Multiple sources neural network direction finding with arbitrary separations

Interference rejection is very important and often represents an inexpensive way to increase the system capacity of cellular and mobile communication systems. This paper presents a modification to the radial basis function-based direction finding algorithm where the DOA problem is approached as a mapping which can be modeled by training the network with input output pairs with multiple angular separations. The network is then able to track a fixed number of sources with arbitrary angular separations using a linear array. A novel training technique is suggested and the performance of the RBFNN algorithm is compared to ideal data.