the signatures of all the other users. In some situations such a requirement looks fairly excessive. One of the most typical involvements of synchronous CDMA is the downlink of mobile radio, and making every mobile know all the signatures currently utilized by the other users would complicate the system dramatically. Fortunately, the correlation matrix of the observed process may be estimated experimentally from the observation itself, provided the observation period is long enough. This is the core idea of blind multiuser algorithms. Suppose that ~yi is an N-dimensional row vector of samples at the chip matched filter output corresponding to the data bit number i ¼ 0, 1, . . ..

Then estimate ^ of the observation correlation matrix may be found as:

Ri

ing to (10.23). Variations and advancements of blind multiuser algorithms are plentiful and may be found in the literature (see [19,100,101] and their bibliographies).

10.1.5 Interference cancellation

One can construe the low-complexity of both decorrelating and MMSE detectors by the fact that they exploit a single-user philosophy, i.e. a linear operation of multiplying the observation vector ~y by a mismatched reference vector u. The interference cancellation strategy is again based on a conventional receiver, supplemented by a loop of subtraction of MAI terms from the output effect (10.7). Suppose that the first user receiver knows the signatures and amplitudes of all users and one way or another has

^ ^ . . . ^

obtained estimates b2, b3, , bK of data bits of the side users. Then one is capable of regenerating all side user signals, subtracting their sum from the observation y(t) and utilizing the result as an input (presumably free of MAI) to the conventional receiver. Certainly, the efficiency of such a detector will dramatically depend on the reliability of knowledge of side signal amplitudes and the accuracy of estimates of side user bits. Among others, the multistage procedure is widely discussed in the literature [19,102]. Its first stage involves successive estimates of user bits transferring from the stronger to weaker signals, and using the already estimated k 1 user bits to remove corresponding MAI terms when estimating the kth user’s bit. After all K bits are estimated by so doing the procedure runs the next stage, where all the same operations are repeated. This time, however, the receiver’s knowledge on MAI is richer compared to the previous stage, and subtraction of the recreated MAI starts from the very beginning, i.e. estimating a bit of the strongest user. Stages like this are iterated as many times as is wished, each starting with an updated MAI recreation and continuing to refine it during the course of user bit pattern estimation. When one or another terminating criterion is met, the procedure outputs the final estimate of the bit of the user of interest.