fully differential operational amplifier

Does anybody have any experience with fully differential operational amplifiers for other purposes than ADC? Like filtering, buffering, summing, splitting  of balanced audio signals?
exemples of FDA's: LME49724, LMH6553

The differential output seems to limit the applications for these. Perhaps an output driver, but I don't see much benefit in filter or sum bus.

JR

And why? Looks very interesting to me for a summing amp.

Because the function of a summing amp is accumulate multiple signals from multiple distant locations, and subtract the distant  ground reference from the distant signal before summing, not convert a SE signal to differential.

Perhaps you could put one of these in every channel post fader before sending to the master sum, while that seems a bit excessive.

..... but what would I know ?

Go for it, and tell us how it works out.  ;D

JR

reanimatorstudio said:

Does anybody have any experience with fully differential operational amplifiers for other purposes than ADC? Like filtering, buffering, summing, splitting  of balanced audio signals?
exemples of FDA's: LME49724, LMH6553

Click to expand...

I'm using an OPA-1632 as the anti-imaging filter for a D/A design.  All differential.  Seems to work well, very low THD+N, excellent output drive capability.

It is prone to that odd oscillation that can't be seen on the scope, but gives high THD until stabilized.


DC

Using fully differential amps (FDAs) in design for for precision Mid Side encoders and decoders (although for good specsmanship, matched components are desirable but expensive)...  Essentially "analog computer" stuff in the stereo audio domain (L+R, L-R, +6dB, -6dB et cetera)...  FDAs are kinda easy to do any single ended to differential conversion for the "math" operations....  Sorta a universal input....

Fully diff stepped gain/attenuator -108dB to +36dB in 0.25dB steps with 0.01~0.03dB accuracy over full temperature range using only 26 DPDT relays....  T-Networks which can be a double edged sword...

Currently experimenting (in the simulator) for EEQ and bass management filters....  Some nice app notes from TI for FDA filter design and their free FilterPro filter building software has FDAs too (although maybe not applicable for EEQ bass frequency purposeful channel crosstalk type designs)....  TINA-TI, Ti's free simulator has the OPA1632 in it...

Fun things with the VoCM pin can be had...  Funny servos and the like...

Twenty Log said:

Fun things with the VoCM pin can be had...  Funny servos and the like...

Click to expand...

Yes, we are using the Vcom pin as a servo input.


DC

Welcome Dave, it's great to see you here!

ruairioflaherty said:

Welcome Dave, it's great to see you here!

Click to expand...

Thanks Ruairi!  I just recently discovered this board.  Have read over 200 PRR posts, and if he has a fan-club I will be the president.

DC

It's a pitty that there are not more fully differential opamps suitable for audio around; particularly for mic preamps and converters, where there is a fully differential signal path from input to output, they are of great use.

It is prone to that odd oscillation that can't be seen on the scope, but gives high THD until stabilized.

Click to expand...

Indeed; how did you fix it? I've found a RC Zobel network across the outputs very beneficial.

Samuel

Thanx for all the replies. (And yes me too, i've read most of PRR more than once. I think I've even copied some of them in a Word document.)

The most interesting article I could find so far:
http://www.ti.com/general/docs/lit/getliterature.tsp?literatureNumber=sloa064&fileType=pdf&track=no

Good tip the OPA-1632, Dave. Farnell does supply it.

I think impedance and oscillation are the biggest designing concerns. Samuel can you give us some more info on your Zobel network calculations or did you do it in a practical way on your workbench. And is it not better to do some filtering in the feadbackpaths?
Filterpro and TINA-TI does nor exist for mac 

Do you guys share your schematics?

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....  The better the match of the components the better CMRR performance but caveat price... I was originally stuck with only using 10k precision resistors as an exercise...  This circuit should be buffered on both L & R main channels (pre vampiring of the alt jumper cables from board to board) to prevent any insertion loss from the real world build out resistances on practical real-world output stages upstream... The circuit was designed to be generic enough to be stamped/repeated and be channel agnostic such that the main channel is the channel of interest, e.g., L or R and the alt channel was siphoned off of the other board to be the opposite channel of interest in the matrix math; one big cross over set of cables to get the "alt" signals jumped over from each separate M/S board)

There is another M/S design around here that uses the THAT chips precision resistors (0.005% match IIRC), that is also pretty good, perhaps better in some aspects than the circuit I show, maybe worse in others - but is also very clever regardless of my brain droppings...

Also, included is a PDF of the basic gain versus attenuation characteristics for FDAs in a proposed stepped gain/attenuator scheme.... And the same singleton resistor values that bridge between the 10k's on the differential lines give the same figure for attenuation as they do for gain...

the caveat is the T- (or is I-???  H-???) networks in the feedback section for gain... This can hinder performance a little bit with noise / distortion et cetera (e.g., TINA-TI also shows a small uptick in these parameters when used in this configuration)... it is sometimes better just to change the resistances for the FB resistors and skip the whole T-network idea, but this is not practical in this example to maintain CMR etc.... 

But if the same ladder of resistors for different step attenuation/gain values (that _could_ be switch between gain versus attenuation positions with only 1 relay --- ah ha) never gets switched into the gain position, then no issue with noise/distortion perhaps....

For example, an 82k resistor will attenuate by 1dB when placed in the attenuation position, and will give 1dB of gain when placed in the gain position... Same for the resistor value that makes 36dB of gain or attenuation.... (try 160R69)...  The lower in resistance value the more precise the resistor has to be, even over temperature range...

As far as the Mac... yes, one can purchase Parallels or VMware for a virtual machine that runs Windows in a window on the Mac....

Thanks for the schematics Twenty Log. Very interesting. Have you build your Fully Diff Gain Attenuator?

I am about to lay out the PCB... The attenuator/gain circuit is part of a MUCH MUCH larger project that is controlled by a DSP running Linux including a 500 pin +/- FPGA and SPI bus relay controllers so there are other things that need to be finished first....  Hey man; this is mastering....

The by-hand calculations match and were verified in the simulator...  The circuit is pretty straight forward and I don't think there will be any issues; T-Networks are well known... Worst case, a resistor value may change if it drifts too far out of spec over temperature but at 0.03dB worst case, it may not be an issue; the artwork for the PCB would still be the same...

The gain/attenuation formula I get (assuming all balanced line resistors are the same and are 10kOhms in this example):

dB = ABS( 20 * LOG(Resistance/(Resistance+10kOhms)) )

Where "Resistance" is the singleton resistor on the "ladder rung" that provides the gain or attenuation

The simplest way to calculate resistor values for the T-Networks is to solve and anti-log the stuff above with 10^X or "10 with an exponent of X" (where X always seems to be -dB/20 in the equation below):

Assuming that all balanced line resistor values are the same (and in this example are 10k):
Resistance = 10kOhm * ( 10^(-dB/20) / (1 - 10^(-dB/20)) )

Where "dB" is the desired dB for gain or attenuation  (I used -dB meaning that desired dB numbers should have a negative sign because I originally derived the formula when looking for attenuation values, but the same values work for gain as previously mentioned)

So for "4 dB" of either gain or attenuation plugged into the formula it looks like:

Resistance = 10kOhm * ( 10^(-4/20) / (1 - 10^(-4/20)) ) = 17.097kOhms

1dB would be 81.95kOhms for example...

One could make a spreadsheet with this formula and also include temperature drift and tolerances of resistors....  Then the fun part is finding resistors (hint: _up to_ 3 resistors in series for each rung of the ladder with standard value 0.1% tolerance; it can be done)...

Of course, a 24 position rotary switch could be used instead of relays for just a simple attenuator....

A caveat though:

One could use lower valued balanced line resistors, say 3k or so, but there are issues with that...

The "Resistance" on each rung would need to be a little more precise especially for large attenuation values (small resistor values), thusly making it harder to spec resistors...

Also there is the possibility of insertion loss in the gain if there is enough "build out" resistance from a previous source output stage upstream....  simple calculation:

Rf/Ri = gain (inverting).... for a 10k Ri and 10k Rf, gain is 1 (inverting) as expected (= 0dB), however, the output of the prior piece of equipment driving the attenuator has 150 ohms of "build out" resistance on its output... thusly Ri is really 10k + 150 Ohms...  So gain is 10k/10.150k or -0.129dB .. not bad.... now do the calculation with 3k for Ri...  -0.423 dB... "insertion loss" as we called in the RF world....

The good news with the 10k + 10k balanced line resistors with the ladder rung "Resistance" in-between, is that the input single line impedance is somewhere around 20k and is usually greater than 10k depending which attenuation value is present...

-------------------------------------------------------------------------------------------------------------------------
BTW... I'm in, for the PRR fan club... I always learn something from his posts and the many others here...

The "build out" resistance can be avoided buy buffering the input by a simple buffer (voltage folower). But than again this will harm the simplicity of your design.

Indeed a buffer would have the highest input impedance that would render the build out restance to have very tiny effect. Or I suppose the build out resistance could be bypassed with a soldering iron

Yes a buffer would potentially be another stage that could overload when the attenuator already has natural padding before the active stage....

I don't know if I can call it "my" design as it seems to me to be use of commonly and well known circuits (even if they are obscure in our digital age, except on google).... FWIW...

Samuel Groner said:

Indeed; how did you fix it? I've found a RC Zobel network across the outputs very beneficial.

Click to expand...

Just with normal frequency compensation.  I am curious what the actual mechanism is in this part than causes a problem (even with my 500MHz scope and an active probe you can't see oscillation) as somehow an internal operating parameter is upset.

As a sidebar, I have some of the Weiss discrete opamps, and I say without hyperbole that they are the best-performing part in the history of the design arts.  Too expensive to use for the time being, but it's a impressive design to say the least.


DC

dcollins said:

As a sidebar, I have some of the Weiss discrete opamps, and I say without hyperbole that they are the best-performing part in the history of the design arts.  Too expensive to use for the time being, but it's a impressive design to say the least.


DC

Click to expand...

And how would you describe it using hyperbole?  ;D

JR

RC network?  Sounds like an impedance issue at frequency...  What are the specs of this oscillation versus the surrounding feedback circuitry?  RF could be sneaking in (maybe through output which may be a friendlier impedance at RF or through power supply pins et cetera) and purturbing the system's phase margin and some such....could show up at LF when the high RF distorts and turns the amp into a non linear system....

Twenty Log said:

I don't know if I can call it "my" design as it seems to me to be use of commonly and well known circuits (even if they are obscure in our digital age, except on google).... FWIW...

Click to expand...

I like that a lot. I haven't seen anything quite like it. Assuming you use 0.1% resistors how much do you think the attenuator would degrade the CMRR of the opamp?