A Shadow-Price-Based Distributed Approach to Dynamic Load Balancing in CDMA Networks

We present a linear-programming approach for dynamic load balancing in CDMA networks. The linear program characterizes the minimum achievable amplifier load for a given user configuration at each time interval, and gives a useful benchmark for the potential gains from optimizing the power assignment. The solution of the linear program also offers valuable insight to the qualitative properties of the optimal power allocation. In particular, the structure of the optimal assignment reflects the critical notion that power allocation should not just be based on signal strength values but also on shadow prices which arise from load considerations. We develop a dual-ascent scheme for solving the linear program in a (mostly) distributed fashion with low communication overhead. Extensive numerical experiments demonstrate that there is scope for significant gains from balancing amplifier loads in typical scenarios.