A State-Space Approach to Robust Multiuser Detection

In this paper, we develop a state-space approach to the blind multiuser detection problem with robustness against mismatches in the desired user signature and the time-varying number of users in the channel. The solution is obtained adaptively using a second-order extended Kalman filter (EKF) and requires only O(L²) operations per iteration, where L is the dimension of the subspace containing the signatures of all the users. We also present a state-space approach to the decision directed multiuser detection problem and an algorithm for switching between robust blind and decision directed detection. The proposed switching algorithm is based on using the normalized innovation square (NIS) of the blind detector to test for its convergence and the NIS of the decision directed detector to detect nonstationarities. Thus, it combines the advantages of both these detection schemes and can achieve an output signal- to-interference-plus-noise ratio (SINR) comparable to that of the minimum mean square error (MMSE) detector without any training, even in the presence of mismatches in the desired user signature. Therefore, it is well suited to practical nonstationary environments where users repeatedly enter and leave the system making the cost of retraining un affordable.