Effect of training sequence bandwidth for Wiener filter based interference cancellation systems

Abstract

The performance of interference cancellation systems based on Wiener filters relies on proper modeling of the channel between the adaptive filter input and the reference signal. Once the optimal condition is achieved the correlated interference signal components are cancelled out and the desired signal can be extracted as the Wiener filter error signal. In practice, for most applications, the signals in concern tend to be non-stationary in nature with fluctuating bandwidths. Thus, training the Wiener filter under such conditions is essential for proper interference cancellation. In our work it was noticed that the use of a narrowband training signal, which is unable to span the channel transfer function, results in incomplete tuning of the adaptive filter. This results in erroneous interference cancellation for bandwidth varying environments. On the other hand, it will be shown in this paper that through proper selection of a training signal that can span the entirety of the channel transfer function, the channel can be modeled properly through the Wiener filter leading to significant performance enhancement. This work also presents an analysis on the impact of different types of training sequences on the performance of interference cancellation systems. This will enable proper selection of training sequences for interference cancellation problems based on application requirements.