Abstract
Future wireless networks will be driven by various ubiquitous broadband services such as portable telephony, mobile Internet, Voice over IP (VoIP), and IPTV. These services motivate the development of smart dynamic power/rate allocation methods with the following goals: 1) throughput maximization, 2) minimization of queueing delay, 3) satisfaction of quality of service (QoS) requirements of all users. In order to achieve these goals, multi-user packet scheduling needs to be based on both the queue state information (QSI) and the channel state information (CSI). Once service rates are determined by a scheduler in the medium access control (MAC) layer, the physical (PHY) layer determines the corresponding power and rate allocation in each transmit dimension. Because of this interplay between MAC and PHY layers, such a combination of queue-channel-aware scheduling and power/rate allocation is known as "cross-layer resource allocation".

This talk discusses cross-layer resource allocation in downlink and uplink orthogonal frequency division multiplexing (OFDM) systems with multiple transmit and receive antennas. A variety of existing scheduling policies such as maximum weight matching scheduling (MWMS) will first be briefly introduced. Next, queue proportional scheduling (QPS) will be presented, whose delay and fairness properties are shown to be superior to other well-known scheduling policies. In the second part of this talk, efficient power/rate optimization algorithms applying QPS and MWMS to multi-user OFDM/MIMO-OFDM systems will be presented. With perfect CSI at the transmitter, geometric programming and Lagrange dual decomposition are utilized to develop efficient algorithms. With only channel distribution information (CDI) at the transmitter, effective power/rate allocation algorithms are developed based on a Gaussian approximation of the MIMO channel mutual information, in conjunction with a successive feasibility check. Simulations demonstrate that QPS significantly outperforms other scheduling policies such as MWMS in terms of average queueing delay and fairness among the users.