📄 Showing the "Simple" version — "Advanced" is not ready yet. Add it to favorites to help prioritize it. Ask for the full read

You are counting sand grains falling in an hourglass, but each time the wind adds or removes a few. In the most sensitive cameras, the task is similar: you need to figure out how many "sand grains" (electrons) the sensor itself added, and how many came from light. This inherent sensor noise is called readout noise, and it hinders seeing weak signals — for example, the light of a distant star.

Usually, readout noise is measured by varying light brightness, but that's like trying to catch the wind. The authors found a way to bypass this dependence. They noticed that if you look not at the signal values themselves but at how they "wrap around" after each integer number of electrons, you get a pattern that depends only on the sensor noise. This pattern resembles a repeating ornament on a clock face — the hand may point to different numbers, but the clock face itself is unchanged.

This approach resembles an old trick: to measure the trembling of a hand more accurately, you look not at its position but at how it passes the same divisions many times. The scientists derived a simple formula linking this pattern to readout noise and tested it on a computer model. Now cameras can be calibrated faster and more accurately, without adjusting for light brightness.

Original abstract

as written by the authors arXiv:2603.01199

Valley-peak modulation (VPM) was introduced as a metric for quantifying read noise in deep sub-electron read noise (DSERN) CMOS sensors. In the original amplitude-domain definition, VPM depends on both read noise and quanta exposure, yet Starkey & Fossum demonstrated exposure-independent approximations that hold in the DSERN regime. In this note we identify the exposure-invariant object those approximations probe. Starting from the standard Poisson-Gaussian model, we apply a phase mapping that quotients out the integer electron count, yielding a wrapped-Gaussian density parameterized only by read noise and admitting both lattice-sum and Jacobi theta-function representations. The fundamental exposure-invariant quantity is shown to be the theta ratio $$R(\sigma)=\vartheta_4(q)/\vartheta_3(q)$$, of which any VPM is a contrast normalization; the existing exposure-independent approximations are then recovered as low-order truncations of the lattice-sum representation of $$R$$. A closed-form inverse expressing read noise in terms of VPM is obtained using elliptic integrals, and a short simulation example illustrates practical estimation of read noise from the VPM in phase space.