Fig. 6: Phase-retrieval problems generated from random complex-valued samples with L = 5 phase filters.
From: Phase retrieval via gain-based photonic XY-Hamiltonian optimization
To solve each problem, the gain-based solver evolved to t = 1000, and RRR ran for 10,000 iterations. All noisy observation vectors \(\widetilde{{{\bf{b}}}}\) contained Gaussian noise. a Phase-retrieval error (RSE) produced by the RRR method (dashed lines) and the gain-based system (solid lines) as a function of Gaussian noise in the measured amplitudes. At each noise level, 20 random complex samples were generated, each with 100 elements whose amplitudes are uniformly distributed in [0, 1) and phases in [0, 2π). The resulting observation vectors were then used for both methods. Vertical error bars indicate the standard deviation in RSE across the 20 trials. b Phase-retrieval error (RSE) versus the dimensionality of the sample vectors, comparing medium noise (SNR = 30, shown in blue) and low noise (SNR = 50, shown in orange). Solid lines again correspond to the gain-based system, and dashed lines correspond to RRR. Each data point represents the average RSE over 20 distinct random instances of the specified dimension, with error bars indicating the standard deviation. See “Fig. 6a” and “Fig. 6b” tables in supplementary data.
