(variable gain) Single ended do differential amp with DC servo

[sdouble]

Hi folks,
I am about to start working on a front end board for a digitizer.
I'll get to the board with a 50 ohms terminated (single ended) signal, including a modest offset (unknown, from -1 and +1V).
I would like cancel (servo) the DC offset, and amplify the signal (gain 1 to 5) and convert to differential (sampling adc input).
input signal is analog, 10 to 100 MHz bandwidth.
I am not really financially constraint but i'm lacking space on the board.
Would you have, dear expert, any circuit topology suggestion ?

[jbb]

How about good old-fashioned AC coupling? What’s the lowest frequency of interest?

[David Hess]

Are you asking about why differential inputs use a DC servo instead of AC coupling at the input to remove the DC component?

The reason is that it is difficult to implement AC coupling at the differential input without compromising the low frequency common mode rejection because this relies on the two time constants being closely matched.  The solution is to take feedback from an integrator driven by the output signal and apply it to the reference input of the differential amplifier or difference or instrumentation amplifier.

[sdouble]

AC coupling is not an option. It would introduce a pole. This is not something I can afford

[David Hess]

AC coupling is not an option. It would introduce a pole. This is not something I can afford

So does a DC servo.

[sdouble]

we are talking about this kind of circuit , right ?
(here single ended output)

[jbb]

David: thanks for the comment of LF CMRR, that could save my ass one day. I think sdouble mentioned a single ended input, so not a killer in this case.

Sdouble: yes, that sort of thing. The DC servo integrator has a low-pass characteristic; when you wire that into a feedback loop I think you effectively end up with a high pass filter. What’s the lowest frequency you’re interested in?

You could instead do a fixed DC offset with an analog subtraction and reference DAC / pot. This is basically what the ‘position’ knob on an oscilloscope does. Of course, it will need adjusting from time to time.

What supply rails do you already have on hand? And how do you need to control the gain? Variable, basic high/low switch?

[sdouble]

THanks jbb for your feedback,
Anything below 100 Hz is not so crucial for me.
On the other side of the spectrum, I need a 3dB cutoff higher than 100 MHz.
THe servo DC seemed to me a good option since I'll have to treat about 1k channels with different offset. S no way to adjust the DC offset individually.
I'll be able to generate the rails I need. I had in mind +-5 V since the max absolute value ot the offset is 1 V and the sampling ADC will get differential inputs of [-1.25:1.25] V  range.
Noise immunity is clearly crucial here. I'm dealing with a 16 Bits ADC (almost 14 effective bits) and the lower end of the ADC range is crucial for my application.
I had in mind the use of a dip switch to toggle from high to low gain (12 dB to 0 dB). Not sure that I can ensure the 100 MHz bandwidth with a dipswitch. the parasitic cap may be a showstopper.

David: thanks for the comment of LF CMRR, that could save my ass one day. I think sdouble mentioned a single ended input, so not a killer in this case.

Sdouble: yes, that sort of thing. The DC servo integrator has a low-pass characteristic; when you wire that into a feedback loop I think you effectively end up with a high pass filter. What’s the lowest frequency you’re interested in?

You could instead do a fixed DC offset with an analog subtraction and reference DAC / pot. This is basically what the ‘position’ knob on an oscilloscope does. Of course, it will need adjusting from time to time.

What supply rails do you already have on hand? And how do you need to control the gain? Variable, basic high/low switch?

[jbb]

1k channels... that’s a lot! I see why you’re concerned about space. Do you need to sample all of them at the same time? Could you use RF MUX chips to select channels before you get to the amplifier / ADC stage? If not your looking at hundreds of Watts for the amplifiers and ADCs, which could be a big pain in the backside.

If you don’t much care below 100 Hz then it sounds like AC coupling is the way to go. You’ll need a suitable coupling capacitor and should stay away from X7R etc. dielectrics (they are nonlinear and change capacitance with DC offset) so maybe film or C0G / NP0 ceramic? (Others of the forum may have experience here.)

You mention a 16 bit ADC over a 100 MHz bandwidth. Have you done a system noise budget? Do you need to process the whole 100MHz at once, or does your downstream application only look at portions of the 100 MHz band?

For gain switching: it’s a bit annoying, but is ‘solder in the right feedback resistor’ an option?

[sdouble]

I was not precise enough.
A total of 1k channels need to be deployed but each board will host 8 channels only.
Typical use would of of 10 to 20 boards but a total of 120 boards will be produced.
AC coupling requires a beefy cap to ensure such a cutoff frequency below 100 Hz, not sure that a NP0 could do the job.
I mentionned, I think that I'll sample at 500 MS/s but a subsequent digital treatment of the digitalized signal will be applied. Depending on the application, we could integrate just a couple of samples... or many (ie thousands) more.
Depending on the use, I may have to toggle between the 2 gain options (12 dB and 0 dB). A jumper might be an option too.

1k channels... that’s a lot! I see why you’re concerned about space. Do you need to sample all of them at the same time? Could you use RF MUX chips to select channels before you get to the amplifier / ADC stage? If not your looking at hundreds of Watts for the amplifiers and ADCs, which could be a big pain in the backside.

If you don’t much care below 100 Hz then it sounds like AC coupling is the way to go. You’ll need a suitable coupling capacitor and should stay away from X7R etc. dielectrics (they are nonlinear and change capacitance with DC offset) so maybe film or C0G / NP0 ceramic? (Others of the forum may have experience here.)

You mention a 16 bit ADC over a 100 MHz bandwidth. Have you done a system noise budget? Do you need to process the whole 100MHz at once, or does your downstream application only look at portions of the 100 MHz band?

For gain switching: it’s a bit annoying, but is ‘solder in the right feedback resistor’ an option?

[Marco]

Hard to solve everything with one IC ... high input impedance single ended to differential amplifiers such as OPA862/ADA4941/lt6350 could just make due with a small blocking capacitor, but don't have the bandwidth, while FDA's have low input impedance.

I would use an opamp as a buffer with a small blocking capacitor (can be small because of high input impedance). Then either a discrete single ended to differential amplifier or a FDA based one.

[David Hess]

David: thanks for the comment of LF CMRR, that could save my ass one day. I think sdouble mentioned a single ended input, so not a killer in this case.

Sometimes a differential input amplifier is used with a singled ended input to remove the effects of ground loops as much as possible.