Every astrophotograph is signal (real light from the sky) plus noise (unwanted variation). The entire acquisition-and-processing workflow exists to separate the two. Noise comes in two families: random noise that differs every frame (beaten by stacking more exposures) and fixed-pattern noise that repeats every frame (removed by calibration frames).
The noise sources covered on this page:
- Photon (Shot) — Fundamental quantum noise. Photons arrive randomly following a Poisson distribution — like raindrops on a window. Brighter areas receive more photons, so they have more absolute noise but BETTER signal-to-noise ratio. This is the only noise that cannot be calibrated away — you beat it by collecting more photons (longer total integration). _Formula:_
SNR = √(signal electrons). _Reduced by:_ More integration time (stacking). - Read Noise — Every time the sensor is read, the amplifier and analog-to-digital converter add a small random error. It's like a toll booth — you pay it every single frame regardless of exposure length. Modern cooled CMOS sensors achieve 1–3 e⁻. This is why longer subs are preferred: you pay the toll once for more signal. _Formula:_
σ_read ≈ constant (e⁻/read). _Reduced by:_ Bias frame subtraction. - Thermal (Dark) — Heat causes electrons to spontaneously appear in pixel wells, even in total darkness. The rate doubles every ~6°C. Some pixels have manufacturing defects causing extremely high dark current — these are 'hot pixels.' This is why astro-cameras are actively cooled to −20°C or colder, reducing thermal noise by 64× vs room temperature. _Formula:_
∝ e^(−Eg/2kT) × time. _Reduced by:_ Dark frame subtraction + cooling. - Fixed Pattern — Each pixel has a slightly different quantum efficiency — how well it converts photons to electrons. Some pixels are 95% efficient, others 107%. This creates a fixed multiplicative pattern that's the same in every frame. Because it's perfectly repeatable, it's perfectly correctable with flat frames. _Formula:_
output = signal × QE(x,y). _Reduced by:_ Flat frame division. - Dust Motes — Tiny particles on the sensor cover glass or filters cast soft, out-of-focus circular shadows called 'dust donuts.' Particles farther from the sensor cast larger, more diffuse shadows. They don't move between frames (unless you touch the camera), so flat frames remove them perfectly. _Formula:_
shadow ∝ 1 − opacity. _Reduced by:_ Flat frame division. - Amp Glow — The readout amplifier chip emits infrared photons that leak into nearby pixel wells. This creates a warm gradient, typically in one corner of the image. It scales linearly with exposure time — longer subs have more glow. Visible in uncooled CMOS and long narrowband exposures. _Formula:_
glow ∝ exposure time. _Reduced by:_ Dark frame subtraction. - Banding — Each column of pixels has its own tiny amplifier with a slightly different electronic offset. This creates vertical (or horizontal) stripes. Some banding is fixed (same every frame), some is random. Dithering between exposures helps average out residual banding after bias subtraction. _Formula:_
offset varies per column. _Reduced by:_ Bias subtraction + dithering. - Quantization — The analog-to-digital converter (ADC) rounds continuous voltages to discrete integer values. With a 16-bit ADC (65,536 levels), each step is tiny and quantization noise is negligible. With 8-bit (256 levels), smooth gradients become visible staircase steps — 'posterization.' Most astro cameras use 12–16 bit ADCs. _Formula:_
σ_q = LSB / √12. _Reduced by:_ Higher bit-depth ADC.