# Are longer integration times equally good or better than averaging over several shorter timespans?

Consider the signal of e.g. a photodiode or a spectrometer, where a dark measurement has been taken to account for baseline noise. The signal we want to measure is quite weak, so we need a decently long integration time to almost saturate the device, lets say 1 second for the sake of simplicity. This gives us a reading every full second.

What if instead we used a 200 or 500 ms integration time and averaged 5 or 2 measurements, respectively, into 1 reading. Are these averaged measurements objectively better or worse, as both read out the signal every full second? Worded differently, one could ask if it is a good idea to always go near saturation if the time allows for it, i.e., if the measured process is static or very slow?

• Please specify what "saturation" is in this case.
– Max
Aug 30, 2023 at 7:12
• In this case, I mean that the sensor is giving out a digital signal of "counts" up to 2^16 for certain wavelength bins, and saturation is reached when this value is hit or exceeded. If you consider a laser emission line with ever increasing power, at saturation you would start to "cut off" the top and get a flat plateau peak. In practice, I try to stay at around 80% of max. signal. Aug 30, 2023 at 7:33
• Ok, so the sensor itself is integrating. Then the answer depends on what you want to measure. Just detection (signal or no signal) or actual measuring (does the actual count matter?). If the former, go for long integration time, if the latter, find a good trade off between SNR and minimal chance of saturation.
– Max
Aug 30, 2023 at 8:18
• Ed V postet a link to an indeed highly detailed answer by Dan Boschen. I don't wanna post an answer here, just give Dan's an upvote.
– Max
Aug 31, 2023 at 6:16
• That's indeed a very useful answer! Upvoted Dan's answer. Aug 31, 2023 at 7:16