I am trying to observe the noise produced by UPS present in our lab, under no load (a normal running condition when its power is on). As I don't know its frequency range, I have randomly chosen 4 sampling rates to see which frequency peak remains constant. But surprisingly, peaks obtained in each graph is almost different from one another. So, in this case how to select sampling frequency, so that I can have a rough estimation of noise frequency coming from ups.
When the input signal bandwidth in unknown, you can begin with the highest possible sampling frequency, then observe the energy in the high frequency region (close to Nyquist frequency; i.e. half the sampling rate) and try to estimate the bandwidth of the signal and then choose the lowest possible rate still maintainig the Nyquist criterion according to this accepted bandwidth.
If the first trial indicates that there's considerable energy towards the Nyquist frequency, then you should try more complex sampling strategies, such as bandpass sampling at higher frequency regions, unless you can afford an even higher frequency sampling device.
Can't comment : modern UPS will have power stage switching in the 10s of kHz to several 100s of kHz. I'd recommend 10Mhz sample rate.