I've just looked up the price of leaded tantalum bead capacitors; they're expensive. But they were popular in the 1980s because everyone had seen wet electrolytic capacitors drying out and failing, so they wanted something different. By the 1990s, people had seen tantalum bead capacitors failing with a little black hole in the side. So I wonder why the designer chose tantalum? It's not for their performance because they have a high (and variable) ESR, which is why they're recommended for bypassing the output of a linear regulator. That phono stage appeared to have literally hundreds of tantalum bead capacitors inside. At £2 each (yes, that was a typical price from Farnell when I looked a moment ago), there might well be £500 of tantalum capacitors in there. Maybe more. Perhaps he got a bulk discount on a bag of 5000.
Well, it's not behaving direct cancelling but more like increased feedback.
Test Circuit:
In this config, the THD is 0.04% for my given signal (300mV, 100Hz) and the modelled input distortion mechanism (injecting an offset voltage going with output current squared).
Without the compensator, THD increases to 1.4%, more than 36 times worse.
So we have 31dB reduction at low frequencies. The effect reduces at higher frequencies when the opamp Aol reduces.
When the distortion is 10 times lower to begin with, let's say 0.14% (by increasing the /100 scale factor to /1000), the compensation yields 0.000447%, 50dB, so 10 times more reduction than at higher distortion levels.
The penalty is increased noise, or course. And the distortion is re-entrant which means while it is reduced it is becoming more complex, just like with increased feedback.
Bottom line, I would say it is quite effective, assuming the distortion matches well enough.
Speaking of Wayne, he did something clever in the WHAMMY: the opamp is fed its supply voltages through 47 ohm series resistors with 220 uF electrolytics to ground. At 6 kHz the equivalent series resistance starts to swamp the electrolytic reactance of 0.1 ohm, but that's okay since the opamp PSRR is more than adequate to deal with any supply bounce caused by output currents. I bypassed those with 10 uF ceramics I had lying around to keep impedance at a minimum, but that's probably gilding the lily since everyone reports very good results with their WHAMMYs.
The cleverness lies in isolating the regulators from the load capacitance: LM7915s are notoriously finicky about having the exact right output decoupling capacitors, otherwise they start oscillating. Sticking in an isolation resistor is a dashed good way of keeping them stable.
Perhaps I'm missing something but I see no balanced output, unless RCA jacks have a new ring/tip/sleeve standard.
Besides, a balanced output is a trivial thing to add if you don't insist on differential signaling: it can be accomplished by having the same value of components on both outputs, so for example the hot line could have a 100 ohm resistor from the opamp, and the cold line could have a 100 ohm resistor to ground. Since both lines have the same source impedance the line will be balanced against interference (until the output impedance of the opamp becomes significant, of course).
As for the "low noise preamp", using 0.1% 1K33 resistors with 1% 2R2 resistors in the feedback loop means the ultimate accuracy will be 1%, not the hoped-for 0.1%. Then there's the feedback capacitor, which I'd bet is not a 1% unit either. Mark's explanation of the noise reduction is a first approximation: noise isn't canceled per se when adding multiple identical signals, it's that identical signals add linearly so the power (which goes as voltage squared) increases by a factor of four for each doubling, whereas noise, being a random process, sees its power increase by a factor of two, for an effective increase in the signal-to-noise ratio of 3 dB for two sections, 6 dB for four sections, and so on.
Hi, KSTR. Referring to your component designations. Let’s arbitrarily assume that U1 alone has an open-loop amplitude gain of 1,000. For the feedback to increase, the open-loop gain of the composite amplifier must increase to something above 1,000. U2, with its output taken from its inverting input, provides a net unity gain contribution to the open-loop gain of the composite. Meanwhile, divider resistors R7 and R8 reduce the open-loop gain following U1 by a factor of 34. Which would reduce the feedback, and INCREASE the distortion. (1,000 / 34 * 1) = 29.4, almost a 30dB reduction. Perhaps, I’m missing something, however.
Last edited:
Thx, that was smart thinking.
The only thing I have against, is that in practice no op-amp in the world will have 1.4% distortion with 300mV@100Hz input and 30dB gain.
In your example things are hugely magnified but they are proving your point with the rather pessimistic distortion model you used.
But in practice most of the advantage will probably be burried in noise, but anyhow it doesn't do any harm.
Hans
@Ken Newton ,
This is not really a composite in that the slave amp U2 doesn't add any gain to U1.
It's easiest to check circuit operation at DC and the error of each amp being the DC offset, also we assume infinite gain.
When both amps have the same DC offset, this means their non-inverting inputs must be at the same voltage as their inverting inputs are tied together.
Now U1's non-inverting input is connected to the signal and U2's non-inverting input is connected to the output voltage, divided down to the same level as the input voltage.
This in turn means the output voltage cannot show the DC offset anymore.
The voltage gain at U2 does not actually enter the equation at this point.
Now let's apply the gain to U2 which means we find the original DC offset times gain at its output.
And from the finite gain of the amps it follows that the non-inverting inputs don't exactly match and thus some small offset remains at the output.
This is not really a composite in that the slave amp U2 doesn't add any gain to U1.
It's easiest to check circuit operation at DC and the error of each amp being the DC offset, also we assume infinite gain.
When both amps have the same DC offset, this means their non-inverting inputs must be at the same voltage as their inverting inputs are tied together.
Now U1's non-inverting input is connected to the signal and U2's non-inverting input is connected to the output voltage, divided down to the same level as the input voltage.
This in turn means the output voltage cannot show the DC offset anymore.
The voltage gain at U2 does not actually enter the equation at this point.
Now let's apply the gain to U2 which means we find the original DC offset times gain at its output.
And from the finite gain of the amps it follows that the non-inverting inputs don't exactly match and thus some small offset remains at the output.
@mhenschel: Wish I had done so. The pre-regulated PSU version replaced the original PCB board with no regulator and, so, it was easy to compare the two by ears alone. And the difference was marked -- I would guess at least 3dB. But that's just a guess and am now at my winter place that does not have the test equipment -- that lives in my sound studio up in the mountains.
Have just about gotten all the way through the Mend It Mark videos and they are well worth the effort.
Have just about gotten all the way through the Mend It Mark videos and they are well worth the effort.
Well, for these reasons I also modelled the thing with reduced distortion, and the lower the initial distortion the larger the reduction in theory.
I would fully agree though, the net benefit, if any, remains to be seen in the actual circuit.
At any rate, it's an interesting circuit idea I have not seen before.
Back in the early 80's I worked for a consultancy company designing electronics systems. At that point I specified tantalum bead capacitors. But the data book for those recommended 6V capacitors on 5V lines and 16V capacitors on 15V lines. And that we now know is a recipe for long term failure. And if any of my systems is still in use, those tantalum capacitors will for certain be failing.
The space grade specification for tantalum capacitors is that they need at least a 30% headroom; so >20V parts on 15V lines etc.
Craig
I didn't bother to derive any equation but stepping the open-loop DC gain of U2 (Aol2) shows that the final offset scales with 1/Aol2. Aol1 also enters the scene of course. The larger both are, the better, but Aol2 has the way bigger influence, basically the only relevant factor once Aol1 reaches any typical value of >= 10k (80dB).
Parametric plot of output offset vs Aol2 (X-axis), with Aol1 as a parameter:
Well, I wouldn't call this clever (with all due respect to Wayne). What is the use of a regulator if you follow it with an RC?
It's an old Philips trick from the 70-ies but it was quickly abandoned after reasonably good quality regulators became available.
Jan
I often use cheap 78/79/317/337 types of regulators for cost constraints and all of those are not happy with brutally low ESR of a modern polymer electrolytic, let alone large MLCCs. Therefore, when you need series resistance anyway you can just as well use it as the series R of an RC-lowpass. With an R of 1 Ohms or thereabouts, of course, not any bigger than needed.
This reduces reduces load regulation, of course. But there is a trick to make the voltage ripple (notably from class-B load currents appearing) on both rails to run in unison so that any opamp PSRR is converted to CMRR which usually is much better: Use small MLCCs (distributed, but not more 1..10uF total) from either rail to GND and add a big polymer (100x as large) from rail to rail.
Last edited:
The manufacturer recommends a bog-standard electrolytic after one of those regs; its ESR assures reg stability.
So it is beyond me why anyone would use a much more expensive polymer cap there and f*ck up the reg stability.
What are they thinking? Are they thinking??
This thread depresses me. After all these years you'd think people would get it by now.
Jan
So it is beyond me why anyone would use a much more expensive polymer cap there and f*ck up the reg stability.
What are they thinking? Are they thinking??
This thread depresses me. After all these years you'd think people would get it by now.
Jan
Jan, the reason is a very engineering one:
Everything is a trade-off, and there are many ways to get to something that works good enough for low enough BOM cost.
- make any ESR explicit, don't rely on the ill-defined and unstable ESR of old-school standard alu 'lytics.
- use that ESR to help line rejection at RF (which is basically non-existent with those regs with 741'ish cores) which is often more problematic than loosing a bit of DC/LF regulation.
- one polymer and a few MLCCs arent that much more expensive than two standard alus.
- total BOM cost is still low and much lower than with modern regs (notably LT3045 class).
Everything is a trade-off, and there are many ways to get to something that works good enough for low enough BOM cost.
Back in the 90’s when I was designing a pcb at HP we were told to use Kemet fused tantalum smt ecaps on our assemblies. A bit later the cost for tantalum went through the roof so we were told to use mlcc as much as possible.
I further played a bit around with this intriguing topology.
1) I placed the distortion at a more usual place, instead of at the input at the op-amp's output and also changed the formula more aggressively to
V=(100*I(Rout))**2.
Distortion reduction in this case followed a -20dB/dec slope versus frequency.
2) Compared to the above distortion injection I found no difference in the distortion reduction slope when reducing the distortion to
V=(10*I(R0ut))**2 .
Distortion reduction followed here still the same -20dB/dec slope versus frequency.
3) I also changed for a different distortion model, where I gave each op-amp 1% H2 and a 0.5% H3 distortion.
And also the distortion reduction in this case was similar to 1) and 2) wit a -20dB/dec vs frequency slope.
As a result, with all different distortion models A) at the input, B) at the output either C) a square function of output current or D) by adding harmonics, the distortion reduction slopes versus frequency where all similar.
And because of the loop gain going down in frequency, distortion rises accordingly, but at the same the distortion reduction becomes less as a logic consequence.
So with this topology it is as if the loop gains of both amps are added, resulting in a significant distortion reduction at LF that is diminishing towards higher frequencies with -20dB/decade.
That means that when at 100Hz THD is already below target, this topology won't bring any further benefit.
Below are just two versions of distortion reduction slopes with different distortion models.
Only the slopes are interesting in this case, the absolute values are unimportant because the amounts of distortion weren't matched
Hans
- Status
- Not open for further replies.