Asymptotically optimal hypothesis testing with memory constraints (Q1175395)

The authors consider a practical communication receiver design problem. Under the conditions considered it reduces to deciding which of two Markov processes is present in a given signaling interval. The classical solution to the problem via the log-likelihood ratio test can not be implemented in electronic hardware due to the virtual impossibility to build an arbitrary mapping from \(C[0,1]\) into \(R\). One has to use the discretized version of the process which is obtained by sampling. The authors develop hypothesis testing designs under memory constraints. The optimality criterion is the Chernoff asymptotic efficiency of tests, i.e. the rate at which the maximum error probability goes to zero as the sample size gets large. It is shown that a unique optimal criterion exists and an algorithm is given for its computation. From the mathematical point of view the problem of finding the optimal criterion leads to determining the solution of a set of equations which are derived from large deviation twistings of the two transition matrices (the one under the hypothesis and the one under the alternative). A comparison with the i.i.d. case on hand of examples is done. This well written paper also contains some ideas and hints for further investigations and generalizations.