# How to figure out SNR of a transducer

This is a novice question.

There is an accelerometer which outputs -10 to +10V voltage output. If I sample the output for constant output(accelerometer is not moving), can I say anything about signal to noise ratio by looking at samples? What should be the method to investigate SNR here or impossible?

If I sample the output for constant output(accelerometer is not moving), can I say anything about signal to noise ratio by looking at samples?

You can only say what the noise is. Here's how you compute the noise RMS level based on taking many samples: - Picture taken from here.

Subtract your static DC offset from the noise samples before squaring each term. Alternatively, if your noise is gaussian in nature you can estimate the RMS value from the peak values: - A reasonable approximation would divide the p-p value by 6.6 to get RMS. Picture directly above taken from here. Here's a picture from TI that shows the reverse calculation: - Regarding your signal, if you know what performance your transducer gives i.e. its transfer function then you can assume that a signal is present in the noise you have measured and calculate SNR.

• They provide only one page data sheet: crlsensors.com/prodDocs/sa-107ln.pdf I need to say something about the noise but I dont know where to start with. I can sample and log the data but I dont know what should I do rest to say anything about the noise. I was thinking to make the accelormeter stay in rest with constant voltage output with some noise on it, but how to quantify noise superimposed on the DC here? I dont know. Would rms of the signal reveal something? I really dont know where to start with. Feb 21 '18 at 12:38
• SNR in voltage terms is the RMS ratios of signal and noise so RMS is the right way. Feb 21 '18 at 12:40
• Is what you measure corresponding with what the DS says: Output Noise <2.5 μV RMS From 0 To 50 Hz? Bear in mind that the p-p noise is theoretically infinite but a lot of people say that multiplying RMS by 6.6 gives the p-p noise to a probability of 1 in 1000. Feb 21 '18 at 12:41
• But lets say signal is f and noise is n. In my sampled logged data i will be seeing (f+n) only. since i dont know what is f or what if n and I only know their sum, how can I find (f_rms/n_rms)? That is my dilemma. Feb 21 '18 at 12:43
• @Andy 6.6 RMS = 6.6 sigma, or approx. 1 part per million Feb 22 '18 at 3:11

It's typical in data sheet specs to list SNR as the ratio of full-scale signal to noise floor. If your accelerometer is actually isolated from all forms of acceleration including seismic noise and tilting, measuring the RMS voltage will give you the noise floor. This is non-trivial with a good (eg. tactical grade) accelerometer.

Data sheets will probably assume a DC signal so the RMS full scale is 10V.

If you want to know what the real SNR is in a particular application, of course you have to know the signal.

• Hello you wrote "If your accelerometer is actually isolated from all forms of acceleration including seismic noise and tilting, measuring the RMS voltage will give you the noise floor." RMS voltage or standard deviation? Feb 16 '19 at 15:49