Generalized Approximate Message Passing Algorithms with Applications to Inter-Cellular Interference Coordination
Event details
| Date | 22.06.2011 |
| Hour | 16:15 |
| Speaker | Prof. Sundeep Rangan, Polytechnic Institute of New York University |
| Location | |
| Category | Conferences - Seminars |
Message passing algorithms on graphical models have been highly successful in inference and optimization problems in a wide range of fields. This talks considers graphical models with linear mixing, which model large systems where the interactions between components are through aggregates of small, linearizable perturbations. For such systems, I present a new class of estimation and optimization algorithms called generalized approximate message passing (GAMP) based on Gaussian approximations of loopy belief propagation. GAMP is computationally simple and significantly more general than earlier AMP methods. Also, for the case of large, random graphs, a state evolution analysis provides sharp, easily characterizable asymptotics with verifiable conditions for optimality. The potential for the GAMP algorithm is far reaching, and in this talk, I will discuss applications to inter-cellular interference coordination (ICIC) in next-generation cellular wireless networks. Dynamic interference coordination has emerged as a central challenge as networks introduce femtocells and unplanned, heterogeneous deployments. I will show that the GAMP method provides a tractable method for distributed interference coordination that is computationally simple, has low communication overhead and can incorporate complex physical-layer mechanisms including beamforming, subband frequency allocation and dynamic orthogonalization. Joint work with Ritesh Madan, Qualcomm. Prof. Rangan's homepage
Practical information
- General public
- Free