Calcs for op amp loading vs THD?

[thor.zmt]

@thor.zmt this is really impressive. I went “wow, -115!” a few posts back and “these concepts should get put together,” and then you did it. This is publishable research.

Well, to me this is mostly "old hats".

Trying to avoid asking any dumb questions here. I don’t have a strong background in discretes, so i’m having a hard time following the way the noise servos work without all the values. Are they just not showing up?

I deliberately hid them. Many people these days are not good with anything not involving triangles with +/-/Out. I do not want to provide ready designs to chinese kopy katz. Anyway, anyone who cannot calculate these should do some remedial...

The noise servo uses a Sziklai or "Compound Feedback Pair". It has also been called "inverted darlington".

In effect the second transistor increases the transconductance of the first one.

The first transistor runs at ~0.5mA collector current and thus has ~ 52 Ohm emitter resistance.
If we ignore base resistance and source resistance this determines the transconductance of the first transistor. A change of 1mV will produce a change of 19uA.

The second transistor runs at ~1.5mA collector current and thus has ~17 Ohm emitter resistance. It also has as a result of a fairly high beta an input impedance of around 5kOhm and the base emitter resistor is 1.2kOhm.

In parallel we get almost exactly 1k, so 19uA change in 1kOhm give 19mV input voltage which, when divided by the 17 Ohm emitter resistance gives around 1mA current change. The circuit also forms a near constant current sink for the first transistor, which now operates at (almost) constant current.

So the "open loop" transconductance of this compound transistor is 1mA/1mV or 1A/!V or 1 Siemens. This means als that the "virtual" emitter impedance of the transistor is 1 Ohm.

Adding a further resistor (in the schematic R8/R19) reduces transconductance and linearises the Transconductance. Bu using a cascode transistor (T5) the whole compound pair also operates at (almost) constant voltage, making it more linear again.

At the same time also the signals from the two separate noise (reduction) amplifiers (one for each Op-Amp) are summed.

The collector of T5 now has a current that is modulated in accordance to the base voltage variations of the first transistor in the two noise amplifiers and it has very high impedance (many MOhm). The circuit is inherently inverting.

The input to the circuit is the inverting input of the Op-Amp.

We are being told that this is a "virtual ground". That is a first order oversimplification that is only true for a "super-ideal" Op-Amp with infinite gain, bandwidth and zero ohm output impedance.

In fact, with a real Op-Amp, even assuming absolutely identical resistors in the inverter, the output will never be exactly Vin * -1. Noise and distortion will make the output different. If so, the noise and distortion must appear on the inverting input. And it is of course amplified, giving the 6dB noise gain of the inverting circuit.

Our Noise (reduction) amplifier simply takes this noise and distortion as it's input signal and converts it into a current of inverse polarity with a certain fixed factor of conversion (~ 62mA/V here).

As said, the current eventually exist the collector of T5 which has a 2.7kOhm resistor to the negative rail to provide 4mA supply current. There is also a 5.1kOhm resistor (and a DC blocker capacitor) to the output, which we can see as 50 Ohm to ground (without signal).

The inverted noise current is now applied to this resistive network (Kirchoff remedial recommended) and produces a voltage across the 5.1k & 2.7k resistors. This voltage in turn causes a current flow in the 5.1kOhm resistor that also enters the load and more crucially the two 100 Ohm build out resistors on the Op-Amp's.

If we have all gains correct and just so, the inverted noise current through the 5.1k resistor produced a noise voltage across the 100 Ohm resistors that when summed with the noise voltage of the Op-Amp nulls out. The same should incidentally happen to any harmonic distortion.

Of course, the Noise (reduction) Amp has it's own noise and distortion, so nulling is imperfect.

The concept is also know as "Error Takeoff" and has been in public domain since the 80's.
If the error "taken off" is not inverted and mixed into the actual inverters output, but instead is buffered, scaled and then mixed into the non-inverting output (or rather cross-coupled between two inverting circuits with a balanced input we get the (not) world famous Jikoda Circuit, named after the japanese "just in time" concept "jidoka" (じどうか)

1. Detect the abnormality.
2. Correct the immediate abnormality "just in time".

This is "feed forward" error correction of the residual error from the feedback circuit to theoretically allow a noise- and distortion-less circuit.

Do I understand correctly that the same concept could be applied to any inverter, and that values are pretty widely scalable as long as the injection ratio matches the Sziklai pair gain?

Essentially yes. But the noise (reduction) amplifier needs "design in".

If the noise sample comes from the inverting input of a diff amp that is being relied upon for CMR,

It does not really work. For that go FDA and noise-reduce both sections or go jikoda.

I saw that you spec’d your original transistor choices and then they went away and am wondering what the factors are in parts choice as you figure this out.

Protecting trade secrets. The input transistor must offer low noise and high beta, not many options that are readily available exist. The second stage needs even higher beta and still low noise.

It would be pretty cool to see this architecture surrounding a solitary LME49724 in a diff-to-diff context.

Yes, that would be a good application.

Not a big difference between 2.1nV/rtHz and the 1.1nV/rtHz of the OPA1612, and it’s on the cheap side. Also note that the 1.1nV/rtHz (!) OPA1633 is in preview; there must be a model available.

Sure, but I ask again, why not just go impedance balanced for line outs? And why not use a discrete line driver that can be made to run on +/-24V, with minimal distortion and noise at a fraction of the cost of a comparable Op-Amp in a can?

Never mind the bragging rights of "+/-24V Class A discrete line driver" in my Gizmo?

Just saying.

Or take a 5534, hang a class P-P stage of Pin 5 with a bootstrapped resistor pair to +V to give the second VAS Transistor a few mA extra current to increase transconductance.

And let's replace the input pair by a Toshiba dual J-Fet with ~1.5nV|/Hz (~ 2nV |/Hz for the full differential pair, 27 cent US if buying 3,000pcs reels from Mouser) and up the input current and second stage differential current a good deal.

I think that is much more fun, never mind performance.

Thor

 

[atavacron]

there’s this now, too, at $1.60/1ku

 

[atavacron]

posting some resources here to read later, for anyone else playing catch-up. (edit: also i think it’s jidoka, from a series of hifi amps, but haven’t yet found the circuit being referenced.)

 

[Admiral-dk]

You can use some very basic "rule of thumb".

Assuming no current or voltage limits, once loading exceeds the limit for Class A Operation, every halving of load impedance will appx. double HD (all harmonics).

So if you have an Op-Amp rated to have current enough to drive +22dBU (10V) into 600 Ohm with reserve (> 50mA short circuit current recommended) but is only rated for 2kOhm for THD, the 600 Ohm THD will be around 3 times or 10dB higher.



Simple, a FDA with differential loading will cancel even harmonics, mainly H2. So you get less THD.

At the same time you get twice the voltage, which if driving the same load will double the current and quadruple the power which will in effect half the load seen by the individual Output.

So distortion and H2/4/6/8... will be around doubled with the output level also doubled, presuming enough current capacity.

On the other hand, if driving the same voltage into the same load with an FDA each half will see half the impedance load but also half the voltage, so HD will be neutral.

Now note all the above is "rule of thumb" and not reliable, it will work most of the time.

Past that...

TI now has "dedicated" drop in replacements for the common audio op-amp's that are fully characterised with 600 Ohm loads at a competitive price, I see no reason not to use OPA1678 (USD 0.25 in 1Ku -> NE5532, TLx2 etc.) and OPA1679 (USD 0.4 in 1kU -> TLx4 etc.) for most positions instead of the "common audio op-amp's".

For specific applications that need lower noise or higher current are best accommodated the same way they are if using the common Op-Amp's, using external discrete components. Looking at Rupert Neve's later IC based designs is a good source for relevant ideas.

Thor

I have used a lot of OPA1678 and they are really great - BUT ...!!!!.... Unavailable since the Chip Crisis - I have a notify when back in Stock, on my TI account - and no news in more than 18 month (and I do check from time to time too)

 

[Newmarket]

Good stuff. But let's bare in mind that this is s a "DIY" forum - so unit prices based on reels of thousands of pieces aren't particularly useful.
Reminds me of a meeting with the pro audio division of Sony in the UK. It was basically "Don't worry about op amp costs - it's pennies for us. And resistors are essentially free".
The wonders of economies of scale.

 

[atavacron]

unit prices based on reels of thousands of pieces aren't particularly useful.

agreed. but “per 1ku” is the only apples-to-apples price comparison that’s reliable, so it gives you a scalable reference point for relative cost in small qty

 

[thor.zmt]

posting some resources here to read later, for anyone else playing catch-up. (edit: also i think it’s jidoka, from a series of hifi amps, but haven’t yet found the circuit being referenced.)

Sorry, yes, jidoka. My lysdexia strikes again. Or may my auto-correct. Or both.

Thor

 

[thor.zmt]

I have used a lot of OPA1678 and they are really great - BUT ...!!!!.... Unavailable since the Chip Crisis - I have a notify when back in Stock, on my TI account - and no news in more than 18 month (and I do check from time to time too)

Yes, it is time to dust off those old analog skills and go discrete with Jelly-beans.

Thor

 

[thor.zmt]

Yes, it is time to dust off those old analog skills and go discrete with Jelly-beans.

Thor

So, in 2023, if I had to suggest a "balanced line driver", what would I do?


Something like this. All transistors are "jelly beans" and SMD with many Duals. The double differential input uses low noise jelly beans to get in effect around 1nV|/Hz noise.

Feedback loop uses 1.2k X 2, so yes, like a "balanced" output this gives 6dB gain and noise gain. As this OPA is not "Rail-to-Rail" anything operation at less than 6dB non-inverting gain is not recommended (in reality it's more like 2dB gain needed, but hey).

Driving 12.86V (due to the buildout resistors) or +24dBu into 600 Ohm with -110dB THD solely H2 and -119.5dBu noise at that for 143.5dB theoretical dynamic range.

Yes, the single line driver (plus PSU) consumes 50mA @ +/-30V with half that in the output stage quiescent current.

So f....g what? It's Rock'n'roll!



Thor

 

[atavacron]

Wow, I just looked at a _lot_ of data sheets to see if any IC op amp exists that can do 24dBu into any load at reasonably low noise on any supply voltage. I was reminded that the ADA4625-1 is rail-to-rail, and is a 36V part, but slew rate induced distortion kicks in around 22.5dBu if I read correctly. It appears that the part I was thinking of is the OPA192/2192/4192, which is a not-too-shabby CMOS amp with 5.5nV/rtHz noise at 1kHz. Noise ramp below that is pretty similar to the OPA1655/1656, which people clearly find acceptable for audio (I’m not a fan). Seems like it will do at least 24dBu, BUT only into a high impedance…so as a line driver it would need a discrete current amp. The BUF634A maxes out at ~2V from each rail, or ~23.3dBu on dual 18V supplies.

 

[thor.zmt]

Wow, I just looked at a _lot_ of data sheets to see if any IC op amp exists that can do 24dBu into any load at reasonably low noise on any supply voltage.

Yes, there are. Many high drive, high voltage op-amp's exist and we could always bootstrap the rails and use that.

Like take a TPA6120, bootstrap the rails to +/- 30V and disable one channel. Remember to run it at least with 6dB gain to avoid exceeding common mode range and go into phase-inversion [ask me one day how I know ;-) ]

Can add OPA1611 (no PSU Bootstrap) for low noise and very low distortion.

With 0.7A peak current and 26V peak output any load can be driven. And with similar noise and lower THD than my example.

BUT, current draw will still be substantial and compared to my discrete made from modern SMD versions of 2N4401/4403, BC547C/557C and BD139/140 it costs a lot more and complexity is only marginally lower.

So why not discrete jelly-beans that never go out of fashion?

Thor

 

[thor.zmt]

The BUF634A maxes out at ~2V from each rail, on dual 18V supplies.

Don't forget the overhead for 100 ohm worth of build-out resistors.

That adds 1/6 of the +24dBu, so at the amp output you need 15V RMS or almost +26dBu.

And something I forgot to mention (obviousness).

Adding a Jensen JT-11-BMCF output transformer (replace build-out resitors with a single Jensen Isolator or a 1uH PCB inductor (air core, spiral) in parallel with 10 Ohm delivers a +26dBu / 600 Ohm capable transformer balanced output. At a price.

Thor

 

[atavacron]

I’m sorry people, but I am confused and need to ask my original question again. Consider this line driver. It’s a +/-14dB differential trim, but I’ve simplified it here to just show minimum feedback current, ie a gain of 5.



In the 600 ohm load scenario, ignoring cable length:

- U1 differentially drives 5K, 5K, 5K, 5K, and 678R. That’s ~440R.

- 39.6Vpp into 440R is ~90mA

Is the next step this?

- Each side of U1 is responsible for ~45mA, swings 19.8Vpp and thus has to dissipate 892mW, so the package has to dissipate 1.78W

Please let me know if this is the right way to do differential load calcs in a way that defines each op amp’s responsibility. I have tried looking at this in a way that treats each side of the signal as its own single ended circuit (ie U1A drives 5K, 5K, and 339R), and it just seems like an inordinately heavy load considering the capabilities of, say, an LME49724 or OPA1632.

If anyone is interested, U1 is an OPA2210, each side of which looks like it can source and sink 45mA at 85deg C with a 20Vpp signal. I would assume that comes with a distortion increase from what is shown on the data sheet for a 600R load (19.8Vpp / 600R = 33mA), but as you all know it’s exceedingly rare to run heavily limited dense program material through a line driver, and the main concern is typically what the peaks of a snare or kick or slappy bass look like at that level into that load. I guess the next step would be to look at the dissipation capabilities of the SON package, as it has a heatsink on the bottom that bonds to a V- pad on the PCB, and you can make that pad pretty big. I’m not gonna bother looking at SOIC here.

But regardless of what amp this is and whether or not it’s integrated or discrete, I just need to get my math right. Thanks in advance…

 

[abbey road d enfer]

Each side of U1 must deliver 90mA. Kirchoff theorem.

I read too fast. Didn't check your figure of 39.6Vpp
Actually half of it.

 

[atavacron]

Each side of U1 must deliver 90mA. Kirchoff theorem.

Yikes. I’ll look it up, cheers. Well, guess I’m looking for the right quad now…

 

[atavacron]

Each side of U1 must deliver 90mA. Kirchoff theorem.

@abbey road d enfer I’m still confused, thus am forced to ask some dumb questions.

What’s the math?

With the Kirchoff reference I assume you’re saying “the load must be supplied, so in a push pull situation, each side supplies the current alternately.” Sí o no?

Why are all of the differential line drivers that claim at least 50Vpp into 600R universally rated at 70-85mA into a short? THAT, DRV, LME. Even if only at room temp, there’s no way they’re doing that cleanly.

 

[Dan Mills]

With ~40V pp across the line (I have not checked that this is +25dBu), so 20V peak across the line, into 600R your peak current is 33mA (Ohms law) into the load, and each opamp is hitting 10V or -10V at peak (Because two driving opposite polarities) at 33mA. You do NOT need to deliver 90mA to drive +25dBm into 600R.

For 50V pp, you are looking at 25V/600R = 41.6mA, so a 75 or 80mA short circuit current looks reasonably sane assuming the current limit is a reasonably hard knee.

 

[JohnRoberts]

@abbey road d enfer I’m still confused, thus am forced to ask some dumb questions.

What’s the math?

With the Kirchoff reference I assume you’re saying “the load must be supplied, so in a push pull situation, each side supplies the current alternately.” Sí o no?

Kirchhoff's circuit laws - Wikipedia Kirchoff's circuit laws are pretty much common sense. The total current entering a node will equal the current leaving a node.

This law, also called Kirchhoff's first law, or Kirchhoff's junction rule, states that, for any node (junction) in an electrical circuit, the sum of currents flowing into that node is equal to the sum of currents flowing out of that node; or equivalently:

Why are all of the differential line drivers that claim at least 50Vpp into 600R universally rated at 70-85mA into a short? THAT, DRV, LME. Even if only at room temp, there’s no way they’re doing that cleanly.

Don't have to drive into a short cleanly... the current is to support driving rated voltage into nominal loads. Current limiting keeps the magic smoke inside the parts.

JR

 

[atavacron]

With ~40V pp across the line (I have not checked that this is +25dBu), so 20V peak across the line, into 600R your peak current is 33mA (Ohms law) into the load, and each opamp is hitting 10V or -10V at peak (Because two driving opposite polarities) at 33mA. You do NOT need to deliver 90mA to drive +25dBm into 600R.

This is what I have previously assumed (take the total current and split it), thus my confusion at both Thor and D’s assessment of the situation.

For 50V pp, you are looking at 25V/600R = 41.6mA, so a 75 or 80mA short circuit current looks reasonably sane assuming the current limit is a reasonably hard knee.

I’m guessing you’re talking about each side here, but the 70-85mA was referencing total output. Examples below are those.

———

We know that the closest to short circuit current one can expect (even with a real hard knee) is not very close, from looking at any source/sink vs Vpp out graph. And at anything but room temp, forget it. (JR, i’m agreeing with you)

If the LME49724 can put out 52Vpp into 600R (86.7mA), and its short circuit current is 85mA, is it just a false claim or is there some bit of math I’m missing? If the THAT 1646/1606 can put out 51Vpp into 600R (85mA), how is its short circuit current 70mA? The DRV134/135 claims 48Vpp into 600R, with a short circuit current of 85mA, so the required 80mA is at least technically possible, but unlikely.

I would love to see someone’s spreadsheet of measured current per side of any differential output made from a dual op amp feeding heavy loads, and/or a canned line driver or FDA doing the same. Surely one has been posted in years past here.

 

[abbey road d enfer]

If the LME49724 can put out 52Vpp into 600R (86.7mA), and its short circuit current is 85mA, is it just a false claim or is there some bit of math I’m missing?

You're forgetting the re-entrant protection scheme. You must read about it.