Özgür, AyferLévêque, OlivierTse, David2007-01-092007-01-092007-01-09200710.1109/ISIT.2007.4557425https://infoscience.epfl.ch/handle/20.500.14299/238735WOS:000257010201120n source and destination pairs randomly located in an area extending linearly with n want to communicate with each other. Signals transmitted from one user to another at distance r apart are subject to a power attenuation of r^{-alpha} and random phase changes. Classical multihop architectures that decode and forward packets can deliver a sqrt{n}-scaling of the aggregate throughput, while recently proposed hierarchical cooperation achieves n^{2-alpha/2}-scaling, which is superior to multi-hop for alpha<3. The study of information theoretic upper bounds has revealed the optimality of multi-hop for alpha>4, while the moderate-attenuation regime (2 < alpha < 4) remains uncharacterized. We close this gap by deriving a tight upper bound on the scaling of the aggregate throughput, valid for all alpha >2. Our result shows that the mentioned schemes are scaling-optimal, namely that no other scheme can beat hierarchical cooperation when alpha<3, nor can it beat classical multi-hop when alpha > 3. The key ingredient is a careful evaluation of the scaling of the cut-set bound.text::conference output::conference proceedings::conference paper