Context-Based Bitmapping in a Multi-Agent Receiver

Re-inforcement learning in a multi-agent environment is applied to a multi-receiver distributed digital communications system. The receivers are given the task of learning the correct bit-to-symbol mapping used in a received bitstream. Using the assumption that a distribution of a proportion of the input characters is known, a reward value is derived and used to train the system. Three different learning policies are compared and suggestions for implementation of the system are presented.