Method and System for Linear Processing of an Input Using a Gaussian Belief Propagation

Dolev Danny, HUJI, School of Computer Science and Engineering, Computer Science

Methods and systems for processing an input. An input vector y is received that represents a noisy observation of Ax, where A is a data matrix and x is a data vector of unknown variables. Data vector x is recovered from the received input vector y via an iterative method. The recovering comprises determining an inference of a vector of marginal means over a graph G, where the graph G is of a joint Gaussian probability density function p(x) associated with noise in the received input vector y.


The method can be applied, for example, for CDMA multiuser detection, linear beamforming, multiuser precoding,  multicell processing in cellular wireless networks, PDE solution, and  other applications. The algorithm is shown to outperform well-known classical solution methods. 

Patent Status

Granted US 8,139,656

Contact for more information:

Anna Pellivert
Contact ME: