Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automatic detection and characterization of random telegraph noise in sCMOS sensors

Arda Özdoğru, Sergey Karpov, Asen Christov, Stanislav Vítek

TL;DR

The paper addresses random telegraph noise (RTN) in sCMOS sensors, a non-Gaussian pixel noise source that limits sensitivity to faint astronomical sources. It introduces a data-driven pipeline that uses the Anderson-Darling normality test on dark-frame pixel histories, with a dithering step to handle integer rounding, followed by Gaussian-mixture fits to characterize RTN levels. Key contributions include a practical method to detect RTN-affected pixels and estimate per-pixel RTN states (1G, 3G, 5G), plus a corrective fitting scheme with iterative re-fitting thresholds. Applied to two Moravian sCMOS cameras, the approach reveals RTN in roughly 1–5% of pixels and provides per-pixel RTN parameters that can inform post-processing and sensor-quality metrics.

Abstract

Scientific CMOS (sCMOS) image sensors are a modern alternative to typical CCD detectors and are rapidly gaining popularity in observational astronomy due to their large sizes, low read-out noise, high frame rates, and cheap manufacturing. However, numerous challenges remain in using them due to fundamental differences between CCD and CMOS architectures, especially concerning the pixel-dependent and non-Gaussian nature of their read-out noise. One of the main components of the latter is the random telegraph noise (RTN) caused by the charge traps introduced by the defects close to the oxide-silicon interface in sCMOS image sensors, which manifests itself as discrete jumps in a pixel's output signal, degrading the overall image fidelity. In this work, we present a statistical method to detect and characterize RTN-affected pixels using a series of dark frames. Identifying RTN contaminated pixels enables post-processing strategies that mitigate their impact and the development of manufacturing quality metrics.

Automatic detection and characterization of random telegraph noise in sCMOS sensors

TL;DR

The paper addresses random telegraph noise (RTN) in sCMOS sensors, a non-Gaussian pixel noise source that limits sensitivity to faint astronomical sources. It introduces a data-driven pipeline that uses the Anderson-Darling normality test on dark-frame pixel histories, with a dithering step to handle integer rounding, followed by Gaussian-mixture fits to characterize RTN levels. Key contributions include a practical method to detect RTN-affected pixels and estimate per-pixel RTN states (1G, 3G, 5G), plus a corrective fitting scheme with iterative re-fitting thresholds. Applied to two Moravian sCMOS cameras, the approach reveals RTN in roughly 1–5% of pixels and provides per-pixel RTN parameters that can inform post-processing and sensor-quality metrics.

Abstract

Scientific CMOS (sCMOS) image sensors are a modern alternative to typical CCD detectors and are rapidly gaining popularity in observational astronomy due to their large sizes, low read-out noise, high frame rates, and cheap manufacturing. However, numerous challenges remain in using them due to fundamental differences between CCD and CMOS architectures, especially concerning the pixel-dependent and non-Gaussian nature of their read-out noise. One of the main components of the latter is the random telegraph noise (RTN) caused by the charge traps introduced by the defects close to the oxide-silicon interface in sCMOS image sensors, which manifests itself as discrete jumps in a pixel's output signal, degrading the overall image fidelity. In this work, we present a statistical method to detect and characterize RTN-affected pixels using a series of dark frames. Identifying RTN contaminated pixels enables post-processing strategies that mitigate their impact and the development of manufacturing quality metrics.

Paper Structure

This paper contains 9 sections, 1 equation, 13 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. Experiment Parameters
  3. Algorithm
  4. Preprocessing
  5. Normality Test
  6. Fitting
  7. Correction
  8. Results and Discussion
  9. Conclusion

Figures (13)

  • Figure 1: Example pixel histograms. Top row: RTN contaminated pixels. Bottom row: normal pixels. Left column is for camera C4. Right column is for camera C5.
  • Figure 2: Block diagram of the RTN detection process.
  • Figure 3: AD statistics for float, rounded integer, and rounded integer with dithered sampling from the artificial normal distributions over a range of sample sizes.
  • Figure 4: Two example pixels from C5, potential candidates for mixture of 5 Gaussians.
  • Figure 5: Random examples of fits for C4 sensor.
  • ...and 8 more figures