The Siglent SDS2104X PLUS convinces me more and more, the firmware is more established, if the 10-bit mode works for me to see the noise, it is an oscilloscope that I like
That's it. Do not believe everything posted here, like claims a DSO being the wrong tool for noise characterization. This might be true for
some scopes with noisy frontends and/or lousy FFT implementation, but not for e.g. an SDS2000X Plus. Of course, an SDS2000X HD would be even better.
When analyzing noise, we don't want a single absolute number like from a DMM with its very limited bandwidth; we want to see the entire spectrum with accurate levels for every single frequency component. Such information also helps us to identify the source of the noise (so we might be able to do something about it).
Once again, instead of just claims and speculations, here is a practical demonstration (older measurement, not the latest firmware), where I happened to measure a weak 10 MHz signal and wanted to find out if the 10 bit mode can improve such measurements even though the FFT in itself provides a resolution enhancement already.
I've found a measurement of the -70 dBm level particularly, which is 71 µVrms or 200 µVpp respectively - far below 600 µV (whatever it should be, peak, p-p or rms).
First the measurement in 8 bit mode.
SDS2354X Plus_LVL_10MHz_1mV_-70dBm_8bit
Of course the signal is not visible in the time domain, because at about 600 MHz bandwidth, the oscilloscope's own noise is 76 µVrms or 685 µVpp. Yet the FFT shows the overall noise floor as well as the single signal quite well.
The frequency step is 119,2 Hz and with Flattop window the resulting RBW (Resolution Bandwidth) is ~450 Hz. I did not measure the noise floor back then, but from the graph we can estimate it to be about -124 dBm = 141 nVrms or 399 nVpp for a 450 Hz wide bin. From this, we can calculate the noise density to 141 nVrms / √450 Hz = 6.65 nV/√Hz. From this, we can calculate the total noise for any desired bandwidth.
Hint: The signal source for this particular test did not include noise deliberately, so this measurement shows the limits of the instrument in this particular configuration and setup. It means that external noise has to be stronger than about 10 nV/√Hz in order to be detectable. Still not too bad.
But even more important than the wideband noise floor is the identification and proper measurement of spurious signals, which can give a strong hint on the noise source.
In this example, with just 8 bits, we measured the 10 MHz signal as -69.8 dBm (72 µVrms, 205 µVpp), which results in an error of +0.2 dB or +2.33 % - not too bad for a microvolts signal level at 10 MHz - try that with any DMM!
Now the same measurement in 10 bit mode.
SDS2354X Plus_LVL_10MHz_1mV_-70dBm_10bit
The signal is still not visible in the time domain, even though the input bandwidth is now limited to 100 MHz because of the 10 bit acquisition mode.
The noise floor is now at about -130 dBm = 71 nVrms or 200 nVpp for a 450 Hz wide bin. From this, we can calculate the noise density to 71 nVrms / √450 Hz = 3.34 nV/√Hz. That means, the noise floor has dropped dramatically by 6 dB – we have gained one bit of dynamic range, or in other words: the ENOB has been increased by one bit, just as expected – and up to 100 MHz there should hardly be any noise sources we could not accurately characterize with this.
Again, we can calculate the total noise for any desired bandwidth.
The measurement of the 10 MHz signal is now spot on as well. We can measure it as -70.015 dBm – an error of just -120 nVrms or -350 nVpp, equivalent to -0.015 dB or -0.17 % - no further comment necessary.