A Game-Theoretic Approach to Decentralized Interference-Avoidance Scheduling for Cellular Systems: Algorithm and Equilibria Igor Stanojev and Umberto Spagnolini Dipartimento di Elettronica e Informazione Politecnico di Milano, Italy {stanojev, spagnoli}@elet.polimi.it Abstract— Dense frequency reuse makes the intercell interfer- ence control a crucial issue in cellular systems. Due to large signalling overhead, multicell processing and resource scheduling are not practicable solutions, at least today. We propose a decentralized transmission scheduling scheme where each base station exploits the knowledge of the intercell interference to locally allocate the resources. The idea is inspired by cognitive radio system where each (secondary) user accesses the spectrum according to the level of interference and thus it schedules the transmission over interference-free frequency intervals. Similarly, here each base station schedules the access to time or frequency resource so as to mitigate the generated interference and max- imize its goodput. In this decentralized approach, the intercell signalling is replaced by the level of interference estimated locally and independently within each cell. Equilibria are analyzed using game theory with each scheduler acting as a player that locally maximizes its objective (goodput) while interacting (or interfering) with others. I. I NTRODUCTION Transmission scheduling [1] is an attractive alternative to access randomization [2] for the mitigation of the intercell interference in wireless cellular networks. Multicell scheduler is expected to address the resource requests from multiple base stations and to allocate the access in time and/or frequency in order to guarantee a close-to interference-free coexistence [1]. In this perspective, power control schemes can be con- sidered as simple multicell schedulers where the optimization is restricted only to the adaptation of the power levels [3]. A straightforward solution to multicell resource management would require a central scheduler that communicates with base stations through a high-speed (optical) backbone. However, this approach requires a signalling overhead (as well as control signalling protocols for intercell control messaging) that cannot be accommodated in current systems [4]. Instead, resource scheduling is conventionally carried out locally by the base station for each cell without any signalling or exchanging transmission parameters. Multicell scheduling without the explicit intercell signalling can rely only on the local measurement of the intercell interference and exploit it as an implicit intercell signalling. In this context, we propose a distributed intercell transmission scheduling scheme where each base station determines the resource access based only on the local estimate of the i i E t Δ scheduling frame () i It 0 T 2 T i i t t T +Δ - i t i i t t +Δ i t T + i i t t T +Δ + Interference at access boundary t Fig. 1. Illustration of interference-mitigating time schedule for ith cell according to the interference power profile I i (t). cell interference. The basic idea is illustrated in Fig. 1 for time scheduling. According to the power profile I i (t) of the interference experienced by the ith cell, the services for the user of interest are scheduled within the interference minima. In particular, each base station schedules the use of the time/frequency resource so as to minimize the effect of interference and thus maximize the cell’s goodput. Any change in the allocation of the resource for a single cell triggers the change in the interference measured by other cells that, in turn, exploit it as the new information to alter their transmission schedules. This loops back and induces the interference change to the first cell. We emphasize that the scheme can be considered as an adds-on to the widely investigated power control paradigm [3], even if here we assume that each transmission is energy limited and adapted in time or frequency. In order to gain insight on practical systems, in this pa- per we might refer to WiMAX or 3GPP-LTE as wireless protocol since it schedules both time and frequency with OFDM (orthogonal frequency division multiplex) multiple access (OFDMA) [2]. To simplify, here we assume that the scheduler decouples the resource assignment in time and frequency as for a dynamic access in decentralized TDMA and FDMA (time and frequency division multiple access, respectively) [5]. Since cells are not necessarily (time and frequency) synchronized, distributed scheduling is constrained not to fragment the time/frequency access as this would raise the intercell interference due to the interference leakage at the This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings. 978-1-4244-4148-8/09/$25.00 ©2009