Channel Equalization using Multiplicative Neural Networks

In this work a multiplicative neural network (MNN) based adaptive equalizer is proposed. Multiplication increases the computational power and storage capacity of neural networks is well known from extensions of artificial neural networks where this operation occurs as higher order units. Multiplicative node functions allow direct computing of polynomials inputs and approximate higher order functions with fewer nodes. The MNN based equalizer is tested on 4 QAM, 16 QAM and 32 QAM signals for AWGN channel noise model. The equalizer is implemented for six different FIR channels using 4 QAM signals.The equalizer is tested on non linear stationary channel model and its performance is compared to Chebyshev functional link artificial neural network (CFLANN). The real valued MNN is further extended to the complex domain. A fully complex single layer multiplicative neural network (MNN) equalizer is proposed. The back propagation algorithm is extended to complex domain to train the Complex Valued NN. The activation functions at the nodes are fully complex. The fully complex fast converging algorithm gives good results even in the presence of substantial noise.