| xiphmont |
Hi folks,
As a sanity check before I go wiring things up... This is discussed alot, but I've not seen the final solution except in rough sketch. I want my details to be correct.
I've purchased LM3886 gainclone kits from chipamp.com. I want isolated/balanced inputs on these guys (they're going into self-powered monitors, thus some potentially long cable runs) and the Cinemag CMLI-15/15B appears ideal for this task.
The LM3886 reference sheet application schematic and the schematic used by the actual amp kit differ slightly. I understand all the differences but one. Neglecting all but the input/output components:
Reference application schematic:
Schematic for the gainclone kit:
The negative input feedback transistors differ because the gainclone kit is setting a higher gain. The extra output Rz/Cz damps output oscillation at high frequency (is it necessary in my application? The drivers will be literally a foot away). The only addition I don't understand is the 22k R2-- I don't see what it is doing aside from slightly attenuating the input and lowering the input impedance.
In any case, my stab at a completely correct balanced input using the Cinemag is:
The cinemag wants a 10k load on its secondary, which the 10k potentiometer (which *will* be there) provides. I saw no point to the 22k R2 and left it out. The faraday shield of the Cinemag should be tied to signal ground-- should the Cinemag case be tied to signal ground or the chassis ground? The Cinemag application notes suggest chassis ground, but the Jensen site states that is wrong and an input transformer's case should tie to signal ground....
Anything else I may have missed? |
|
|
| AndrewT |
Hi,
do you need to go with the transformer isolated balanced input?
R3 could be returned to source and achieve the balanced input.
Rane's paper on how to ground is the more reliable. There was a newer version posted here in the last day or so.
Look up the schematics for balanced opamps.
You do need very good matching between components on the inverting to non-inverting inputs.
Have you looked at BWRX version of the buffered chipamp?
Have you calculated or measured the change in output offset as the volume control is adjusted?
Keep the output Zobel.
Insert RF attenuation. |
|
|
| juergenk |
| quote: | | The only addition I don't understand is the 22k R2-- I don't see what it is doing aside from slightly attenuating the input and lowering the input impedance. |
R2 provides a ground path in case the pots wiper lift off
some people may find it unnecessary, I find it a wise thing to do :) |
|
|
| juergenk |
ok, I admit, this would be my intention adding the resistor
the designer may have thought otherwise |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
do you need to go with the transformer isolated balanced input?
R3 could be returned to source and achieve the balanced input.
|
Not quite true. You get a differential input that way, but it is not balanced. The impedence cannot be the same on the two inputs as the output is affecting the negative input.
Regardless, the transformer gives me isolation [and a potential ground lift], which I do want.
| quote: | Originally posted by AndrewT
Hi,
Rane's paper on how to ground is the more reliable. There was a newer version posted here in the last day or so.
Look up the schematics for balanced opamps.
You do need very good matching between components on the inverting to non-inverting inputs.
Have you looked at BWRX version of the buffered chipamp?
Have you calculated or measured the change in output offset as the volume control is adjusted?
Keep the output Zobel.
Insert RF attenuation. |
I'll look for Rane's paper. I had not seen the BWRX; I'll go look for that too.
I don't see why the output offset would change; the feedback gain is not being altered, only the input attenuation. (I know, I left out the offset cap on the negative feedback loop for simplicity, I will probably actually use it in the built circuit). I would think the output offset voltage is dependent on the input offset voltage, not the input offset current...
As for Zobel, as I do more reading, yes, it should stay.
| quote: | Originally posted by juergenk
R2 provides a ground path in case the pots wiper lift off
some people may find it unnecessary, I find it a wise thing to do :) |
Ahhh! Yes, *that* makes perfect sense. Thanks! I'd think something a little bigger than 22k though... |
|
|
| xiphmont |
| quote: | Originally posted by juergenk
ok, I admit, this would be my intention adding the resistor
the designer may have thought otherwise |
It's still a great reason. :-) Certainly good enough to have a resistor there. |
|
|
| xiphmont |
Sorry for the doublepost... I'm under moderation as a new user, so I can't edit the previous one.
| quote: | Originally posted by AndrewT
Hi,
Rane's paper on how to ground is the more reliable. There was a newer version posted here in the last day or so.
|
I had thought you meant a user here named 'Rane', not the website :-) I've read the Rane notes before and went back to review. They address the suboptimal case in great detail-- what to do if you need to shoehorn input transformers on externally as part of interconnect cabling-- but do not cover at all where to ground the input transformer case (nor do their transformer examples even have Faraday shields...)
Furthermore, the Jensen docs are not consistent about where the input transformer case gets grounded. I'm sure there is a reason why the case shield keeps getting moved around (sometimes chassis, sometimes signal, sometimes input shield) but they don't explain so it's not possible to generalize the thinking.
They are explicit and consistent about where the Faraday shields should go (signal ground of that side of the circuit).
The Cinemag docs are entirely consistent-- the input transformer case is grounded to chassis ground. I'm just not sure that's right and that's why I ask. |
|
|
| AndrewT |
Hi,
output offset depends on stage gain, input terminal source resistance and input offset current.
In the circuit you posted, the input terminal resistance is variable and this leads directly to variable output offset.
Post the wrong schematic and you'll get wrong replies. |
|
|
| BWRX |
| xiphmont, you may want to consider using the LM3386 as a differential amp with the transformer winding connected to the differential amp inputs. You would probably want to load down the transformer with a 10-15k resistor (or two 5-7.5k resistors in series with their center connected to signal ground) and use input resistors around 1k. Then you can set the feedback resistors to get your desired gain. Both of the input and both of the output resistors would need to be matched well (preferably within 0.1%). |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
output offset depends on stage gain, input terminal source resistance and input offset current.
|
Ah, so output offset depends on the input offset current *not* the input offset voltage? The chip docs are not clear on that point.
I was under the impression that, like with most other opamps, the input offset voltage is somewhat constant, entirely determines the output offset (along with the feedabck circuit) and any input offset current is merely an annoyance, eg, when taking an input from an insanely high impedence source in microamperage circuits.
| quote: | Originally posted by AndrewT
In the circuit you posted, the input terminal resistance is variable and this leads directly to variable output offset.
Post the wrong schematic and you'll get wrong replies. |
Fair enough. Howver, what you say about input offset current would apply whether or not the optional Ci capacior is there or not, right? The cap just removes stage gain from the equation.
| quote: | Originally posted by BWRX
xiphmont, you may want to consider using the LM3386 as a differential amp with the transformer winding connected to the differential amp inputs. You would probably want to load down the transformer with a 10-15k resistor (or two 5-7.5k resistors in series with their center connected to signal ground) and use input resistors around 1k. Then you can set the feedback resistors to get your desired gain. Both of the input and both of the output resistors would need to be matched well (preferably within 0.1%). |
I certainly will consider this, but I don't think I've ever actually seen a diagram of this sort of setup (got any URL pointers?) How does it escape from feedback influencing impedence on the inputs unequally? Well matched resistors alone don't change the fact that the negative input's impedence varies significantly input/output differential state. |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
output offset depends on stage gain, input terminal source resistance and input offset current.
In the circuit you posted, the input terminal resistance is variable and this leads directly to variable output offset.
|
Ok, reviewing my input offsets...
The op amp is sourcing small currents from both inputs (the bias currents). In an ideal opamp, these would be zero, but no op amp is ideal. If the input stages were perfectly matched, the bias currents would be identical. They're not. That difference is the offset current.
The offset voltage is not a first-order phenomena, but arises from the imbalance in the offset currents and impedences looking out of the op-amp inputs. The currents flowing out of the inputs have to drop to ground somewhere and the difference in the voltage drops is the input offset voltage. Biasing the inputs with trimmable imedances could cancel out the offset, but anything from temperature change to age of components will pull things back out of trim.
In any case, the important thing is that the inputs act like [very small] current sources; the bias current is fairly constant. The input offset current does not directly cause the output offset voltage, except when viewed through the input stage impedences when looking out of the op-amp inputs.
(Long-winded way of saying AndewT is correct and I was confused)
Anyway, eliminating the voltage offset a) Will never entirely happen without a servo and b) is just not that important anyway unless it's large. Because the input stage impedences we're talking about are relatively small (<15k), voltage offset is unlikely to be a concern even if it does vary slightly when the volume knob is turned (after all, all of NatSemi's own literature doesn't particularly care....)
That also answers the earlier question about R2; it's cant be that large or the volume knob wiper lifting would see a sudden large increase in DC offset.
As for where to ground the input transformer case, I guess I'll just have to measure the CMRR both ways meself. It should be reasonably easy to set up. |
|
|
| AndrewT |
Hi,
an amp configured for a DC gain of 30 will multiply the input offset by the same 30.
Swing the input resistance from zero to 50k with an input offset current of 3uA and the output offset could swing as much as 4.5V.
One could choose an input stage with an input offset current around 0.3uA and Zin=20k and DC gain=1. The output offset=6mV. What is the guaranteed input offset current of a chipamp? are they designed to be low? are they designed to remain stable? Will the chip temperature affect the current? Chipamps test the DC offset severely simply because they are integrated.
If one fits DC blocking caps to both the input and the NFB lower leg then the change in output offset DUE to changing the volume control can be eliminated.
The input offset voltage at the non-inverting pin is balanced by an equal offset voltage at the inverting pin when the output offset is set/trimmed to zero. It's that opamp rule again: the difference in the voltages at the input pins is amplified to appear at the output. If the difference is zero then the output is zero. |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
an amp configured for a DC gain of 30 will multiply the input offset by the same 30.
Swing the input resistance from zero to 50k with an input offset current of 3uA and the output offset could swing as much as 4.5V.
|
Specifically, we're talking about a swing of 1k to 6k (through the pot and transformer) and a .2uA typical bias current for a total output offset swing of 32mV. Worst case input bias for a 3886 at the edge of operating range and manufacturing tolerances is 1uA, so 160mV if you have a borderline out of spec unit being driven at the limits of its range.
The National application notes suggest an 'optional' DC blocking cap on the NFB leg that eliminates stage gain from the offset so all that remains is the very small input swing. That I didn't show in my original schematic because it doesn't appear to be strictly necessary (although it might be a good idea). It seems most chipamp builders don't bother with the cap. But let's crunch those numbers because the above is only the swing and doesn't account for the input bias on the negative input and not the absolute output offset values.
In the chipamp.com circuit gain is configured to 30dB (32x) and Ri is 680 ohms. The input bias current voltage drop on the negative input will be a worst-case .68mV and a typical case .136 mV. The worst case largest drop on the positive input will be 6mV. The typical case smallest drop will be .2mV. Thus the output DC offset is guaranteed to be in the range of -15.4mV to 188mV without Ci. That should account for the 10k volume pot, transformer, variations in 3886, temperature, the works. The only thing that doesn't account for is resistor tolerance. Typical case, the output offset without Ci will be in the range of 2mV to 34mV.
So, the very largest magnitude output offset I'd suffer with the circuit I originally suggested (the third schematic) would be 188mV; more likely closer to 34mV. With the 'optional' Ci on the NFB lower leg, the worst case maximum offset magnitude would be just under 6mV worst case and just over 1mV typical case.
I think we can also declare that it is solidly not worth decoupling the positive input from the volume knob.
I chugged though the math out loud to spot obvious errors. We've been violently agreeing with each other for a while I think. |
|
|
| AndrewT |
Hi,
our numbers only differ by a factor of a hundred or so. Violent agreement?
And what if the next source you plug in has a DC blocking cap on it's output? or you inadvertently fit a source that has no DC block and zero output offset?
And that example of the wiper lifting due to age/dirt/pitting? |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
our numbers only differ by a factor of a hundred or so. Violent agreement?
|
Our numbers differ because you stated hypothetical values that did not match the circuit I'm working on. I had assumed you stated them as a cautionary example for me to run numbers from my specific circuit. Your caution was heeded, and I did run those numbers. Every number above was from the spec sheets for the parts sitting on the desk in front of me.
| quote: | Originally posted by AndrewT
And what if the next source you plug in has a DC blocking cap on it's output?
or you inadvertently fit a source that has no DC block and zero output offset?
|
My input is and always will be transformer isolated. As in 'permanently mounted inside the case, you will need a screwdriver and wire cutters to plug in 'the next source' without it'. So this scenario will never happen.
| quote: | Originally posted by AndrewT
And that example of the wiper lifting due to age/dirt/pitting? |
That's what R2 is for, which I will be using. I left it out of the original schematic and asked "I don't know what this is for, most designs leave it out." Both you and juergenk cued me into its importance (the wiper lifts and suddenly you need a path to ground to avoid huge DC offset). I already said more than once it will be there.
"You need to do this."
"Yes, I agree, I will do it."
"No, you don't understand, you need to do this."
"Uhh... OK, yeah, I see that now. I'll do it."
"Listen to me! It is vital that you do this."
"Ummm....." |
|
|
| AndrewT |
Hi,
we are almost at the end.
using the 10k pot which presents a range of resistance between 0k0 and 2k5 (=10k/4) and 1uA (maximum @25degC) the swing in offset will be only 32*1*10^-6*2.5*10^3=80mV, until the wiper lifts.
The maximum resistance on the input pin varies from 1k0//22k to 3k5//22k compared to the inverting pin with 680r//22k. It is likely that the offset will always be in one direction and change from just (a few) mV to 80+ (a few)mV. But, the extreme offset case is the infinite (or 22k) resistance if that wiper lifts. What happens to your speaker cones if the wiper makes intermitant contact as you adjust volume? (704mV should not be a killer, but it will be loud)
I used a passive into a conventional power amp and came to no harm. That was before I came the realise the implications. I have recently lost another bass/mid. It appeared to be OK when I switched off but was immoveable when I switched on. Don't know what to blame other than listening too loud, or could it have been something like what we are discussing?
What if one were to move the pot in front of the transformer and wire it as a variable resistor across the balanced inputs? |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
What happens to your speaker cones if the wiper makes intermitant contact as you adjust volume? (704mV should not be a killer, but it will be loud)
|
I would say that R2 is intended as a failsafe, not normal operating procedure. If the wiper lifts normally, that's a bad potentiometer.
As for a pot that is dying from age, OK, hmm, the crackle wouldn't be of the same magnitude as the input, it would be the full offset. Oh, I have another thought...
| quote: | Originally posted by AndrewT
I used a passive into a conventional power amp and came to no harm. That was before I came the realise the implications. I have recently lost another bass/mid. It appeared to be OK when I switched off but was immoveable when I switched on. Don't know what to blame other than listening too loud, or could it have been something like what we are discussing?
|
The LM3886 Application Notes address startup and shutdown transients. In short, speaker protection is built into the 3886 (it mutes on power on and power off). The 1k on the positive input is there to protect the input-side from spikes in input bias current when the device is powering up and down. It's probably there entirely for the sake of a non-isolated input and protecting the source, actually... I bet R1 could be outright replaced with a good decoupling cap and R2 changed to a very low value such as 1k!
Yes, I *definitely* need to think about that idea some more. I seem to recall seeing something similar in an alternate application note. R1-> .1uF decouple, R2-> 1k-ish...
FWIW, drivers do die. They don't have an infinite lifetime. All coils want to explode, and eventually some do (as a sound reinforcement guy for live theater, I can state that actors can find a way to break *anything*. They just have to *emote* at it...)
| quote: | Originally posted by AndrewT
What if one were to move the pot in front of the transformer and wire it as a variable resistor across the balanced inputs? |
Hrm, I don't like where this idea is going, but let's run with it for now...
Directly across, so that the input impedence changes? It would need to be exponential, not logarithmic. And at high attenuation, the input impedence would be impractically low.... So I guess that's not what you meant.
The only way that seems like it might work is in series with the positive input... Taking the impedence the source sees from 15-ish K to infinity.... That seems like it would mess with the function of the transformer and alter the sonic behavior of the source as it will see varying input impedence. The LM3886 will not suffer any sonic consequences from a slight swing in output offset and, taking your 80mV offset figure, that's a steady state of slightly less than a milliwatt being 'wasted' in the speaker. |
|
|
| peufeu |
If you want low offset, be sure that the + and - inputs see the same DC resistance. For the - input you have both feedback resistors in //.
For the + input it depends on the Pot setting. So, you put the Pot on the other side of the transformer, wired as a balanced mode shunt volume control, and add a proper resistor. No offset ! |
|
|
| xiphmont |
| quote: | Originally posted by peufeu
If you want low offset, be sure that the + and - inputs see the same DC resistance. For the - input you have both feedback resistors in //.
|
That ignores offset current, you still need a trimmer. ...until temperature variations, etc, get it all out of trim again. It's not the offset we care about, it's the sweep of the offset if the pot lifts. Eliminating the offset for the sake of eliminating the offset is a red herring. Simply the fact that it's there affects nothing. However an aging/noisy potentiometer would introduce noise into the input far in excess of the actual input program energy (the scratchiness will always be the same volume, whether the input is quiet or loud) in this config.
These are monitors and will have the attenuators on the back (not for constant fiddling) so I'm not sure I care, but it's worth at least thinking about.
| quote: | Originally posted by peufeu
For the + input it depends on the Pot setting. So, you put the Pot on the other side of the transformer, wired as a balanced mode shunt volume control, and add a proper resistor. No offset ! |
Doesn't a 'shunt volume control' severly stress sources by making the input impedence *very* low on highly attenuated volume settings? Is it actually simpler/better than just replacing R1 with a decoupling capacitor (and balancing +/i inputs by approriate choice or R2)?
Perhaps I'm misunderstanding what's being proposed. The only hits Google is finding me are in marketspeak feature lists.
[edit: finally found it. Yeaaaah, something that whacks around input impedence doesn't sound like a superior solution to just decoupling the pot to me...] |
|
|
| BWRX |
| quote: | Originally posted by xiphmont
I certainly will consider this, but I don't think I've ever actually seen a diagram of this sort of setup (got any URL pointers?) How does it escape from feedback influencing impedence on the inputs unequally? |
It doesn't escape from feedback influencing the input impedances unequally. In fact, the two input impedances can't be matched. But this configuration does what you were looking to do - convert an unbalanced or balanced signal to a balanced signal in a relatively simple way,
This is where the three op amp instrumentation amp is worth mentioning. Adding two op amps as a differential buffer in front of the differential amp ensures not only a low source impedance to drive the inputs of the differential amp but also matched (and really high) input impedances because both inputs are the positive input terminals of the differential buffer op amps.
Here is a schematic of the instrumentation amp in this post: http://www.diyaudio.com/forums/show...141#post1049141
I had suggested replacing IC1 with the input transformer and loading resistors but you can keep the differential buffer and add the input transformer and loading resistors in front of it too. That configuration would have excellent common mode rejection. |
|
|
| xiphmont |
| quote: | Originally posted by BWRX
It doesn't escape from feedback influencing the input impedances unequally. In fact, the two input impedances can't be matched. But this configuration does what you were looking to do - convert an unbalanced or balanced signal to a balanced signal in a relatively simple way,
|
Ah, OK. That's what I thought. Much like the ESP solution: "Not strictly correct, but no more practically flawed than most aspects of any given real-world design".
For my purposes, a $38 high-end transformer also counts as 'simple change' in that it adds only one passive. $38! I had originally completely discounted even looking at transformers because I had expected them to be $300+!
| quote: | Originally posted by BWRX
Here is a schematic of the instrumentation amp in this post: http://www.diyaudio.com/forums/show...141#post1049141
I had suggested replacing IC1 with the input transformer and loading resistors but you can keep the differential buffer and add the input transformer and loading resistors in front of it too. That configuration would have excellent common mode rejection. |
Thank you much for the pointer. I had in fact seen the thread before, but didn't know that was the one being referred to. I originally thought 'BWRX' was an amp design as opposed to a username. |
|
|
| xiphmont |
| quote: | Originally posted by gootee
Maybe something like Figure 6-33 would be good, which is a line transformer buffered by a simple differential line receiver. Otherwise, maybe Figure 6-30. Figure 6-26 is also interesting.
|
Oh, 6-33 *is* tantalizing... in that diagram, could U1 simply be the 3886, or does that run into the same problem of having the output influencing the impedence on the negative input? Or does it not matter because both sides are being driven by the transformer?
[edit: I meant 6-33 not 6-30] |
|
|
| BWRX |
| quote: | Originally posted by xiphmont
Oh, 6-33 *is* tantalizing... |
That's exactly the configuration I was describing back in post #9. And yes, U1 would be the LM3886 with external feedback resistors. |
|
|
| xiphmont |
| quote: | Originally posted by BWRX
That's exactly the configuration I was describing back in post #9. |
I see that now :-| :-(
Does the differential (not quite balanced) input actually give me anything here? The single-ended version (that is, the balancing transformer is feeding the 3886 as a single-ended amp) is inside the case with the input transformer an inch away from the + input. It is only single ended for that distance.
It seems like the differential version (transformer feeding differential inputs rather than + and siggnd) is just going to require better balanced resistors (ie, be a pain).
The single-ended gainclone kit doesn't bother to balance the branches. The input side is 1k/22k and the output 680/22k! The reference sheet's 'Typical Application' doesn't even pretend to set up a symmetrical positive input (R2 is missing entirely). So... for my purposes... it looks like single-ended non-differential is actually better. The single-ended does not care about balance at all (or is that just because there's no point in caring-- it's that much worse?)
The original point of a balanced input was CMNR to the box, not pursing an ideal of balanced signal path all the way to the chip. So unless I screw up with wire/component placement, the single ended 3886 configuration should be just as good as the differential, yes?
[edited for more thinking]
[edited again for much more thinking] |
|
|
| BWRX |
The differential configuration gives you better common mode rejection and lower output offset. The common mode rejection figures do depend on how well the resistor pairs are matched. The closer they are the better the CMR will be. Matching the resistors isn't that difficult. You can buy 0.1% tolerance resistors or match them as close as possible with a multimeter.
I can't say for sure which configuration would sound better because everyone has a different set of ears :) If it were my choice and I had a balanced source I would definitely try the LM3886 as a differetial amp with input transformer. |
|
|
| xiphmont |
| quote: | Originally posted by BWRX
The differential configuration gives you better common mode rejection and lower output offset. The common mode rejection figures do depend on how well the resistor pairs are matched. The closer they are the better the CMR will be. Matching the resistors isn't that difficult. You can buy 0.1% tolerance resistors or match them as close as possible with a multimeter.
I can't say for sure which configuration would sound better because everyone has a different set of ears :) If it were my choice and I had a balanced source I would definitely try the LM3886 as a differetial amp with input transformer. |
...at what point would the resistor mismatch affect CMR such that it's no longer better than single ended? This is mostly a question of idle curiosity. (And in fact-- most of the inputs will be pseudobalanced or direct-balanced). Also, when you say 'improve the CMR', you're talking about the CMR of just the amp input or of the the transformer too? The input transformer should already be providing an additional 70dB-120dB of CMR even into a single-ended amp input according to the cinemag and Jensen application notes. I thought the transformer was, in some ways, a *replacement* for the CMR of the amp.
The other problem---where would the attenuator pot go? I really don't like the idea of radically altering the input impedence (shunt mode attenuator on the source side of the transformer). I really don't trust my sources enough to try to pull that.
Stab in the dark-- shunt mode across the differential inputs to the 3886? There's gotta be something elegant that isn't too huge a tradeoff....
All the above referring to the schematic in figure 6-33:
...with values altered to suit the design (about 10k/220k instead of 12k/6k). Hmm, would this topology/value change affect the original output Zobel?
[edit: more questions about CMR and Zobel] |
|
|
| AndrewT |
Hi,
that last schematic could have your chipamp in there instead of the ssm2143. Just add the 0.1% matched resistors and adjust them to give the gain you require.
As for volume.
Insert series resistors in between each of the xlr poles to each of the transformer inputs.
Then insert the variable resistor(or even a twelve way switched resistor) after the series resistors. Now you have a balanced input and volume control and no extra chips nor extra expensive components.
The series resistors could be 1k0 to 10k depending on whether your balanced source can drive 600ohm load. If not then the higher value would be suitable. |
|
|
| gootee |
| quote: | Originally posted by xiphmont
...at what point would the resistor mismatch affect CMR such that it's no longer better than single ended? This is mostly a question of idle curiosity. (And in fact-- most of the inputs will be pseudobalanced or direct-balanced). Also, when you say 'improve the CMR', you're talking about the CMR of just the amp input or of the the transformer too? The input transformer should already be providing an additional 70dB-120dB of CMR even into a single-ended amp input according to the cinemag and Jensen application notes. I thought the transformer was, in some ways, a *replacement* for the CMR of the amp.
|
See Figure 2.2, and Equations 2-2 and 2-2, in section 2 ("Specialty Amplifiers") at the link I gave for the Op Amp Application Handbook. It looks like for 0.1% tolerance resistors, the worst-case dc CMR would be 66dB, but with some assumptions. With three 0.1% resistors and one mis-matched by 1%, CMR drops from 66dB to 46 dB.
Those figures assume R1=R2, and that the amplifier's CMR > 100 dB. For other R values, see the cited equations.
And note that it's actually the RATIOS of the two resistors in each leg that need to match closely. i.e. The top two resistors' ratio needs to match the bottom two resistors' ratio.
You can get IRC 0.1% resistors, 25ppm/degC (as opposed to 50ppm for the common 1% metal film type), for less than $1.00 qty 1, at http://www.mouser.com . Those 25ppm/degC ones have part numbers like 66-RCLF-D-value, at mouser.com, IIRC. There are also some others, there, that have 5 and 10 ppm/degc tempcos, but aren't offered with nearly as many values.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
- |
|
|
| xiphmont |
| quote: | Originally posted by gootee
See Figure 2.2, and Equations 2-2 and 2-2, in section 2 ("Specialty Amplifiers") at the link I gave for the Op Amp Application Handbook. It looks like for 0.1% tolerance resistors, the worst-case dc CMR would be 66dB, but with some assumptions. With three 0.1% resistors and one mis-matched by 1%, CMR drops from 66dB to 46 dB.
|
OK, so you are in fact talking about additional CMRR provided by the amp *over* the CMR provided by the transformer. The transformer is already handing me 120dB of DC rejection (less at HF). It seems like going for more is a bit unwarranted; I better not have common mode noise greater than 0dB to worry about!
(In my case, 'long runs' are 50 feet of signal level at home, not 500 ft of mic level run in a theater space with a messed up power grid shared with 500kW of dimmerpacks. In the theater, I'd probably want more then 120dB of CMR).
Again, thinking aloud not arguing, I'd like to know if I'm wrong there. |
|
|
| xiphmont |
Let me tie all the threads so far together and post a new schematic. As always, this is thinking aloud and I want to hear when I'm wrong.
Point 1: transformer into single ended or pseudobalanced
Most people running Gainclones as 'super high fi' amplifiers are doing so single ended and as such have a low CMRR (somewhere from 0 to 30dB-ish). Because these are going to be in self-powered monitors, the line level runs will be a little longer, and so a little more hum/buzz rejection is needed. The Cinemag CMLI-15/15B is being used as an isolation and rejection stage. The idea is that it gives substantial enough improvement over a naked single-ended input that the additional noise rejection of going to a differential/pseudo-balanced topology doesn't make much sense in the context of the rest of the system mostly because it complicates the placement of an attenuator.
Point 2: attenuator placement
Assuming a single-ended input and attenuator on the input, as is illustrated in the NatSemi application notes and as configured in the majority of gainclone kits, the wiper 'lifting', either through poor design or age of the pot (think about the scratchiness potentiometers develop with age) will cause swings in the output DC offset. The concern is not that it will damage the speakers, but that even a relatively small DC offset will cause alot of noise as the pot ages, and that noise is in proportion to the max DC offset not the input program. It will be loud, even if the amp is 'idle'.
A few solutions have been offered:
a) Choose not to care. This is a popular solution as the majority of the small kits go this route. Don't use an attenuator, or use the best one you can afford. Replace it if it gets icky. Potentiometer noise once it shows up is annoying but not going to damage anything; a scratchy potentiometer is going to be noisy anyway. Optional Ci reduces the liability further by cutting down the output offset by 30dB.
b) Run the circuit as-is in the NatSemi notes but decouple the + input by replacing R1 with a coupling cap and make sure R2 is there to provide a path for bias current to ground. Disadvantages: another capacitor in the signal path (is this *really* something people care about with good modern capacitors?) Advantages: Input completely isolated from bias current changes, can be trimmed to near-zero offset. (Hmm, C1 and Ci are insanely large given the resistor values, aren't they?)
c) shunt-mode attenuator in front of the transformer. I'm going to reject this one straight out because of the inherent impedence/attenuation tradeoff on the input (either you have large impedence swings, or you have alot of inherent signal loss through attenuation) and the fact that the Cinemag transformer's design notes state explicitly that the transformer wants to be 'looking into' a low impedence. Adding the shunt attenuator in front of the transformer gauarntees the source will appear to be high impedence.
d) Use a potentiometer shunt-mode as a gain-pad, not an input attenuator. That is, the potentiometer replaces Rf1 and directly alters the circuit gain rather than altering the input attenuation. Disadvantages: can only reduce gain down to unity, not below. In my circuit, it has a 0dB to 30dB range. Advantages: eliminates several passives, S/N is just as good at low gain as at high gain.
(Actually, it's not entirely clear to me that this greatly reduces the output offset voltage swing on noisy pot, just that it simplifies the circuit if you're willing to give up 'attenuation' range. Can somone convince me choice D actually gives me nothing?)
 |
|
|
| AndrewT |
Hi,
post 27 shows the balanced input opamp.
Add this topology to pic 3 from post 31.
Now delete the transformer.
You have a balanced attenuator, with balanced chipamp and retain much of the CMRR of the electronically balanced circuit (not quite as good as the isolating transformer, but a lot cheaper and as you explained may be enough for a domestic environment.
Is it worth experimenting with?
If you still reject this option, no hard feelings, it's your amp not mine.
BWRX,
have you added local attenuation to your balanced circuit? |
|
|
| xiphmont |
| quote: | Originally posted by xiphmont
(Hmm, C1 and Ci are insanely large given the resistor values, aren't they?)
|
...and in fact, back of the envelope reactance calculations suggest .2 to 1uF is a better choice (given that the amps will be dedicated to a single driver/aperiodic enclosure their entire lives, and there's no point to a flat response below 40Hz).
Xc = 1/(2pi*f*C), setting f to 8 (40Hz/5) and Xc to 100kOhms, we get C = .2uF
.2uF might even be a good choice for making sure the amp can't push out as much bass below where the driver and air mass decouple and thus serve to limit excursion as well...
Whee, integrated design requirements. |
|
|
| AndrewT |
Hi,
C1 can be scaled for F-3db=8Hz. That's OK.
But Ci should be scaled for about half an octave lower, say 5Hz. Ci starts to get very big with just 3k above it (your 10uF value hits this target exactly). or DC couple both input and NFB.
Why do you show the lower leg going to audio ground rather than tying directly to the transformer?
Then you can omit an extra pair of ground connections at the amplifier input. |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
post 27 shows the balanced input opamp.
Add this topology to pic 3 from post 31.
Now delete the transformer.
You have a balanced attenuator, with balanced chipamp and retain much of the CMRR of the electronically balanced circuit (not quite as good as the isolating transformer, but a lot cheaper and as you explained may be enough for a domestic environment.
Is it worth experimenting with? |
In fact I probably will build it for educational purposes.
My fondness for the transformer solution mainly stems from it being an affordable CMRR Solution-In-A-Can(tm). The pseudo balanced circuits with .1% tolerances will give me 40 or 50dB CMRR, and the $38 can gives me 120dB... It seems worth it to me mainly because it does cost me some extra money, but it is importantly also not costing me practically any additional time (this is the tipping point-- time does equal money. All this design work is fun. Sitting down to build a fleet of these things will quickly become monotonous. I can do no real extra work for 40dB or no real extra work for 120dB is how I'm looking at it.)
The transformer does wreak some havok WRT placement of the attenuator. I am currently leaning toward either solution A or B. Solution D was to see if anyone thought it is any better than than A.
[edited for clarity] |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
C1 can be scaled for F-3db=8Hz. That's OK.
But Ci should be scaled for about half an octave lower, say 5Hz. Ci starts to get very big with just 3k above it (your 10uF value hits this target exactly).
|
Ah, it 'sees' the 3.1k R3, not the 100k Rf? OK.
| quote: | Originally posted by AndrewT
Why do you show the lower leg going to audio ground rather than tying directly to the transformer?
Then you can omit an extra pair of ground connections at the amplifier input. |
Because it's running single ended and these all should reference to ground.
Do you just mean they should be tied together, then tied to ground, or are you suggesting the transformer negative output and NFB lower leg should tie toether and float? I don't see what the point of that would be as the circuit is single-ended... (The Faraday shield still goes to signal ground in any case, yes?) |
|
|
| AndrewT |
| quote: | Originally posted by xiphmont
Because it's running single ended and these all should reference to ground.
Do you just mean they should be tied together, then tied to ground, or are you suggesting the transformer negative output and NFB lower leg should tie toether and float? I don't see what the point of that would be as the circuit is single-ended... (The Faraday shield still goes to signal ground in any case, yes?) |
Hi,
look again at post27.
It's a balanced input to unbalanced output.
The gain is set with the integrated resistors built into the SSM package.
Substitute the chipamp for the ssm and select your own resistors to duplicate the schematic.
Now adjust your resistors to get the gain you need and to ensure the chipamp has enough gain to remain stable.
Note that the inputs are NOT connected to ground. The reference leg on the non-inverting connects to ground and it can provide the RF attenuation you can optionally add.
So, yes you can let the inputs float (well almost float, they go all the way back to the source via your balanced interconnect). They then measure and amplify the input signal and don't get contaminated by noisy grounds/shields/etc.
Go to Walt Jung's site and download all his audio articles from the archive. Read all about how to get the best out of opamps (and cascaded opamps) and particularly the tricks to ensuring electronically balanced works as intended. |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
look again at post27.
It's a balanced input to unbalanced output.
[...]
Go to Walt Jung's site and download all his audio articles from the archive. Read all about how to get the best out of opamps (and cascaded opamps) and particularly the tricks to ensuring electronically balanced works as intended. |
I'm trying to figure out if you're helping me with the single ended designs (eg A/B) or trying to convince me to ditch it and do it another way.
Eg, Are you suggesting something like the following will work?
I see that the inputs don't need to reference ground (within limits, they only reference each other), but the above, in which I am attempting to faithfully follow your instructions, does not seem to make any sense.
"the tricks to ensuring electronically balanced works as intended" aren't that interesting since a balanced circuit is not my goal! I'm looking to apply the inherent CMRR of the transformer in a preexisting circuit, avoid mistakes doing so, and if there are any easy freebies that will also improve the noise rejection, I'll apply them too. Tossing out the existing PCBs is not an option I will consider.
I keep saying 'single ended' you keep saying 'you're doing balanced wrong.'
[edit: OK, I lied, Walt's articles on balanced line receivers are interesting (I've read them before) but they're not that useful in direct application to my end goal] |
|
|
| xiphmont |
Elaboration on 'does not make sense'--
It makes sense but I'd think the inputs (looking through their 3.1k resistors) would have to be 'driven' by a low impedence source, not the transformer negative and pot wiper. |
|
|
| AndrewT |
Hi,
post38 does not look nice.
Not sure if this idea is workable.
add two resistors between the transformer and the pot.
convert the pot to variable resistor.
add two caps//100k
add a third cap across the variable resistor.
The transformer will reflect the source impedance back to this receiver. | quote: | | I keep saying 'single ended' you keep saying 'you're doing balanced wrong.' |
the title says balanced. You have not confirmed if your source can drive 600ohm. |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
Hi,
post38 does not look nice.
Not sure if this idea is workable.
|
I didn't think it was. I was trying to follow your suggestion, but didn't entirely understand what you wanted (a floating NFB leg on a single ended input?)
| quote: | Originally posted by AndrewT
You have not confirmed if your source can drive 600ohm. |
I don't have a specific source in mind, and I would not assume the source could drive a 600 ohm load, no. Most can, but I don't know how many do it all that well.
| quote: | Originally posted by AndrewT
the title says balanced. |
You have a point. I see where that would confuse the issue. What I meant: The input to the box would truly be balanced due to the transformer, and the whole thing inside the box (including the transformer) would be 'a gainclone'.
Even most of the folks building three op-amp buffers with instrumentation amplifiers then feed that signal into a chip amp single ended. They still call the entire thing, including the buffer, 'a gainclone'. I'm using a transformer instead of a solid-state buffer.
Perhaps I should start a new thread as 'balanced input to single ended gainclone'. Posting all the schematics was an attempt to make it clear what I was trying to do.
|
|
|
| xiphmont |
A pirate walks into a bar with a giant transformer hanging out of his pants by the wires...
It's the punchline that's on-topic. |
|
|
| gootee |
| quote: | Originally posted by xiphmont
A pirate walks into a bar with a giant transformer hanging out of his pants by the wires...
It's the punchline that's on-topic. |
This thing's drivin' me nuts!
(Or, a guy walks into a psychiatrist's office with a steering wheel sticking out of his pants, and says, "It's drivin' me nuts.") |
|
|
| gootee |
If you have any ways to measure what you're after, perhaps it's almost time to do some breadboard experiments. The parts don't look all that expensive, even with 0.1% resistors (or add a multi-turn trimpot somewhere). And it looks pretty simple to set up the different variations that have been discussed.
Or, if you're a spice user, maybe you should first try some simulations of the various topologies. (If you're not a spice user, you could download LTspice, free, from linear.com.) For a "generic" chipamp model that might be better than nothing, there's the OP541E model from ti.com. And there are speaker system and cable models you might want to try, such as the ones at http://sound.westhost.com/cable-z.htm . Transformer modeling is covered somewhere, too. Sorry, I don't have the URL handy.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
- |
|
|
| BWRX |
| quote: | Originally posted by AndrewT
BWRX,
have you added local attenuation to your balanced circuit? |
No. I take care of volume control with a buffered potentiometer to avoid all the issues you guys have been discussing ;) However, the high input impedance of the instrumentation configuration means there shouldn't be any problem using a potentiometer between the source and the amp.
| quote: | Originally posted by xiphmont
Even most of the folks building three op-amp buffers with instrumentation amplifiers then feed that signal into a chip amp single ended. They still call the entire thing, including the buffer, 'a gainclone'. I'm using a transformer instead of a solid-state buffer. |
I incorrectly called my implementation an instrumentation gainclone. I've since just called it an instrumentation configuration with a chip amp. Technically, a gainclone is a clone of the gaincard, which is just a barbones configuration. |
|
|
| xiphmont |
| quote: | Originally posted by gootee
If you have any ways to measure what you're after, perhaps it's almost time to do some breadboard experiments. The parts don't look all that expensive, even with 0.1% resistors (or add a multi-turn trimpot somewhere). And it looks pretty simple to set up the different variations that have been discussed.
|
Sure, and I even have most of these parts.
This discussion mostly boils down to 'I had it mostly right the first time' and a few people offered alternatives to consider (which is something I wanted, and I took some time thinking about it). It got a little sidetracked when AndrewT kept suggesting differential topologies and I thought he was suggesting some improvement to the single-ended strategy that didn't seem to make sense.
| quote: | Originally posted by BWRX
I take care of volume control with a buffered potentiometer to avoid all the issues you guys have been discussing ;) However, the high input impedance of the instrumentation configuration means there shouldn't be any problem using a potentiometer between the source and the amp.
|
Right, more actives is not a bad strategy. However I decided I was going to pursue a strategy of minimalism and a signle passive or two would do all the same for me. It is a convenient coincidence that it also means it will take much less assembly work :-)
| quote: | Originally posted by BWRX
I incorrectly called my implementation an instrumentation gainclone. I've since just called it an instrumentation configuration with a chip amp. Technically, a gainclone is a clone of the gaincard, which is just a barbones configuration. |
Sure, but common usage has gone beyong calling just the replica of the gaincard 'a gainclone'; you knew what you meant, I knew what you meant, AndrewT apparently didn't, so I went into more detail to clear up the confusion. I wasn't complaining about anyone being loose with their terminology.
Anyway, for me, it boils down to choice A sans Ci or Choice B fully capacitor coupled. A scratchy pot will make either noisy, but it will make A much noisier than B. I'd want to be using at minimum a fully sealed CP pot anyway (eg, Mouser 72-PA16NP103MAB15), will that pot give me a lifetime such that I can eliminate the coupling caps in the signal path? Or is a decent coupling cap (metallized PP) at a volt and microamps just not something anyone worries about using these days? Heck, the gainclone kit, as shipped, is using a Panasonic electrolytic as Ci.... |
|
|
| xiphmont |
| quote: | Originally posted by xiphmont
Heck, the gainclone kit, as shipped, is using a Panasonic electrolytic as Ci.... |
(Hm, I seem to have sold the FCs a little short, even if the tan-delta is almost 5%...) |
|
|
| gootee |
| quote: | Originally posted by xiphmont
<snipped>
I'd want to be using at minimum a fully sealed CP pot anyway (eg, Mouser 72-PA16NP103MAB15), will that pot give me a lifetime such that I can eliminate the coupling caps in the signal path? Or is a decent coupling cap (metallized PP) at a volt and microamps just not something anyone worries about using these days? Heck, the gainclone kit, as shipped, is using a Panasonic electrolytic as Ci.... |
In case anyone else is interested, that pot's datasheet is at:
http://www.vishay.com/docs/51036/p16pa16.pdf .
That potentiometer design is actually very interesting. They put the pot inside the knob, so there's almost nothing behind the panel. Cool...
So let's see... "PA" instead of "P" means it's the "professional audio" version, which makes it conductive plastic instead of cermet (and also apparently cuts the power rating in half and multplies the temperature coefficient by ten, neither of which should matter, here), and lowers the contact resistance variation from 3% to 2%. Conductive plastic is much quieter than Cermet, I believe. Good choice.
"NP" means it has a black plastic knob. Probably OK. But they do offer it in metal, too (the "NM" version), if you're interested.
"M" is the 20% tolerance. But it's a _variable_ resistor, anyway.
"A" means "linear law" instead of "logarithmic law". Is that really what you want? [But their "logarithmic law" plot looks like two straight line segments, with a breakpoint at (50% rotation, 10% of total resistance), which you might be able to get by adding a fixed resistor, if you want it.]
Regarding your original question about lifespan: They give "load life" and "rotational life" test results. But they only give one data point for each one, for which the pot doesn't appear to be too degraded, at all. So they seem to not really say how long it might last. I guess you can probably count on the 25000 rotational cycles without much change, if your power load isn't too high.
All in all, it seems like a pretty good choice, especially for $8.98 qty 1. But you still might want to do at least one search (probably for both P16 and PA16 series, maybe at diyaudio.com and at http://groups.google.com), to see if anyone who has some experience with that model line has anything to say about it.
About eliminating the coupling caps: I'd say go for it. But that's just my impulsive opinion. Here's an idea I just thought of: You could (at least) geometrically reduce the chance of total pot failure (i.e. wiper eventually lifting) by using a DUAL pot, with the elements wired in parallel. For example, you'd get a dual 20K pot and wire the two 20K pots in parallel to get your 10K pot. Then, if one wiper lifts, you've still got the other one, for double the resistance instead of infinite resistance (assuming I'm correctly imagining how you'd want to wire them). And then you'd probably notice it and realize it was time to replace the pot, before the other half failed (essentially making it failure-proof, as well as fail-safe, for that failure/amp-damage mode at least). Two parallel pots might even be electronically better than one, too; maybe sort of like a wider trace on a PCB, giving lower inductance. But I have no idea whether or not some other, possibly "bad" effects of having two pots in parallel might be more significant. My guess would be that probably nothing about it would be significant except the fail-safe feature (and the slightly bigger size and cost, of course).
Sorry to "ramble on" about that, for so long.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
- |
|
|
| xiphmont |
| quote: | Originally posted by gootee
"NP" means it has a black plastic knob. Probably OK. But they do offer it in metal, too (the "NM" version), if you're interested.
[...]
"A" means "linear law" instead of "logarithmic law". Is that really what you want? [But their "logarithmic law" plot looks like two straight line segments, with a breakpoint at (50% rotation, 10% of total resistance), which you might be able to get by adding a fixed resistor, if you want it.]
|
Oh, if I linked the linear one I brainoed. I intended to link the log one. As for the plastic knob, I'll be using plastics on the back panel anyway (power, breaker, XLR gasket, lift switch).
| quote: | Originally posted by gootee
About eliminating the coupling caps: I'd say go for it. But that's just my impulsive opinion.
|
Having looked over real research (eg, done by EEs whose job it is to design audio sans pandering to superstition) I see no compelling reason to avoid the coupling caps, or even the FC on the NFB leg.
For every audiophile who claims electrolytics ruin the sound, there's some other superstitious whack who claims BlackGates on all his inputs 'smooths out' the sound, revealing 'hidden nuance' and 'refined detail'. :-)
National gave us the chip amps, they're damned good, and they recommend coupling and filter caps. I'm inclined to believe National on this one.
| quote: | Originally posted by gootee
Here's an idea I just thought of: You could (at least) geometrically reduce the chance of total pot failure (i.e. wiper eventually lifting) by using a DUAL pot, with the elements wired in parallel. For example, you'd get a dual 20K pot and wire the two 20K pots in parallel to get your 10K pot. Then, if one wiper lifts, you've still got the other one, for double the resistance instead of infinite resistance (assuming I'm correctly imagining how you'd want to wire them). And then you'd probably notice it and realize it was time to replace the pot, before the other half failed (essentially making it failure-proof, as well as fail-safe, for that failure/amp-damage mode at least). Two parallel pots might even be electronically better than one, too; maybe sort of like a wider trace on a PCB, giving lower inductance. But I have no idea whether or not some other, possibly "bad" effects of having two pots in parallel might be more significant. My guess would be that probably nothing about it would be significant except the fail-safe feature (and the slightly bigger size and cost, of course).
Sorry to "ramble on" about that, for so long.
|
No worries, was easy to tell what you were getting at. Either way 'fail' wasn't really going to be a worry. a 470mV (worst case) 'crackle' wasn't going to hurt or bother anything except the listener ;-) |
|
|
| gootee |
| quote: | Originally posted by xiphmont
Having looked over real research (eg, done by EEs whose job it is to design audio sans pandering to superstition) I see no compelling reason to avoid the coupling caps, or even the FC on the NFB leg.
For every audiophile who claims electrolytics ruin the sound, there's some other superstitious whack who claims BlackGates on all his inputs 'smooths out' the sound, revealing 'hidden nuance' and 'refined detail'. :-)
National gave us the chip amps, they're damned good, and they recommend coupling and filter caps. I'm inclined to believe National on this one.
|
Well, I did say that that opinion was "impulsive". :-)
I AM an EE. So maybe I should be embarrassed. Just for the record, I'm sure that my tendency toward disdain of "audio superstitiousness" is as at least as great as anyone's. And I am sure that it is safer to use the coupling capacitors, in most cases.
However, even National states that a coupling capacitor will degrade the audio quality. For example: "For best audio performance, the input capacitor should not be used.", after Equation 6 on Page 14 of the LME49810 datasheet, which I just happened to be reading, a little while ago. But it's already obvious that using coupling caps will always tend to limit the low-frequency response, in any amplifier circuit. The _only_ significant positive aspect of their use is in blocking DC (and very low frequencies, if they happen to be a problem).
So using them or not boils down to a "trade-off" decision, as does sizing them if you do use them.
Many times, an affirmative decision about whether or not to use coupling capacitors is based on the simple fact of having no way of knowing whether any possible future input device might have some "dangerously-high" level of DC present, with its output. But in your circuit, no DC at all should ever be coming in through your coupling transformer. So if your chipamp's inputs' impedances were balanced, or at least if input-current-induced voltages at the chipamp's inputs were not different by "too much", then DC offset at the output should not be a problem, except in the case of something equivalent to an "open wiper" failure in your potentiometer (and depending on the values of the input and feedback-divider resistances used, etc), in which case you would usually get a very large DC voltage at your chipamp's output pin, which would definitely be "a BAD thing".
But, if it's just a matter of not trusting the pot wiper forever, and since your "Choice A" schematic already has always-unbalanced (and varying in difference with pot setting) input impedances for the LM3886 inputs, anyway, it sure SEEMS like you could just add a fairly-large resistor to ground between R1 and the + input, in your "Choice A" schematic, to guard against an open pot wiper, and still omit the coupling caps. (You'd probably want to keep the resistor as small as possible but without sacrificing too much gain, to limit the noise produced by the resistor. Another trade-off.)
I don't know the relative failure rates of pot wipers and fixed resistors and solder joints. But even in your "Choice B" schematic, i.e. WITH the coupling caps, R2 failing open (or a solder joint in that vicinity failing open) would be at least as catastrophic as the pot wiper failing open for "Choice A".
It's very difficult to protect against all kinds of failures (or even some, or just one). I suppose you could use two parallel resistors, in place of each resistor, to get closer to "safe", in this case.
Is any of it worthwhile? Another trade-off. It's your amp.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
- |
|
|
| xiphmont |
| quote: | Originally posted by gootee
I AM an EE.
|
...as am I. But I come from a different set of specialties. Analog high-fidelity just isn't my forte, I'm not going to pretend it is. I have to deal with 140dB deep stuff all the time-- but only in software dsp ;-)
| quote: | Originally posted by gootee
However, even National states that a coupling capacitor will degrade the audio quality. For example: "For best audio performance, the input capacitor should not be used.", after Equation 6 on Page 14 of the LME49810 datasheet, which I just happened to be reading, a little while ago. But it's already obvious that using coupling caps will always tend to limit the low-frequency response, in any amplifier circuit.
|
I agree there are reasons to use coupling caps carefully (eg, factor 5/50 in low end rolloff so LF attenuation isn't really a worry). The distortion figures for modern caps are damned near objectively indetectable for foil/film (which I will use for the input couple just to be sure-- it's also nearly immortal) and even the modern low-ESR eletrolytics are very good. Power supply grade electrolytics even look to have better characteristics in some ways in that they move charge *fast* for their size. Way better than big metallized (which is why I'm no longer quick to dismiss the FC).
Can I hear the difference? I'm going to build it both ways myself and listen. Golden ears are one thing I do have. Better use them while they last.
| quote: | Originally posted by gootee
But, if it's just a matter of not trusting the pot wiper forever, and since your "Choice A" schematic already has always-unbalanced (and varying in difference with pot setting) input impedances for the LM3886 inputs, anyway, it sure SEEMS like you could just add a fairly-large resistor to ground between R1 and the + input, in your "Choice A" schematic, to guard against an open pot wiper, and still omit the coupling caps. (You'd probably want to keep the resistor as small as possible but without sacrificing too much gain, to limit the noise produced by the resistor. Another trade-off.)
|
AndrewT already prompted me into doing the math on that, and a pot failure would just be annoying, no worse. 100-500mV is not a big deal (although the noise would be loud, the driver I'm using has thermal/mechanical properties well in excess of what this amp will be able to drive).
What I am *actually* thinking about in choice B is limiting self-noise as the pot ages. If the CP gets even a little bit 'scratchier' with age like carbon does, limiting the gain on that scratchiness is probably worth doing. I don't want the attenuator on this thing to sound like wind into a microphone in five years. Choice B cuts down on the potentiometer self-noise by 30dB in the worst case.
| quote: | Originally posted by gootee
I don't know the relative failure rates of pot wipers and fixed resistors and solder joints. But even in your "Choice B" schematic, i.e. WITH the coupling caps, R2 failing open (or a solder joint in that vicinity failing open) would be at least as catastrophic as the pot wiper failing open for "Choice A".
|
Well, if we're going to decide not to trust solder joints over time, there's no saving any design :-)
Anyway, looking for the log version of that Vishay pot, no one stocks it. Mouser quotes a minimum order of 20 (at $20) and a 16 week lead time to get it, despite it appearing in the catalog. Bourne precision and a knob it is! The Bourne 91 series are cheaper and unsealed, but claim a 100k rotation life. The 51 series 9also available) are about 50% more expensive and sealed, but claim half the rated life. |
|
|
| gootee |
| quote: | Originally posted by xiphmont
...as am I. But I come from a different set of specialties. Analog high-fidelity just isn't my forte, I'm not going to pretend it is. I have to deal with 140dB deep stuff all the time-- but only in software dsp ;-)
|
Very good. Sorry for my "speechifying". I'm probably much farther from knowing a lot about analog audio than you are.
| quote: |
I agree there are reasons to use coupling caps carefully (eg, factor 5/50 in low end rolloff so LF attenuation isn't really a worry). The distortion figures for modern caps are damned near objectively indetectable for foil/film (which I will use for the input couple just to be sure-- it's also nearly immortal) and even the modern low-ESR eletrolytics are very good. Power supply grade electrolytics even look to have better characteristics in some ways in that they move charge *fast* for their size. Way better than big metallized (which is why I'm no longer quick to dismiss the FC).
|
(Sorry if some of this is too far off-topic. I just wanted to ask you about capacitor choices.)
I have never understood the "reasoning" of the "no caps in the signal path" crowd. And thanks for the further insights about capacitors.
I have spent a fair amount of time selecting capacitors for analog signal-processing circuits, and for switch-mode power supply circuits, over the last couple of years, and have even implemented a dielectric-absorption correction circuit (similar to what Bob Pease suggested) for an integrator (to try to get the last little bit of linearity improvement for a precision ramp generator). I remember that I was never able to find a source for Teflon capacitors, and usually used polypropylene, or polystyrene, instead, when low dielectric absorption was needed. But for very small cap values, Bob Pease had a great suggestion for making Teflon ones: Just get some Teflon-dielectric coax and trim the length until it has the desired capacitance!
Regarding the "power supply grade electrolytics" that you mentioned: Do you have any specific ones in mind, that you like? I usually ended up using Nichicon's UHE series, for medium-to-low values, probably mostly just because Mouser (and Digikey, IIRC) carried them. Their 3300 uF 35V UHE-series model has an ESR spec (given for 20 degC/100 kHz) of .013 Ohms max (which is similar for the same 16x40mm size for lower voltages and higher capacitances), and a ripple current rating of 4.08A RMS (@ 105 degC/100 kHz). I also used their 2200uF/50V UHE model, which has a .017 Ohm ESR spec.
But I also used smaller UHE-series electrolytics as DC blocking caps (e.g. 220uF/10V), in audio-frequency circuits (in test/measurement equipment) where very low distortion and very level frequency response were both important, but frequencies could be as low as 60 Hz (and up to at least 22 kHz). Now I'm wondering if there is a better choice than the UHE caps. The signals, in one case, were triangle waves, typically less than 350 to 500 mV P-P, with 220uF/10V cap current in the 180 uA P-P range. Unfortunately, the cap diameter could only be 6.3mm max. In another case, the signal is sinusoidal, 5V P-P, with cap current of about 250 uA P-P. In that case, I also wonder if it's really necessary to use two caps, back to back, to form a non-polarized version, if a 50 uF value is needed.
| quote: |
Can I hear the difference? I'm going to build it both ways myself and listen. Golden ears are one thing I do have. Better use them while they last.
|
I hope you'll share your results.
| quote: |
AndrewT already prompted me into doing the math on that, and a pot failure would just be annoying, no worse. 100-500mV is not a big deal (although the noise would be loud, the driver I'm using has thermal/mechanical properties well in excess of what this amp will be able to drive).
|
That's only if you do have the coupling caps, I assume.
| quote: |
What I am *actually* thinking about in choice B is limiting self-noise as the pot ages. If the CP gets even a little bit 'scratchier' with age like carbon does, limiting the gain on that scratchiness is probably worth doing. I don't want the attenuator on this thing to sound like wind into a microphone in five years. Choice B cuts down on the potentiometer self-noise by 30dB in the worst case.
|
I just don't know if five years would be a problem, with a conductive plastic pot. "It depends." But ten or fifteen years might be, especially if it's not sealed. Some of the guys in the TekScopes group, at yahoogroups.com, would really know for sure.
| quote: |
Well, if we're going to decide not to trust solder joints over time, there's no saving any design :-)
|
They can be saved, if you throw suitcases of cash at them.
The probabilities of problems being caused by such failures can sometimes be reduced geometrically for each level of redundancy used, in cases where it might be worth the cost. I used to work in aerospace/defense, analyzing and designing guidance and control systems (but not at the electronic board/component level, in my case). You might be surprised at the amount of effort put into planning for such scenarios, in certain types of systems. In a lot of cases, a single-point failure of _any_ kind was not allowed to cause any degradation of functionality, or near-term system survivability.
Sorry to have blathered-on for so long, again.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
- |
|
|
| xiphmont |
| quote: | Originally posted by gootee
But for very small cap values, Bob Pease had a great suggestion for making Teflon ones: Just get some Teflon-dielectric coax and trim the length until it has the desired capacitance!
|
Heh, that's funny and something I'd probably not have thought of. Too bad most coax is actually polyethelyne or polypropylene though.
| quote: | Originally posted by gootee
Regarding the "power supply grade electrolytics" that you mentioned: Do you have any specific ones in mind, that you like? I usually ended up using Nichicon's UHE series, for medium-to-low values, probably mostly just because Mouser (and Digikey, IIRC) carried them. Their 3300 uF 35V UHE-series model has an ESR spec (given for 20 degC/100 kHz) of .013 Ohms max (which is similar for the same 16x40mm size for lower voltages and higher capacitances), and a ripple current rating of 4.08A RMS (@ 105 degC/100 kHz). I also used their 2200uF/50V UHE model, which has a .017 Ohm ESR spec.
But I also used smaller UHE-series electrolytics as DC blocking caps (e.g. 220uF/10V), in audio-frequency circuits (in test/measurement equipment) where very low distortion and very level frequency response were both important, but frequencies could be as low as 60 Hz (and up to at least 22 kHz). Now I'm wondering if there is a better choice than the UHE caps. The signals, in one case, were triangle waves, typically less than 350 to 500 mV P-P, with 220uF/10V cap current in the 180 uA P-P range. Unfortunately, the cap diameter could only be 6.3mm max. In another case, the signal is sinusoidal, 5V P-P, with cap current of about 250 uA P-P. In that case, I also wonder if it's really necessary to use two caps, back to back, to form a non-polarized version, if a 50 uF value is needed.
|
I've usually used the Nichicons you mentioned or one of the Panasonics. Like you, I suspect that's mostly because it was what Digikey has. The hobbyist electronics scene has entirely dried up from when I was a kid. There used to be two good electronics parts stores in each of the malls near where I was a kid, now Digikey and Mouser are all that's left. I miss Olsens and Active.
I have wondered the same about polarization given that the gainclone kits are using polarized caps in their filter and output stages-- definitely not polar circuits! I've been assuming my worry is unfounded and there's something I'm missing.
(Oh, and it is odd how the current 'style' in virtually all caps is low and squat :-)
| quote: | Originally posted by gootee
The probabilities of problems being caused by such failures can sometimes be reduced geometrically for each level of redundancy used, in cases where it might be worth the cost. I used to work in aerospace/defense, analyzing and designing guidance and control systems (but not at the electronic board/component level, in my case). You might be surprised at the amount of effort put into planning for such scenarios, in certain types of systems. In a lot of cases, a single-point failure of _any_ kind was not allowed to cause any degradation of functionality, or near-term system survivability.
|
My design durability criteria really isn't all that much more than "I want my stuff to survive long enough that my kids will be old enough to appreciate the workmanship." :-)
| quote: | Originally posted by gootee
Sorry to have blathered-on for so long, again.
- |
No worries, I *live* for this kind of geeking. |
|
|
| AudioFreak |
| quote: | Originally posted by xiphmont
I have wondered the same about polarization given that the gainclone kits are using polarized caps in their filter and output stages-- definitely not polar circuits! I've been assuming my worry is unfounded and there's something I'm missing.
|
The AC component mostly passes thru the cap so it is really only the polarity of the standing voltage that is of concern. |
|
|
| xiphmont |
| quote: | Originally posted by AudioFreak
The AC component mostly passes thru the cap so it is really only the polarity of the standing voltage that is of concern. |
Correct, it is the persistent bias of the circuit in question. However, earlier in this thread, we established the bias can be positive or negative; the DC offset is not necessarily positive (or negative); individual variation in 3886 is enough to change that. |
|
|
| AndrewT |
| quote: | | the DC offset is not necessarily positive (or negative); individual variation in 3886 is enough to change that |
this should not be the case.
If the front end is NPN type (or Nch) then the bias is negative.
If the PNP (or Pch) the the bias is positive.
If the front end is dual complementary National would advertise it as such and this is extremely unlikely. A bias compensated front end could swing either side of zero bias. |
|
|
| xiphmont |
| quote: | Originally posted by AndrewT
this should not be the case.
If the front end is NPN type (or Nch) then the bias is negative.
If the PNP (or Pch) the the bias is positive.
If the front end is dual complementary National would advertise it as such and this is extremely unlikely. A bias compensated front end could swing either side of zero bias. |
No, I mean: Look at the Zobel on the output stage. The DC output offset, given input bias offset variation, could end up being negative or positive in the most common designs. But the Zobel is using a polarized cap.
Even if the bias is positive, |
|
|
|
