Efficient block and time-recursive estimation of sparse Volterra systems

We investigate the application of non-convex penalized least squares for parameter estimation in the Volterra model. Sparsity is promoted by introducing a weighted ℓ_q penalty on the parameters and efficient batch and time recursive algorithms are devised based on the cyclic coordinate descent approach. Numerical examples illustrate the improved performance of the proposed algorithms as compared the weighted ℓ₁ norm.