Diff-out instrumentation amplifier?

[jt]

So, I've had the recent pleasure of being tasked to design an analog front end (only happens every few years in my profession) and I was surprised to find that there doesn't seem to be any ASIC differential-output instrumentation amplifiers while working through my part selection trade.

It seems like this wouldn't be an uncommon desirement - it would allow you to keep your analog signal chain fully differential after a high-impedance sensor.  Diff-input capable A/Ds seem to be the norm after all. 

In my googling I came across an article that suggests a circuit technique to achieve this from standard IAs, but again, I'm surprised no one seems to have an ASIC for this.  Is there something I'm misunderstanding about the benefits of keeping a signal chain fully differential?

See page-78 for referenced article:
http://m.eet.com/media/1134731/15496-81904di.pdf

Are there pitfalls to the referenced article topology other than increased complexity?  Perhaps reduced IA bandwidth implications of driving the REF input dynamically? 

[jt]

Thanks EmmanuelFaure,

I had browsed through fully differential amplifiers and had come back with the impression that these were fully differential op-amps, not so much instrumentation amplifiers.  The important distinction being high input impedance. 

To elaborate on this further, if I were to look at the THS4531 device as an example, the typical application note on page 32 shows a diff-in, diff-out amplifier, but requires feedback resistors on the inputs that will be unsuitable for may high-impedance sensors. 
http://www.ti.com/lit/ds/symlink/ths4531.pdf

I guess what I am looking for is a Instrumentation amplifier, per your picture, but with the last opamp stage being fully differential.  I am hoping something exists in an integrated circuit; circuit board space is more of a concern than BOM cost in my application. 

[jt]

I didn't known the input impedance was a concern (Not with this importance I mean).

Yes, I'm sorry I did not elaborate more.  I guess in my mind, high input impedance is the defining characteristic of an instrumentation amplifier, so I didn't think to specify this explicitly.

you havent mention a single spec of your application and you requested ASIC... it can be anything... 

I didn't think to get into specifics, I intended this would be more of a general discussion about diff-out instrumentation amplifiers (and whether or not they exist) and not a request for a specific part number, but if it helps here are a few specs of interest:
1.  >500kHz GBWP
2.  >>1M-ohm input impedance @ 100kHz
3.  Resistor configurable gain up to at least 10
4.  Single supply capable with reference input
5.  input noise PSD to be minimized, PSRR and CMRR to be maximized (I know this is vague, I haven't done the signal chain analysis yet)

[T3sl4co1l]

Yeah, I don't know of fully diff IAs.  You'll probably have to make one with a dual op-amp followed by a FD.  Which really isn't that bad, unless you were expecting like, GHz bandwidth where having everything on chip helps keep things clean and stable.

Tim

[Mechatrommer]

there are but... to meet the spec, one need to browse each datasheet...
http://www.linear.com/parametric/Differential_Output_Amplifiers#!1021_>=100!gbw_100!vsmax_!sr_!ismax_!vos_!enoise_!vosdrift_!lfnoise_!ibias_!inoise_
http://www.analog.com/en/products/amplifiers/adc-drivers/fully-differential-amplifiers.html
the few that i opened only few kiloohm input...

[dom0]

If it is just about being able to represent zero and negative values accurately, there are instrumentation amps with a reference input / output. "Quasi differential."

[jt]

Hey, I found one!  Although it is NRND...

http://www.intersil.com/content/dam/Intersil/documents/isl2/isl28617.pdf

Also another relevant article:
http://www.edn.com/design/analog/4426097/Differential-I-O-low-power-instrumentation-amp

[Marco]

Diff-input capable A/Ds seem to be the norm after all.

Their CMRR isn't necessarily great though, some are and some aren't. Take the superstar ADC LTC2378 for instance, which has relatively poor CMRR. Doesn't matter if it's AC coupled through a decent transformer of course.

Here's an example from Linear on how to do it with discrete opamps (and with a differential ADC with good CMRR).

It makes far more sense to me than the EDN one, where the AD8476 needlessly loses you CMRR. The EET one is cute, but using VREF on a differential amplifier in a feedback loop is a bit of a bodge (and not fully differential obviously).

[tszaboo]

You can use two IAs, just swap the + and - input on the second one, connect reference together. No loss of performance or added offset voltages whatsoever. Matching of the gain resistor is very critical, use a network or matched high quality resistors.

[Marco]

Two inamps obviously works, but it seems worse in pretty much all ways than the Linear approach. The CMRR completely falls to a single stage, the ADC, no matching necessary.

[jt]

Marco, with reference to your comment towards the EDN link: 

the AD8476 needlessly loses you CMRR

Can you elaborate? To my naive eyes, it looks like it is just the classic 3-amp instrumentation amplifier topology, except that last differential amplifier circuit stage is made to be fully differential.

Thanks for the Linear link.  It would be nice if tweaking the gain was just one-resistor replacement, but I suppose swapping an array isn't too much worse. 

[Marco]

Can you elaborate? To my naive eyes, it looks like it is just the classic 3-amp instrumentation amplifier topology, except that last differential amplifier circuit stage is made to be fully differential.

Yes it's fully differential, but the two sets of resistances setting the amplification inside the AD8476 have to track to maintain CMRR (it's significantly worse than for instance the ADC Linear used). That said, as they say if the pre-amplifier has significant gain the CMRR of the AD8476 becomes less relevant.

[C]

Guess I am missing something



A center tap of Rgain should be your common mode signal. Why can you not use this signal to remove more common mode signal from the outputs?

[Marco]

A center tap of Rgain should be your common mode signal. Why can you not use this signal to remove more common mode signal from the outputs?

It's not the task of the differential amplifier to remove common mode, it just has to amplify the differential voltage relative to the common mode (the precision with which the common mode is maintained isn't very critical on the other hand).

[andy1]

I was searching amplifiers for different application and  ISL2863X series would maybe fit yours if the voltage range is compatible.
Also 9 settable gains by pins seem quite nice https://www.intersil.com/en/products/amplifiers-and-buffers/all-amplifiers/amplifiers/ISL28633.html

[suicidaleggroll]

I was about to mention the ISL286xx series, it seems to hit all your bullet points.  I just used the 28635 in a design a few months ago, though it hasn't been fabbed yet.

10 Gohm input impedance, 1.6-2.3 MHz GBP, differential out, etc.

[jt]

Thanks to both to you on the ISL286xx recommendation.  At first glance this could be exactly what I was looking for (although smaller package size would be nice). 

[sarepairman2]

I think that two gain programmable instrumentation amplifiers is an interesting idea. It is the first thing that came to mind.

I think you can keep the very high CMRR the instrumentation amplifier has, the problem goes to summation of gain differences and offset differences.

If you use a chopper stabilized fixed gain of 10 or 100 instrumentation amplifier (2x) you might have a viable solution. The offset will be in the nanovolts and the gain should be matched very well since it is trimmed to 10 at the factory. This of course lowers your op amp choice very much however.

Maybe you can use high resolution DACS (4 of them) to match the offsets and gain of two gain programmable instrumentation amplifiers too. I believe they are just programmed by a resistor which biases some kind of control system inside of them, it might be replaceable with a DAC.

If you use dual output DACS you have thermal stabilization of the biasing of both op amps.

You could use some seriously high speed shit if you did it this way (I believe they have some fairly wide band gain programmable instrumentation amplifiers.) The chopper compensated ones will of course have low bandwidths. And you would not be at the mercy of modern low voltage op amp supply rails 

[Jay_Diddy_B]

Hi,

To the OP.

>>1M-ohm input impedance @ 100kHz 

Means the input capacitance is << 1.6pF

Jay_Diddy_B

[jt]

>>1M-ohm input impedance @ 100kHz 

Means the input capacitance is << 1.6pF

Good point, I guess that requirement was artificially driven by a desire to neglect input impedance - I suppose I should reassess.

[Materiewelle]

Probably not relevant anymore for the author of the original post, but here is a part that performs exactly the requested function, is very high performance, and easy to use:
https://www.analog.com/en/products/ltc6373.html

In case you find this while searching for such a part, like I did :-)

[hanakp]

Just in case anyone was interested, there is a new part ADA4255:

https://www.analog.com/en/products/ada4255.html

It's not just simple amplifier though, it contains overvoltage protection, input multiplexer, gain selectable from 1/16 to 128 and integrated charge pump that allows it to work from single supply. ADA4254 is similar, but instead of the pump, it has current sources for RTD and similar sensors:

https://www.analog.com/en/products/ada4254.html

Apparently, TI didn't want to be left behind:

https://www.ti.com/product/INA851

This one has even larger GBW of 22 MHz and can work with ±15 V supply.

Edit: there is also AD8295:

https://www.analog.com/en/products/ad8295.html

This is much simpler and cheaper IC, its gain is set by a resistor. Its output is not truly differential though, the datasheet shows a trick (figure 64) how to make it with one internal auxiliary opamp. But since the aux opamp feeds the REF pin, I think the resulting phase errors (induced by its propagation delay) should be smaller than the usual inverting scheme (figure 66). The same trick can be done with AD8222, either with second instrumentation amp in the package or an external opamp:

https://www.analog.com/en/products/ad8222.html

Edit 2: another (surprisingly old) parts are PGA280 and PGA281, I missed them because TI has them listed in programmable gain amp section, not instrumentation amp section:

https://www.ti.com/product/PGA280
https://www.ti.com/product/PGA281

Newer but very similar is PGA855:

https://www.ti.com/product/PGA855