Independent Vector Analysis via Log-Quadratically Penalized Quadratic Minimization

We propose a new algorithm for blind source separation (BSS) using independent vector analysis (IVA). This is an improvement over the popular auxiliary function based IVA (AuxIVA) with iterative projection (IP) or steering (ISS). introduce adjustment (IPA), where we update one demixing filter and jointly adjust all other sources along its current direction. Each involves solving non-convex minimization problem that term log-quadratically penalized quadratic (LQPQM), think of interest beyond this work. In general case, show global minimum corresponds to largest root univariate function, reminiscent modified eigenvalue problems. simple procedure on Newton-Raphson efficiently compute it. Numerical experiments demonstrate effectiveness proposed method. First, it decreases value surrogate function. further synthetic mixtures, study probability finding true matrix convergence speed. method combines high success rate fast convergence. Finally, validate performance reverberant speech task. find AuxIVA-based methods perform similarly in terms acoustic BSS metrics. However, AuxIVA-IPA converges faster. measure up 8.5 times speed-up runtime compared next best method, depending number channels signal-to-noise ratio (SNR).