Automatic regularization for Kronecker product MMSE filters
Authors: Daniel Gomes de Pinho Zanco, Leszek Szczecinski, Jacob Benesty, Eduardo Vinicius Kuhn
Organizations: INRS–Institut National de la Recherche Scientific, Montreal, QC, H5A-1K6, Canada. · Federal University of Technology - Paraná, Toledo, Paraná, 85902-490, Brazil.
In this work, we propose a method to automatically find the regularization parameter for low-rank MMSE filters based on a Kronecker-product representation, using an approximate leave-one-out (ALO) cross-validation metric. We formulate the problem in a constrained form that covers both error minimization and interference suppression, and explore the link between regularization and rank selection. The proposed method is validated on system identification and antenna beamforming problems, showing significant gains over commonly used methods.
Figures & tables
Figure 1: Simulated room impulse response considered in the experiment.
Figure 2: (a) Misalignment, (b) rank, (c) ALO/GCV metric, and (d) nuclear norm of solutions obtained by varying α for different regularizers. Solutions are obtained by switching the regularization function in ( 5 ), for the unconstrained case ( d=0 , c=0 ). To compute the rank and nuclear norm, ℓ2 and ℓ1 solutions are reshaped to (20,25) . The small circle markers in (a) indicate the misalignment obtained by choosing α from (c). The vertical dotted lines in (d) indicate αmax for ℓ1 and nuclear norms, from left to right, respectively [see ( 16 )]. Here N=1000 and SNR=5 [dB].
Figure 3: SINR as a function of number of snapshots N for different methods, with construction rank (a) R=1 and (b) R=4 .
Figure 4: Estimated α as a function of number of snapshots N for different methods, with construction rank (a) R=1 and (b) R=4 .
Hyperparameter optimization selects model and training settings to improve predictive performance, efficiency, or reliability under limited evaluation budgets.
Graduate School of Information Science and Technology, The University of Tokyo, Tokyo 113-8656, Japan · National Institute of Technology, Kagawa College, Takamatsu, Kagawa 761-8058, Japan · Yamaha Corporation, Hamamatsu-shi, Shizuoka 430-8650, Japan