Performance analysis of SIMO space-time scheduling with convex utility function: Zero-forcing linear processing

摘要

In a multiple-antenna system, an optimized design across the link and scheduling layers is crucial toward fully exploiting the temporal and spatial dimensions of the communication channel. In this paper, based on discrete optimization techniques, we derive a novel analytical framework for designing optimal space-time scheduling algorithms with respect to general convex utility functions. We focus on the reverse link (i.e., client to base station) and assume that the mobile terminal has a single transmit antenna while the base station has nR receive antennas. In order that our proposed framework is practicable and can be implemented with a reasonable cost in a real environment, we further assume that the physical layer involves only linear-processing complexity in separating signals from different users. As an illustration of the efficacy of our proposed analytical design framework, we apply the framework to two commonly used system utility functions, namely maximal throughput and proportional fair. We then devise an optimal scheduling algorithm based on our design framework. However, in view of the formidable time complexity of the optimal algorithm, we propose two fast practical scheduling techniques, namely the greedy algorithm and the genetic algorithm (GA). The greedy algorithm, which is similar to the one widely used in 3G1X and Qualcomm high-data-rate (HDR) systems (optimal when nR = 1), exhibits significantly inferior performance when nR > 1 as compared with the optimal approach. On the other hand, the GA is quite promising in terms of performance complexity tradeoff, especially for a system with a large number of users with even a moderately large nR. As a case in point, for a system with 20 users and nR = 4, the GA is more than 36 times faster than the optimal while the performance degradation is less than 10%, making it an attractive choice in the practical implementation for real-time link scheduling.