Chapter 18: Learned OAMP and Deep Unfolding for Imaging

Learning Objectives

• Formulate algorithm unrolling as converting iterative algorithms into fixed-depth learnable computational graphs
• Implement Learned ISTA (LISTA) and ALISTA and analyze their convergence acceleration over ISTA
• Design unrolled OAMP with ProxNet exploiting Kronecker structure of the sensing operator
• Compare LISTA, Learned ADMM, Learned Primal-Dual, and unrolled OAMP architectures for imaging
• Derive theoretical recovery and generalisation guarantees for unrolled networks under RIP conditions
• Apply hierarchical soft-thresholding with structured sparsity to OTFS channel estimation and OFDM/OTFS sensing