Oscilloscope Bode Plot Speed Comparison

[Martin72]

Hi,

Anyway, maybe someone with various instruments (1000, 2000, 5000 & 6000 series) at hand could do a comparison of Bode Function speed? Would be interesting to see if the newer Siglent instruments are significantly faster than the entry and mid level devices. This might hint at where the bottleneck is regarding speed.

Hehehehe....
I got a SDS1104X-E and my SDS2540X-HD at home, and access to a SDS2104X+ at work.
As I´ve tested the batronix demoboard, bodeplot with the 1104X-E was the fastest.
Could re-test it again, making sure the settings were all the same.

https://www.eevblog.com/forum/testgear/oscilloscope-recommendation-ds1054z-sds1104x-e-gds-1104b-etc/msg4919146/#msg4919146

To keep the other thread "clean", I open up a new one.
This is not meant to be a "mine is longer than yours" thread, just a matter-of-fact observation of how quickly each scope processes a single sweep pass.
The conditions should be as equal as possible, otherwise it makes no sense.
I can contribute measurements with an SDS2104X+, an SDS2504X HD and an SDS1104X-E.
The means:
-Demoboard from Batronix with lowpass and bandpass filters(or similar filter)
-External frequency generator via USB, also for the two SDS2000s, since the SDS1104X-E has no internal one.
-Frequency range: 200Hz...10Mhz
-Frequency mode: Decade
-Number of measurement points per decade: 105
-Amplitude: 0.71Vrms
-1st pass with one DUT output (lowpass)
-2nd pass with two DUT outputs (lowpass and bandpass)
-3rd pass with three DUT outputs (LP,BP and "Linear", i.e. signal input)

Today I did this at work with a SDS2104X+, here are the results:

1.pass: 7min37sec
2nd pass: 8min46sec
3rd pass: 10min21sec

At the weekend the 2504X HD and the 1104X-E will follow, I will make a table.
Gladly seen are further results, from other models, like e.g. SDS5000X, SDS6000, Rigol MSO5000 and the R&S models.
Conditions are clear, the same means must be applied.
Then we can see what other tests we can do under what conditions.

[jjoonathan]

Anritsu MS4630B:
10Hz-10MHz
1st pass: 200ms
2nd pass: 200ms
3rd pass: N/A (not enough channels)
Points: 501
Dynamic Range: 109dB (noise floor with input terminated, settings unchanged -- you can eek out more by sweeping slower)

This is not meant to be a "mine is longer than yours" thread

Exactly, it's a "mine is shorter than yours" thread!

Kidding aside, I would love to save the space and get rid of my low frequency network analyzer, even if it meant settling for something significantly less good. Unfortunately the integrated oscilloscope bode plotters I've seen are so terribly slow that I am not even tempted. Hopefully this will change, or has changed and we'll see proof in this thread -- but live updates are something I won't live without. They calibrate your intuition as to the quality of your setup and the sensitivity to environmental factors, fixturing, etc that you just don't get with "go get a coffee" sweep times.

[Martin72]

Hi,

Anritsu MS4630B:

Nice to know how fast it can go - when you don´t use a scope for it.
Not really comparable in the sense of this thread.

[jjoonathan]

Scopes could plot this fast if they tried. I suspect they could hit the dynamic range too, possibly with a few caveats on the limitations of process gain. Scope bode plots don't suck because the architecture forces them to suck, they suck because they aren't trying very hard. What we have now are bode plots written by interns using SCPI functions, not bode plots written by graybeards using verilog.

Now that I've said my piece I'll stand aside and enjoy the sloth race 

[nctnico]

Scopes could plot this fast if they tried. I suspect they could hit the dynamic range too, possibly with a few caveats on the limitations of process gain. Scope bode plots don't suck because the architecture forces them to suck, they suck because they aren't trying very hard. What we have now are bode plots written by interns using SCPI functions, not bode plots written by graybeards using verilog.

I don't think that is entirely true. What oscilloscopes lack is steep filters in hardware so there is quite a bit of processing involved so sift through the incoming signal. The R&S RTM3004 needs about 80 seconds to do a sweep from 200Hz to 10MHz which is over 5 times faster compared to the Siglent scope. I doubt the RTB2004 is slower as the basic platform is mostly the same.

BTW: I have an Anritsu MS4630B as well which I use for my frequency analysis needs (paid around $450 for it which was money well spend).

[thm_w]

Scopes could plot this fast if they tried. I suspect they could hit the dynamic range too, possibly with a few caveats on the limitations of process gain. Scope bode plots don't suck because the architecture forces them to suck, they suck because they aren't trying very hard. What we have now are bode plots written by interns using SCPI functions, not bode plots written by graybeards using verilog.

I don't think that is entirely true. What oscilloscopes lack is steep filters in hardware so there is quite a bit of processing involved so sift through the incoming signal. The R&S RTM3004 needs about 80 seconds to do a sweep from 200Hz to 10MHz which is over 5 times faster compared to the Siglent scope. I doubt the RTB2004 is slower as the basic platform is mostly the same.

BTW: I have an Anritsu MS4630B as well which I use for my frequency analysis needs (paid around $450 for it which was money well spend).

There is no filtering processing is there, I don't think?
Scope sets gen to 1kHz -> waits a bit for signal to stabilize -> measures amplitude and phase on CH1 -> scope sets gen to 1.1kHz -> etc. fairly simple process.

Its mostly the waiting around bit that is taking the time. Which obviously goes way up when you put a large number of points in.

[Fungus]

I was told very sternly by forum members that a Bode Plot should show phase as well, otherwise I can just feed some white noise into my 'scope and do an FFT to get a real time display of frequency response.

[Martin72]

Its mostly the waiting around bit that is taking the time. Which obviously goes way up when you put a large number of points in.

Yap,you can watch this on the display of the gen - tick...tick...tick and from 1Mhz on it goes faster.
I don´t have a problem with the "speed" of this process, I´m not in a hurry.
And I already knew it from other applications like ARTA ( Speaker measure program suite).
Where the frequency response is measured very fast (due impulse) and thd is real "slow".
There are other things around this bode mode that are annoying.
But this is not a point here.

[nctnico]

Scopes could plot this fast if they tried. I suspect they could hit the dynamic range too, possibly with a few caveats on the limitations of process gain. Scope bode plots don't suck because the architecture forces them to suck, they suck because they aren't trying very hard. What we have now are bode plots written by interns using SCPI functions, not bode plots written by graybeards using verilog.

I don't think that is entirely true. What oscilloscopes lack is steep filters in hardware so there is quite a bit of processing involved so sift through the incoming signal. The R&S RTM3004 needs about 80 seconds to do a sweep from 200Hz to 10MHz which is over 5 times faster compared to the Siglent scope. I doubt the RTB2004 is slower as the basic platform is mostly the same.

BTW: I have an Anritsu MS4630B as well which I use for my frequency analysis needs (paid around $450 for it which was money well spend).

There is no filtering processing is there, I don't think?
Scope sets gen to 1kHz -> waits a bit for signal to stabilize -> measures amplitude and phase on CH1 -> scope sets gen to 1.1kHz -> etc. fairly simple process.

No, it is not simple as that  The software needs to apply a steep filter (single tone FFT / Goertzel filter) to retrieve the transmitted tone from the signal and determine magnitude & phase from that. Otherwise the result is affected by noise, auxilary frequencies (switching power supplies for examples) and harmonics from non-linear DUTs. It is possible the tone that is measured is swamped by a tone at a different frequency.

[Someone]

Looks like a rehash of:
https://www.eevblog.com/forum/testgear/bode-plot-computational-time-for-various-dsos/?all

[thm_w]

No, it is not simple as that  The software needs to apply a steep filter (single tone FFT / Goertzel filter) to retrieve the transmitted tone from the signal and determine magnitude & phase from that. Otherwise the result is affected by noise, auxilary frequencies (switching power supplies for examples) and harmonics from non-linear DUTs. It is possible the tone that is measured is swamped by a tone at a different frequency.

Is that confirmed on all scopes though? There is a post from Someone here stating cursor measurements may be used: https://www.eevblog.com/forum/testgear/bode-plot-computational-time-for-various-dsos/msg4510168/#msg4510168

For the ones that use FFT, a comparison might mostly depend on whatever memory depth/timing parameters the manufacturer decided to go with. Although R&S did have a fast FFT so it could help.

[Anthocyanina]

the AD2 does it in under 20 seconds. this is an active low pass filter, the same test conditions, 200Hz to 10MHz, 105 points per decade. It shows the bandwidth warning, and in the yellow trace you can see the measured wavegen gain compared to what it is supposed to be, it goes to -3db at around 9MHz and remains flat up to about 2.2MHz, but still, that doesn't affect the bode plot speed, only its accuracy. I'll do keysight 1102g next and report! 



1102G takes 5 minutes 24 seconds, and things get very weird near -18dB

[Berni]

I was told very sternly by forum members that a Bode Plot should show phase as well, otherwise I can just feed some white noise into my 'scope and do an FFT to get a real time display of frequency response.

The Agilent 89410A will do a bode plot just as fast as that and can display phase or group delay too.

On top of that it will also do 85dB of dynamic range. See signals down  to -140dBm and will also compensate for the DUT loading it down the signal source as it has 2 inputs and plots the difference between the two inputs.

It really is a bode plotting machine. Tho due to this thing being from the 90s it has taken a lot of hardware to pull this off, making it a huge 21kg boat anchor and draws 100s of watts of power.

[skander36]

... I doubt the RTB2004 is slower as the basic platform is mostly the same...

Right! RTB2K 74.1 s for 200Hz-10MHz, 105 points, low pass filter.
Rigol 5K - 5min 45s same parameters.

[rf-loop]

Some common comments about FRA aka BodePlot. Bit OT because this do not include speed test but some comments relative to this and specially relative to comparing.



There are different ways to do it in normal oscilloscope.  And when do compare then need carefully think what we compare... ok now speed but without  caring any other things...


But also there are many differences in performance and example in speed. 
Different oscilloscopes may use very different way to do BodePlot what leads then also very different performance. .

Also user can set most optimal settings for different cases (after he well know how it works and how to setup)




Some tiny examples.



Full dynamic range when Channel Gain = Auto 
From displayed noise floor to +40dBm. Yes it can do whole range (up to 140dB)  in single sweep. (when 1M input or External 50 ohm load what can handle these max powers)








Example: one 455kHz IF filter (lot of  better than in normal cheap radio)
There are many oscilloscopes what have BP... many of them can not do this, not even close.
Also this Siglent oscilloscope FRA can not measure where is real stop band... because oscilloscope noise level is far too high. But with this we still know stop band is least what is displayed.







Simple example: One 84.050kHz resonator. Even in this simple example dynamic range is over 100dB
(in this picture one div is 5 Hz)
There are many oscilloscopes what have BP... many of them can not do this, not even close. IN many oscilloscopes BodePlot is just for checkbox feature for marketing. They can do some RLC filter etc very simple cases, lot of fun but in bit more challenging cases they fails.  Of course there are also oscilloscopes what have good FRA also for bit more difficult things than just for easy playing... also example some Nnao like VNA's are out as...snow man in summer.




Then what it mean it is based to frequency selective receiver...
Systems what do not have frequency selectivity can not do good job when there exist strong harmonics or spurs what also DUT can produce and specially in many cases when we do control loop analysis is some PSU etc.



Some random example about Siglent FRA frequency selectvity when it sweeps.
Keep in mind that this is not normal BodePlot sweep when BodePlot control Generator.
Here BodePlot "sweep" over 200kHz to 700kHz frequency band, but input to Channel 4  is 455kHz fixed frequency signal so we can see BodePlot selectivity figure in this point.





About speed.

It is good to remember that with Channel Gain Auto sweep is much slow than Ghannel Gain Hold.
Also speed may be different depending input levels when Channel Gain Auto (some may call it as AGC).
Also speed may be different depending communication between oscilloscope and generator.

It is good to realize that after every frequency hop first it need solve level and adjust optimal V/div using input channels fine steps. After every step it can do full adjust for all channels in use (500uV to 1V/div when 50ohm and 500uV - 10V/div when 1Mohm)
Naturally when AGC is off (HOLD) measurement dynamic range is reduced. Still enough for most cases (but it need also that user carefully set first this optimal settings for input channels so that input signals do not clip in any case.)

Also for normal passive low order filters we really do not need so many steps as is used in OP test example (naturally this is ok for speed test purpose as used well ok)

With low frequencies this is still slow. It simply can not do it faster as long as we keep same measurement performance because this IS time domain oscilloscope and all is still based to it.

Example for low frequencies (10Hz  - 100Hz)
100ms/div horizontal speed takes 100ms/div. 10 div take 1s and then... whole this 100kpts need FFT  and repeating... repeating until enough for solve level and specially phase, and then go to next frequency step. Higher frequencies naturally t/dif is faster and FRA steps are then faster.
If reduce this, also performance is reduced. 
Example some AD2 can do it much faster. AFAIK it do not work as frequency selective sweeping receicver. If there is some other signals, say example harmonics or non harmonic spurs it also listen these.

Naturally there can do quite fast in oscilloscope BodePlot  if use more or less different method, and reducing reducing performance. Also Siglent can do. But 100dB or max 140dB dynamic range (with also good harmonics and spurs attenuation) is not so easy peace of cake.

So, When compare do not compare oranges and shoes. Not even apples and oranges or let us also then know all about performance differences. Speed is just one thing.


Then as @Berni well said... some real "bodeplotters", as example Hewlett-Packard 89410A (Sadly Lady Fiorina destroyed Hewlett-Packard, later known as Agilent and Keyshitg)
These instruments are very extremely different "animals" and expesive and they do just this but, and well, but not lot of other things. These BodePlot performance is really in different level.. as also price.

I remeber in old times I have some R&S dedicated "bode plotter". One was heavy and expensive R&S SWOB IV. Fast and "nice" to use for some filters etc. But not nice to move (heavy).

[nctnico]

I was told very sternly by forum members that a Bode Plot should show phase as well, otherwise I can just feed some white noise into my 'scope and do an FFT to get a real time display of frequency response.

The Agilent 89410A will do a bode plot just as fast as that and can display phase or group delay too.

On top of that it will also do 85dB of dynamic range. See signals down  to -140dBm and will also compensate for the DUT loading it down the signal source as it has 2 inputs and plots the difference between the two inputs.

It really is a bode plotting machine. Tho due to this thing being from the 90s it has taken a lot of hardware to pull this off, making it a huge 21kg boat anchor and draws 100s of watts of power.

The MS4630B can do all that (and much more as it is a network analyser) but jjoonathan just didn't show it.

[tautech]

The MS4630B can do all that (and much more as it is a network analyser) but jjoonathan just didn't show it.

Just as the 4 port 4.5 GHz SNA5004A then hacked to 8.5 GHz I had could do too but where does one stop ?


I believe it would be useful for the OP to add 'Oscilloscope' to the topic title of this thread.

[Berni]

Tho now that i look back at this, most of what makes the Agilent 89410A so fast could be done on any modern scope too.

The 89410A has a weird semi digital architecture that actually has quite a bit in common with digital oscilloscopes. The architecture of it is basically a SDR, the analog side is just an analog front end feeding a ADC. All the rest happens in the digital domain where it has dedicated DSP hardware for processing all that data in real time as it streams out of the ADC. All the spectral analysis features are actually done in the digital domain.

Sure the ADC it uses has more bits (14bit i think) and the front end is lower noise, but apart from that it is basically the same as a scope. With how much more powerful CPUs have gotten since the 90s we could probably run all the processing in software now.

The way it does bode plots is by generating a sine sweep waveform and feeding it into the arbitrary signal source. Then it synchronizes the acquisition with the source and does FFT on what it got back. So the sweep is basically just a more predictable replacement for 'white noise' in other to get a broad frequency signal to excite the DUT. This means it only needs to do one FFT per sweep, so it can do multiple sweeps per second.

I don't see why oscilloscopes don't do the same thing given that they have ARB signal generators and already have FFT capability. All of the scopes with bode plot functionality seam to do the thing where they step trough frequencies one by one and do a one shot amplitude/phase measurement at each one, this is obviously very very slow.

[jasonRF]

Unless you know the gory details of how the scope is doing the analysis, these will all be apples to emu comparisons. 

For example, using my picoscope 5244B with the FRA4Picoscope app I can change the time by orders of magnitude by adjusting the parameters.  If I assume it is a low-noise environment and ask for at least 16 cycles of the waveform, then it takes about 24 seconds.  On the other hand, if I assume a high-noise environment and ask for at least 10 Hz resolution bandwidth, I get just a couple of samples per second even at the highest frequencies.  Of course at those high frequencies it needs to collect (and transfer over usb 2) huge numbers of samples. I'm sure if I go to even narrower resolution it will slow down accordingly. 

I have never dug into the details of how that app works, so don't even know if it does averaging.  In principle, this is another knob that can be adjusted (if it isn't already).  And then the app has an adaptive mode where it adjusts the signal amplitude, in order to use it with SMPS where too large of a signal can of course cause problems.  I have never worked with that, but presumably that can make it even slower. 

When writing my own code to do FRA with simple, shallow-memory scopes (Owon vds1022 and Pico 2204a), I usually assumed a low-noise environment so did no averaging.  I still get ~70 dB dynamic range (~40 from the scope, ~30 from the several-kSample FFT).  In this case at low frequencies I just wanted X cycles of the waveform and maximized the sample rate given the number of samples available; but at high frequencies where I need to use the max sample rate, I end up with a 'constant resolution bandwidth' approach since I don't want use smaller FFTs at the higher frequencies.  If you go down to, say, 1 Hz, then collecting data and needing to do things like adjust the vertical scale just slows things down a ton, so I always start with the highest frequencies and work backwards to minimize those low-frequency vertical scale adjustments.  The AD2 has the advantage of not needing any such adjustments, which can help speed it up. 

jason

[mawyatt]

I believe it would be useful for the OP to add 'Oscilloscope' to the topic title of this thread.

Agree, this should be about Scope Bode Function, FRA or whatever one wishes to call this capability, not all inclusive of other instruments. We all know, or should know, that the dedicated instruments mentioned (likely many more) are going to perform better than the DSO under Bode display & analysis.

A DSO is a Time Domain instrument, designed as such, however does respectable service (at least some do) when asked to perform some service in the Frequency Domain (FFT or Bode).

To put this in perspective, would anyone consider comparing a SA, VNA, Frequency Response Analyzer, Bode Analyzer to display Time Domain waveforms and expect it to perform on par with a modern DSO 

Best,

[nctnico]

Unless you know the gory details of how the scope is doing the analysis, these will all be apples to emu comparisons. 

Logic dictates that every FRA implementation uses narrow bandwidth filtering aka selective frequency detection. Otherwise any DUT other than a simple RC filter will give false readings.

[jasonRF]

Unless you know the gory details of how the scope is doing the analysis, these will all be apples to emu comparisons. 

Logic dictates that every FRA implementation uses narrow bandwidth filtering aka selective frequency detection. Otherwise any DUT other than a simple RC filter will give false readings.

Agreed, but even in this framework you can adjust parameters that can change the time dramatically by changing how “narrow” you desire, determining how much averaging you need, etc.  Of course some measures of performance will also change with these decisions  (dynamic range, immunity to noise, phase estimate accuracy, etc).    With the picoscope  example, with 1% resolution bandwidth (at least 100 cycles) it takes < a minute.   Requiring at least 10 Hz absolute resolution takes about 10 times as long.  I would call both “narrow”, but they do not yield identical performance.   

Jason

[mawyatt]

Had a few moments late last night and decided to give the PicoScope 4262 a try, this has 16 bit ADCs and a high quality built-in AWG, however the AWG is limited too <20KHz, and the scope to <5MHz.

The FRA software for the PicoScopes was created by Aaron Hexamer, we are using version 0.7.3b (need to look for a later version if available). This is quite a nice implementation by Aaron, and definitely utilizing a Frequency Selective sampling scheme, with various parameters controlling such available to the user.

The only criticism of this software is the lack of a linear frequency scale, however the hardware (4262) has very limited AWG frequency range of <20KHz, so the ability to work with an external AWG would be nice.

We conjured up a Peltz Oscillator at ~10KHz for the Bode plot of the Injected Oscillator. After some fiddling around with the various parameters working with a Noise Rejection BW of 3Hz with high levels of oversampling and Noise Rejection parameter, this was very slow but produced some outstanding results 

This morning we decided to try the Analog Discovery 2, only to "Discover" that we couldn't download the basic scope operating software for either a Mac nor PC after jumping thru a bunch of hoops and getting an "access denied error messages" when attempting the download from a Mac or PC. This sent us searching for an answer on their user site, only to find nothing specific regarding such, which then required site registration and a very unusual method of verification which employs getting various letters from certain words and then added another with other letters which we failed to understand or execute properly, after several failed attempts finally getting thru. So not very impressed with Diligent so far

If we get the AD2 working we'll attempt a Bode plot of the mentioned oscillator.

Best,

[Fungus]

...this was very slow but produced some outstanding results 

All this test gear but no stopwatch or enough disk space for a screenshot?

[2N3055]

...this was very slow but produced some outstanding results 

All this test gear but no stopwatch or enough disk space for a screenshot?

No need to be disrespectful and rude...