Abstract—Wireless relays extend coverage, improve spectral ef�ciency, and enhance reliability and rates of wireless cellular communication systems. In this work, we introduce the fundamental notion of asymmetric cooperation among cooperating relays in cellular downlinks- different relays are party to different but overlapping knowledge about the messages transmitted from the base station. We argue that asymmetric cooperation arises naturally in most two-phase protocols in which the base station �rst transmits information to multiple relays that then cooperatively forward the information to the recipient mobile stations in the cell. For a system in which two relays are of the decode-and-forward type and cooperate using linear precoding to communicate with two mobile stations, we formulate the general, but complicated, throughput optimization problem and derive several results that considerably simplify the optimization. We show that under different channel con�gurations and fairness criteria, asymmetric cooperation is often the throughput-maximizing option. Under typical con�gurations, a 20-30 % throughput enhancement is achieved compared to conventional full-cooperation systems. Index Terms—Relays, fading channels, cooperation, linear precoding, asymmetry, SDMA, MIMO. I.

Abstract — We consider a relay that simultaneously assists multiple source-destination pairs that do not have a direct link, the so-called gateway channel, and explore the sum capacity of this network in the presence of quasi-static fading. In the absence of transmitter-side channel state information (CSI), we study superpostion as well as orthogonal channel access. In the presence of transmitter CSI, we consider opportunistic channel access with full CSI, as well as limited CSI via a 1-bit feedback (per user). In each case, the outage capacity and the diversity-multiplexing tradeoff are calculated. It is observed that orthogonal channel access is almost as good as superposition coding, and that opportunistic access provides signficiant gains. It is shown that a 1-bit feedback per user captures most of the gains available in opportunistic communication.