be used in managing the flow of information in the packet network improve overall performance. Interacto tions of packet voice synchronization techniques with other network designissues are also discussed.

PLAY A

PLAY B

PLAY C (MISSED)

. PLAY D
L

TIME

Fig. 1. Packetvoicesynchronization.

11. THE PACKET VOICE SYNCHRONIZATION PROBLEM

I. INTRODUCTION
RGUMENTS have been advanced to show that packet switching can be an effective technology to integrate voice and data in a single network [l], [2], [3]. A packet network can exploit the bursty nature of dataand voice to achieve a reduction in the transmission bandwidth needed to carry a particular mix of traffic over a circuit network. Packet switching offers many other advantages, including a more flexible allocation of bandwidth to individual calls, and interfaces that are well matched to modern terminals. A number of significant technical problems exist in sending voice through a packet network.One of the most significant involves the reconstruction of a continuous stream of voice from a set of packets sent through the network [4],[5],[6]. With packetized voice, each packet contains a segment of speech plus some header information. Packets may either be fixed or varying in length. Because of the statistical nature of packet switching, each packet can encounter a different amount of delayin traversing the packet network. If the packets of a given call are routed independently, where each may follow a different path through the network, packets may even arrive at the destination out of order. The significant issue involved is, thus, how to design an appropriate voice reconstruction strategy that reproduces acceptable quality speech from packets that arrive with varying transit delayand may, in somecases, arrive out of order. The remainder of this paper discusses the packet voice synchronization problem and some potential solutions.
Manuscript received March 31, 1983; revised April 11, 1983. The author is with Bell Laboratories, Naperville, IL 60566.

A

As discussed, the fundamental problem that we are addressing is to reconstruct a continuous stream of speech frompackets that amve with varying transit delay. This is illustrated by Fig. 1, which shows the timing relationships in a typical'packet voice network. Packets are produced at the packet voice sender (PVS) and sent through the network. Typically, packet voice systemsproduce packets at a regular time interval. The selection of the packet production interval is governed by the desire to keep bufferingdelay and overhead small. Low delay requires small packets, while low overhead requires large packets. Packet intervals of 10-50 ms are typical [l], [7].'We assume that packets are not produced during silent intervals in the call. As the packets pass through the packet network, each can encounter a varying amount of queueing delay in the statistically multiplexed links. The variation in delay...