A neural network approach to design of smart antennas for wireless communication systems

Signal processing with neural networks has become popular. The inherent merits of neural networks make it attractive in many applications. We propose the use of cross-correlation neural network models which makes use of the cyclo-stationary property inherent in many communication signals to perform blind beamforming. The proposed approach is based on two sets of linear neurons with cross-coupled Hebbian learning rules orthogonalized to each other. Taking the array data and its time-frequency translated version as inputs, the neural network extracts and separates the desired signals simultaneously. This approach may have advantages in multi-user wireless communications where the co-channel interference condition is severe or the number of interferences is larger than the number of array elements.