The generalized CMAC

The cerebellar model arithmetic controller (CMAC) is a neurological model for human-like nonlinear controller design. The CMAC can approximate a wide variety of nonlinear functions by local training. The conventional CMAC uses receptive fields with equal width along all dimensions of input space without considering the variation of the target function in different areas. To avoid this limitation, we propose to adapt the width of the receptive fields depending an the input data distribution. The present paper describes this strategy and explores the application of the proposed algorithm in a nonlinear digital communications framework.