Bursty data over CDMA: MAI self similarity, rate control and admission control

We study bursty data communications in the downlink in code division multiple access (CDMA) systems. We first present a new model that simultaneously takes into account the traffic burstiness and time-varying fading for studying the multi-access interference (MAI), and characterize the MAI from a stochastic process perspective. This new approach enables us to understand the temporal correlation structure. Our finding reveals that the MAI exhibits scale-invariant burstiness and is "self similar" across multiple time scales. The MAI self similarity indicates the existence of a nontrivial predictive MAI structure, which we exploit to conduct resource allocation for interference management. In particular, we utilize the MAI temporal structure to construct a multiple time-scale interference predictor, which is used to predict the MAI level. Rate adaptation is then carried out based on the predicted MAI. Our results show that this rate control scheme achieves better performance than that of the packet-level predictor, and can yield significant performance gain. We also devise a joint rate control and admission control scheme. Specifically, observation time windows are divided into slots, and rate control based on interference prediction is conducted in each slot. Then, the corresponding throughput in each observation window is used for admission control. We also investigate the impact of feedback delay and data burstiness on the system performance.