Orthogonal multiple access and information fusion: how many observations are needed?

In this paper, we consider a central estimating officer (CEO) scenario, where sensors observe a noisy version of a binary sequence generated by a single hidden source (the phenomenon). Distributed source coding is not used; rather, the correlation is exploited at the access point (AP), whose goal is to estimate, by properly fusing the data received through AWGN or block fading channels, this sequence. We assume that each node uses a classical channel code to transmit its information to the AP, where decoding is followed by fusion. In the decoding block, joint channel decoding (JCD) or separate channel decoding (SCD) are considered: in the former case, correlation is exploited in the decoding process, whereas in the latter case it is not. We first investigate the ultimate achievable performance limits with JCD (i.e., considering the multiple access channel), in terms of: (i) feasible capacity region in the additive white Gaussian noise (AWGN) channel case; (ii) outage feasible capacity region, in block faded channel case. This analysis provides preliminary insights on the impact of the number of observations. Then, we investigate the overall system performance in terms of bit error rate (BER), probability of outage (only in the block faded case), and probability of decision error (after fusion).