# Noise from quantization and sampling frequency

How to estimate if noise of sampled sound signal can be heard? I mean the noise that appear when an analog wave is sampled.

Similarly if I use 16bits, 24 bits, and 32 bits resolution, how to calculate noise in dB. I found this eq. SNR = 6.02*NoBits but I am not certain is it correct.

• There are some rule of thumbs, but it really depend on other things if it can be heard. For example, a CD is 16 bits, so that is a quite good estimate that it can not be heard (or can it?). However, compression schemes like MP3 utilize the fact that they can "add more noise if there are strong sounds". The quantization noise is not "plenty of square waves". The SNR-equation is correct (sort of, for sinusoidal input). – Oscar Mar 2 '15 at 15:30
• The quantization noise is not "plenty of square waves" what I meant here was sampling noise and sampling that changes a smooth wave. Because it is sampled periodically, effectively we end up with something that reminds stairs, squared signal. A signal like this introduces additional harmonics and therefore sampling noise. I would like to know how to estimate it. – Celdor Mar 3 '15 at 0:52
• It really depends on the input signal. For a sinusoid your equation is correct apart from adding 1.76 dB. For dynamic range one tends to use that equation. Google SNR and look at the Wikipedia-page (scroll down to digital signals and fixed-point). There is no direct relation between the SNR and if you can hear it (because of psycho-acoustic effects). – Oscar Mar 3 '15 at 10:57
• there are more than rules of thumb. this is essentially what we get in electrical engineering class when we study statistical or stochastic communications systems (i.e. signal processing regarding signals with models that are random processes) and some practical electronics regarding design of and specifications of electronic devices (i.e. what we get outa these devices) that we usually call "A/D converters" and "D/A converters" and that we sometimes mathematically model as "quantizers". today is Town Meeting Day in Vermont, i'm busy, i'll return to this later. – robert bristow-johnson Mar 3 '15 at 11:04
• rather than repeat what i have typed recently, i will just refer you to this answer that describes uniform quantization and suggests a simple method that sometimes helps reduce bad effects of quantization. – robert bristow-johnson Mar 5 '15 at 15:28