Low complexity dsp implementation of recurrent neural network based multi-user detection

In modem multi-user communication systems the role of proper detection handling has gained much importance [3]. The research of multi-user detection started in the late So??s, and developed into a wide coverage scientific field till the end of the century Nowadays the proposed methods are implemented in DSP devices to exploit the performance improvements of multi-user detection in modem mobile communication devices. This paper introduces the idea how to implement the recurrent neural network based detectors with moderate computational complexity. On the one hand the dimension of the network is reduced according to practical considerations. On the other band the applicability of both soft decision function and additional noise sources inside the neurons are analyzed.