Optical Equalization for Reduction of Fixed-Point DSP Word Length
Year: 2026
Type: preprint
Abstract: We present a mixed-mode signal-processing framework in which analog optical spectral-phase modulation is followed by coherent detection and digital phase extraction to generate a bounded representation of high- dynamic-range temporal waveforms that requires a substantially shorter fixed-point word length for downstream DSP. Motivated by the VEViD algorithm in the Physics-inspired Computer Vision (PhyCV)—itself inspired by photonics time stretch—the method exploits the natural equalization that arises from near-field spectral phase modulation followed by coherent detection. In this representation, the resulting in-phase (I) channel closely preserves the original signal profile, while the quadrature (Q) channel provides a spectrally filtered response that functions as a highly accurate dark-field measurement. The temporal phase derived from the I and Q channels follows a nonlinear, amplitude-compressed transform that maps a wide-dynamic-range input onto a bounded output. Because this output spans far fewer distinguishable amplitude levels, it can be represented at the same relative fidelity using fewer bits, reducing the word length needed in fixed-point DSP hardware. This in turn facilitates efficient hardware DSP through narrower multipliers, smaller registers and memory, lower power, and higher clock rates. To evaluate whether this concept can be realized in hardware, we emulate an LCOS-based programmable waveshaper as the spectral phase modulator. The model incorporates hardware-informed characteristics of the waveshaper, including spectral-phase discretization imposed by the LCOS pixel grid, finite frequency resolution and accuracy. Applying this module to representative high-dynamic-range waveforms, such as LiDAR, shows that the spectral phase engineering provides a natural equalization of the signal while preserving temporal structure and enhancing low-amplitude returns, yielding an output that needs fewer bits of word length than the input. These proof-of-concept results suggest that this may be a promising physical-computing primitive for real-time preprocessing that lightens the word-length burden on DSP hardware in high-dynamic- range sensing and communication systems.
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7150875271