Adaptive optimization of tracking algorithms: applications to adaptive antenna arrays for randomly-varying mobile communications

Adaptive antenna arrays are used for reducing the effects of interference in mobile communications. The adaptation typically consists of updating the antenna weights by a recursive least-squares algorithm. We add another adaptive loop that greatly improves the performance when the environment is randomly-varying. Consider a single cell system with a (receiving) antenna array at the base station. Algorithms for tracking time varying parameters require a balance between the need to track changes (needing a short memory) and the need to average the effects of disturbances (needing a long memory). Typical algorithms seek to recursively compute the antenna weights that minimize (at times kh, k=1, 2..., for small h) E/spl Sigma//sub l=1//sup k/ /spl alpha//sup k-l/e/sub l//sup 2/:e/sub l/ are the reception errors and /spl alpha/<1. This minimization is used only to get good weights. The performance is measured by the sample average bit error rate, which depends heavily on /spl alpha/. The optimal /spl alpha/ can change significantly in seconds. The method can be used to improve algorithms for tracking parameters of time varying systems. The additional adaptive loop, based on a natural "gradient descent" method and of the stochastic approximation type, tracks the optimal value of /spl alpha/. The antenna weights and the value of /spl alpha/ are adapted simultaneously. Simulations under a variety of operating conditions show that the algorithm is practical and tracks the optimal weights and value of /spl alpha/ very well. In terms of average bit error rates and for all of the scenarios tested, the new system always performs better (sometimes much better) than an algorithm that uses any fixed value of /spl alpha/.