Signal-to-noise ratio determines how clean an image is.
Sampling determines how that signal is distributed across pixels.
You cannot understand one without the other.
Pixel scale is the angular size of sky covered by a single pixel.
It is usually expressed in arcseconds per pixel (″/pix).
Pixel Scale (arcsec/pixel) = 206.265 × Pixel Size (µm) / Focal Length (mm)
Pixel scale defines how finely the image is sampled.
Seeing describes atmospheric blurring caused by turbulence in the air column above the telescope.
It sets a hard resolution limit, regardless of optics or sensor.
Typical seeing values:
Excellent site: ~1.5″
Good site: ~2″
Average UK conditions: ~2.5–3.5″
Poor nights: 4″+
Seeing is usually quoted as FWHM (Full Width at Half Maximum) of star images.
Sampling describes how many pixels are used to represent a seeing-blurred star.
Under-sampling: too few pixels
Critical sampling: optimal
Oversampling: too many pixels
Sampling is the relationship between:
Pixel scale (″/pix)
Seeing (″)
The Nyquist criterion states that to preserve information:
Pixel Scale ≤ Seeing / 2
(2″ seeing → 1″/pix)
This is the theoretical maximum needed to preserve spatial information.
In real astrophotography, this is rarely optimal.
In long-exposure deep-sky imaging:
Seeing varies during the exposure
Guiding error adds blur
Optical aberrations contribute
Noise dominates faint detail
As a result, slight under-sampling is usually optimal.
Seeing defines the atmospheric PSF.
To fully sample this PSF, Nyquist theory requires a pixel scale of seeing / 2.
In deep-sky imaging, however, signal-to-noise limits which parts of the PSF are measurable.
While finer sampling preserves all information in principle, coarser sampling can improve
per-pixel SNR and increase the detectability of faint extended structures.
Oversampling spreads the same photons across more pixels.
Each pixel carries:
Less signal
The same read noise
The same sky noise
If signal is split across 4 pixels:
Signal per pixel ÷ 4
Noise per pixel does not ÷ 4
This directly reduces per-pixel SNR.
For extended objects (nebulae, galaxies):
Signal per pixel scales with pixel area
Noise per pixel scales with √(signal + background)
Halving pixel scale (oversampling by 2×):
Pixel area ÷ 4
Signal per pixel ÷ 4
Noise ÷ √4 = 2
SNR per pixel ÷ 2
This is why oversampled systems look noisy even with long integration.
Moderate under-sampling:
Improves per-pixel SNR
Reduces read-noise impact
Is often invisible for faint structures
Under-sampling becomes problematic when:
Stars become blocky
Star shapes distort
Astrometry suffers
For wide-field and deep imaging, under-sampling is often a feature, not a bug.
Focal length does not determine resolution on its own.
Resolution is limited by:
Seeing
Sampling
SNR
A long focal length system that is oversampled:
Does not always resolve more detail
Has worse SNR
Requires longer exposures
Demands more integration time
Shorter focal lengths often go deeper.
Binning combines neighbouring pixels into one larger pixel.
2×2 binning:
Pixel area ×4
Signal ×4
Noise ×2
SNR ×2
Binning effectively:
Increases pixel scale
Improves SNR
Matches sampling to seeing
This is why binning works — and why CFALD works.
CFALD creates a high-SNR luminance layer by:
Combining Bayer pixels before de-mosaicing
Preserving spatial coherence
Increasing effective pixel area
Reducing read-noise impact
The result:
Improved luminance SNR
Sampling better matched to seeing
More recoverable faint structure
This is sampling optimisation, not magic.
Signal-to-noise ratio determines how much information you capture.
Sampling determines how efficiently that information is stored.
Too fine:
You waste signal
You amplify noise
Too coarse:
You lose spatial information
The goal is not “maximum resolution”.
The goal is maximum information per photon.