Precision, exposure and signal-to-noise tradeoffs

Count signal and noise for the same aperture

Exposure planning begins with a defined measurement. A count integrated over a ten-pixel aperture is different from the charge in its brightest pixel, a background-subtracted rate or a spectral-line ratio. The scientific precision target must refer to the same quantity as the noise calculation.

For an ideal linear CCD model, source, sky and dark arrivals contribute counting variance, while every readout adds read-noise variance. The ACS handbook’s exposure-time equation makes those contributions explicit. A known, subtracted background mean still leaves the fluctuations of the background arrivals in the recorded exposure.

A signal-to-noise ratio is a signal measure divided by a specified noise scale. Here it will be the expected source electrons in an aperture divided by the standard deviation of a summed, background-mean-subtracted measurement. It is a conditional forecast. It is not a universal discovery probability or a complete uncertainty budget for a calibrated physical flux.

All following detector rates, overheads and charge limits are original synthetic teaching inputs. They are not ACS specifications or a recommended observing mode. Their role is to expose a design tradeoff that must be recalculated for a real instrument and target.

The wall-clock budget is larger than the integration

Allocate 1800 seconds of wall-clock time. Stipulate n equal frames, each with 30 seconds of overhead, and no other overhead. The total open-shutter duration is T = 1800 − 30n seconds, and the integration per frame is t = T/n. Use only positive durations.

In a fixed ten-pixel target aperture, stipulate source rate s = 2 electrons s⁻¹, sky rate b = 5 electrons s⁻¹ and dark rate d = 0.2 electrons s⁻¹. These three rates already refer to the entire aperture. They must not be multiplied by ten again.

Let read noise be 3 electrons per pixel per readout, independent across the ten pixels and all frames. Its variance contribution is n × 10 × 3² = 90n in squared-electron units. Assume independent Poisson arrivals, constant rates, a linear unsaturated target response and an exactly known background mean.

The expected source signal is S = 2T electrons. Its measurement variance is V = (2 + 5 + 0.2)T + 90n in squared-electron units, using the numerical electron-count convention for the Poisson terms. The forecast SNR is 2T/√(7.2T + 90n). Uncertainty from estimating the background or calibration is omitted from this particular toy.

Dynamic range can constrain a faint target’s design

A bright comparison star elsewhere in the same frame has a peak pixel with gross rate 30 electrons s⁻¹. This is a separate object and pixel, not a pixel contributing more counts than the faint target’s entire aperture. Assume a design rule that its expected accumulated peak charge must not exceed 30000 electrons in a frame.

With one frame, t = 1770 s and the expected comparison-pixel charge is 53100 electrons. The target SNR is about 31.2480, but this design fails the supplied comparison-pixel rule. The largest unconstrained target SNR in the listed designs is therefore not the best feasible choice.

With two frames, T = 1740 s and t = 870 s. The comparison-pixel expectation is 26100 electrons, which satisfies the declared limit. The target source expectation is 3480 electrons and V = 7.2 × 1740 + 180 = 12708, giving SNR ≈ 30.8703.

The 30000-electron rule concerns expected charge. It is not a guarantee against stochastic excursions or a real full-well or linearity specification. Actual observations need instrument-specific limits and margins, along with source variability and calibration constraints.

More frames have a cost under the fixed budget

The table retains all six listed designs. Increasing n removes more wall time from integration and adds more readouts. Under these assumptions, two frames provide the highest target SNR among the listed designs satisfying the expected comparison-pixel limit.

For six frames, only 1620 seconds remain open-shutter time. The expected target signal is 3240 electrons, and V = 7.2 × 1620 + 540 = 12204. Its SNR is about 29.3288. Although the variance is lower than for two frames because fewer photons are collected, the useful signal has also fallen.

For thirty frames, half of the wall allocation is overhead. The 900 seconds of integration provide 1800 expected target electrons and V = 6480 + 2700 = 9180. SNR falls to about 18.7867. Short exposures are not free repetitions when overhead and readout noise matter.

This ranking is limited to the declared candidates, noise model and constraint. Multiple frames can also help reject cosmic rays, avoid artefacts, sample variability, dither around bad pixels or obtain reference measurements. Those benefits are not quantified in this SNR-only table.

A useful precision forecast carries its exclusions

If counting variance dominates and rates remain constant, signal grows approximately with T while its standard deviation grows approximately with √T. SNR then grows approximately with √T. This scaling changes when read noise, observing overhead, changing background, detector effects or a different estimator matters.

An instrumental calibration uncertainty common to every exposure is not generally reduced by the same factor as independent arrivals. A variable source also cannot be treated as a constant rate simply to preserve the convenient forecast. The exposure plan should distinguish these effects in its uncertainty budget.

The ACS handbook’s worked examples illustrate that changing frame counts and analysis apertures changes exposure forecasts. Its actual numerical recommendations belong to their documented instrument conditions. The appropriate aperture must balance collected source light, sky variance, neighbouring sources and the desired measurand rather than be chosen solely to maximize a formula.

Return to the scientific question: is the feasible precision sufficient for the intended ratio, timing measurement or model comparison? If not, another band, observing condition or reference strategy may be more useful than splitting the same budget differently. The next lesson accounts explicitly for the time and noise cost of measuring a background or comparison.

Worked examples

Try the reasoning

1. The supplied sky rate is 5 electrons s⁻¹ for the whole aperture. Should it be multiplied by ten pixels again?

   Show the answer and reasoning

   No.

   That would count the aperture area twice.

2. Does subtracting an exactly known sky mean remove its arrival noise?

   Show the answer and reasoning

   No.

   The realized sky photons still contribute counting variance.

3. Why is the single frame excluded despite its larger target SNR?

   Show the answer and reasoning

   It violates the supplied expected-charge rule for a separate bright comparison pixel.

   A feasible design must satisfy the declared constraints as well as its precision objective.

4. Does an expected peak below 30000 electrons guarantee no saturation in a real observation?

   Show the answer and reasoning

   No.

   This is an invented expected-charge constraint; real stochastic margins and detector limits require their own treatment.

What to carry forward

• Define signal, noise and aperture consistently before forecasting precision.
• Include overhead and every readout when comparing exposure splits.
• Choose a feasible design for the scientific measurement, retaining unmodelled calibration, variability and detector limits.