Experimental demonstration of OCC-based optical over-the-air computation

We propose and demonstrate an optical camera communication (OCC)-based novel optical over-the-air computation (OAC) method. In this method, a defocused camera image of an LED array realizes the optical aggregate computation, where the receiver estimates an aggregate function of the emitted intensities directly from the pixel-domain optical observation. Hence, it eliminates the need to decode each LED data stream separately, reducing the processing complexity significantly. The proposed direct OAC method is experimentally demonstrated using a proof-of-concept camera receiver. Across a wide range of transmission powers and aggregation weights, a single linear read-out attains a normalized mean-square error (NMSE) of 10^-5-10^-4 (coefficient of determination R²>0.94), substantially outperforming channel-blind image summation. Because a learned linear read-out makes direct aggregation and separate-then-compute mathematically equivalent, the direct method matches the learned separate-then-compute baseline in accuracy while using a single aggregate-specific model instead of K per-emitter models. This complexity advantage grows in a strongly overlapped reflection channel, where the direct read-out stays accurate while a calibrated zero-forcing per-emitter inversion degrades by up to two orders of magnitude, suggesting a simple optical front end for aggregate computation.