Post Go back to editing

30 MHz amplifier to drive isolation transformer


I am trying to design a low-noise amplifier with 30 MHz bandwidth that would be able to drive an RF transformer (1:1) to go across an isolation barrier. The source signal is differential and low impedance, low level (mV) with a huge and fast varying common mode (0 to 1kVp in 10-50 us), hence the need for isolation before going to a signal acquisition system.

First I must say I am new to design at these frequencies, I am more familiar with sub-MHz designs... I was thinking about using two ADA4899-1 in non inverting configuration (like the first stage of an instrumentation amplifier, without the difference amplifier 2nd stage) and directly drive the transformer between the outputs of these opamps. I have also considered the ADA4895-1 as an alternative with higher gain. Of course the power supply (dual 5V) is isolated too via a DC/DC converter. I am currently using 10kOhm bias resistors to floating ground on both positive inputs, though it might be too large (?).

However, I am running into a couple of issues. I never managed the ADA4895-1 to be stable (gain 25, Rf=240 Ohm) without adding load capacitors (at least 10nF behind a 24 Ohm series output resistor) and feedback capacitors (typ. 4.7pF). Is that expected behavior ? The bandwidth flatness is also very far from my expectations above 1 Mhz, although the transformer is supposed to be flat (for 50 Ohm:50 Ohm) between 10kHz to 20MHz at least. I am really getting resonances (plus noise peaking) above 4 MHz.

So first, is the chosen design usable, or should I consider something completely different ?

Thanks for any help !

Top Replies

  • FormerMember


  For the png you attached, the op amps see a common mode gain of +1, so with a decompensated op amp that is stable at a gain of 10 or greater, you will always have oscillations.

Is this a pc board…

  • Hi,

    Do you happen to have a specific part number for the transformer? Additionally, are you actually loading the secondary of your transformer? If not, you're only presenting the magnetizing inductance to the two amplifiers. In that case, you have setup an LC resonator tank with the 22pF caps.

    Additionally, how much signal swing are you intending to get out of the amplifiers? Supposing that each could hit the full rail, you would present a differential voltage of 10V into 100ohm (assuming terminated output of the secondary) results in a peak current of 100mA. This exceeds the linear output current for the parts (though it is less than short-circuit peak current).

    I would like to suggest to try using a 50ohm resistor where the primary winding is now (removing the transformer) and see how well the design behaves. If things begin to behave, then that would lead me to place further scrutiny on the transformer portion of the design.



  • David,

    Thanks for your answer, and sorry for slow reaction - I have been very busy on that circuit.

    I haven't been accurate in my description, true, I am indeed loading the transformer output with 50 Ohm. And yes, even before plugging the transformer, I had a first try with a single 100 Ohm load instead of primary. That was definitely unstable at gain 25, which is not quite I would have expected from the specifications. On input, there is a 50 Ohm impedance too. The level was rather low, like 20 mVrms in, which is 500 mV out, far from rails.

    Anyway, I have noticed some needed changes that improve the behavior :

    - bias current is huge (11 uA) but quite nicely matched (offset typ 20 nA). I do not care about small DC offsets, so did not bother about impedance matching - how wrong ! It seems to be very touchy on this point, and matching both inputs of the opamp strongly reduced the observed oscillation...

    - even with matched input impedance, and with a resistive load only, I haven't been able to get stable output whatever I put in the feedback. So I had to add a small RC pole on the load (like 10Ohm 10pF in series). I am a bit surprised, as this is never mentioned anywhere in datasheet, nor from simulations.

    I also gave up with differential input configuration, and turned back to single-ended with a single amplifier - not absolutely sure it was needed, but I considered I had to go back to simpler configuration at some point, to have it under control.

    With a small capacitive load and matched inputs, I get decent behavior from gain 10 to 25. But I still have some less than optimal behavior just below cutoff, and have the feeling it could still be improved. I am a bit disappointed too about PSRR, I have pairs of bypass caps (100nF 0805 and 10nF 0603), still I keep several mV on output ripple that come from my supply ripple of 20mV at 1MHz - where has the expected 60dB PSRR (for 1 MHz) gone ?

    Anyway, if anyone has some experience to share with this opamp I would be glad to here it, in the meantime I will satisfy myself with what I got...

  • Thank you Harry, this is just the kind of answer I was sort of expecting.... I had the feeling this differential architecture was ill-formed, but could not understand exactly how/where/why !

    Is this what you are suggesting (See PNG) ? I have just copy-pasted nearby components, the 10Ohm and 4.7pF values are random and probably have to be tuned - right ? Would this remove the need for the capacitive load ? However I have found a small capacitive load to be necessary even with a very basic single ended / single opamp design (Rf // Cf and Rg only) - is that expected ?

    My PCB has 2 layers and already tons of modifications made in last couple of days , I am already thinking about version 2.

    Thanks for other post on PSRR too, I'll have a careful look...

  • FormerMember
    0 FormerMember


  For the png you attached, the op amps see a common mode gain of +1, so with a decompensated op amp that is stable at a gain of 10 or greater, you will always have oscillations.

Is this a pc board?  how many layers?  how easy to make changes?

You need to split the 10 ohm resistor and add a series RC to ground to fool the op amps.


  • FormerMember
    0 FormerMember


      With respect to the PSRR, It might be sneaking in from anywhere.  See figures 10 and 11 in the article(s)

    The dual op amp used had -120 dB channel seperation, but the customer measured -60 dB.

    The problem was a common ground return.

    Look at every single ground on every component and visually where the currents go.

    Are there any common paths?


  • FormerMember
    0 FormerMember


      Yes, but your noise gain is still a little shy.  drop the 20 ohm to 10-15 ohms.

    Also, my feeling is you are doomed with two layers.  You need a minimum of four

    and I would start with six personally.

      The only paper I have seen is the one by Rod White, which is almost impossible to find.

    Our IEEE subscription is for in house use only, so I can't send it to you, but it is reference three

    in MS-2178 on our web site.


  • This question has been assumed as answered either offline via email or with a multi-part answer. This question has now been closed out. If you have an inquiry related to this topic please post a new question in the applicable product forum.

    Thank you,
    EZ Admin