fully differential operational amplifier

[reanimatorstudio]

Assuming you use 0.1% resistors how much do you think the attenuator would degrade the CMRR of the opamp?

copied from a That note I read somewhere: Most active line receivers, including the simple differential amplifier have common-mode input impedances in the 5 kS to 50 kS range, which is inadequate to maintain high CMRR with real-world sources.
With common-mode input impedances of 5 kS, a source 
imbalance of only 1 S, which could arise from normal 
contact and wire resistance variations, can degrade CMRR
 by 60 dB. The solution is bootstrepping.

 

[Twenty Log]

I use the generic rule of thumb for CMRR with component match of CMRR=20*LOG(Tolerance)...  0.1% would be 60dB CMRR for the 10k balanced line resistors...

However, 0.01% tolerance resistors are available for sub dollar each... cheap.... Bummer is that one has to purchase 1000 units to get the price...

They are also film and not metal, so current induced noise performance would be "compromised", but one might ask, how much current goes through a 10k resistor with, say, worst case 15V on the input to make current noise an issue when kRT noise from 10k may be dominant at temperature?  Or with lesser currents under normal operating currents with nominal voltages?  Yes, one can purchase Vishay foil resistors (or foil resistance 3-pin ratios) too for reduced current noise, for tens of dollars each, which may not be out of the question depending on needs of the design....  Have fun with a surround sound attenuator for 22.2 or 23.1

Optionally, and as an exercise in lunacy, 0.01% tolerance parts can be further hand matched to 0.001% (100dB CMRR!?!?! but actually slightly less with the hypothetical mismatch build out resistances in the design calculations here;  and it is probably closer to 93dB with resistors in series and 0.001% tolerance)... Should take about 21 bins for the binning operation to match to 0.1 Ohm (0.001% for 10k)

This would involve a mini metrology lab; not out of the question for DIY, but could take time and a little cash to implement....  A mini oven would be a bonus to check ppm temperature drift...

of course, it is not the actual value that matters... rather it is the ratio of the values in the match that matter and the newer 0.001% match would be superb especially when sets of input resistors or feedback resistors are all pulled from the same bin (regardless of which bin it may be) when stuffing the board... it would certainly preclude the need for trimmers... Trimmers of course may have 200ppm/C drift (or even 100ppm) too, so calibration on the bench with the box open may have a different operating temperature than when in service, in the rack...  I have found by design calculations that 5ppm is really good; 100 ppm is abysmal over temperature range....  Of course if one limited the temperature range of operation in the specs, then it would be fine... say if the range were 1 to 2 degrees

There are some other long-term stability and resistor mechanical issues that also are proposed for implementation in this design to maintain specs over time....

I wonder of commercial units that purport 100dB of CMRR and trimmers and the need to tweak some of them once a year or so... well....  I also think they may use the Birt of the BBC circuit .... just a guess from verbiage in the manual and a rough PCB outline in an older version of said manual highlighting trimmer positions and functions although I have never seen one of their boxes in person.... The Birt of the BBC circuit, I have not analyzed for CMRR fun and recreational math yet...  it is sort of a hybrid between balanced and unbalanced....  Really cool concept/circuit in reality....

Meow... I may have let the kitty out of the bag....

---
Sent from my iPun

 

[Twenty Log]

Yes boot strapping and other "active" line termination techniques have been around...  Seen it in many other industries other than audio.. TI has some obscure app notes that have similar bootstrapping that is not necessarily simple cross coupled outputs (to account for active line termination for CMRR et cetera), but I can't remember... might have been for high speed acquisition or fiber optics....  Good stuff indeed....  I have simulated an experimental CMRR servo and am wondering what it will be like in the real world... might need a respin of copper or two when the time comes...

 

[reanimatorstudio]

All very intersting, but we are getting off the topic: Does anybody have any experience with fully differential operational amplifiers?

 

[dcollins]

And how would you describe it using hyperbole?  ;D

JR

That I liked them.


DC

 

[Twenty Log]

All very intersting, but we are getting off the topic: Does anybody have any experience with fully differential operational amplifiers?

Yes... Input and (simple) Ouput line stages... conversion between bal and unbal....  physical copper with OPA1632 including thermal vias and pad underneath SMT die attach pad with solder stencil on PCB with 1000 parts +/- in 60 square inches or less....  Created custom footprint in Altium for 3D PCB design when board to board mezzanine connectors are sandwiched nearby onto the PCB and clearance issues abound...

> And how would you describe it using hyperbole?  ;D

That I liked them.

俳句 の OPA1632? 
(Haiku no OPA1632 desu ka?)

OPA1632 haiku poetry anyone?

(please forgive my weak Nihongo/Japanese and Romaji translations)

 

[reanimatorstudio]

this one is nice also: http://www.shine7.com/audio/bpre.htm

 

[Twenty Log]

This is an interesting design... 

I personally like to avoid non-inverting buffers, and it usually does not cost me anything to do so (I seem to recall something from my analog VLSI microchip days about slewing issues with non-inverting, but with today's modern opamps, who cares)... I like to avoid them due to alleged and purported "cross over distortion" (different from class A/B output stage cross over distortion perhaps), where operating points for the diff pair input stage internal to the opamp are brought away from their static designed operating points (like say for example putting 5V on the (+) pin of the non-inverting buffer, then the feedback makes the (-) pin also 5V (both of which are not at nominal internal IC bias conditions of 0V as in the inverting buffer/opamp case))....

Of course, "your mileage may vary", and I am sure with low distortion amps in an attenuation configuration like this that it is probably OK...

It does not cost anything to avoid said buffers, but sometimes they just need to be used...

But the trade off between the design submitted above for consideration and this website design is better insertion loss remediation....  And the output FDA has lower resistances (which I wonder of current carrying capabilities, as the output has to carry both feedback current and output current to the next device in the chain --  which probably is not much)...  Yes the lower resistances account for lower noise, but the stuff upstream may have different thoughts...  But in attenuation, should not be bad....

 

[JohnRoberts]

This is an interesting design... 

I personally like to avoid non-inverting buffers, and it usually does not cost me anything to do so (I seem to recall something from my analog VLSI microchip days about slewing issues with non-inverting, but with today's modern opamps, who cares)... I like to avoid them due to alleged and purported "cross over distortion" (different from class A/B output stage cross over distortion perhaps), where operating points for the diff pair input stage internal to the opamp are brought away from their static designed operating points (like say for example putting 5V on the (+) pin of the non-inverting buffer, then the feedback makes the (-) pin also 5V (both of which are not at nominal internal IC bias conditions of 0V as in the inverting buffer/opamp case))....

I don't think this is called crossover distortion. This would worry me far more in the random DOA than with well engineered commercial opamps, but very low voltage gain non-inverting topology will have to eventually deal with input stage operating point. back in the day, the few DOA I made (back when it still made sense to me) were used inverting. 

I also question some of the uber low distortion specifications for the modern uber opamps, that are so clean that they have to measure them at high gain, because this high gain conveniently simplifies input stage operating point issues.

Of course, "your mileage may vary", and I am sure with low distortion amps in an attenuation configuration like this that it is probably OK...

It does not cost anything to avoid said buffers, but sometimes they just need to be used...

That's why we have test equipment... If it messes up you should be able to see measure it.

For line level stages it is possible to use all inverting topologies, albeit with lower input impedance and other issues.

JR

But the trade off between the design submitted above for consideration and this website design is better insertion loss remediation....  And the output FDA has lower resistances (which I wonder of current carrying capabilities, as the output has to carry both feedback current and output current to the next device in the chain --  which probably is not much)...  Yes the lower resistances account for lower noise, but the stuff upstream may have different thoughts...  But in attenuation, should not be bad....

 

[reanimatorstudio]

where operating points for the diff pair input stage internal to the opamp are brought away from their static designed operating points 

Very intersting, never thought of that before. Please teach me more Twenty Log!!

the output FDA has lower resistances (which I wonder of current carrying capabilities, as the output has to carry both feedback current and output current to the next device in the chain --  which probably is not much)

In most "tipical applications" that's what you see, isn't it?

but the stuff upstream may have different thoughts

I agree this design has some points of discussion. But it was just another example.

 

[Twenty Log]

Ah... I meant to say it is "common mode" distortion (where the common mode voltage is deviated from the bias points/zero point in a non-inverting buffer)... not cross over distortion... sheesh.... sorry for the confusion... I was thinking of it being as a cross over point (or inflection point) where one transistor deviates from one operating point versus, say a long tail operating point while the other.... well anyway....

Obviously I don't consider myself an expert, but I have been collecting design tradeoffs and experience for a couple of decades...  I also learn a lot from other folks here....

I dunno what other wisdom to impart for FDAs that may be helpful to what you may want to accomplish...  What are you trying to accomplish with the original question?

What ya buildin' ?

 

[JohnRoberts]

Ah... I meant to say it is "common mode" distortion (where the common mode voltage is deviated from the bias points/zero point in a non-inverting buffer)... not cross over distortion... sheesh.... sorry for the confusion... I was thinking of it being as a cross over point (or inflection point) where one transistor deviates from one operating point versus, say a long tail operating point while the other.... well anyway....


What ya buildin' ?

Thank you, yes that's better. Pretty much understood as a design issue by opamp or discrete circuity designers, and one reason for using current sources instead of resistors to set operating current density for input stage LTPs when significant CM voltages are expected.. 

Worst case at some input voltage the opamp stops working properly and reverses or latches its output. More commonly, input stage operating point shifts cause more subtle errors despite corrective negative feedback.

JR

 

[Twenty Log]

Cheers... I seem to recall the Output Voltage "phase" or polarity reversal in some of the Analog Devices op amp datasheets from the 1990s and early 2000s; unity gain stable and common mode diodes on the input diff pairs for limited differential voltage (perhaps not an issue here in this instance)... They had a nice write up about the phenomena in some of these data sheets... E.g., the OP275 (and other OPxyz) family IIRC....

 

[reanimatorstudio]

I dunno what other wisdom to impart for FDAs that may be helpful to what you may want to accomplish...  What are you trying to accomplish with the original question?

What ya buildin' ?

Well I recently discovered FDA's. I couldn't find a lot of audio examples. So I was wondering if it's the right option anno 2012 for high end stuff.
I'm designing something in between a summing box and a mixing console. Till now I mix on an analog desk and I calibrate the whole analog system (desk, inserts, routing) before I start. I use the console as a big static routing box. We have to recall a mix these days accurately and fast! Automatisation is ITB.
So now I want a personal apparatus that can do all I need for mixing instead of my console. Soundwise it has to be an upgrade.

 

[Twenty Log]

Ah... I see.. I am big into digitally controlled analog, but it is really expensive (sort of), with relays, unless one wishes to get into FET switches instead of relays... Yes there are methods with FET switches that can reduce artifacts, of which I have very limited experience, but it may still pale in comparison to some of the more modern FDAs specifications with distortion measured in ppm and ppb (straight-ish wire with gain)...  Depends on the design goals of course...  Other folks can chime in...

I am into mastering, so a lot less of the "studio math" (where one takes a single channel, at say, several hundred money units, and then multiplies it by number of channels for a full mixing desk, including cable costs, connectors et cetera)...

I also shy away from just being limited to a 24-position hard wired rotary switch for only 24 different settings on one parameter... with software control it can be more like 128 positions/settings with relays on one parameter or...  if the software is modified, up to 16k+ setting positions (although more than likely overkill given the resolution of some of the other components)... yes the 16k+ scenario actually doubles the amount of relays (relays for 128 steps is 7-bit,e.g., 7 relays so 2 X that) for twice the cost (about $40 instead of $20) but the doubled relays are already there for other reasons and the cost of a 24-position Grayhill 2-deck may be around $40 anyway...  There are power supply penalties too for relays of course....

Almost all designs in progress here are FDA balanced throughout and I have been digging into the details for quite some time based on previous experience with differential stuff out in industry...

As you know, the caveats with FDAs might be more precision components on both balanced lines (slightly more cost) as desired for design goals...  I do know of local mastering engineers that actually use unbalanced on purpose for short runs inside their entire signal chain from processor to processor (they made their own processors so they had control over input/output stages, power supplies and everything else from scratch); most processors may be unbalanced anyway internally so it saves many conversion stages... balanced of course being good for long haul / longer distances, rejection, long distances to the summing node (even if pseudo balanced)...

 

[reanimatorstudio]

FDA's as attenuators: www.ti.com/lit/an/slyt336/slyt336.pdf

 

[Twenty Log]

Now that would have been handy a couple of years ago instead of deriving my own for the "gainuator"....  It is more of an industrial application it seems. There are other app notes more apropos I suppose for audio with MFB and a capacitor near the input of the A/D to keep a charge reservoir for the switched inputs of said A/D..

Nice app note though. 

 

[thor.zmt]

http://www.groupdiy.com/index.php?topic=37821.msg466618#msg466618
Although non-FDA amps are shown, an FDA was really there before and can be dropped in.... This circuit is bizarre and can be improved upon/simplified depending on the application....

What I dislike is that there is a lot of excess noise.

Better to use binary weighted series resistors in the NFB loop and select the for the correct gain/attenuation.

Say have 2.2k input resistors.

In unity gain conditions the resistor noise and noise gain cause a noise around -114dBu.

Attenuate by 6dB by switching the NFB to 1.1k and noise drops by ~5dB (reduced resistor noise and reduced noise gain).

Now that the 8 x 10k circuit. In unity gain conditions the resistor noise and noise gain cause a noise around -103dBu.

Attenuate the input by 6dB and noise gain is actually increased to 8dB and maxes out at 12dB.

So the apparent easy adjustment comes at the cost of a lot of excess noise, for 6dB attenuation we are looking at ~ -119dBu for the feedback adjusted circuit and -103dBu for your suggestion.

Even rescaling to 8 x 1.1k merely gets equality at unity gain but is worse at any gain below unity.

Thor