Estimation of MIMO channels using complex time delay fully Recurrent Neural Network

Estimation of Multi Input Multi Output (MIMO) channels can be performed by Artificial Neural Network (ANN)s such as Multi Layer Perceptron (MLP)s. However, the cost of training overload in case of time varying MIMO channels is the main bottleneck of such ANN architectures for which a viable alterative, namely, the Recursive Recurrent Network (RNN) is explored. Although for tightly coupled real and imaginary components of a transmitted signal RNN cannot provide a satisfactory solution, nevertheless, a split - complex activation RNN approach can be adopted to deal with such cases averaging the output obtained for a given time length. The results demonstrate better performance as well as computational simplicity compared to MLP architectures with temporal characteristics.