Hello all!

In the design of a low-speed analog pulse receiver, I  have come across with the need of some sort of "programmable differential voltage reference" for feeding and setting the threshold (trip) point of a dynamic differential comparator. It's the kind of comparator found in pipeline ADCs, in which both the input signal and the comparation threshold voltage are fed in the form of differential voltages, and whose trip point corresponds to the condition:

Vin(=Vin_plus-Vin_minus) = Vref (=Vref_plus-Vref_minus)

Specifically, I need a circuit able to set two voltages "Vref_plus" and "Vref_minus" with a digitally-programmable difference in the [0,400mV] range, and with the added complication that the common mode of these voltages should be as constant as possible, and equal to the common mode of the signal input, 1.2V (this is because the comparator's offset is heavily dependent on the common-mode mismatch between the signal and threshold reference inputs). I am not designing an ADC, and this is why I need to be able to "program" the threshold (trip) point of the dynamic comparator.

Currently I have devised a solution that uses a current mirror with two copy-branches loaded with diode-connected MOS whose Vds sets the two reference voltages. The copy-side transistors are source-degenerated by parallel MOS resistors in a binary-weighted fashion, and in this way I can I vary the current through the copy-branches and thus the reference voltages. With 4 bits I get 16 levels and can cover the [0,400mV] range. The main problem with this solution is that the voltage range is covered in a highly-nonlinear fashion, and the circuit is too sensitive to temperature and mismatch!

I have done an extensive search in the literature, but so far I haven't found anything useful (of course, the point here is not to use a pair of DACs - I am looking forward to a solution of the same order of complexity as the previously described). I also searched for some kind of "programmable differential bandgap" in the literature, but from what I learned about bandgaps now I guess that such a thing cannot really exist!. I really don't need high precision reference voltages, anything that covers the [0,400mV] differential range in 20-30mV steps with a mostly-constant common-mode is fine.

Well, any ideas/references/clues are welcome. Thanks in advance for any help, and sorry for the long post!

Cheers,

Jorge.

Thanks for your reply, Tlaloc!
So basically, you mean building some sort of current-steering DAC based on a stable current source and with matched resistors as loads? Seems a good idea... I'll start working on it right now!

Any other suggestions? I would like to avoid using resistors as much as possible because it's a very size-critical design...

Hi Jorge,

that not really an exotic problem. Almost every ADC is differential and therefore needs a differential voltage reference.
In my opinion, rf-design gave you the best solution. Every bandgap does (or can easily modified) provide a current output. Feed the current into the circuit in the attachment. As rf-design said, use the same resistors (for matching) that are used in the bandgap.
To digitally program the reference voltage you can either program the resistor value by e.g. parallel resistors, or select different resistor tabs.

Regards

Hi Berti! Thanks so much for your detailed explanation. I think I have gotten the idea behind rf-design's suggestions. It would be great if you could please give me your opinion about some doubts that still remain...

I really need to keep the circuit as simple as possible and small as possible because I'll be integrating 8 or 16 of these "differential reference voltage generators" in my application, and all of them should be programmable independently. Thus I was thinking that:

1. The resistive trimming option looks good. I have included an schematic of what I have understood about this option (see figure "Option 1" below-is this what you meant?). With regard to this option...
-How could I actually implement the switches in this circuit such that their ON resistance doesn't interfere with that of the resistors? I am afraid that if I use small devices the ON resistance won't be small enough compared to that of the resistors, and that I might end up having too big resistors to overcome this.
-Do you think the resistors could be implemented with MOS transistors working in triode? for the expected maximum differential voltage their Vds would be 200mV max, much more than their overdrive voltage, so they should be in deep triode... Do you think it would be practical to use MOS resistors instead of actual resistors? This way I could get rid of the switches altogether, but I am unsure about the precision/reliability/nonlinearity issues that might arise...

2. Leaving aside for a moment the first solution, I was thinking that maybe I could use fixed resistors and vary the current by connecting binary-weighted current legs, as shown in the schematic below ("Option 2"). Do you think this approach might work? The benefits would be the resistors are fixed and that the current consumption could be somewhat lower. Do you see any practical shortcommings of this idea?

Well, thanks again for any help & ideas!

Cheers,

Jorge.

P.S.
I discarded the multi-tap resistive ladder idea because of the added logic overhead that it entails (decoders, multiplexers, etc), but  I am really unsure if the area requirements of this solution would be significantly larger than those of the other solutions... do you think this makes sense?

Quote:


I agree that option 1 is probably not the best. I also wouldn't use MOS transistors working in triode because they a) are non-linear and b) don't match with the
resistor in the bandgap.

Quote:


As long as the amplifier loop which regulates the common-mode remains stable, I think that this is a good solution.

Quote:


Often you don't care that much about resistance in series to the reference (because the reference voltage will be buffered). In this case I think that this is still a reasonable solution since you can have moderate switches, and I guess that the decoding logic will be small.

Regards

Since low speed can usually accepted for the common-mode regulation loop I would use a simple single-ended differential pair (tail-current source, diff. pair, current-mirror as load and single-ended output).

Regards

Just for my own two cents worth, I like R-2R ladder trees for this type of application.  Then, your power consumption is independent of your trim code, and the switch resistance is unimportant (with a buffer, which you would have to use anyway).  Also, resistors match much better than current mirrors.  The penalty is the full decoding needed and the amount of resistors.  The decoding should be small; the resistors may not be.