Accuracy of CGH encoding schemes for optical data processing

We consider 1-D and 2-D error diffusion (ED), modulated error diffusion, random encoding, and direct quantization computer generated hologram (CGH) encoding schemes. Multilevel (L transmittance levels), macro pixel (N2 CGH pixels per sample) and one CGH pixel per sample, and amplitude and phase encodings are considered. Three diverse applications (matrix-vector processors, spectrum analyzers, and general interconnections for neural nets etc.,) are considered. We find random encoding is best for matrix-vector macro pixels (due to the high local accuracy required, i.e., the fact that one matrix cell maps to one output point), modulated ED (or ED) is best for spectrum analyzers (where good global signal representation is required), and multilevel ED is needed for complex lenses and interconnections (where a very complex global function is required).