Context and Why This Matters
A faint source is not simply "a star you cannot quite see"; it is a source whose signal sits among several independent noise sources. Whether a signal counts as a detection is a statistical judgment, not a visual impression, and that judgment depends on the noise model, not just the raw counts.
Learning Goals
- Write the CCD signal-to-noise equation including source, sky, dark, and read-noise terms.
- Compute SNR for a bright and a faint source using identical instrumental conditions.
- State the SNR threshold conventionally used to claim a secure detection.
Opening Phenomenon
Two sources produce very different counts on the same field with an identical 3×3 pixel aperture: one at 2500 electrons, one at only 300 electrons. Both sit on the same sky background and read noise, but only one is a secure detection.
Terms Developed in Context
These terms form the measurement chain for Noise, Signal-to-Noise, and Detection Limits: each connects an observable to a model choice, and each has a boundary that must be respected when making a scientific claim.
- Poisson noise
Counting uncertainty that arises because independent photon arrivals fluctuate with variance approximately equal to the expected count. It establishes a fundamental noise floor even for a perfectly calibrated detector.
For \(N\) detected electrons, \(\sigma\approx\sqrt{N}\); source, sky, and dark counting variances add in the SNR denominator. Poisson noise is random scatter, not a removable calibration pattern.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- sky background
Photons recorded from airglow, scattered light, unresolved sources, and other non-target illumination. Sky counts add noise inside the source aperture and can dominate a faint-source detection.
The sky level is estimated from nearby pixels and contributes \(n_{pix}N_{sky}\) to the aperture variance. Subtracting the estimated mean sky does not subtract its shot noise.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- read noise
Random electronic uncertainty introduced when the detector charge is measured and digitized. It limits faint-source measurements and makes many very short exposures less efficient than fewer longer exposures.
For an aperture of \(n_{pix}\) pixels, the variance contribution is \(n_{pix}R^2\), added in quadrature with photon-noise terms. Read noise is not removed by bias subtraction; the mean bias pattern can be removed, but stochastic read noise remains.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- dark current (SNR term)
The expected thermal-electron contribution per pixel over the science exposure, included as a variance source in the SNR calculation. It separates the uncertainty produced by thermal charge from the calibrated source and sky signals.
For \(n_{pix}\) pixels, dark variance contributes \(n_{pix}N_{dark}\) when the counts are treated as Poisson. This term is not the same as the residual mean after dark subtraction; the mean can be removed while its counting noise remains.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- aperture
The chosen set of detector pixels whose background-subtracted counts are summed to measure a source. Aperture size controls the tradeoff between captured source flux and added sky/read noise.
The SNR numerator grows with encircled source flux, while background and read-noise variance grow roughly with the number of included pixels. A larger aperture is not automatically better and must not be confused with the telescope’s physical entrance aperture.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- detection threshold
A stated signal-to-noise or false-alarm criterion used to decide whether a measured signal is distinguishable from noise. It turns “I can see something” into a reproducible statistical decision.
A common convention is SNR near 5 for a secure isolated detection, but the false-positive rate also depends on how many locations or trials were searched. A threshold is a convention tied to a noise model, not proof that the source is real under every systematic error.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- shot noise
The Poisson fluctuation in the number of discrete photons or electrons counted during an exposure. It explains why relative precision improves only as the square root of the signal.
For signal \(S\), the fractional shot-noise floor is approximately \(1/\sqrt{S}\). Shot noise is not detector malfunction and cannot be calibrated away after the photons have been counted.
Developed in this lecture: The term helps interpret this evidence: Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error. It also constrains the working model: Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
Evidence
- Repeated measurements of the same faint source scatter by an amount consistent with the square root of the total counts (Poisson statistics), not a fixed absolute error.
- Increasing the aperture size to include more sky pixels increases the noise term even when it does not increase the captured source signal.
- A source with SNR near 3 appears in some exposures and not others of the same field — exactly the behavior expected of a marginal detection.
Model
- Random noise sources (source shot noise, sky shot noise, dark current shot noise, and read noise) combine in quadrature, not by simple addition.
- SNR above about 5 is conventionally treated as a secure detection; SNR between about 3 and 5 is marginal and requires independent confirmation.
- A quoted flux without an SNR or an aperture specification cannot be evaluated for reliability.
Working Equation
\[ {\rm SNR} = \dfrac{S}{\sqrt{S + n_{\rm pix}\,(B_{\rm sky} + D + R^2)}} \]
Worked Example
- Aperture has n_pix = 9 pixels; sky background B = 150 e⁻/pix, dark D = 5 e⁻/pix, read noise R = 8 e⁻ (R² = 64).
- Noise term per pixel-summed background: n_pix × (B + D + R²) = 9 × (150 + 5 + 64) = 1971 e⁻².
- Bright source (S = 2500 e⁻): SNR = 2500 / √(2500 + 1971) = 2500 / 66.87 = 37.4 — a secure detection.
- Faint source (S = 300 e⁻): SNR = 300 / √(300 + 1971) = 300 / 47.66 = 6.3 — marginal; still above 5 but close enough that a single frame should not be over-interpreted.
Common Misconception
Equating "visible in the displayed image" with "securely detected." Image display stretches can make an SNR-3 fluctuation look identical to an SNR-30 source.
Active-Learning Guidance
Recompute the faint-source SNR after doubling the exposure time (S and background terms both double) and explain why SNR improves by less than a factor of two.
Lab / Observing Connection
Lab 02 (Signal-to-Noise Budget) requires computing SNR for several source/background combinations and identifying which are secure detections.
Synthesis Questions
- What was measured directly in this lecture’s worked example, and what was inferred from the model?
- Which uncertainty or systematic would most change the interpretation?
- How does this technique connect to the technical observing report due at the end of the term?
References
- OpenStax Astronomy 2e, Chapter 6.3 (detectors) and Chapter 17.1 (the brightness of stars).
- Local reference copy:
references/openstax-astronomy-2e.pdf.