MIT Department of Electrical Engineering & Computer Science

SPECIAL SEMINAR

Monday, February 13, 1995
RLE Conference Room, 36-428

Refreshments at 1:45 PM
Talk at 2:00 PM

Mismatched Decoding and the Multiple-Access Channel

Amos Lapidoth
Stanford University

The multiple-access channel models a communication link where multiple users independently communicate with a single receiver. For example, the "reverse link" found in wireless mobile communications is a multiple-access channel through which the different mobile users transmit to a common base station. The theoretically optimal receiver for these channels performs joint maximum-likelihood decoding in order to minimize the average probability of error. Joint maximum-likelihood receivers, however, are almost never implemented since they require exact knowledge of the probability law that governs the channel's behavior. Furthermore, such receivers often suffer from lack of robustness and can be of prohibitive complexity.

Given that the theoretically optimal decoding rule is seldom implemented, the following question arises. What is the best we can achieve with a non-optimal decoding rule? In this talk we shall derive an achievable rate region for a multiple-access channel with a given non-maximum-likelihood ("mismatched") metric-based decoding rule. For the special case of a multiple-access channel with additive noise of a given power but of otherwise arbitrary distribution, we show that Gaussian codebooks and nearest-neighbor decoders are robust and achieve the corresponding Gaussian capacity region, irrespective of the noise distribution. We then extend this result to the sometimes more realistic case of fading channels with side information at the receiver.

Using some of these multi-user results, we derive improved lower bounds on the capacity of the single-user mismatched channel. With these improved bounds, we answer an open problem posed by Csiszar and Narayan concerning the single-user mismatched capacity. We shall conclude with results for some dual problems related to mismatch in lossy source compression.