OPA698: Settling time of OPA698

Hey Steve, perhaps you have the clamps too close together, can you try increasing their limit voltages then attenuating the output back to where you need it? 

Hi Mike,

I have VH at +300mV and VL at -300mV so VH-VL = 600mV datasheet says 400mV minimum so it should be ok. My suspicion is that the performance may not be so good with larger levels of overdrive or if limit values are in the hundreds of millivolts but without data I’m guessing.

I could set limits higher and attenuate as you suggest but to go from +/-2V to +/-300mV would mean adding 16.5dB of attenuation so instead of having a linear gain of 6dB I’d have a linear loss of 10.5dB so would need an external amplifier. I couldn’t put it in front of the OPA698 or I’d be exceeding the common mode input range of +/-3.1V and if I put the amp downstream then the noise would be worse and the amplifier would be saturated on peaks long before the clamps become active.

Do you know anyone within TI who might have the information I need ? Happy to sign an NDA if necessary.

Best,

Steve

Just did some more measurements, with the identical signal source and the in house limiter we've designed using fast op-amps and schottky diodes we get 40nS settling to 60mV. With two of the same op amps at lower gain (so faster and correctly compensated) each driving a cascaded OPA698 with +/-300mV limits and identical small signal gain the settling is 220nS to 60mV. I also notice if I drive the first OPA698 harder and the second one less during limiting this improves to 180nS. There's also a weird step that appears to be caused by the second OPA698 being overdriven by 350mV for 5nS. The pulse delay is around 30nS then it appears to stretch the 5nS pulse to create 30nS plateau. If true these number for low limit voltages are a lot worse than the few nS for +/-2V limit with 2x overdrive presented on the datasheet ! Anyway this points back to needing more information on the part with lower limit voltages. Also ideally the effect of differet overdrive levels.

Not an easy issue, that will have to go back into the product group for support. 

I figured so.  What is the process for TI applications support ? Do you refer, do they ever read the forums ? Or do I have to do something ? I ask because one thing that makes me shy away from using TI parts wherever possible is the lack of support. Competitors offer phone  support and even involve designers or do bench testing to answer tough questions by comparison.

Hello Steve,

I have been reading through your post (and others!) this morning and I am going to check my thoughts on your question with my team today.  I look forward to getting a response to you and help you resolve this behavior you are observing.

On the notion of product support and customer relations, I do understand your frustrations.  I remember during my undergraduate studies I found it difficult to get any engineering support from an array of manufacturers and designers.  The opportunity to be listened to and properly supported is important; this need actually informed my choice of career.  My team and I do read these forum posts and we are ready initiate email or phone conversations if the post can be better discussed via those channels.  

I would like to help you with your issue and hopefully reset some of the existing sentiment you may have developed from other groups and entities when looking for quality support.  If I cannot answer or figure out what you need, I can always ask another team member (or even a designer ).

Please do allow me some time to confer with my team.

Best Regards,

Alec

Hi Alec, thanks for your prompt response. I've been doing EE design for 40 years now so my questions tend to be the tough ones that can't easily be answered from the datasheet   I'd perfectly expect you to need to contact the part design team and so I appreciate this may take a few days. By the way Mike is always helpful if he picks up a question, but other times issues have fallen into the void. Anyway I will look forward to hearing what you can find because even if I could get your part to work with say 5-10nS settling from a 300mV limit with any drive level any gain I'd be happy.

Hi Steve,

A good tough question every so often helps keep the mind fresh and curious.  I would like to ask you for any waveforms or other data you have to help with my analysis.

Best,

Alec

what you really need is the full test setup with source and load, supplies, decoupling etc etc etc. 

Thank you Michael, that would be very helpful indeed.

Alec

Each stage is pretty simple, each part has a 4.7K to 300Ohm divider  to  set the upper/lower reference voltages. Supplies are +/-5V each with 10 Ohm/ferrite bead and 100u/4.7u/10n decoupling. Rf =499 Ohms, Rg=499 Ohms. Av = 2 (non inverting). First stage is loaded with 100 Ohms to virtual earth plus Rf/Rg in series. Second stage ididentical  loaded by just Rf/Rg each of which is 250 Ohms again Av=2. I did try uploading a sketch but your app tells me it is invalid.

Steve,

Thank you for the extra description.  The process of inserting a file or image is rather tricky; I have found it best to use 'Insert -> Image/Video/File -> Upload/Browse -> Select File -> Hit Ok/Submit.

Any inline copy/paste usually causes an error.  This always reminds me of issues copy/pasting between virtual machines...you have to do it a certain way for it to work the best.

Alec

Any high frequency decoupling on the clamp reference dividers? 

Just the +/-5V supply decoupling. The assumption being that this is a high impedance reference input. Also we don't care what's going on during clamping, the issue comes after clamping when the part is back into the linear region and within 60mV of ground.

Well I think we got better recovery with them, but it has only been 20year since I did these, 

Hi Mike, from your drawing you were using +/-2V for the part. The clamp is only active when A*Vin>Tclamp so I'd be surprised but will try with the capacitors added just for thoroughness. I suspect the more likely culprit it the settling with lower clamp voltages so hopefully Alec can find some additional data.

Steve,

I am taking a look at these updates and I will reconnect soon.

Best,

Alec

Hi Michael, So we added the capacitors as suggested but it didn't make much if any difference to the settling time. As I mentioned the other day the clamp voltages would effect performance during clamping so even if the clamp reference input current changed during limiting I'd expect it to be small. What we are seeing is the effect once the part is well out of limiting and into linear operation, looks more like a short (180nS) thermal tail as it approaches zero. So will wait to see if Alec can come up with more information. Best Steve

Thanks Alec, so unfortunately Michael's suggestion wasn't the silver bullet so we will look forward to hearing back from you. So to recap our fixture with discrete limiting now settles to 60mV in 40nS, but the version using the OPA698's set to x2 to do the limiting and proportionately reducing the gain of the amplifiers slows the settling from 40nS to 180nS with +/-300mv clamp voltages and Mike's 0.1uf decoupling capacitors on the clamp reference voltages. So it's definitely looking like the OPA698's are a lot slower at x2 gain +/-300mV clamp and 500% overdrive than x2 gain, +/-2V clamp and 200% overdrive given on the datasheet. But anything you can give me in terms of additional characterization of ways to speed the part up while still being useful would be most helpful. Thanks.

Hi Alec, Fortunately you have a Pspice simulation model for this part so I was able to load it into my LTSpice simulation for the surrounding circuitry. While it doesn't show the 180nS delay that I'm getting with the parts on the board it does give me some important clues. Just driving a single OPA698 driven with a 1nS rise/fall pulse I see a delay from input falling edge to output fully settled of around 20nS this is true whether clamp voltages are +/-2V or +/-300mV. Measuring on the bench I see around 50nS of delay due to the OPA698, but then there's a long settling tail to <6mV which takes over 100nS.  I do notice this is worse when the output of the OPA698 is loaded with a x10 probe so am wondering if the part may be heavily effected by load ? I do have scope shots if you send me your email.

Hi Steve,

I have sent you a message via E2E so we can connect.  That is interesting about the behavior with a x10 probe; I will take a look at those scope shots once I have them.

Alec