a20mc1
plot first 256k samples of data, (0-14 ms).
plot power spectral density, (DC-10MHz, semilog).
plot power spectral density, (DC-10MHz, loglog).
Suppose we calculate the power spectral density, (PSD), we would expect due to quantization noise alone. Let w denote the quantization noise process, with Fourier transform W, and power spectral density P. I estimated P by averaging |W|^2 over 244 successive segments of size L=4k, then dividing by L^2. The oscilloscope was set to 50mV/div on this channel, so the quantization step size, given 8-bit resolution, is q= (50e-3V/div)*(8div)/(2^8)=1.5625e-3V. So w is uniformly distributed over (-q/2, +q/2), and its variance is (approximately) s=2e-7 (V^2). Assuming the quantization noise samples are uncorrelated, the expected value of |W|^2 is L*(s^2), so the expected value of the PSD is P=(s^2)/L=5e-11, (for all frequencies). This compares very well with the noise floor, (roughly horizontal, with values between 5e-11 and 7e-11), observed in the plotted PSDs.