Optimal mitigation of random telegraph noise for improved photometry at high frame rates

TL;DR

This paper investigates random telegraph noise (RTN) in CMOS sensors for astronomy, compares mitigation strategies, and introduces an effective correction algorithm that improves photometric precision at high frame rates.

Contribution

It provides a detailed analysis of RTN in specific CMOS sensors and introduces a new correction algorithm that outperforms masking in many scenarios.

Findings

RTN significantly increases read noise in certain CMOS sensors.

The correction algorithm improves SNR by over 5% for faint sources.

The algorithm is more effective than masking, especially with undersampled PSFs.

Abstract

Random telegraph noise (RTN) is a major contributor to read noise in many CMOS image sensors considered for astronomical use. While scientific CMOS image sensors deliver lower read noise than traditional charge-coupled devices, mitigating RTN would widen this gap and enable more precise photometry when using the fast readout rates achievable by CMOS image sensors. We report the levels of RTN in three CMOS image sensors used in astronomical instruments: the Sony IMX455, Gpixel GSENSE400, and Fairchild Imaging HWK4123. For the IMX455 in a high gain mode, RTN is the dominant source of pixels with high read noise and increases the overall read noise floor by >20%. RTN is present in the GSENSE400 and HWK4123 but to smaller effects. We compare two strategies for RTN mitigation: masking pixels exhibiting RTN or using a new algorithm for correcting RTN jumps. For faint (< 3 e-/pix/frame) observations of a stellar field with the IMX455, both masking and our algorithm improved the signal-to-noise ratio (SNR) of light curves by >5% on average. Larger improvements were achieved for sources falling on multiple RTN pixels. Our algorithm outperforms masking, especially when the point spread function is undersampled, masked pixels are near the source center, or read noise and shot noise are comparable. In such cases, masking may even deteriorate photometric precision. In other cases, masking remains an effective RTN mitigation technique. We have made available our software for identifying RTN pixels, parametrizing their bias level distributions, and applying our correction algorithm.