Aberration-Free Optical Spectrometer

Abstract

Optical spectrometers are fundamental to scientific analysis, yet achieving high performance at low cost remains challenging because uncorrected aberrations rapidly degrade spectral resolution and typically necessitate complex, expensive optics. Moreover, to preserve spectral resolution, many compact designs remain fundamentally throughput-limited in terms of having a high f-number and a narrow slit. Here we present SHADES (Stochastic High-throughput Aberration-free Deep-Encoded Spectrometer), a general framework that mitigates the effects of optical aberrations using a stochastic grating array (SGA) coupled with physically grounded deep learning (DL), while substantially increasing optical throughput using encoded multi-slits. We develop a theoretical framework establishing aberration resilient spectroscopy in compact, highly aberrated systems, enabling miniaturization without sacrificing spectral resolution and optical throughput. SHADES utilizes an arbitrary spectrum generator (ASG) for hardware-in-the-loop calibration with a DL-based reconstruction pipeline. We further leverage transfer learning (TL) to reduce calibration data and computation for scalable deployment of SHADES. Experimentally, a micro-SHADES prototype achieves a spectral resolution of 2.4 nm over 450-700 nm and accurately reconstructs fluorescence spectra for chemical identification. Collectively, SHADES provides an aberration-free, high-throughput, low-cost spectrometer architecture suited for compact and scalable sensing applications.