Evolutionary Programming for the Optimization of Trellis-Coded Modulation Schemes

This paper presents some results on evolutionary optimization of trellis-coded modulation (TCM) schemes. The TCM-encoder and output mapping used in a TCM system are represented as a nite state machine whose parameters are optimized using various genetic operators. Through extensive simulation it is shown that it is possible to consistently arrive at hand-generated (Ungerbck) codes for AWGN channels which are known to be close to the theoretical optimum as determined by the Shan-non bound. This allows the conjecture that the method can be used to generate codes for channel characteristics for which no nearly optimal codes are known. It is also shown that this optimization process yields a rate of convergence which is faster than for pure random search, thus proving that solutions are not simply arrived at by chance. This paper deenes a new genetic operator, joining, and also adds the crossover operator to the set proposed by Fogel (1994). Also, demes are used to concurrently run the evolutionary programming algorithm on several smaller populations which only exhibit a low probability of exchanging genetic information.