11:50 - 12:10 | Fri 17 Mar | Main Room | S6.5
Convex-optimization algorithms, such as SDP's, have been widely used over the last couple of decades for the recovery of various modulated signals (mPSK, mQAM, etc.) in MIMO systems. These algorithms are known to outperform conventional methods, such as zero-forcing and decision-feedback (BLAST). However, until very recently, an exact analysis of the performance such methods did not exist. We leverage recent results from machine learning to give the first exact BER expressions for a wide class of convex-optimization-based algorithms used in MIMO. In particular, we show that for square systems, the so-called "box relaxation" comes to within 3db of the performance of the celebrated matched filter bound. The results rely on the strengthening of a classical comparison lemma, due to Gordon, developed by the speaker and his collaborators.