Investigation of a voice-data switch with optional silence suppression [thesis]

Sandeep Chandra
1991
The combined behaviour of voice and data traffic is studied for the Fixed and Movable Boundary schemes of multiplexing when DSI (Digital Speech Interpolation) is optionally used. A combined speaker model is developed to model voice traffic. Results are compared with the popular Weinstein's Dynamic Talker Activity Model. It is asserted that our model is simpler and caters to miscellaneous probability distributions of talkspurt and silenceperiods. At the outset a simple case of a TDM multiplexer
more » ... tructure with 2 slots/frame is analyzed. Voice is assumed to have pre-emptive priority over data. Some results are shown and verified by simulation. Next a typical frame size of 15 slots is examined. For both Fixed and Movable Boundary schemes three frame structures are considered i.e. with 20%, 50% and 80% of total capacity for data traffic. Each is studied in two stages: without and with DSI. In the first stage the number of active speakers is kept fixed and equal to the maximum number of voice slots in a frame. It is seen for both schemes that average data delay reduces as average number of data slots increase. Further, for a given frame structure the Movable Boundary scheme causes lower average data delays compared to the Fixed Boundary scheme. In the second stage, the relationship between the number of active speakers, probability of voice clipping and average data delays is studied. It is seen that increasing the number of active speakers causes the average data delays to increase and, voice clipping to occur if this number goes beyond a corresponding threshold. Based on these results, the third frame structure is chosen for further study. The relationship between the average data delay and buffer size for different data loads is then studied. A retransmission policy is then formulated which takes into account the time-out intervals of the data sources connected to the switch. Finally an analytical model for integrated voice and data traffic is put forth. A similar model based upon the on-off characteristics of voice exists in the literature. Our model takes into account the talkspurt-silence characteristics of voice. Using this model approximate expressions for the expected data queue lengths are presented. At low to moderate 99%) data utilization close agreement is found between the simulated and analytical results. Reasons for departure from the simulated results at higher utilization are discussed. It is suggested that this model in conjunction with the above work be used to calculate the required data buffer of the switch.
doi:10.26190/unsworks/4890 fatcat:rke6b5w3ybawtjoiopfymjmu3i