Low complexity cross-layer design for dense interference networks

We considered a dense interference network with a large number (K → ∞) of transmitter-receiver pairs. Each transmitter is endowed with a finite buffer and accepts packets from an arrival process. Each transmitter-receiver link is a fading vector channel with N diversity paths whose statistics are described by a Markov chain. We investigate distributed algorithms for joint admission control, rate and power allocation aiming at maximizing the individual throughput defined as the average information rate successfully received. The decisions are based on the statistical knowledge of the channel and buffer states of the other communication pairs and on the exact knowledge of their own channel and buffer states. In the case of a finite number of communication pairs this problem is computationally extremely intensive with an exponential complexity in the number of users. Assuming that K,N → ∞ with constant ratio the algorithm complexity becomes substantially independent of the number of active communications and grows with the groups of users having distinct asymptotic channel statistics. The cross-layer design is investigated for different kind of decoders at the receiver. The benefits of a cross layer approach compared to a resource allocation ignoring the states of the queues are assessed. The performance loss due to the use of policies designed for asymptotic conditions and applied to networks with a finite number of active communications is studied.