THS3491: Higher Output Voltage Swing

Hi Eric,

your picture didn't come through. Please post again by clicking the "Insert File" button (paperclip) in the header of input box.

Kai

Well this supply bootstrapping approach needs amplifiers that are faster than your desired output signal, 

Cant describe you signal requirements? 

I am trying to achieve settling as fast as possible to low levels, and an output voltage swing of ~+/-24V with a gain of 3. The load is a small capacitance in the pF range. I believe the amplifiers driving the supply rails need to be faster than the main signal amplifier and I may need to reduce the bandwidth of the main signal amplifier by increasing the feedback resistance slightly but that would also increase the settling time. So it doesn't seem like an ideal topology for fast settling, but Id like to see how fast this type of design could work.

Also, with regards to simulating this sort of design, I took a look at the spice model and below is an image showing how the PSRR is handled. This type of bootstrap design seems like it may not simulate well due to the non-typical application, but I am not an expert with the spice models that were put together for this part so I thought it would be good to ask on the forum.

Hello Eric,

As you suspect, the THS3491 model does have issues when trying to create a bootstrap circuit and you will find some trouble when trying to simulate. I believe the DC and AC analysis may work, however you may run across issues when trying to get the transient response. I can provide a few tips that will help with this design as we are working on releasing a reference design similar to what you are proposing. 

- You are correct in that the supply amplifiers will need to be faster or at least have the same bandwidth as the main amplifier. Otherwise there will be a delay in the loop contributing to oscillations in the circuit. 

- There will be a tradeoff between your maximum output voltage and the frequency of your signal. You should be able to achieve a 1MHz signal with an output voltage of 50Vpp.

- The board will need a lot of thermal relief, otherwise you will notice negative performance of the overall circuit.

- I would recommend using Riso resistors on the supply amplifiers of around 3 to 5 ohms when working with the THS3491. This resistor will help maintain the stability of your supply amplifiers, however making it too high will contribute to the slowdown of your amplifiers.

- In addition to Riso resistors it may be beneficial to include a resistor going into the non-inverting input of the amplifier. It may help to add a test points to the supplies of each amplifier. These will help if debugging is needed to test each amplifier separately.

Best,

Hasan Babiker

Hi Hasan,

Thanks for the tips, that all makes sense but I am not too sure about the resistor going to the non-inverting input of the amplifier. The THS3491 has a low input capacitance, but I believe I need to reduce the bias resistor divider values to reduce any delay associated with that RC without loading down the output stage significantly. 

I am interested in a fast step response for my application.

I understand you haven't released it yet, but if you have any preliminary information on work you have done for a bootstrapped design would you be able to share? We can do it over email instead if you don't want to post preliminary information on the forum.

It sounds like making a board to experiment with the configuration would be the best way for me to proceed since I will have issues in the transient response for the simulation model. Any more information on work you have done at TI could potentially be a big time saver for me so I don't have to start from scratch.

You keep saying you need fast step and settling times, but no numbers? Actual targets would help frame the question a lot better, 

One of the basic problems you run into this type of circuit is the high speed amplifier really wants to see low impedance looking out of the supply pins, hence the usual requirement for decoupling caps - if you use those decoupling caps, then the bootstrap supply amplifier has to to drive those - those get into a peak dI/dT issue as well as stability for those amplifiers. If you start with just the amplifier driving the supply pins, it will have a low DC impedance that will look inductive as its LG rolls off, if you try to compensate for that with a cap on the supply pin, then that supply amplifier has to drive that - not an easy problem. 

You will also need to define your cap load for the higher speed amplifier - this kind of thing used to come up occasionally in wide range VCO loops - long time ago, the CLC210 was developed for that purpose - at which point I spent a lot of effort understanding and explaining thermal tail. Also, if this is VCO diode driving app, that Cload varies with voltage. 

Incidentally Eric, I was working quite a lot on a very similar request last June with a guy named Steve Smith - if that rings a bell, you might also check with him for what was going on there. I took what I was doing then a little further into a publication ready solution - but have not yet published it yet. That is mainly about thermal tail correction for a large step using the THS3491. 

Sort of touched on these settling and thermal tail issues recently in this set of articles - we were doing a lot of this kind of work back in the late 80's at Comlinear Corp. The original current feedback amplifier source. 

https://www.planetanalog.com/separating-linear-from-slew-limited-performance-in-amplifiers-part-1-of-2-the-signal-sped-up-insight-14/

https://www.planetanalog.com/separating-linear-from-slew-limited-performance-in-high-speed-amplifiers-part-2-of-2-the-signal-sped-up-insight-15/

https://www.planetanalog.com/slew-limits-create-settling-time-issues-in-high-speed-amplifiers-the-signal-sped-up-insight-16/

Hi Michael,

Yes, I know Steve. Thanks for the article links, that is a good reference for settling time details. Interesting to see the work done on CLC400 and unfortunate information like that is not published with some of the more recent amplifiers especially with the number of 16 -20 bit DACs available now.

It would be interesting to read about your solution for thermal tail correction for a large step using the THS3491 once you publish as well.

 Sorry for the vague specifications, a target would be 50V step with settling to 2mV in 50ns. I think with having an isolation resistor as Hasan mentioned in the range of 3-5 ohms I can get away with close to 1nF on the supply to the amplifier, so I will probably try that as a starting point after running a few simulations. Point taken about the load capacitance, I believe if the capacitance on the supply rails is larger than the load that would provide a good starting point for trying out some settling time experiments, just need to make sure the supply amplifiers can still operate fast driving the additional capacitance as you point out.

After reading your article, my understanding is that limiting the rising edge could also improve settling down at low levels as that would prevent some of the fast slew rate circuitry from turning on, but I can look at that in more detail if this design seems to give promising results.

Good settling data is not easy to get, and few can reproduce it, 

The large step with thermal tail article might publish someday if one of the mags picks up up, none so far - not much interest apparently in that kind of thing anymore, 

1. work your 2mV in 50V back to a #of time constant to target the min allowed main amp bandwidth

2. Calculate the 2nd order step peak dV/dT as 2.85*50V*that F-3dB - this sets your actual response dV/dT peak, make sure amp is faster

3 The bootstrap amps probably can be faster than that, make sure you sim phase margin with any Cload before you get too far, 

Thanks Michael for the information, appreciate the tips. I may be interested in that solution you came up with, and will contact you directly if I want to discuss that specific item further.

Hasan, are you able to provide any further information on the work you have done with the bootstrap configuration? I am curious if you have any more comments on the 3-5 ohm series resistor on the supply rails and their effects on the amplifier performance along with the feedback resistance you used to optimize bandwidth in your configuration. Also, did you find a need to use capacitors on the supply rails or reduce the amplifier bandwidth by increasing the feedback resistance? I am curious because these effects probably wont simulate well as per our previous comments.

I was thinking keep the supply rail impedance lower than 3-5 ohms over the operating bandwidth of the main signal amplifier would be a good starting point, and adding additional capacitance to the supply rail  and compensating for it only if necessary to keep the supply rail impedance low enough, but would be interested to hear your first hand experience with trying it out.

I actually had one more application question with regards to this topology, I noticed the THS3491 did not have any information about unity gain operation on the datasheet.  Should I be able to trust simulation results for this part at unity gain operation, or do you think it will be unstable at this gain regardless? I noticed the recommended value of Rf increases significantly for a gain of 2, especially for the DDA package. Also it was surprising to see that no where in the datasheet they talk about unity gain operation, so I was wondering if that was for a particular reason.

Well Eric, 

When I was defining the targets on the THS3491 one degree of freedom I gave the designer was non-inverting gain of 1 was a bit of a don't care - if you can produce the 9000V/usec slew rate at the input for gain of 1 (that the THS3491 has) then why do you need the device except for load drive - what this meant, was the input headroom is relatively high so your gain of +1 I/O range is input limited - that tradeoff allowed cascoding the current mirrors into the V+ input buffers to improve CMRR. 

It will work at gain of 1, but it will not be flat across frequency (but not unstable) - the very high R's at gain of 2 are apparently brute forcing (compensating) a flatter response for what is probably a peaked intrinsic response - will do better inverting bypassing the buffer across the inputs. Kind of chasing your tail there, as the buffer sees a lighter load, it is probably peaking and using the higher Rf in the overall CFA loop to compensate, less than ideal pulse response when that is going on. 

Not sure the sim model is good enough to show all this, but you could try it. 

Alright, thank you for the detailed response. Could you suggest a feedback resistor value for the best transient response for unity gain operation for trying out on the bench?

Probably should come from Hasan, he has really good test equipment while I have only TINA. 

waveform parametric targets would help - the table you sent is trying to get a flat response, do you need that, you can have a peaked response with ok steps if you drive with a ramp instead of a sharp edge. 

Ok thanks, yes if Hasan could provide that information it would be very helpful. 

A flat response would be nice to minimize overshoot on the supply rails, but it is being driven from the main signal amplifier so if that causes excessive overshoot on the step response we may have to reduce the rise time as you point out.

Also, adding a small capacitance(100p -1n range) on the output after a small series resistor it would ideally still not oscillate with the choice of the feedback resistor at unity gain.

Hello Eric,

- In my case an RGT amplifier was used for the main amplifier and DDA amplifiers were used for the supply. 

- The 3 - 5 ohm resistors at the output were to isolate the output of the THS3491 from the load, in order to maintain stability. Without these resistances, the board was drawing more quiescent current than expected likely due to some oscillation from the supply amplifiers. There were no capacitors added to the supplies of the main amplifier.

- You should be able to determine the stability of the THS3491 using TINA, the issue will lie however in determining the load to provide. 

- In my case 2.2kohms was used for RF in a unity gain configuration. The stability and bandwidth of the supply amplifiers will be determined by the isolation & feedback resistances.

Best,

Hasan Babiker