Optimizing the Physical Front End: Microring Neuromorphic Sensing for Weak RF Signals

Detecting weak radio-frequency (RF) signals is fundamentally limited by noise introduced at the analog front end, which irreversibly degrades information before digitization and constrains subsequent digital or machine-learning–based processing. We introduce the microring perceptron (MiRP), a sensing architecture built on an RF-photonic platform that performs learned analog preprocessing directly in the physical domain. MiRP uses a high-χ⁽²⁾ microring driven by a programmable optical pump whose temporal waveform is optimized end-to-end, enabling extraction of task-relevant features from the incoming RF waveform prior to homodyne detection. An array of MiRPs produces a high-dimensional optical representation that preserves informative structure even at picowatt-level input powers. Numerical studies using state-of-the-art transduction parameters show that MiRP improves performance across multiple classification and regression benchmarks compared with conventional direct RF and RF-photonic receivers. MiRP is also compatible with quantum-enhanced operation via phase-sensitive amplification, which further increases the detection signal-to-noise ratio within the same architecture.