MIMO channel modeling with cluster configuration of Complex Time Delay Fully Recurrent Neural Network

Artificial Neural Network (ANN)s like the Multi Layer Perceptron (MLP)s with temporal characteristics demonstrate architectural complexity during estimation of Multi Input Multi Output (MIMO) channels for which alternative ANN options are required. The Recurrent Neural Network (RNN) with the ability to track time-dependence of input signals emerge as a choice for such applications. But a standoff surfaces regarding the approach in which RNNs are to be trained to deal with signals with real and complex components. Signals with separate real and complex components can be used to train RNNs better with the responses of each block combined and optimized with Self Organizing Map (SOM) enabling them to show satisfactory performance with tightly coupled transmissions. The present work deals with the formation of a cluster of Complex Time Delay Fully Recurrent Neural Network (CTDFRNN)s optimized with SOMs with inherent temporal characteristics to deal with MIMO channel estimation. The performance derived is superior to statistical and MLP-based approaches and provide diversity gain.