Stochastic MIMO Channel Modeling Using Split-ANFIS Framework With GA Optimization

Channel modeling of multi input multiple output (MIMO) wireless systems continues to offer considerable challenges for being infested with uncertainty and random behaviour. Though traditional methods of modeling the MIMO channels have already proven their worth, soft-computational approaches have also received attention due to the fact that these tools learn from the environment, retain it and use the knowledge acquired for subsequent processing. The computational complexity of such systems are a bit higher but with other statistical and evolutionary aids, considerable improvement can be achieved. Here, we propose such an approach based on fuzzy systems and Recurrent Neural Network (RNN). Fuzzy systems are useful for modeling uncertainty while RNN is a form of Artificial Neural Network (ANN) which due to the presence of multiple feedback loops, track timedependent variations in input patterns. These considerations catalyze the formation of a hybrid set-up designed as an adaptive neuro-fuzzy inference system (ANFIS) constituted by fuzzified RNN (FRNN) blocks configured to model MIMO channel characteristics. The proposed architecture is formed by split ANFIS blocks to deal with in-phase and quadrature components of received signals combined by a Self Organizing Map (SOM). During the fuzzification stage, a genetic algorithm (GA) block selects the optimal set of parameters (center, slop and spread) of the Bell-membership function (MF) considered. The GA assisted fuzzification adds to the precision of the system. The experimental results derived establish superior ability of such an architecture compared to reported fuzzy based approaches with lower processing speed and improved precision during recovery of transmitted data through severely faded MIMO channels. Key–Words: Formant, Phoneme, ANN, KMC, LPC, DWT.