Performance evaluation of packet data services over cellular voice networks

We develop a Markov chain modeling framework for throughput/delay analysis of data services over cellular voice networks, using a dynamic channel stealing method. Some effective approximation techniques are also proposed and verified for simplification of the modeling analysis. Our study identifies the average voice call holding time as the dominant factor affecting the data delay performance. Especially in heavy load conditions, namely when the number of free voice channels becomes momentarily less, the data users will experience large network access delay in the range of several minutes or longer on average. We also examine the data performance improvement by using a priority data access scheme and speech silence detection technique.