DSO Quad bandwidth

I have to say that I never expected 72MHz bandwidth, as it wouldn’t be of much use with 72Msps. That being said, I had assumed that it would be somewhere north of 7.2MHz, say 10MHz? I’ll wait and see what it ends up being before getting judgmental.

Neat little package, starting to find my way through the menus, but not at the point where I trust it yet… the first signal I tried it on happened to be 24 MHz, so not a good place to start. But with judicious choices (one input active, scale adjusted favorably) I could see that there was a signal…

I have tested that just yesterday using an HP signal generator, a LeCroy scope an the DSO Quad.
I must confirm that a 5Vpp square wave is altered as low as 1 MHz. At 2MHz is much shaped as a sine than a square.
I don’t know whether my own DSO sample was particularly unlucky, but I can’t believe you see a 3.5M square wave with “reasonable” shape…Also depends on what you mean for “reasonable”.

Glad to hear from Seedstudio that there was some mistaken.
I also didn’t expect having a pro-scope, because it would meant that in our labs, by purchasing 4-zeroes-Euro scopes, we would have thrown money.
Being the sample freq to 72M (or 36M), it is also true that the cut-off frequency must no higher than the half. Practically speaking no more than a third or even a quarter.
I have a peek at the schematic, and I think something could be done, but not so far. Probably 15MHz will be a dream using this hardware.
Some observation about:
the OPAs have a very large bandwidth only by keeping the gain close to the unity (look at the GBW). Can’t expect having any substantial amplification over 10MHz. That’s the way I have tried with a strong signal yesterday (5Vpp). That is a complex task to solve with a small circuitry…
To mutate the analog input circuitry, you are using HCMOS switches. Even having a low parasitic capacitance, it cuts off a lot the final BW. It might be expensive, but the opto-switches could perform better, I guess.

I don’t think this scope is a fault, neither I feel to be betrayed. Maybe Seedstudio wanted to build an amazing scope, but I’d expect some additional consideration before initiating the production, mechanical also.
I want also wait for some concrete new about this instrument, then I will see what to do.

A thought: instead of retouching the existent hardware, I’d consider a almost-total bypass of the analog section of the DSO (almost straight to the ADC) and using an external amplifier/adapter. As for me would be a decent solution, if that will meet some concrete goal.

I do not take measurements and then fudged the results. I am very precise in what I post on forums. There is no magic or luck here. If it looks like a square wave, walks like a square wave and quacks like a square wave, then it must be a square wave at 3.546Mhz. Reasonable means that is has a small rise-time of less than 10% of the signal’s period and a slight sag (top and bottom of square wave slopes a little (which indicates a small DC gain loss in the front-end circuits)). You apparently haven’t learned how to maximize the DSO Quad which has a small capture buffer (4-Kbytes per channel) and sample rate limitations. You absolutely must optimized the T/Div for the highest sample rate as described in a previous post. When you have a more costly scope with a larger capture buffer and much faster sample rate (1GS/s) then you can be sloppy with your T/Div selections. :astonished:

The analog section is not the problem, the problem is the low sample rate. The front end op-amps and the DAC are not bottle-necks at 15Mhz with DAC overclocking for 144Ms/s. That stray trace capacitance would most likely not affect frequencies below 15Mhz. That only leaves the sample rate as the true bottle-neck. Your solution would have no affect on the sample rate whatsoever, therefore it would be ineffectual at best and the added stray capacitance of your additional circuitry would most likely have adverse effect.

That’s probably true: I’d love if you explain me how to display a 3.5MHz well-shaped square wave.
Let me know what and how to connect the Quad and I will try.

Absolutely NOT agree at all.
The sampling rate is enough to shape even a 10MHz square wave: just consider that there would be about 7 samples per period.
Anyway, let’s take a 3.5M square: that is 20 samples per period. It is a super-sampled signal!..Keep in mind that CD audio waves are sampled at 2.2xBW!!!
The real problem is just the analog section: take a look at the schematic and the OPAs datasheet!
The proof is what I see when connect as low as 1MHz square: an exponential rise and fall, instead of an edge.

Not sure to understand: do you mean a pro-scope is worser than a 150$ Quad???


CD audio captured data is then integrated to return back to a usable analog signal and this has nothing to do with viewing captured data.

The attached drawing demonstrates how 10, 8.5, and 4.25 samples per waveform distorts the results. As the samples per waveform decrease, then the distortion increases. Prior to SYS v1.31 you could only get 36MS/s for two channels, my 3.5xxMhz signal sampled at 36MS/s is more than 10X sample rate and displays properly on the DSO Quad as shown in the attachment below. Do your own sample rate drawing and demonstrate your point, instead of just talking about it.

I am aware of the hardware op-amp specs and they are 100Mhz band-width and 250Mhz upper limit. The front-end circuits appear to have gains of 7 or less for all V/Div settings. 100/7 = 14.285Mhz that the op-amps can pass and that is probably why Seeed Studio expects about 15Mhz with over clocking of 144MS/s. The analog mux switches that do the V/Div scaling have a rise time of 1.67ns which should easily pass 500Mhz, so that is not an analog limitation. The DAC is fed by unity gain op-amps for 100Mhz signal capability (at unity gain) and the DAC itself has a 470Mhz analog bandwidth but a limited sample rate of 100MS/s using the designed clock rate. So that is not an analog limitation. The only analog limitations would be the distributed trace capacitance and the capacitance of the frequency compensation caps in the front-end circuits, and most would agree that this band-width limiting would be minimal if any at all, below 15Mhz.

The only appreciable analog band-width limiting factor at 15Mhz is the sample rate and the resulting display distortion.

It also appears that you may be unaware of the concept of a single sweep capture at the fastest T/Div setting required to get the required sample rate, and then scroll through that capture buffer such that you can view the entire captured waveform. I have DSO Nano video lessons on YouTube that include this topic which is also applicable to the DSO Quad. Just search this Forum or YouTube for DSO Nano videos.

I have no object of insulting you, but your ignorance of this topic is no excuse for you to accuse me of falsely reporting 3.5xxMhz square waves on the DSO Quad that are of reasonable shape. The attached waveforms clearly show that you are wrong. In all this discussion (at least on my part), valid knowledge has been imparted to other Forum Users, and that is why I have continued this discussion with you. I am not sloppy in my testing and I report the scientific and repeatable facts as they are observed. You can have the final word if you need to, because I will no longer respond to your unsubstantiated remarks. I stand on the facts that I have already presented and I remain devoted towards expending my energy in helping others who want to understand the DSO Nano and the DSO Quad.

This post was edited on 4/23/2011 to clarify and to add more sample rate demonstrations in the attachment.
dso example.jpg

I never would have hurt you!..I really apologize if you feel that!
I was only asking you in which way did you measured/show a good shaped 3.5M square wave, nothing all.
Could you post any DSO snapshot of that wave?
I have explained what in the lab I have tried, which instrumentation and how much was the amplitude of the signals. I guess anyone can test it with a quite common scope with 100-200MHz of BW.
The wave I have seen (under several context) is far from being square.

The picture you have attached is partially correct.
It is true that having a perfect signal in the ADC input (better: the track-and-hold input) the equivalent wave will be distorted.
It is also true that never we could sample a signal having a BW over fs/2, because it generates aliasing. That is a fundamental math rule of discrete signals (Shannon). That mean we cannot feed any signal higher than 36MHz (theoretically speaking), let’s say 10-20MHz practically.
Otherwise the equivalent sampled signal would be a mess, without any chance to reconstruct correctly.
This is clear represented by the frequency spectrum that would see the baseband and the lower-modulated band overlapped.

When I write “signal” here, I mean SINE waves, not square or else.
You know that the only “pure” signal is just a sine wave. Any other periodic signal can be thought as composition of harmonics.
In the case, a (perfect) square wave is one of the most “tedious”, because it has only odd harmonics, decaying hyperbolically (so really slowly).

For example, let’s consider a 1MHz perfect square wave.
The main sine wave is at 1MHz, of course, and it is the greater in amplitude.
It has a 3rd harmonic, whose amplitude is about 1/3 the main. Then there is the 5th, having 1/5 of amplitude, etc.

If the scope would had a 15MHz of BW (as HugeMan is going to do), then I would have attenuations over the 13-15th harmonic…my wave would be a good square anyway.
Since I do NOT see that and I see a oddly shaped wave, I’d suppose the BW is much slower.
How to prove it?
Well, simple…that was a typical high-school task.
Just feed the DSO with a 1MHz (or less) sine: no matter what’s the amplitude. On the DSO display I see a good-shaped sine having about 500mVpp.
How is defined the BW? The cut-off frequency is defined as the point where the amplitude falls to 1/sqrt(2)=0.707 the nominal.
That’s good…so step up the input frequency to 2MHz: how is the sine on the DSO display?..It is still a good-shaped sine, but it’s amplitude is about 75% as before.
That is: the BW of the analog section of the DSO is 2-3MHz.

Hope that was clear enough.

I survived Vietnam so I am sure that you can not hurt me emotionally or physically, I just don’t want valid facts to be discredited. Although against better judgment, I will do one or two more responses on this matter if you will limit yourself to provable facts and the 3.5xxMhz display as the point of discussion. I no longer have a DSO Quad because my engineering hardware v2.2 was returned to China for exchange for a v2.6 model and that is the only reason you have not seen a picture here already. When I made those measurements for the benefit of those needing more info (before the betas were available), pictures were not included because the engineering model firmware has trigger, capture, and measurement issues that need to be addressed, and those fixes would most likely change the screen displays. I was just providing the bare facts at that time.

Which part of my previous attachment is incorrect? Be specific and don’t generalize because that serves no purpose. All square waves have some rise-time (both the generator and the display) due to the lack of infinite odd harmonics, and that is not shown for simplicity.

These parameters are not relative to a 3.5xxMhz square wave being viewed on the DSO Quad with SYS v1.30 or earlier.

These parameters are not relative to a 3.5xxMhz square wave being viewed on the DSO Quad with SYS v1.30 or earlier.

When you replace your word “slower” with the word “narrower”, then we are approaching the crux of this misunderstanding. What is your T/Div setting when you conducted this measurement? If you don’t remember, then repeat your measurements and tell me what you used for T/Div for the same results.

Unfortunately you are now talking about analog display roll-off and not band-pass roll-off. This is probably where your misunderstanding takes place. You have also switched off topic to a sine wave at a different frequency. BenF has already cited the artifact of display amplitude roll-off due to sampling rate choices. This display roll-off may be negatively influencing your viewed results. I did not compare the amplitude of the square wave at different frequencies such as 1.88Mhz like you are doing so we are now mixing apples and oranges again. Refer to page-5, my second post and BenF’s following response at viewtopic.php?f=12&t=1793&start=40 for more details about display roll-off.

It wont be clear enough until you place a 3.5xxMhz square wave signal into the DSO Quad and use the faster T/Div settings to view the signal with both analog channels active, and then tell me what you see. Then we will both be on the same page. Sorry about being short and direct with you, but we have to discuss the same issues in order to communicate successfully.

Hi Venarim & Lygra,

i do not wan’t to play as devil advocate here, but i want to add my two cents.

I have to agree with venarim in everything related to signal distortion & bandwidth (Hugeman has also indicated that bandwith is somewhat crippled).

But, i have to say that stray capacitances play it’s role here. The wave you see on screen depends directly on the probe you have. If you want to measure medium frequency signals (>10MHz), you have to use a compensated probe. Even in a Lecroy Wavemaster.

More importantly, the probe impedence must be adequate to use with that oscope. But what does a compensated probe actually do? It’s purpose is to reshape the leading edge and trailing edge overshoot/undershoot by matching the probe stray capacitance to the oscope stray capacitance. If you adjust the compensated probe for a nominal leading edge shape, then the trailing edge of that square wave is also compensated. The fundamental frequency of the square wave (and hence the bandwidth displayed) is not affected by probe compensation and has no effect upon the fundamental square wave frequency band-pass of the DSO Quad. So you have strayed considerably from the topic at hand. Probe compensation simply tries to compensate the higher order odd harmonics of the fundamental square wave frequency to compensate the square wave leading edge which is typically distorted when the probe stray capacitance does not match the oscope stray capacitance.

None of this has any appreciable effect upon sampled data collection bandwidth and the associated displayed fundamental signal band-width as shown on the DSO Quad. The absence of a compensated probe would only affect the leading and trailing edges of the 3.5xxMhz square wave as viewed on the DSO Nano and also as found in the capture buffer of the DSO Quad. The fundamental frequency of the square wave would not be affected in the absence of probe compensation.

You have failed to address one of the most important issues of my previous post, that issue which discusses display amplitude roll-off due to T/Div selection and display cell averaging. This can effect apparent band-width roll-off as viewed on the DSO Quad.

My attachment in previous post clearly shows how sample rate affects the reconstructed display output. At 4.25 times sampling rate, the signals have become clearly distorted. Why don’t you provide a rate sampled drawing showing what happens when the sample rate is only 2X, the value that you appear to agree with. If you were to take the time to confirm by doing this simple drawing, then you would find that 2x sample rate of a square wave results in a display that looks like a triangle wave. And I would have to say that a triangle displayed waveform is severe distortion of a square wave signal and is only caused by the low sample rate. The attached drawings do not include amplitude variation of the displayed square wave (reduced amplitude due to T/Div and averaging of each display position). I could not include this because I have not yet been informed of the “venarim” T/Div that was used during his measurements.

Hi Lygra,

I try explain myself better (also in shortform again :frowning::

This is what i was refering as compensation problems (gain error and X1 probe limited bandwidth): http://scope-probe-schematic.blogspot.com/

I know that average & T/Divs make signals appear bad sometimes (specially with T/divs over 1us), but if you have a 1Mhz signal (sampling @ 36Msps) and gets 1Vpp and for a 2MHz signal you get 0.75Vpp then the input stage (analog) bandwidth is 2MHz (-3dB).

Analog input stage seems to be ok (for a “modest” DSO) except for the 8/1000 gain branch. OPA2354 is a high speed one, but for high input amplitudes. In the case of small signal bandwidth (Vo = 100mVpp, see datasheet) the f(-3dB) is reduced a lot (90MHz for G=2), so could be noticeable @ G = 8.

I’ve seen a lot of strange things in electronics, so it’s very possible that both of you are right.

Disclaimer: i don’t own a DSO QUAD yet, so i’m speaking based on schematics and my field experience.

Hi Slimfish, being right is not a requirement for me as I have been wrong before. I am just trying to conduct a reasonable argument that we can all (including me) learn from. I have failed in my attempt to point out how T/Div in a DSO scope can impact the display amplitude roll-off and appear the same as band-width limiting would appear on a non-DSO scope. You have both ignored this argument as if I never made it.

I do appreciate your participation, but any future discussion without repeating the measurement on a Quad is futile. I don’t understand why Venarim has not repeated my measurement and reported his results. So all I can do now is wait until I get a Quad back into my hands and then take pictures.

I am surprised that another Quad owner has not responded with pictures of a 3.5Mhz square wave as depicted by a Quad. Oh well, I guess this just proves that no one with a Quad cares about this topic anyway. Thanks to both of you for participating.

LOL Id love to but i dont have a sig-gen :frowning:


Hi Lygra,

I also want to learn. But i didn’t think you failed to make your point.

Maybe someone provide that measurement

The problem with the community is that usually, someone will do for me if i wait enough time. Or i don’t have enough skills. Or that the threads evolve so fast and are enough complex that i could not follow them. But 300s QUADs have been sold in beta, so the interest is there. Don’t lose your faith. Your work is very interesting (i follow old DSO V1 threads) & instructing. Keep on going.

you don’t need an RF generator. I don’t know if signal generator in QUAD can work (square wave) at the same time scope do. If that works, set output to square 50% @ 2 & 4 MHz and measure changes in amplitudes.

Or if you have a PIC or an ATMEL lying arround, you can configure it’s internal clock in two different frequencies (say 4 & 8 MHz). Both architectures have a clkout pin which usually delivers clk/2 or clk/4. Use that to check QUAD.

According to the user manual, you do have such capability. The manual specifies that the gen output is capable of 8Mhz (for square waves), and I had previously reported that it looked good on another scope at 8Mhz.

I only used a function generator to test (because of better frequency adjustment resolution), but using the Quad’s gen output may be possible. I don’t remember if I tried that or not at 3-Mhz or 4-Mhz. I do know that I looked at it at 8-Mhz, but that was about 4-samples per waveform at 36MS/s and it did not look good. Maybe you could try at 4, 6, and 8Mhz and see what you get with 72MS/s and 144MS/s (in single enabled channel @ sysv1.31 or higher).

One word of caution is not to connect the input probe ground to the gen output leads. By not connecting the input probe ground at all, you can not get it reversed by accident and short out the gen output signal. This has happened on V2 Nanos that use a TTL output chip, so make sure that you do not do that.

This Quad uses an opa354 op-amp for the gen output stage, and it is spec’d for continuous output short circuit for one output pin per package, but I still would not take the risk. It appears from the schematic that U7 is a single stage package, but why take the risk? You don’t go around slamming autos into each other at 5-MPH to test that 5-MPH bumper and this is no different.

Another warning has to do with the Digital input jacks. Please be aware that those jacks are tied directly to the FPGA chip and an over/under-voltage here might prove fatal to the Quad’s FPGA. The digital channel maximum input voltages are not found in the User Manual nor in the FPGA data sheet. It appears that any voltage between Quad GND and 2.8V would be acceptable, but the current User manual does not specify, so don’t go there at this time.

If my warnings haven’t scared you away, then maybe you can let us know about your results as described above. :smiley:

There are 5V clamping diodes on the digital inputs. Used to protect against ESD but with 11kOhms in series can protect against an overvoltage.

I had thought of that but testing an unknown with an unknown was a bit of a waste of time too many variables. I was not sure of the accuracy or the precision of the sig-gen built into the quad. My reference was of unknown quality.

Since you have confirmed for me the quality of the output of the sig-gen then I can use it for all sorts of calibration and testing, including one presumes the matching of the probes to the quad.

My problem is that i am just a hobbyist with an interest in electronics and a capability to write software. I am not a field engineer nor a mathematician so my contributions to this debate are only the observations of a layman and i do not have access to secondary test equipment with which to calibrate/test my test equipment (my shinny new quad)

Cheers Pete.

This protection is likely to fail due to the very low power capability of the CESD5V0AP ESD Protection Diodes. They are designed to keep ESD off cell phone input circuits and the like. They are not designed to clamp over-driven input signals that can generate more than 1-ma for longer than a few micro-seconds. Many sampled wave forms can easily exceed these requirements and burn open the shorted CESD5V0AP ESD Protection Diodes. Now the full signal is applied to the unprotected FPGA with the exception of the 11k-ohm current limiting which may or may not provide adequate current limiting for input voltages exceeding digital levels. If you do the math it only takes 11V to push 1-ma through 11K-ohms. When this exceeds a few micro-seconds in duration, then the spec has been exceeded. There is also the issue of optional R39 & R40 which appear to tie back to both the STM and the FPGA when and if they are installed.

If these were 2-watt clamping zeners, then I would have to agree with you. But they are only 225mW clamping zener diodes designed to protect from narrow spike ESD and they can not protect the DSO Quad digital input circuits from signal supply voltages or large signal sources. The 225mW power capability will not stand up to this abuse.

I have done some other test.
I am sorry but I can’t made any picture, so you should believe what I write…
Generator: HP33120A
Reference scope: LeCroy LC358 1GHz
I have placed a 50 Ohms load on the generator output, then a 2-feet coaxial cable toward a T-plug. One side is for the LeCroy scope (with its probe), and the other for the DSO (with its probe).
Nothing else.
Hereinafter I will not mention what on the reference scope I see, because I have checked that is always as on the generator has been set. On the reference scope the square wave has a 20ns rise time, as well a 20ns falling time. This is constant over the various test below. The shape of the square wave is almost perfect, at a glance.
The waves outputted from the generator are perfectly balanced on zero Volts, that is a 1 Vpp sine looks like a Sin(t) function, having an amplitude of 500 mV.

On the DSO is always use only the channel A, the others are disconnected (without any probe). The channel B is hidden, but trying to enable it I didn’t notice any significative modification of the A track.
Over the tests, I have kept the DSO with DC coupling; also switching to AC no noticeable change of behavior/shape.
The triggering I have tested is NORM or AUTO, always using a rising (i.e. positive) slope.
I have also noticed that the amplitude measured with DSO is slightly less than the real one: it shows about 90% of the reference scope. That for low frequencies, so there is not any bandwidth limitation.

Test #1: 100KHz square wave @ 1Vpp
The DSO shows a well-shaped square and pretty stable. With NORM triggering it seems working well, although you have to serach carefully the stability level. That is strange because the wave is tall enough to find the voltage step. It seems there is no problem even switching the timebase to 100ns/div.
By using the AUTO triggering, the things are gettings worser: from 1us/div and above it seems working okay, but under 1us/div the wave is getting crazy: unstable and heavily deformed, not recognizable as a square. It looks much more a digital message, instead of a perfect wave. However, by adjusting the triggering level you may find a stability point. What the level does in AUTO mode?
When the square wave is stable, at 100ns/div, the rising/falling step of the wave is measurable: let’s say around 200ns, maybe slightly less.

Test #2: 500KHz square wave @ 1Vpp
Same as above. The wave displayed begins to be “rounded”, because of the rising/falling time. However not a bad shape.

Test #3: 1MHz square wave @ 1Vpp
Same as above.

Test #4: 2MHz square wave @ 1Vpp
Same as above.

Test #5: 3.5MHz square wave @ 1Vpp
Same as above. The wave period is shorter than the rising+falling time and the wave looks much more a sine than a square. The trigger ability to catch the signal seems begin faulting: it is always more hard to find the right point to keep the wave stable.

Test #6: 100KHz sine wave @ 1Vpp
The sine looks perfect.

Test #7: 1MHz sine wave @ 1Vpp
The sine still looks perfect.

Test #8: 5MHz sine wave @ 1Vpp
The sine still looks like a sine, but its aplitude is about 700mVpp. No matter what t/div you choose. It is also valid all the considerations for the trigger modes and their instability, as seen for the square wave.

Test #9: 10MHz sine wave @ 1Vpp
The sine is step-shaped, but it is absolutely normal because of the sampling. At 100ns/div you may easiliy count 7 steps for each sine period. Every step is really well shaped, that is the track-and-hold section is performing its task very well.
The amplitude of the wave is almost 400mVpp, that is less than an half of the real input. The funniest thing is that the DSO is set at 200mV/div, but it seems that is not possibile to magnify. By trying to set 100mV/div or less, the wave (i.e. the scope) don’t care at all. There is not any perceivable magnification. This strange bahavior is firing me a suspect: will the actual amplitude read by the ADC about 400mVpp or less? Ya, because if I switch to 50mV/div, the wave displayed is still about 2 divs…so it is almost 400mV or almost 100mV???

Test #10: elevating the generator amplitude
It seems there is no significative way to get better performance, even by feeding a 5Vpp signal on input.

That’s all.
I hope that clarify some doubt about the analog section of the DSO.

I made some measurements today with unfortunately broken Rohde & Schwarz sine wave generator but it was checked at other two scopes for accuracy. I made more than 20 screen captures and will upload them somewhere soon with comments.

In summary, the hardware should be capable to display good waveforms at the usual standard 10 samples per period but there are many bugs or firmware limitations which make it less useful. As of now when you turn on both analog channels, it seems to sample at 72 Msps, which was still OK at my 7 MHz test measurement. The wave is crude but for information display it’s good. For me it the 0.5 V/DIV or greater resolution was working well.

I forgot to add I’m using latest firmware SYS 1.34 and APP 2.33.

Thank you for expending the time and effort to conduct that valid and productive test. Only four additional parameters could make this post better, the T/Div for each test result, your SYS version number, T/Div sweep for each test, and your final conclusions.

I have learned that if you sweep the T/Div several steps smaller and the waveform amplitude does not change, then display amplitude roll-off due to time compression and display sample averaging is not a factor, and instead band-width limiting may very well be the culprit. Maybe you could provide those four parameters to your very informative post.

If you like you could add the additional info as an edit to your original post to keep all the results within the same post for continuity.

In test #9, did you try to increase the V/Div to see if the amplitude halved? This could prove noteworthy to see if the T/Div front-end circuit calibration is causing the same error on other higher V/Div steps. Of course the whole front-end calibration procedure is not well documented at this time and such non-calibration may be involved in the test result errors. But that is a whole separate topic.

In your test #5, the rise time will cause the sample to miss the top of the rise time and instead fall on the next sample time, and that would account for the narrowed square wave alternations. I have added another attachment to more clearly show how the rise time adds to the square wave distortion by missing a sample interval. I considered this when I called it a reasonable square wave display, but it does change the wave shape as shown in my attached drawing.

Once again I would like to thank you for taking the extra time and effort to help clarify this DSO Quad issue of band-width and display accuracy. You have done a good thing here. :smiley:

dso example2.jpg