Inferring loss causes to improve link rate adaptation in wireless networks

Most of the link rate adaptation algorithms reduce transmission rate on a failed transmission assuming the cause of the loss to be due to radio channel impairment. However, if this failure is due to collision, then lowering the transmission rate in response is unnecessary. This paper proposes a solution to address this problem by augmenting the rate adaptation algorithm with a loss classification component. The main idea underlying this approach is to determine if channel condition is improving or deteriorating by identifying trends in the signal strength. On encountering frame loss, the estimated value of signal strength is used to infer the cause of loss. If estimation indicates a deteriorating channel condition, then the loss is classified as a likely 'channel loss'. Otherwise, the loss is classified as a likely 'non-channel loss'. Thus, in the former case, the rate adaptation algorithm lowers the transmission rate whereas in the latter case, it continues using existing transmission rate. Simulation based evaluation reveals that the use of such a simple heuristic significantly improves performance compared to the case of not using any loss classification mechanism. This was further confirmed by measurements conducted in an experimental test-bed.