Ask five imagers how long their subs are and you will get five answers, from 10 seconds to 10 minutes. Most of them are right for their own setup. The length that matters is not a matter of taste: it is set by two numbers you can look up, your camera's read noise and the brightness of your sky. Below that length you lose signal to noise. Above it you gain almost nothing.
What a sub actually contains
Every frame holds three kinds of noise that matter here:
- Shot noise from the target, which you cannot avoid and which grows with exposure.
- Shot noise from the sky background, light pollution, airglow and moonlight, which also grows with exposure.
- Read noise, added once each time the sensor is read out, however long the exposure was.
Stack 60 one-minute subs or 6 ten-minute subs and you collect the same target light and the same sky light. The difference is read noise: the first stack pays it 60 times, the second only 6. So longer subs always win on paper. The question is by how much.
When read noise stops mattering
Noise adds in quadrature, the square root of the sum of squares. Once the sky has delivered enough electrons to each pixel, its shot noise is so much larger than the read noise that read noise barely moves the total. A common target is to let read noise add no more than 5% to the noise in a sub. That happens when the sky electrons per pixel reach about ten times the read noise squared:
sky electrons per sub >= 9.76 x RN² (5% added noise) shortest useful sub = 9.76 x RN² / sky rate
The sky rate is how many electrons per second the sky drops on one pixel. It depends on how bright the sky is, how wide your filter is, the area of sky one pixel sees, the aperture of your telescope and your camera's quantum efficiency. Pixel size and focal ratio do most of the work: a pixel at f/4 collects about twice the sky of the same pixel at f/5.6.
Worked numbers
Take a common setup: a ZWO ASI2600MC Pro color camera at its high-gain setting, about 1.5 e- of read noise, on a 72 mm f/5.6 refractor (400 mm focal length). Here is the shortest useful sub at each sky, with no filter and with a 7 nm dual-band filter:
| Sky | SQM | No filter | 7 nm dual-band |
|---|---|---|---|
| Bortle 2, remote dark site | 21.6 | 96 s | 23 min |
| Bortle 4, rural edge | 20.8 | 46 s | 11 min |
| Bortle 5, suburb | 20.1 | 24 s | 5.7 min |
| Bortle 7, bright suburb | 18.6 | 6 s | 86 s |
| Bortle 8, city | 18.1 | 4 s | 55 s |
Three things stand out.
A brighter sky needs shorter subs, not longer ones. In a city the sky swamps read noise within seconds. Long subs there only fill the wells with light pollution.
Narrowband changes everything. A 7 nm filter passes a tiny slice of the sky's spectrum, so the sky rate falls and the subs needed rise by a factor of about 14. That is why narrowband imagers shoot five to ten minute subs while broadband imagers in the suburbs shoot thirty seconds.
Gain matters through read noise. The same camera at gain 0 has about 3.3 e- of read noise. Sub length scales with read noise squared, so at Bortle 5 the shortest useful sub jumps from 24 s to about 2 minutes.
How much you lose with short subs
Under the Bortle 5 sky above, read noise adds 11.7% to the noise in a 10 second sub, 2.0% in a one minute sub and 0.7% in a three minute sub. Past the 24 second mark the curve is almost flat. Going from one minute to three takes about 1% off the noise, and leaves you with a third as many frames.
Why not just shoot as long as possible
Once read noise is swamped, longer subs carry real costs:
- A satellite trail, gust of wind or guiding glitch ruins a whole sub. Short subs lose less when that happens.
- Bright stars clip sooner and lose their color.
- Stacking software rejects satellites and planes by comparing many frames. With fewer than about 20 subs, rejection struggles.
A practical rule: use the shortest useful sub, or a little more, and round to a number your capture software makes easy. Go longer only when you have a reason, such as dim narrowband targets under a dark sky.
What sub length does not fix
Sub length only controls read noise. It does nothing about the sky itself. For faint detail, total integration time is what counts, and a bright sky needs much more of it. The Bortle 5 sky above is about 4.4 times brighter than a dark site at SQM 21.7, so matching a dark-site image of a faint galaxy takes about 4.4 times the hours. A filter helps with emission nebulae because it blocks sky light while passing the nebula's light. For galaxies and reflection nebulae, which shine across the whole spectrum, no filter can separate them from the sky.
Assumptions behind the numbers
The figures come from the exposure calculator, which uses a sky flux of 10,080 photons per second per square centimetre per nanometre for a magnitude 0 source, 81% combined optics and filter transmission, and splits broadband light three ways on a color sensor. It ignores dark current, which is small on a cooled camera. Read noise figures come from the manufacturer's published curves. Your sky meter reading, filter and gain setting will move the answer, so run your own numbers before a session.