A high resolution oscilloscope can be a great tool for certain tasks and some expect it to be particularly suitable for audio work. While the theoretical 12-bit dynamic range of ~72 dB should be a major improvement compared to the ~49 dB of an 8-bit system, the pronounced 1/f-noise of a general purpose Oscilloscope might get in the way especially at audio frequencies.
This is a general statement, and I wanted to quantify it for particularly the SDS2000X HD.
Look at the first screenshot
SDS2504X HD_NSV_50_Normal_BW20MHz
This is the noise spectrum from 1 Hz to 1 MHz, at a vertical gain of 1 mV/div with 20 MHz bandwidth limiter, open input at 50 ohms impedance and normal acquisition.
We can see there is an 1/f characteristics below 1 MHz, with a moderate increase of 5 dB/decade from 100 kHz to 1 MHz. This is much more pronounced below 100 kHz, where we see a major step down to 1 kHz and at 100 Hz the noise gets worse by more than 30 dB compared to 1 MHz.
Simple conclusion: working with small LF-signals can be challenging.
Of course, it is still possible to reduce the noise. Average acquisition mode is very effective for this. The frequency response remains unchanged, hence this mode is perfectly suitable for static signals – it just suppresses signal changes like modulation. There is up to 14 dB improvement with just 16x averaging and up to 35 dB with the maximum of 1024x, so it is almost possible to compensate for the excessive 1/f-noise, but at a slow acquisition like 100 ms/div it obviously takes a lot of time.
SDS2504X HD_NSV_50_Avg1024_BW20MHz
The 1024x averaging now suppresses spurious signals to levels we would never see otherwise. Take note of the strongest spur at around 622 kHz with a level of about -142 dBV. Of course this is really nothing. -140 dBV is already only 100 nVrms. The lowest visible spurs are just -162 dBV, so we’re able to measure signals below 10 nVrms - if we have a strong copy of the signal we can trigger on.
The thing is that we need very slow timebase settings in order to capture low frequencies. At 100 ms/div we get a record length of one full second, which allows a lower bandwidth limit of 1 Hz. But this also means that 16x averaging takes at least 16 seconds just for the acquisition, and likewise it’s 1024 seconds (more than 17 minutes) for 1024x averaging. Of course, it can get faster if we make do with a higher lower bandwidth limit.
Some folks put high hopes in resolution enhancement techniques, but a higher resolution does not reduce the frontend noise. Any noise reduction effect comes from the lowpass filter effect – but this obviously won’t help at low frequencies. ERES2.0 increases the ENOB by 2 bits in theory, but the resolution is increased by 4 bists, so it makes for a total of 16 bits. At 20 MSa/s it starts to show a slight effect at 100 kHz and gets quite effective at 1 MHz.
The following table summarizes the results. It also includes the measurements for 1 meg input impedance, which doesn’t make much of a difference at most frequencies though.
Norm Avg 16 Avg 1024 ERES 2.0
Frequency 1M [dBV] 50 [dBV] 1M [dBV] 50 [dBV] 50 [dBV] 1M [dBV] 50 [dBV]
-------------------------------------------------------------------------------
10 Hz -120.076 -119.272 -130.687 -130.018 -147.835 -118.742 -118.294
100 Hz -118.343 -120.791 -133.380 -130.826 -147.019 -119.660 -121.530
1 kHz -124.813 -127.056 -139.074 -136.987 -162.968 -125.774 -126.184
10 kHz -131.727 -141.982 -141.682 -150.861 -173.416 -131.434 -143.085
100 kHz -145.991 -149.912 -157.580 -160.940 -185.850 -149.188 -155.733
1 MHz -150.905 -150.224 -162.034 -158.489 -173.304 -166.809 -169.508
-------------------------------------------------------------------------------
EDIT: corrected the statement about low level spurs.