Synthesis of self-learning neural-net-like system for coherent images reconstruction

A neural-net-like adaptive system has been synthesized. The system has self-learning ability, programmed in a rule of synapse weights transformation. The system is able to work in a quasi-analog regime. Modified annealing, which prevents slowing-down (phenomenon of short-term memory) of the iterative procedure near local maxima of the PDF, is possible. The general processing scheme is parallel-sequential and is based on the use of an interference filter, lenses, arrays of matched coherent filters, and amplitude-phase-tunable modulators. The adaptability of the system to phase defects or wave fields permits nonideality of the mirror profile and, consequently, the use of less expensive large-aperture telescopes