A genetic approach for downlink packet scheduling in HSDPA system

High speed downlink packet access (HSDPA) system currently under development in 3rd generation partnership project (3GPP) [1] employs number of parallel shared channels and adaptive modulation and coding scheme (MCS) to enable a high rate packet data transfer from base station (Node-B) to user equipment (UE). To decide how to share the available radio channel capacity amongst the active UEs, a packet scheduling and channel assignment algorithms are employed at Node B. Packet scheduling techniques such as max carrier to interference ratio (C/I) or round robin (RR) fails to take into account all the aspects of the quality of service (QoS) provisioning. In this paper, the fundamentals genetic algorithms and conventional wireless scheduling techniques are combined and the weaknesses of the existing known techniques are exploited to propose a novel hybrid genetic packet scheduler (HGPS) for the HSDPA system. A combination of random and intelligent diversity of population and comparative nature of the selection process of genetic engine contribute to its robustness. The proposed HGPS outperforms Max C/I packet scheduler in terms of total delivered throughput within low delay thresholds. Unlike conventional packet scheduling technique HGPS does not rely only on the current or past status of the scheduling process. By treating the possible solutions as points in a search space, the proposed HGPS through a genetically guided search visits and examines the possible solutions and estimates the impact of each these solution on overall performance of system in terms of fairness, throughput or QoS without actually performing a transmission. Subsequently, the solution that achieves the best estimated overall performance is chosen for the actual transmission. By means of computer simulation, performance of the HGPS algorithm is characterized in Rayleigh fading and shadowing radio channel conditions. The impact of imperfect reporting on the performance of HSDPA system is evaluated. We examine the joint impact of reporting latency, imperfect channel estimation and the corruption of reports in the feedback channel on the performance. It is shown that the proposed class of intelligent parallel random schedulers is highly robust against the imperfect reports from UEs.

Authors and Affiliations

1. Fujitsu Laboratories of Europe LTD. (FLE), Hayes Park Central, Hayes End Road, Hayes Middlesex, UB4 8FE, UK

   S. Abedi & S. Vadgama

Corresponding author

Correspondence to S. Abedi.

Acknowledgment The authors would like to acknowledge the valuable assistance provided by Mr Pierre Coulon of Fujitsu Laboratories of Europe Ltd.

This work was presented in part at IEEE WPMC’2002, and will be presented in part at IEEE IST’2003.

Reprints and permissions