Rigol DHO914S Bode plot

[gf]

I determine the (complex) ratio between CH1 and CH2. From that I determine the phase and amplitude. Both phase and amplitude can be averaged over multiple measurements taken into account phase wrapping in the right way.

Do not average magnitude and phase separately, but average the complex ratios fft(CH1) / fft(CH2) of multiple measurements in complex space (vector averaging), and finally determine magnitude and phase from the (complex) average of the ratios. The vector average is unbiased with regard to the noise of CH1, while separate magnitude and phase averages are significantly biased if the SNR of CH1 is low.

[ The noise of the reference signal CH2 (divisor) unavoidably introduces a bias in any case, but it seems to be virtually negligible as long as the SNR of CH2 is at least 20dB (after utilizing the detector's processing gain). ]

[TheoB]

For the moment I will not use averaging as there is much better way to increase the SNR of the measured signal.
I start with a square wave signal. The fundamental is set to 0dBfs (fill the entire screen). The fundamental has a SNR of more than 100dB. Harmonics attenuate by 1/N or 20dB/decade. This means an S21 (or 'gain') can be measured over a 4 decades frequency range and you have still 20dB of dynamic range left at the end.
For example when measuring an audio amplifier flatness from 10Hz to 20kHz. At 10Hz the SNR is >100dB and at 20kHz the SNR is lowered to 34dB. Is 34dB at 20kHz enough? That all depends on how much attenuation you want to see at that highest frequency.
With the current hardware I have (happy with the cheap DH0804) I have a 12 bit DAC running up to 1.25Gs. To prevent aliasing and noise folding, it's best to limit the frontend bandwidth to 20MHz and choose a sample rate of at least 40Ms/s. The Rigol has 62.5Ms/s at 1ms/div horizontal timebase (sampling rate cannot be set manually but follows from the horizontal timebase setting and the number of channels enabled).
Memory depth of 25Ms is the max when two channels are enabled (what I need).
25Ms/(62.5Ms/s)=0.4s That means FFT bin size is 2.5Hz (and the lowest bin also starts at 2.5Hz). You can measure the entire amplifier over 4 decades (10k) with at least 34dB SNR.

An extra addition is to capture twice. Once with 10Hz and once with 10kHz. The result will be merged into one plot ranging from 10Hz to 100MHz (or actually 20MHz due to channel bandwidth filtering).
The user should do this:

• Put generator at 10Hz
• Press capture
• Put generator at 10kHz
• Press capture

The FFT bins are exactly the same so this should be doable to extend the useful range.

[TheoB]

Implemented the dual capture idea idea and indeed it seems to work nicely:

The plot shows a single capture with 10Hz on the left which becomes limited by noise at high frequencies (and I therefor do not use that data).
On the right an extra run at 10kHz which shows the LRC notch at 110kHz.
It's just proof of concept (and hobby  )

[gf]

The 3rd decade already looks quite noisy. Maybe you should not cover more than 2 decades with one fundamental frequency.

That all depends on how much attenuation you want to see at that highest frequency.

In order to set a challenging aim, let's say 100+ dB.
Maybe user configurable, in order that the algorithm can be adapted according to the user's needs.
At the end it is a trade-off with measurement speed.

[TheoB]

Well, if you use just one decade for each frequency, the dynamic range is between 110 and 80 dB. Limited by the scope performance. And you get only 10 points per capture. Still 10 times faster than when using a single tone.
Drawback is also that you have to change the frequency more often. Two times feels like ok for manual control.
It's nicer if you can control the signal generator from a PC. In that case you have best of both worlds. Start with a fast frequency transfer using one or more square wave tones to cover a large band quickly. For high Q or high dynamic range use a classic slow stepped sine wave.
In all cases, use FFT on the raw data. It provides very low RBW for noisy environments like switched mode power supplies. And it gives the frequency, amplitude and phase of the external stimulus.  I've read the discussion earlier in this thread about using the scope's internal counter for frequency. That does not make sense if you can get that info from the frequency domain directly.

[gf]

In all cases, use FFT on the raw data.

How long does a full 25M point FFT take on the DHO800's CPU?

I guess there is neither an API to the hardware-accelerated FFT engine that could be used by custom apps, nor does it support that many points?

[Lathe26]

I think I have a guess as to why the Bode Plot feature has weird bends or corners in it (the "wiggles" issue is separate).  They are simply drawing the points at the wrong horizontal position in the graph.  The measurements themselves are fine; it's just displayed wrong.

For some reason, at each decade boundary, the 1st sample to the right of the boundary has an unusually large horizontal jump that is inconsistent with the ones that follow it.  In contrast, the 1st sample to the left has a small jump that is consistent with the ones that precede it. 

In the 3 photos, 300 points per decade was used.  This made the delta Hz between samples small.  The 1st photo shows the cursor at 9.92 kHz, the 2nd at 10.00 kHz, and the 3rd at 10.08 kHz.  Notice the cursor make a small horizontal jump from 9.92 to 10.00 kHz while it is large for 10.0 to 10.08 kHz.  The 2nd jump should instead be slightly smaller than the first due to the logarithmic scale.

The Excel file and chart were taken directly from the Rigol's measurements generated by the same Bode plot.

I'll report this to Rigol's support.

Apologies if someone else already covered all this.

Just to follow up on this bug after almost 2 years, with firmware v1.04, this bug still has not been fixed.

I pinged them again the other day, but it is the holidays...

[TimFox]

I think I have a guess as to why the Bode Plot feature has weird bends or corners in it (the "wiggles" issue is separate).  They are simply drawing the points at the wrong horizontal position in the graph.  The measurements themselves are fine; it's just displayed wrong.

For some reason, at each decade boundary, the 1st sample to the right of the boundary has an unusually large horizontal jump that is inconsistent with the ones that follow it.  In contrast, the 1st sample to the left has a small jump that is consistent with the ones that precede it. 

In the 3 photos, 300 points per decade was used.  This made the delta Hz between samples small.  The 1st photo shows the cursor at 9.92 kHz, the 2nd at 10.00 kHz, and the 3rd at 10.08 kHz.  Notice the cursor make a small horizontal jump from 9.92 to 10.00 kHz while it is large for 10.0 to 10.08 kHz.  The 2nd jump should instead be slightly smaller than the first due to the logarithmic scale.

The Excel file and chart were taken directly from the Rigol's measurements generated by the same Bode plot.

I'll report this to Rigol's support.

Apologies if someone else already covered all this.

Just to follow up on this bug after almost 2 years, with firmware v1.04, this bug still has not been fixed.

I pinged them again the other day, but it is the holidays...

In my (much) earlier posts, I showed the discrepancies in the Bode plot by graphing the data directly from .csv files using an external graphing program.

[Martin72]

Just to follow up on this bug after almost 2 years, with firmware v1.04, this bug still has not been fixed.

This firmware is from 2024.
The next one might work, because the new MHO900 series no longer has the Bode plot problem.
(It works now, but you can't scale it, and the strange display (horizontal lines) remains.)

[Lathe26]

In my (much) earlier posts, I showed the discrepancies in the Bode plot by graphing the data directly from .csv files using an external graphing program.

Cool.  I must have missed that post (the thread is long so I had to skim) since I posted the same thing here on Feb 11th, 2024.  Glad I'm not the only one who is annoyed by this bug.

[Lathe26]

This firmware is from 2024.
The next one might work, because the new MHO900 series no longer has the Bode plot problem.
(It works now, but you can't scale it, and the strange display (horizontal lines) remains.)

Do we know if Rigol will continue releasing new firmware update for the DHO900 or is v1.04 the end of the line?  Given that it has been over 1 years since the last version (none in 2025 for the DHO900), I have some concerns that they might have moved on to the next shiny thing.  If they back-port MHO900 fixes into a DHO900 release, that would be great.

[Martin72]

They should definitely do that, because not only has the Bode plot been repaired, but also the FFT function, plus the additional FFT modes (Average, Peak)...
https://www.eevblog.com/forum/testgear/rigol-mho98-and-mho900-oscilloscope-series/msg6080453/#msg6080453
https://www.eevblog.com/forum/testgear/rigol-mho900-testreview-(mho984)/msg6089599/#msg6089599

I can't think of any reason why Rigol shouldn't do that.
It's the same software platform, after all.

[Lathe26]

Just to follow up on this bug after almost 2 years, with firmware v1.04, this bug still has not been fixed.

I pinged them again the other day, but it is the holidays...

Following up again: no reply from Rigol so I created a new email thread with Rigol.  Hopefully, they will reply to this one.