Multichannel Extension of the Self-optimizing Narrowband Interference Canceller

The problem of cancellation of a nonstationary sinusoidal interference, acting at the output of an unknown multivariable linear stable plant, is considered. No reference signal is assumed to be available. The proposed feedback controller is a nontrivial extension of the SONIC (self-optimizing narrowband interference canceller) algorithm, developed earlier for single-input, single-output plants. The algorithm consists of two loops: the inner, control loop, which predicts and cancels disturbance, and the outer, self-optimization loop, which automatically adjusts the gain matrix so as to optimize the overall system performance. The proposed scheme is capable of adapting to slow changes in disturbance, measurement noise, and plant, characteristics. It is shown that in the important benchmark case – for disturbances with random-walk-type amplitude changes – the designed closed-loop control system converges locally in mean to the optimal one.