Dynamic Load Balancing Through Coordinated Scheduling in Packet Data Systems Suman Das, Harish Viswanathan, Gee Rittenhouse Wireless Technology Research Lucent Technologies, Bell Labs. 600 Mountain Ave, Murray Hill, NJ 07974. {sumand, harishv, gee}@lucent.com Abstract— Third generation code-division multiple access (CDMA) systems propose to provide packet data service through a high speed shared channel with intelligent and fast schedul- ing at the base-stations. In the current approach base-stations schedule independently of other base-stations. We consider scheduling schemes in which scheduling decisions are made jointly for a cluster of cells thereby enhancing performance through interference avoidance and dynamic load balancing. We consider algorithms that assume complete knowledge of the channel quality information from each of the base-stations to the terminals at the centralized scheduler as well as a two- tier scheduling strategy that assumes only the knowledge of the long term channel conditions at the centralized scheduler. We demonstrate that in the case of asymmetric traffic distribution, where load imbalance is most pronounced, significant throughput gains can be obtained while the gains in the symmetric case are modest. Since the load balancing is achieved through centralized scheduling, our scheme can adapt to time-varying traffic patterns dynamically. I. I NTRODUCTION Third generation cellular systems are currently being de- signed with the emphasis of providing high speed packet data service on the downlink through a shared high data rate channel. The base-station arbitrates this channel among the different users based on channel condition feedback and backlog [1], [2], [3]. Variable rate transmission at fixed power is a key characteristic of these systems. In the current ap- proach, the scheduler resides at the base-station and operates independently across different cells. In other words, a cluster of contiguous base-stations transmit signals to different users independent of the activity or the scheduling decisions made at the neighboring base-stations. We consider a system in which a cluster of neighboring base-stations schedule trans- missions to users in their cells jointly. We refer to this as coordinated scheduling. The coordinated scheduling can be implemented through a centralized scheduler entity or through signaling between the base-stations on the backhaul network. With coordinated scheduling we can potentially incorporate interference avoidance to some degree (depending on the cluster size) by turning off some base-stations in the cluster while other base-stations are transmitting to users at the edges of their cells. Furthermore, the coordinated scheduling can adapt to traffic imbalances dynamically since any base- station within the cluster can transmit to any user within the cluster. We propose some coordinated scheduling algorithms and evaluate their potential gains in this paper. Our results show that for asymmetric traffic scenarios significant gains are obtained through load balancing from coordinated scheduling. We do not consider phase-coherent transmission of radio- frequency signals across base-stations or joint encoding of the information signals across base-stations. These would require considerably more coordination between base-stations but can potentially provide more gains. Load balancing has been considered in the literature for both time division and code division multiple access networks. Traditional time division multiple access (TDMA) cellular networks employed frequency reuse to guarantee sufficient signal-to-interference-and-noise-ratio (SINR) at the receiver for transmission of information. At the time of deployment the available spectrum is assigned to the different cells according to the desired frequency reuse. A major disadvantage of such networks with fixed frequency assignment was the inability to adapt to asymmetries in traffic patterns that could potentially be dynamic. As a consequence dynamic channel allocation (DCA) has been studied extensively in the literature for such TDMA voice networks [4], [5]. With the advent of code-division multiple access (CDMA) technology frequency planning was no longer required because of universal reuse. Nevertheless, one could consider designing the network to match the traffic conditions at the time of deployment through base-station location, antenna down-tilt and base station trans- mit power [6], [7], [8]. To date most of these optimization techniques have been applied to optimize the network for non- shared channels. Furthermore, these optimization procedures are performed only at the time of deployment and hence are not dynamic. The coordinated scheduling considered in this paper provides a method for dynamic load balancing in packet data CDMA networks. Inter-cell coordination with variable power and variable rate transmission was considered in [9]. However, they did not consider scheduling in combination with inter-cell coordination. The main contributions in this paper are the proposed algorithms for coordinated scheduling, characterizing their po- tential gains and understanding of the impact of channel fading on the gains. The paper is organized as follows. A detailed description of the different system architectures considered in this paper that makes it possible to do coordinated scheduling are described in Section II. The coordinated scheduling algo- 0-7803-7753-2/03/$17.00 (C) 2003 IEEE IEEE INFOCOM 2003