DSO Quad bandwidth

If the gain resistors are reduced and front end optimized as is then theoretically one can get up to 10MHz overall flat response @ G=10 according to the OPA2354 datasheet which makes sense. The corrected circuit is shown below. The C9 and C11 10pF caps should be lifted as well as C7A. Feedback resistors should be reduced as much as possible while maintaining the necessary gain.

DSOQ_FE_Schematic_Single_Stage_Optimized.jpg

According to the plots these changes shift the 3dB point for both signals up to 12MHz which in reality is overly optimistic. I’d say 10MHz BW is more realistic but still hard to achieve as it is highly dependent on the PWB layout.

DSOQ_FE_F_Response_Single_Stage_Optimized.jpg
DSOQ_FE_Phase_Single_Stage_Optimized.jpg

In order to achieve higher BW multiple smaller gain stages are needed. The example below shifts the cutoff to about 27MHz. I think I will try the first solution as it is fairly straight forward given the parts are not something smaller than 0805…
DSOQ_FE_Schematic_Dual_Stage_Optimized.jpg
DSOQ_FE_F_Response_Dual_Stage_Optimized.jpg
DSOQ_FE_Phase_Dual_Stage_Optimized.jpg

Hi,

i also did the same simulation (i wasn’t aware of geshsoft ones) as geshsoft but with TINA-TI (Texas Instruments). And looking at the simulations i came with the same conclusions. Getting rid of C9, C11 & C73 (or at least use lower capacities) will improve the circuit performance. Also the use of small valued resistors (i.e. 2,7k instead of 27k and keeping the same ratio for others) will also improve bandwith (but that’s a few small resistors -16- to solder).

That said, the only fact that disturbs me from simulation is that i didn’t find the strange channel behaviour venarim described. I suppose this is due to the stray capacitances not modelled in simulator. Sooo, it would be interesting to know how accurate are this simple simulations in order to tweak the circuit.

I leave the circuit as an attachment just in case someone wants to play with it.
QUAD.zip (7.86 KB)

Unfortunately simulators don’t tell the truth at these gain and frequency levels. The biggest problem would be getting stability and I suspect that C9 C11 and C73 were added later to acheive stability. Removing them is probably not an option. PCB layout is everything and having gain switching really makes the layout difficult because the + and - input nodes of the op amps need to be as short as possible. It should be possible to get 15nS rise time with the 27MHz circuit above but it will be difficult. Might be an idea to build up a simple fixed gain circuit first.

Thanks for pointing out that the circuit board traces have resistance, inductance, and capacitance that can greatly affect these circuits. I also suspect that those caps were added through trial and error.

Is there any way to quickly determine if there is a ground plane layer on this board?

It would be interesting if we could find a v2.2 schematic to compare any differences in the front-end circuits.

Yes, pick a powerfull light source and put the PCB in between the light and you. If you could not see through the PCB, then there is probably (could be a VCC one or other - but it’s very unlikely at best) a ground plane. :wink:

I just tried removing C73 (33pF) and it made a remarkable difference to bandwidth on the 0.5v range of ch.A I am now getting -3dB sinewave response to 10MHz. I then used a good quality scope to check the squarewave response at U6, AIN and there was slight overshoot but no instability or ringing.

I have noticed that my DSO Quad will not trigger on timebase ranges of 0.2uS and 0.1uS when set for channel A only (ie. with all other channels in HIDE) Has anyone else seen this?

Yes I agree with the fact that the simulation is just a rough approximate of the real world performance which is not accommodating for parasitics, and I did state so. The gerber files would be something I’d definitely want to see.

I opened up the scope yesterday and the R and C are definitely smaller than 0805, so I will order the parts I need to tweak the front end soon enough. I am planning on running a frequency sweep on the front end only and monitoring with my Tektronix - this way i won’t be limited by any firmware / software faults to properly determine what are its actual capabilities.

As far as slight overshoot - C37 is killing AC performance at higher frequencies just like C9, C11, C10 and C12 do. Removing the capacitors opens up the BW so the op-amp will perform closer to its specs which do show slight overshoot at the falling edge on the step response plot. The behavior is slightly worsened probably by parasitics. As bielec pointed out these capacitors were there as an afterthought or maybe even planned as space holders as compensation network which is fine. I have the feeling that the final values placed in our units did not get tested before hand however. In other words some fine tuning needs to be done to get the max benefit from this very simplified front end.

Like bielec I also noticed the triggering problem at 0.1us and 0.2us. Actually the signal fails to reconstruct properly on the screen once I exceed 2.3MHz or so. I will need to rerun all this once the analog BW is opened up.

I removed C9 and C11. This allowed much better bandwidth but overshoot was present and it varied depending on the input voltage setting. So what is needed is a different overshoot trimming capacitor across each gain setting resistor. That is 6 capacitors for channel A across each of R13,R17,R21,R11,R15,R19. These could be soldered on top of the resistors when proper values are determined.

For the sake of the experiment, I made capacitors from two short lengths of thin teflon coated wire, twisted together and fitted these across the R’s. I acheived acceptable compensation on all ranges and got bandwidth up to about 10MHz with good pulse response enabling me to easily display a 100nS pulse with 20nS edges.

A note on the probe compensating capacitors is in order. For ch A. the influence of C3A and C5A interact depending on voltage setting, making setting up difficult. The method I used was to inject a 10KHz squarewave and adjust for best shape while switching between the 0.5v and 1v ranges.

i am not so skillfull about the circirt , but i think, maybe you stimulation lacks a inportant item, the cpc1017 , which is not a wire when shorted, it contains a C about 30pF between the 2 legs in circuit , and also C between the 2 legs with the ground. that is what the designer (bure) told me, these C make the whole story different…

As far as datasheet states, cpc1017 has 30pf parallel to the switch and 1pF between input/output. I included in the simulation and the effect is negligible at the opamp inputs in terms of gain. Probably it will affect in terms of overshoot.

I updated simulation circuit to reflect the changes.
QUAD v2.zip (8.12 KB)

I have modified my scope to increase the bandwidth as follows…

For ch A.
Remove C73
C11=1.5pF
Across R17 3.3pF
Across R21 22pF
C9=1.5pF
Across R15 3.3pF
Across R19 12pF

For ch B.
Remove C74
C12=1.5pF
Across R18 2.2pF
Across R22 15pF
C10=1.5pF
Across R16 2.2pF
Across R20 12pF

I now get about 10MHz sinewave bandwidth on all ranges and good pulse response with minimal overshoot and rounding. Trigger and good display of sub 100nS pulses is now possible. As delivered, the scope had less than 1MHz B.W. and poor risetime which varied depending on range.

I would suggest the above values are used as a guide, for experimentation, and I hope it may help others get better performance.

Greetings all,
I’ve been monitoring all your comments and investigating your data concerning the Quad bandwidth topic.

I ordered a Quad, but not received it yet.

I have the schematic, component datasheets, and have been using Slimfish’s “Quad.zip” file for analysys. Thanks Slimfish, you saved me the trouble.

This is my $.02

In my mind, the first logical step is the 72Ms/s sample rate issue.
I like lygra’s take on this and the sampling chart.
I personally would want an accurate graphical representation of a signal, otherwise, want’s the point?
My Instek GDS-1062A has a sample rate of 1Gs/s and 60MHz bandwidth.
With a 60MHz signal, that works out to 16 samples per cycle, and that many samples per cycle on the Quad would be for a 4.5MHz signal.
I don’t see the Quad being useful beyond 10MHz.

Now for the OPA2354 opamp issue.
With all range switches open, the amp has a gain of 10, the datasheet shows a small signal bandwidth f-3db = 10MHz at a gain of 10.
Also I noticed the power supplies are 3.6 and -3.0 which exceed the max specified voltage of 5.5V, but not the absolute max rating of 7.5V.
I’m wondering if this will effect the slew rates/bandwidth because operating a transistor close to breakdown voltage increases leakage current possibly causing slower response.

As far as testing an actual Quad, I have a scope and a Parallax Propeller proto-board, there is a demo for synthesizing square waves up to 120MHz

Tina-TI testing confirms a lot of what everyone has been saying.
I plan to modify Slimfish’s circuit to include a scope probe and function generator impedence.
I’m aware that spice programs don’t include parasitic signals and stray capacitance, Spice is for before/after picture, looking for improvement. The designer should know the limits of spice and compensate for it in the real world.

I’ll post more when I have more, Thanks

Nice find on the op-amp supply voltage specifications. The Quad -3V comes from a non-regulated TC1221 which provides -5V light load, and slumps with more current draw. I want to confirm that it is -3V when I get my replacement Quad (supposedly next Monday according to DHL). I will also confirm the 3V6 voltage with a fully charged battery.

Hi,

Bielec, twisted wires as small capacitors… that’s a very neat trick (also an old school one :slight_smile:). Good work!!!

My question (as you are very handy with hw) is whether using small valued resistors could avoid the use of the three compensation capacitors… (as in U17). Other thing is if you have 0603 resistors of proper values…

Unrelated, ADC decoupling is not bad (22uF+1uF), but for high frequency (>3 MHz) one expect also a smaller capacitor (10nF) which would provide the fast transient current for conversion peaks.

Today I attempted to do some Quad measurements (fresh out of the box sys 1.31, app 2.30. fpga?? (I wanted a base-line before I changed anything)) but I ran into a couple of snags:

  1. All my waveform displays appear to be 1/10 of the sine wave input signal. Where do you set the 1x or 10x probe type? I can’t find a menu choice for that in the Quad Manual 0.91b. Is there a 10x probe selection option?

  2. The Auto trigger mode is very flaky. It has trouble finding sync and many times displays garbage when it appears to be synched, but after a few seconds it may lock in and display properly. If it doesn’t lock in properly, then you must move the trigger level out of the signal and then bring it back and try again. Success rate appears to be about 30%

  3. The Normal mode trigger on the other hand appears to lock properly and always displays an appropriate waveform.

  4. The Quad-measured Vpp does seem to track the Quad displayed waveform amplitude properly, but the Quad-measured Vpp appears to be off by a factor of 20, 1/20th of the actual input signal; or to put another way, 1/2 of the 1/10 amplitude observed waveform.

  5. All amplitude and frequency measurements of the input signal are conducted off a TEE’d BNC cable to a Tektronix TDS-210 60Mhz scope with 1GS/s sample rate. The input to the Quad uses an MCX to BNC adapter and BNC cable from the TEE that also feeds the Tektronix scope.

Once I get this 1/10 issue resolved, then I will conduct accurate bandwidth measurements at 1Mhz increments. My initial findings using only the display results shows the roll-off (-3db)to start between 3 and 4 Mhz and at 10Mhz the amplitude falls to 40% of 1/10th of the input signal. Once again, all these measurements were conducted with sine wave signals.

Found my 1/10 problem to be the 10x setting (by mistake) on the Tektronix. :blush:

The test conditions listed below took about 5 hours:

  1. Quad FW = App 2.35, Sys 1.34, FPGA 2.5, used “Normal” trigger mode for sinewaves, “Auto” trigger mode for calibration DC voltage measurements.
  2. Performed front-end compensation for best waveform as per factory procedures.
  3. Used Tektronix TDS-210 60Mhz 1GS/S scope to monitor sinewave Quad inputs.
  4. Used a digital multimeter to measure the DC voltages supplied to the Quad.
  5. Using the same DC calibration voltage as an input results in errors. This indicates that the calibration process is not working properly. When you use the same DC voltage to calibrate and then measure, you would expect consistent results, but not the case with these FW versions.
  6. Used the Quad Measure Vp-p function and display waveform for results measurements.
  7. The 50mv and 0.1v scales have bandwidth issues.
  8. The 0.2v scale provides consistent and flat bandwidth for the tested frequencies. :smiley:
  9. My Heathkit IG-102 RF Generator can’t produce output voltage to properly test the remaining scales.

Conclusions:

  1. Throughout these tests I never felt comfortable with the Normal Trigger, however the Auto trigger was much worse. It got so bad that I would turn off the Nano for each new test because I didn’t trust that the Vpp was always updating properly. Trigger detection and associated measure updates have been measured and found lacking.
  2. Calibration FW and/or procedures still have issues.
  3. Flat bandwidth can be obtained on the 0.2v/Div scale, although the DC accuracy if different from RF accuracy.
  4. Need to repeat these tests using AC coupling.

Test Results:

[code]Calibrated Chnl-A 50mv/Div with 286mvDC

applied 286mvDC to input and quad = 268mvDC, same with waveform, accuracy = 94%

Tektronix 100Khz sinewave @ 156p-p mv, quad = 110p-p mv, display = 2.2 div = 110p-p mv, accuracy = 70%
Tektronix 1Mhz sinewave @ 156p-p mv, quad = 62p-p mv, display = 1.1 div = 62p-p mv, accuracy = 39%
Tektronix 2Mhz sinewave @ 156 p-p mv, quad = 42 p-p mv, display = .85 div = 42p-p mv, accuracy = 28%
Tektronix 3Mhz sinewave @ 156 p-p mv, quad = 34 p-p mv, display = .7 div = 34 p-p mv, accuracy = 22%
Tektronix 5mhz sinewave @ 152 p-p mv, quad = 28 p-p mv, display = .5 div = 28 p-p mv, accuracy = 18%
Tektronix 10Mhz sinewave @ 152 p-p mv, quad = 18 p-p mv, display = 18 p-p mv, accuracy = 12%

Calibrated Chnl-A 0.1v/Div with 559mvDC

applied 559mvDC to input and quad = 492mvDC accuracy = 88%

Tektronix 100Khz sinewave @ 156p-p mv, quad = 112p-p mv, display = 1.1 div = 112p-pmv, accuracy = 70%
Tektronix 1Mhz sinewave @ 156p-p mv, quad = 100p-p mv, display = 1 div = 100p-p mv, accuracy = 64%
Tektronix 2Mhz sinewave @ 156 p-p mv, quad = 76p-p mv, display = .8 div = 76p-p mv, accuracy = 48%
Tektronix 3Mhz sinewave @ 156 p-p mv, quad = 68p-p mv, display = .7 div = 68p-p mv, accuracy = 44%
Tektronix 5mhz sinewave @ 152 p-p mv, quad = 56p-p mv, display = .6 div = 56p-p mv, accuracy = 37%
Tektronix 10Mhz sinewave @ 156 p-p mv, quad = 36p-p mv, display = .4 div = 36p-p mv, accuracy = 23%

Calibrated Chnl-A 0.2v/Div with 1161mvDC

applied 1161mvDC to input and quad = 984mvDC accuracy = 85%

Tektronix 100Khz sinewave @ 156p-p mv, quad = 120p-p mv, display = .6 div = 120p-p mv, accuracy = 77%
Tektronix 1Mhz sinewave @ 156p-p mv, quad = 120p-p mv, display = .6 div = 120p-p mv, accuracy = 77%
Tektronix 2Mhz sinewave @ 156 p-p mv, quad = 120p-p mv, display = .6 div = 120p-p mv, accuracy = 77%
Tektronix 3Mhz sinewave @ 156p-p mv, quad = 120p-p mv, display = .6 div = 120p-p mv, accuracy = 77%
Tektronix 5mhz sinewave @ 156p-p mv, quad = 120p-p mv, display = .6 div = 120p-p mv, accuracy = 77%
Tektronix 10Mhz sinewave @ 156p-p mv, quad = 120p-p mv, display = .6 div = 120p-p mv, accuracy = 77%
[/code]

Attached are pictures of a 3.55Mhz square wave signal from a function generator. All connections are same as previous post, except that a function generator is generating the square wave signal.

  1. The first picture shows the Tek view by itself with both the tee’d BNC cables connected and the Quad MCX adapter disconnected. The rounded edges are due to the stray capacitance of the cabling which does not match the stray capacitance of a standard 1x probe.

  2. Second picture shows the Tek view with Quad connected.

  3. Third picture shows Quad view with both connected. Please note that calibration has not been conducted on the 0.5V scale. the waveform is the subject of concern here, not the amplitude. My original Engineering model Quad presented with a much better waveform for the same condition as this Beta model. Note that this is a multiple sweep display because the camera shutter allows more than one sweep to be viewed in this picture. This accounts for the strange look due to multiple display sweeps.

  4. Fourth picture can be found on the next post, because only 3 attachments are allowed. This picture shows the Quad after being disconnected from Quad MCX adapter and shows the measured remnants from picture #3. This is very disturbing because it is a false measurement in real-time, this is why I suspect all measurements when changing the input signal.

Conclusions:

a. Quad stray capacitance is limiting the input signal and causing Tek measurement errors.
b. Measurement values are not properly updated when signal trigger is lost.
b. Those boxes in picture 3 are due to multiple sweeps. One sweep shows steps, not boxes. What is interesting is that these steps are in the same time domain as the switching noise of the interlace mode. This may indicate that display issues are causing these steps in the interleave mode, and not the interleave mode itself.
tek view 3.55Mhz tek only.jpg
tek view 3.55Mhz both.jpg
quad view 3.55mhz on both.jpg

Picture #4 goes with previous post.
quad view 3.55Mhz no Quad.jpg

Lygra wrote…

“What is interesting is that these steps are in the same time domain as the switching noise of the interlace mode. This may indicate that display issues are causing these steps in the interleave mode, and not the interleave mode itself”

I think these steps are simply the sample steps. I notice you have both chA and chB on so the unit is not interleaving two sets of samples. To see the interleaving effect, you have to “HIDE” chB. Then you will see the sawtooth effect discussed elsewhere and you will notice an apparent doubling of timebase rate. (which I consider advantageous)

Also apparent on jpg 4 is poor risetime and rounding due to the effect of excessive compensation capacitors discussed earlier. After adjusting my values, I acheive a risetime within 2 sample periods for a 3.5MHz waveform of this amplitude. Go on Lygra, get you soldering iron in the back there!