Optimal link adaptation over partially observable Gilbert-Elliot channels

In this paper a communication system operating over a Gilbert-Elliot channel is studied. The goal of the transmitter is to maximize the number of successfully transmitted bits. This is achieved by choosing among three possible actions: (i) betting aggressively by using a weak code that allows the transmission of a high number of bits but provides no protection against a bad channel, (ii) betting conservatively by using a strong code that perfectly protects the transmitted bits against a bad channel but does not allow a high number of data bits, (iii) betting opportunistically by sensing the channel for a fixed duration and then deciding which code to use. The problem is formulated and solved using the theory of Markov decision processes (MDPs). It is shown that the optimal strategy has a simple threshold structure. Closed form expressions and simplified procedures for the computation of the threshold policies in terms of the system parameters are provided.