Lightweight complex-valued U-Net CGH for joint distance–wavelength coverage of the ASM parameter plane

Angular-spectrum-method computer-generated holography produces self-imaging revivals as the propagation distance or wavelength changes, so a model trained at one operating point degrades sharply at neighboring comb minima. Rather than proposing a new architecture, we study how to condition an existing lightweight complex-valued U-Net CGH backbone, adopted unchanged from prior work, on the operating point ( z , λ , pitch) through an input-side spatial bias, so that a single 44k-parameter weight covers z  = 145–205 mm across 450/532/638 nm. Ablations confirm that explicit conditioning is necessary and that the spatial-domain form is sufficient. In numerical simulation, the worst self-imaging valley exceeds 25 dB and 8-bit phase quantization costs less than 0.08 dB; the per-color footprint is 0.20 MB. All results are based on simulated ASM propagation.