Bandwidth limit on Siglent SDS2000X Plus oscilloscope

[rf-loop]

Another problem is that the "0.35" factor only applies when the scope has a Gaussian frequency roll-off.  A scope with a more flat or brick wall type frequency response would require the use of a higher number factor (usually 0.4 - 0.45).

You should avoid the term "brick wall" here, as no serious scope frontend will have a frequency response that is even remotely brick wall.

But still "brick wall" is commonly used (for Flat-response meaning in oscilloscopes) when professionals talk because they know context.
It have widely used in old HP-Journals, and HP-Agilent-Keysight appnote etc papers. All we know (I hope) what it mean when we talk oscilloscopes and brick-wall type freq response and still we do not mean (in any cases) ideal Brick-wall filter aka rectangle filter because it do not exist in practice... just as all we know that there is not perfect square waves or true DC but still we talk using these...of course.

Or is it only allowed for engineers from A-group manufacturers. At least they quite often use that term.
   

----
Common sidenote... HP-Journal, Oct 1993 is still good to read. The basics don't expire (mostly).

[pdenisowski]

But still "brick wall" is commonly used (for Flat-response meaning in oscilloscopes) when professionals talk because they know context.

Keysight:  "Responses that have brick-wall filters are desired as they produce less noise"
https://www.keysight.com/us/en/assets/7018-04317/application-notes/5991-4088.pdf

LeCroy:  "the tailored brick-wall frequency response"
https://cdn.teledynelecroy.com/files/whitepapers/enhanced-sample-rate-whitepaper.pdf

R&S:  " The user can choose between Gaussian or brick wall filter characteristics to optimize the oscilloscope step response"
https://scdn.rohde-schwarz.com/ur/pws/dl_downloads/pdm/cl_brochures_and_datasheets/product_brochure/3608_6919_12/RTO6_bro_en_3608-6919-12_v0502.pdf

Tektronix:  "Most real-time oscilloscopes today have a rather sharp roll-off (e.g. "brick wall")"
https://www.tek.com/en/datasheet/dpo7oe-series-optical-probes

Or is it only allowed for engineers from A-group manufacturers. At least they quite often use that term.   

Well, I've worked at both HP/Agilent and R&S, so that might explain it 

[BillyO]

You can tell that I've never worked with that formula, always used accurate frequency response measurements instead.

That formula is based on the physics of the system.  Provided you can mathematically characterize the system's response you can work out a version of that formula for it.  It should be quite accurate.   Gaussian response is well characterized and the resulting 0.35/risetime is accurate.  For moderately flat response systems (the top end of most mid-range PGA/VGA) the response is also well characterized and 0.4/risetime is also accurate. 

Using a sine wave generator has it's perils.  Unless you have calibrated it and are confident in it's performance it can also give very skewed results.  Ideally you would measure it's output and normalize it before taking each data point of the DUT to ensure it is flat at the frequency being tested.  The rise time is a single measurement and is accurate as long as the character of the response of the DUT is known.

There were several reasons that got my thinking wrong - ridiculous bandwidth claims being one of them.

Like the over 600MHz for a un-corked SDS2000X-P?  As an example mine has a rise time of 660ps, not the 800ps claimed by Siglent.  Using the formula (0.4/risetime) we get a BW of 606MHz.  Using a sinewave sweep @ I got 650MHz .  That 606 MHz does not look like a ridiculous claim to me.  The difference is probably got to do with my not being able to determine if the sine wave generator was flat.

It looks very different for a higher bandwidth scope like the SDS6000A. The 0.35 factor fits for the 1 GHz model, but nothing else matches.

I imagine the 500MHz license on the 6000 series has much better performance than advertised.  As for the 2GHz version, it is most likely using more advanced PGA/VGA with more agressive flattening.  Tektronix claim .45 for their "maximally flat" response so 0.46 seems very much in line with that.

[mawyatt]

Amplitude Brickwall Filters are not realizable but the analog IIR high order Chebyshev, Inverse Chebyshev and Elliptical filters produce a good approximation at the expense of highly non-linear phase response, FIR digital filters do even better. One issue for scope use of one of these analog filters would be the highly non-linear Group Delay (derivative of phase with frequency) associated with these type filters, whereas the less Brickwall like Bessel and Gaussian have a much better Group Delay which is very important for scope use and why they are usually employed. Of course the Digital FIR filters can disconnected the phase/delay and amplitude responses, which is a huge benefit in some applications.

Seems the swept frequency is the overall best way to accurately characterize a DSO. Since one doesn't know the exact relationship between the frequency response and step/impulse response, one uses the usual fudge factors, i.e. 0.35, 0.4 and so on, if that's important! The issue with the swept frequency response; Does the signal generator remain "flat" across the entire band in question? This is easily resolved with a quality SA, where the amplitude variation can be accurately measured across the band of interest. The tracking generator in our SA is pretty good and we've used such to "characterize" our uncorked SDS2000X+, which demonstrated over 600MHz BW

This is not to say the DSO inherent rise time and channel characteristics aren't important, it is and one must always consider the displayed waveform rise time/pulse waveform characteristics as being distorted by the DSO, how much so depends on the waveform and DSO. One must consider that the scope and waveform measurement at hand are non-coherent (generally a good assumption), so the display is the result of the root-sum-squared (RSS) response if the DSO is moderately well behaved. So for a 10% DSO displayed contribution/degradation the DSO rise time needs to be better than ~3.2 times the waveform under consideration, similar concept applies for waveform bandwidth measurements.

Best,

[pdenisowski]

Using a sine wave generator has it's perils.  Unless you have calibrated it and are confident in it's performance it can also give very skewed results.  Ideally you would measure it's output and normalize it before taking each data point of the DUT to ensure it is flat at the frequency being tested. 

Agree completely that you would want to be sure your signal generator amplitude is either "flat" or has a known frequency response (output level as a function of output frequency).

Most "professional" sig gens should have very good spectral flatness over typical scope bandwidths (100s of MHz up to low tens of GHz), especially as many monitor their own output level and can internally adjust output level as needed to maintain good "flatness" over their specified operating range.

[Martin72]

Once again I'm too stupid to do it and can't get it to work on my HD, it looks just like it did on the X+.
But since 2N3055 showed it with an HD(didn´t find the thread so far), I assume it's my stupidity.

Yepp it was...

https://www.eevblog.com/forum/testgear/math-problems-on-sds2k-(trying-to-display-bandwith)/msg4972270/#msg4972270

Now I´m curious if this will also function on the SDS2000X+ after several firmwareupdates.
Will check it when I´m back to work or someone could check it before.

[tautech]

Once again I'm too stupid to do it and can't get it to work on my HD, it looks just like it did on the X+.
But since 2N3055 showed it with an HD(didn´t find the thread so far), I assume it's my stupidity.

Yepp it was...

https://www.eevblog.com/forum/testgear/math-problems-on-sds2k-(trying-to-display-bandwith)/msg4972270/#msg4972270

Now I´m curious if this will also function on the SDS2000X+ after several firmwareupdates.
Will check it when I´m back to work or someone could check it before.

Try on P2 in the SDS2000X Plus thread:
https://www.eevblog.com/forum/testgear/siglent-sds2000x-plus-coming/msg2787168/#msg2787168

[Performa01]

You can tell that I've never worked with that formula, always used accurate frequency response measurements instead.

That formula is based on the physics of the system.  Provided you can mathematically characterize the system's response you can work out a version of that formula for it.  It should be quite accurate.   Gaussian response is well characterized and the resulting 0.35/risetime is accurate.  For moderately flat response systems (the top end of most mid-range PGA/VGA) the response is also well characterized and 0.4/risetime is also accurate. 

Nobody said the formula isn't exact, yet as long as the frequency response is unknown (i.e. as long as it is not properly plotted with a levelled signal generator), the factor in that formula is unknown as well. This is where the cat bites their own tail.

Using a sine wave generator has it's perils.  Unless you have calibrated it and are confident in it's performance it can also give very skewed results.  Ideally you would measure it's output and normalize it before taking each data point of the DUT to ensure it is flat at the frequency being tested.  The rise time is a single measurement and is accurate as long as the character of the response of the DUT is known.

Come on. I was talking about signal generators, not toys from ali express. Especially higher end ones have excellent flatness, all the more so when only the range up to about 500 MHz is really important, as for the usual entry level DSOs discussed here.

I even demonstrated the accuracy and amplitude flatness for both the Siglent SDG6052X AWG and my Anritsu MG3633A signal generator here. I think this should be flat enough for characterizing a scope bandwidth:
https://www.eevblog.com/forum/testgear/siglent-sdg6000-series-awg_s/msg2621457/#msg2621457

There were several reasons that got my thinking wrong - ridiculous bandwidth claims being one of them.

Like the over 600MHz for a un-corked SDS2000X-P?  As an example mine has a rise time of 660ps, not the 800ps claimed by Siglent.  Using the formula (0.4/risetime) we get a BW of 606MHz.  Using a sinewave sweep @ I got 650MHz .  That 606 MHz does not look like a ridiculous claim to me.  The difference is probably got to do with my not being able to determine if the sine wave generator was flat.

I think I have alredy stated that the actual bandwidth exceeds the specifications, especially in the entry level DSOs.
As I've posted many times before, my own measurements resulted in ~570 MHz - and I have not taken the cable loss into account, which was about 1 dB @ 500 MHz. There was no need to, in my book, because this gives some safety margin for "guaranteed results", so to speak. If we add that one dB and look at -4 dB, we are at about 605 MHz.

650 MHz on the other hand sounds not very realistic, yet not ridiculous either.

It looks very different for a higher bandwidth scope like the SDS6000A. The 0.35 factor fits for the 1 GHz model, but nothing else matches.

I imagine the 500MHz license on the 6000 series has much better performance than advertised.  As for the 2GHz version, it is most likely using more advanced PGA/VGA with more agressive flattening.  Tektronix claim .45 for their "maximally flat" response so 0.46 seems very much in line with that.

The hardware is the same for all SDS6000A devices.
I've tried to explain numerous times, why artificial bandwidth limits have high tolerances. So you can bet that pretty much all SDS6204A will have an actual bandwidth of about 2.2 GHz, but there might be quite some variation for the artificially bandwidth limited 1 GHz and 500 MHz models. The only safe bet is that the bandwidth of those will exceed the specification - but the amount could vary a lot.

[Performa01]

Amplitude Brickwall Filters are not realizable but the analog IIR high order Chebyshev, Inverse Chebyshev and Elliptical filters produce a good approximation at the expense of highly non-linear phase response, FIR digital filters do even better. One issue for scope use of one of these analog filters would be the highly non-linear Group Delay (derivative of phase with frequency) associated with these type filters, whereas the less Brickwall like Bessel and Gaussian have a much better Group Delay which is very important for scope use and why they are usually employed. Of course the Digital FIR filters can disconnected the phase/delay and amplitude responses, which is a huge benefit in some applications.

That's the key. We cannot do anything but Gaussian (or Bessel, at most) at the analog side, otherwise the step response gets awful and pulse fidelity is lost. Yet you can do it digitally – a FIR filter can actually approximate a brick wall and still have constant group delay. High End scopes combine analog and digital filters to optimize their frequency and phase response. But this requires a sufficiently high sample rate to bandwidth ratio – the usual factor of 2.5 in modern top models (within their class) will not allow an effective digital filtering, if only because of the ineffective AA-filtering in the analog frontend.

[gf]

With modern DSOs and their advanced features we can even let the scope plot its own frequency response, see the example for the SDS2504X Plus below.

Apparently just a FFT plot. This, of course, requires a stimulus with a flat spectrum in the frequency region of interest. May I ask what stimulus you used for this test? Sinc pulse? Chirp pulse? If you used a chirp pulse, how did you tweak it to avoid Fresnel ripples?

[Performa01]

I use a levelled signal generator (usually Anritsu MG3633A or an AWG like SDG6000X and SDG7000A), slowly sweeping over the desired bandwidth. FFT in peak hold collects the measurements over a few minutes.

This method has proved to be just as accurate and still way faster than manual data acquisition.

The big advantage is that we can tweak it a lot.
We can use very slow sweeps in order to get a nice looking graph right away. We can use a faster sweep so we get a first result pretty quickly, and the let it sweep some more times to fill the gaps in the plot.
What resolution in the frequency domain (RBW) do we need? I've chosen about 920 kHz, (8192 FFT-points @ 2 Gsa/s), but we can alter that if required. More FFT points will slow it down, so we need longer sweep times and/or more sweeps to get a nice contigous plot.

[gf]

... slowly sweeping over the desired bandwidth. FFT in peak hold ...

Thank you, that makes a lot of things clearer.

[ I was errenously assuming that the stimulus was some kind of wide-band pulse fitting into a single FFT window of only 4.096µs, and that would have been quite challenging, with good SNR. ]

[Martin72]

I use a levelled signal generator (usually Anritsu MG3633A or an AWG like SDG6000X and SDG7000A), slowly sweeping over the desired bandwidth. FFT in peak hold collects the measurements over a few minutes.

I had bought an SML01 (1.1Ghz) to recreate this for my 2504X HD.
But I have never done this before, how could a suitable sweep setting for this purpose be parameterized?
500Mhz bandwidth has the scope according to the model name, do you start at 300Mhz and stop at say 800Mhz?
And how could "slow" turn out?

[Performa01]

I use a levelled signal generator (usually Anritsu MG3633A or an AWG like SDG6000X and SDG7000A), slowly sweeping over the desired bandwidth. FFT in peak hold collects the measurements over a few minutes.

I had bought an SML01 (1.1Ghz) to recreate this for my 2504X HD.
But I have never done this before, how could a suitable sweep setting for this purpose be parameterized?
500Mhz bandwidth has the scope according to the model name, do you start at 300Mhz and stop at say 800Mhz?
And how could "slow" turn out?

For a frequency response plot from 1 MHz to 1 GHz I use the following setup:

Signal generator
Sweep mode: linear
Sweep time: 300 s
Sweep frequency setting: 1 MHz to 1 GHz
Sweep direction: up
Carrier amplitude: 0 dBm

DSO
Acquisition
Timebase: 500 ns/div
Trigger: edge, 0V, auto - or mains
Input: 50 ohms, 100 mV/div

FFT
Max points: 8 kpts
Window: Flattop
FFT mode: Max-Hold
Vertical: 2 dB/div, Ref = +2 dBm, Unit dBm, Ext. Load = 50 ohm
Horizontal: 0 - 1 GHz

The trace might not be perfect after the first pass, but you can just wait and let consecutive passes accumulate to get an optically nice result.

See attached example.

SDS2504X HD_FR_Full_0dBm_split

Of course you can choose a different frequency range, but this doesn't make much sense, because we always want to see the level at low frequencies as a reference. Usually, we take 1 MHz as reference and with the settings given above, this is just possible, because the RBW is still below 1 MHz. For faster scans we need to use wider RBW (= less FFT points) and then we also need to switch to 10 MHz as the reference frequency.

[Martin72]

Thank you very much