A Homotopy—Controlled Mean Field Annealing Partially—Connected Neural Equalizer for Pan—European GSM Syst5em

摘要

Channel equalization is essential in the Pan-European GSM mobile communication system.The maximum likelihood sequence estimation(MLSE) using the Viterbi algorithm(VA)iscommonly recommended for the dqualization,which can only accommodate the channels with limited time delay spread.In[1],we presented a mean field annealing(MFA)partially connected neural equalizer for the GSM system,in which the complexity is linearly proportional to the time delay spread and therefore relatively fast convergence speed is achieved.But the annealing coefficient of the MFA equalizer is fixed,which is not flexible in timing-varying circumstance such as mobile communications.To decrease the computation of MFA approach so as to make it more easy for practical use,the MFA approach is reated as a homotopy problem.The ordinary equations which the MFA approach should obey are derived.These equations can be used to reflect the deviation of the iteration result from the track of MFA approach.Based on this tesult,an adaptive annealing control algorithm is proposed,which can dynamically control the annealing coefficient according to the iteration deviation.Computer simulations show that our approach can provide a much higher convergence speed and performance improvement over 16-state and 32-state VA's which are usually suggested for practical applications.