A neural approach for compensation of nonlinear distortion effect in up to 1600bps Data Communication over voice-dedicated channels

Data communication over Voice-dedicated (DoV) channels of wireless telecommunication networks provides possibility of transmission of bit stream obtained from speech encoder over desired channels. On the other hand, compression techniques of wireless networks, generally by vocoders, cause severe non-linear distortions to DoV systems. Indeed, it is network that compensates the distortion which is produced by its channel and produces distortion to upper system of DoV by its vocoder. In channels with minimum phase or linear one; Linear Equalizer (LE) with LMS is a quite practical solution. Even though normal channels with fading or non-minimum phase have destructive effects, equalizers with Viterbi decision device such as Maximum Likelihood Sequence Detection (MLSD) equalizer are able to compensate distortions. Nevertheless, LE and MLSD are not able to combat with the whole distortions of DoV systems, theoretically and practically. Although Decision Feedback Equalizer (DFE) is a proper choice for non-linear channels, due to its capability of non-linear modeling, but it would be preferable to improve for applications of DoV. This paper considers a neural based DFE with fast and reliable neural networks. We represent the results of proposed method for DoV in an experimental state. The final empirical results on GSM network show that proposed method is reliable for DoV systems.