Non-linear mobile-radio channel estimation using neural networks

Channel estimation is a crucial problem in any equalization algorithm in time varying environments such as those encountered in mobile radio systems. Referring to the European GSM mobile radio system, an approach based on automatic tracking of the channel variations based on the use of neural networks (NN) is presented which exhibits good bit error rate (BER) performance is presented. The NN architecture employed is a modification of Kohonen's (1988) self organizing map (SOM) with a recursive structure (output folded to the input). This dynamic network performs classification by vector quantization of the input stream using the information contained in the middamble of the GSM burst with an unsupervised control scheme. The neural receiver is demonstrated to converge very quickly to stability and to have a lower computational complexity than the Viterbi algorithm.