PERFORMANCE EVALUATION OF DIFFERENT GOSSIP ALGORITHMS FOR DISTRIBUTED QR FACTORIZATION IN WIRELESS SENSOR NETWORKS

The paper treats the problem of distributed QR factorization for orthogonalizing a set of vectors in wireless sensor networks (WSN). The orthogonalizing algorithm used is based on classical as well as modified Gram-Schmidt (GS) orthogonalization, both of which are suitable for WSNs. The whole computation is made distributed using gossip algorithms to compute sums that are inherent in GS orthogonalization. We focus on three different iterative gossip averaging/summing algorithms. These are the gossip-based push-sum algorithm (PSA), the geographic gossip algorithm (GGA), and the greedy triple-wise gossip algorithm (G-TGA). A simulation-based comparison between these three algorithms, in the context of QR factorization in WSNs, is presented regarding robustness against link failures, communication overhead, convergence speed, and power consumption. Alongside, results for the centralized version of QR factorization are also presented for comparison.