Achieving the Capacity of the $N$-Relay Gaussian Diamond Network Within $\log N$ Bits.
Authors: Bobbie Chern, Ayfer Özgür
We consider the -relay Gaussian diamond network where a source node
communicates to a destination node via
parallel relays. We show that
several strategies can achieve the capacity of this network within
bits independent of the channel configurations and the operating SNR. The first
of these strategies is partial decode-and-forward: the source node broadcasts
independent messages to the relays at appropriately chosen rates, which in turn
decode and forward these messages to the destination over a MAC channel. The
same performance can be also achieved by compress-and-forward,
quantize-map-and-forward or noisy network coding if relays quantize their
observations at a decreasing resolution with
, instead of quantizing at the
noise-level. The best capacity approximations currently available for this
network are within O(N) bits which follow from the corresponding capacity
approximations for general Gaussian relay networks.
