Due to the U.S. Thanksgiving holiday, please expect delayed responses during the week of 11/22.

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

THS3092 20Vpkpk Issues

Other Parts Discussed in Thread: THS3092, THS3091, OPA695, LMH6321, LM7171

Hello TI,

I was hoping you would have some tips...

I'm having difficulty getting a clean 20Vpkpk Sine Wave @ 15 MHz out of my THS3092 hooked up in a unity gain buffer (with Rf=1.1K) into a high impedance load (playing with load impedance; somewhere between 1KOhm and 10KOhm).

@1 Kohm load I can get a clean sinewave up to around 16-18Vpkpk. The THS3092 output swing is specced up to ~24Vpkpk over temperature.

This application is a peak detector, so Sinusoidal linearity (Distortion) and voltage dynamic range are important.

I have browsed the forums and seen several similar posts/discussions about this, as well as the THS3091 datasheet which is a bit more thorough.

Some issues/ideas/questions below:

0) Does TI think I should be able to get a clean 20Vpkpk @ 15MHz with the THS3092 into a ~1Kohm-10KOhm load?

1) There are several "THD vs Frequency" plots which show @ 20Vpkpk/15MHz only ~-35dBc; typically into 100Ohms. I would expect this to get better into a higher impedance load (e.g. 1KOhm to 10KOhm) since it would drive down the current load.

2) I tried load-sharing as recommended in THS3091 datasheet but it didn't seem to help (which implies it is not a current-loading problem).

3) This could potentially be thermal... I noticed my THS3092 getting warm and I naively used the SO-8 package. Some handy freeze spray seemed to help a tad. I've got the SO-8-EPAD part on order and will try that when it arrives with some nice copper for heat sinking.

4) Would hooking this into an Inverting topology help for any reason? (i.e. this would help input range, but I don't think it would help the output which is where I believe the issue is).

5) Should I switch to using the THS3091 since it has slightly better performance in THD?  (I was using the THS3092 since I need 2x amplifiers)


  • I tried configuring the EVM as a buffer to drive either a 1MOhm (Hi-Z scope) load, or 50-Ohm load (50-Ohm Scope). (see waveforms)

    I also tried the EVM in gain as -1 and saw no improvement. (see waveforms)

    Your datasheet claims up to ~ +/-12V and -45dB @ 15MHz. I could be wrong but I would think -45dB would be enough to prevent this distortion I am seeing. I will try to load-share the THS3091 but am not expecting improvement as I already tried it with the THS3092...

    GREEN=INPUT

    BLUE=OUTPUT

  • Hello Ernest,
    An inverting configuration should work better since there is no input common-mode swing. I will need to check this out in the lab here and get back to you.I will be trying things out on a package with a power pad because of the large power being burnt.
    -Samir
  • Hi Samir,

    Some even more useful information... I'm out of ideas now. (Inverting topology did not help based on screenshots in prior post).


    1) I tried Load-Sharing with the THS3091 which did not seem to help (i.e. 2x THS3091 in parallel).


    2) I mimicked the test in the THS3091(Figure 72-75). This did work! It seems I can indeed get ~+/-12V when the circuit is configured in a GAIN=5 topology.


    Can you explain why I get better output swing at GAIN=5 as opposed to GAIN=1 @ say 15MHz?


    Can you recommend a way I can get better output swing in GAIN=1 topology? This was the best CFB amplifier I could find for my application.

    Sorry forgot waveforms... Added below

  • Ernest,
    Do you have the ability to check the frequency response of the amplifier in unity gain configuration to see if there is any peaking in the response. When I look at your inverting operation with 1MOhm scope impedance (figure 3 in your 1st email) the amplitude seems to be closer to 24V rather than 20V. This suggests peaking in the response which can then result in slew rate limiting and other undesirable effects. On the same figure the input (green curve) seems to have < 20Vpp input. I find that behavior very strange. The buffer configuration doesn't seem to show this large amplification but does show the distortion. If you have access to a network analyzer you can look at the frequency response to see if there are any anomalies.

    For a G= 5V/V operation, there are a couple of things happening:
    1. For a current feedback amplifier, the loop-gain stability is determined by Rf + Ri*Noise Gain, where Ri is the output resistance of the input buffer. See the OPA695 datasheet, section 9.1.5 for a more detailed description. In a gain of 5 the amplifier tends to have more phase-margin which gives a better response shape. In addition to this in a G=1 configuration, Rf will interact with the input capacitance of the amplifier to produce a pole in the feedback factor (Beta) which further tends to degrade phase margin. So in G=1 operation one would think that by increasing Rf we can improve phase margin (because LG = Rf+Ri*Noise Gain), however as you increase Rf the pole formed by the input capacitance will move to a lower frequency which tends to degrade phase margin.....so you end up being caught between a rock and a hard place.
    2. Another factor that makes G=5V/V operation better is that the input buffer between the non-inverting and inverting terminal will only have to swing 4V as opposed to 20V in the G=1V/V case. This too improves amplifier performance.
    -Samir
  • Hi Samir,

    Indeed the Inverter had some gain (as the waveforms show). I was just trying to see if I could get the output swing of 20Vpkpk. After some more thought I don't think an inverting topology can be used in this application anyways (since I need High-Z input impedance to interact with sensor circuitry), so let's focus on the buffer application...

    I'll try to get a good frequency response plot of my buffer. Again it seems to work fine at lower amplitudes and/or lower frequency, so I may need to do a few different frequency sweeps.

    After an additional survey of other Amplifiers out there, I notice they all suffer from significant THD with large signal (20Vpkpk) around 15MHz. I still continue to think the THS309x is my best bet.

    And unfortunately although the CFB's will be more stable with some gain, gain here results in a reduction of dynamic range of my overall sensor circuit on the low-end (minimum detectable signal).

    Obviously it works perfectly in Spice (waveforms below @ 24Vpkpk), so I assume Spice model doesn't include THD.

  • Hi Ernest,
    Spice unfortunately will not simulate linearity issues. Also, this is quite an old part and the model may not be very accurate so issues like slew rate limitations on the input buffer stage may not be properly modeled. With regards to the frequency response, I suggest a small-signal response (100mVpp) and a large signal (as high as your instrument will allow you or 20Vpp). We should be able to figure things out with those two results.
    -Samir
  • Hi Samir,

    I hooked up the THS3091 EVM to a network analyzer and did indeed see that there was some gain peaking.

    I increased Rf from 1KOhm to 1.78KOhm (from datasheet) to improve the gain peaking, and it certainly improved the gain peaking response.

    However when I look at large signal I still see the same large signal distortion. What's next?

  • Hi Samir,

    Got any other ideas for me to try? (I somehow blew my last THS3091 so I'm ordering a few more; I still have plenty of dual-channel THS3092 though).

  • Hello Ernest,
    Based on the information so far, I think the problem is with the input buffer of the CFA. I just don't think the input buffer has the slew rate needed for a 20Vpp pulse in in unity gain mode. One critical piece of evidence here is looking at the Slew-Rate spec in the datasheet.
    For a 10V pulse the specified gain is 2V/V -> indicating a 5Vpp pulse on the input. The SR under this condition is 5000V/us.
    For a 20V pulse the specified gain is 5V/V -> indicating a 4Vpp pulse on the input. The SR under this condition is 7300V/us.
    The way these specs were defined makes it appear as if the output had better swing capability than the input so as the output swing increased the gain was increased as well.

    Also, if you look at Figure 17 of the datasheet it looks like the SR was only defined for a 5Vpp output in a gain of 1V/V.

    Doing a SR to BW conversion - a 20Vpp, 15MHz sine wave requires around 950V/us of slew-rate. While this is 1/2 the SR spec shown in figure 17, the one thing that is not captured here is the linearity performance degradation as the SR approaches its limit. I believe we are running into such a barrier in your application.
    -Samir
  • Hi Samir,

    I agree with your comments; and obviously the reason I went with the THS3091 was the very high slew rate and wide dynamic range. I thought the THD might be the "non-linearity performance degradation" I was interested in, and the THS3091 performs reasonable well in this aspect.

    Since this issue does not show in simulation, and the linearity degradation is not really shown in the datasheets, it makes it hard to determine if TI has an alternative... Some specc sheets have "Undistorted Output Swing vs Frequency" which can be useful.

    I should add that it really needs to be an Op Amp (ruling out the LMH6321), as the application has an additional slow feedback loop.



    Since it is not working, do you have another part that you can recommend that might perform better?
  • Hello Ernest,
    Can you point me to a datasheet which has "Undistorted Output Swing vs Frequency". What we usually do is to have a distortion vs output swing graph in the datasheet similar to what you described above. The difference is that in the case of a current feedback amplifier, we rarely do a G=1V/V configuration since things tend to get quite ugly because of the dependency of the Phase margin on Rf and the also the effect of the input cap (interacting with Rf).

    The LM7171 may be a good choice for you. It seems to have a 20Vpp swing graph on Page 1 in a G of 2V/V. Since the LM7171 is a voltage feedback amplifier, the unity gain configuration should be easier to tame.
    -Samir
  • Hi Samir,

    The reason I ruled out the LM7171 is because it was not unity gain stable, "The LM7171 is stable for gains as low as +2 or −1" (Pg 1 Description).

    Although since this is a true VFB amplifier with High-Z input Vin-, I could potentially use this as an inverter in my application... Let me dwell on this. I will order some.


    Also the "Undistorted Output Swing vs Frequency" was something LTC had in their datasheets; you are correct :).

    Thanks,
    -Ernie