| In this thesis, optimal and near-optimal power allocation strategies for Automatic Repeat reQuest (ARQ) protocols operating in wireless environments are studied. These policies minimize the long-term average power expended, while guaranteeing Quality-Of-Service (QoS) constraints such as average throughput. Power is initially adapted based only on the Channel State Information (CSI) made available through feedback. The limitations of such a scheme motivate us to study policies that take into account window dynamics of the ARQ protocol as well. We study two sub-optimal schemes that differ only in the degree of interaction between data link and physical layers. The optimal policy is formulated using dynamic programming; this has an exponential complexity in the sender window size. The intractability of the optimal policy motivates us to study sub-optimal policies that are based on limited state information. For small window sizes, sub-optimal policies' performance is observed to be close to the optimal policy in numerical simulations.; Initially we consider Independent and Identically Distributed (i.i.d.) fading channels and next model cases when the fading is correlated from slot-to-slot. Renewal theory arguments and Markov chains are used in the analysis of the sub-optimal schemes operating in such channels. For correlated channels, the dimensions of the Markov chain increases with the product of the sender window size and number of channel states. This motivates us to study an i.i.d channel-approximation for large sender window sizes. We also outline the penalty in transmission power when the transmitter ignores the channel memory altogether. Numerical results seem to suggest that ignoring memory either completely or partially does not deteriorate performance appreciably, as long as the first-order statistics are known.; Finally, we study two kinds of iterative schemes when channel statistics are not completely known to the transmitter. The first algorithm exploits the structure of the optimal policy and uses a bisection search procedure to arrive at the desired QoS value. The second algorithm is an adaptive power control scheme based on Lyapunov drift arguments. |
