Influence of RNN structural parameters on blind equalization of shortwave channels
Guobao Ru, Wenqiang Sun, Liangcai Gan · 2019
Aiming at the problem of severe inter-symbol interference and high bit error rate in short-wave fast time-varying channels, this paper designs a short-wave channel blind equalizer based on RNN, and focuses on the analysis of the accuracy rate, cross entropy loss and the number of hidden elements in the RNN training process on the equalizer training process and the equalization error rate. By simulating two typical short-wave Rayleigh flat fading and frequency selective fading channels, the results show that: (1) When RNN is used for short-wave fast time-varying channel equalization, improving the SNR can enhance the accuracy of the training process and the stability of the loss curve. Meanwhile, the accuracy of classification can be improved, and the convergence speed is further accelerated. (2) Increase the number of hidden units can improve the convergence speed of RNN; But, excessive hidden units can easily lead to a sharp jump in the training process curve, which not only reduces the stability of the RNN model during training, but also causes over-fitting problems in RNN. (3) For short-wave channels, the BER of the RNNE is lower than that of the CMA equalizer at the same SNR.