Computation over Mismatched Channels

We consider the problem of distributed computation of a target function over a two-user deterministic multiple-access channel. If the target and channel functions are matched (i.e., compute the same function), significant performance gains can be obtained by jointly designing the communication and computation tasks. However, in most situations there is mismatch between these two functions. In this work, we analyze the impact of this mismatch on the performance gains achievable with joint communication and computation designs over separation-based designs. We show that for most pairs of target and channel functions there is no such gain, and separation of communication and computation is optimal.