You could use any PCB design software, e. g. http://www.cadsoftusa.com/index.htm.en or spice modelling software, e. g. http://www.linear.com/designtools/software/.
Thanks, I'll look into it. Sure would be simpler if there was a fixed-width font for ascii drawings...
In the last measurement we saw how the 5532 (and, it turns out, the NI 4562 and other bjt opamps too) have better symmetry as inverters with the + input grounded. Now we can see it's not so for fet opamps.
Here's the plot for the AD746, a longtime fave of mine. There is a HUGE improvement, even as an inverter, with balanced impedances at both inputs. (The OPA2132 curve looks similar though not as dramatic as the 746, which has higher GBW and OLG.) Listening tests last year showed a clear preference for the balanced config, but harmonic distortion measurements showed no difference.
Looking at this curve, it's very tempting to conclude that the symmetry has more to do with available feedback than it does with power supply Z/phase, especially at high freq's. And the fet opamps appear to have 4-6 dB worse symmetry at 10kHz than the bjt's. Maybe that's what gives the fet units that characteristic "edgy" or "better defined" sound.
Noise in a 30kHz BW is the same whether grounded or balanced, so it's not tracing the noise floor.
This measurement is bringing out some very interesting info, but not yet what I was hoping it would. Yet...
In the last measurement we saw how the 5532 (and, it turns out, the NI 4562 and other bjt opamps too) have better symmetry as inverters with the + input grounded. Now we can see it's not so for fet opamps.
Here's the plot for the AD746, a longtime fave of mine. There is a HUGE improvement, even as an inverter, with balanced impedances at both inputs. (The OPA2132 curve looks similar though not as dramatic as the 746, which has higher GBW and OLG.) Listening tests last year showed a clear preference for the balanced config, but harmonic distortion measurements showed no difference.
Looking at this curve, it's very tempting to conclude that the symmetry has more to do with available feedback than it does with power supply Z/phase, especially at high freq's. And the fet opamps appear to have 4-6 dB worse symmetry at 10kHz than the bjt's. Maybe that's what gives the fet units that characteristic "edgy" or "better defined" sound.
Noise in a 30kHz BW is the same whether grounded or balanced, so it's not tracing the noise floor.
This measurement is bringing out some very interesting info, but not yet what I was hoping it would. Yet...
Attachments
Code:
39pF
|-----| |-----|
|----/\/\/\---|
| |\ 20k |
13k8 | | \ |
in ----/\/\/\--+-|- \ | 75R
| \----+------/\/\/\--- out
------|+ / |
| | | / |
/ | | / /
\ | |/ 10k \
8k2 / - /
\ - 39pF \
| | |
| | |
| | |
------------------------- GndBest thing is a good software, LTspice is KILLER. I can hardly imagine to do anything without it when it comes to electronics design and analysis. No more guessing... or at least very much less. Dislcaimer: many device models, opamps especially, suck, except for the newest ones -- and even those do not model distortion, for good and practical reasons.
Your findings seem to back what people like Walt Jung and others have been pointing out for quite some time: match you impedances and use inverting configs (or bootstrap the supplies with the input pin level to avoid common mode errors if inverting mode is not feasible). Even with bjt inputs this makes sense. Only thing is that we then have a small noise penalty, which I consider less a problem than distortion.
- Klaus
Klaus, thanks for the CODE tags, that at least makes it decipherable.
I'm going to have to pass on learning SPICE right now, I have enough on my plate with this project and normal workflow...
"The LT1469 employs bias current cancellation at the inputs.
The inverting input current is trimmed at zero common
mode voltage to minimize errors in inverting applications
such as I-to-V converters. The noninverting input current
is not trimmed and has a wider variation and therefore a
larger maximum value. As the input offset current can be
greater than either input current, the use of balanced source
resistance is NOT recommended as it actually degrades
DC accuracy and also increases noise."
And that's partly what I see with the 5532 and 4562: increased DC offset and worse symmetry at low freq's with the balancing RC. And so it made me wonder if the DC offset was causing it. In fact, one could imagine that's sort of the underlying concept behind the symmetry error; a frequency-dependent, dynamic DC offset, where the waveform fidelity is preserved (i.e. the signal swing is the same) but is offset with respect to the 0V baseline. Which just happens to be the reference point for the power supplies. Hence my attempt to find some correlation between the two.
(Sorry it's taken so many weeks to come up with a cogent 100-words-or-less explanation. Sometimes it takes the right frame for the words to come together...) 🙂
BTW, DC removal is part of my post-processing. In some cases it's been as much as 50% of the error waveform. But I have chosen to remove it so it doesn't leak into the first few freq bins in the Hanning-weighted FFT.
I'm going to have to pass on learning SPICE right now, I have enough on my plate with this project and normal workflow...
Yes, I also remember Gerald Graeme (sp?) saying so in one of his early books. And I've done it with fet opamps (both inverting and non-) for some time. But then a while back I read this paragraph from the LT1469 data sheet, and it made me wonder about bjt opamps:KSTR said:
"The LT1469 employs bias current cancellation at the inputs.
The inverting input current is trimmed at zero common
mode voltage to minimize errors in inverting applications
such as I-to-V converters. The noninverting input current
is not trimmed and has a wider variation and therefore a
larger maximum value. As the input offset current can be
greater than either input current, the use of balanced source
resistance is NOT recommended as it actually degrades
DC accuracy and also increases noise."
And that's partly what I see with the 5532 and 4562: increased DC offset and worse symmetry at low freq's with the balancing RC. And so it made me wonder if the DC offset was causing it. In fact, one could imagine that's sort of the underlying concept behind the symmetry error; a frequency-dependent, dynamic DC offset, where the waveform fidelity is preserved (i.e. the signal swing is the same) but is offset with respect to the 0V baseline. Which just happens to be the reference point for the power supplies. Hence my attempt to find some correlation between the two.
(Sorry it's taken so many weeks to come up with a cogent 100-words-or-less explanation. Sometimes it takes the right frame for the words to come together...) 🙂
I'd prefer that tradeoff too. But in this case, there wasn't a measurable noise difference. The error in the 5532 measurement is not noise-related. And certainly not in the AD746, whose voltage noise at low freq's is 4+ times higher.
BTW, DC removal is part of my post-processing. In some cases it's been as much as 50% of the error waveform. But I have chosen to remove it so it doesn't leak into the first few freq bins in the Hanning-weighted FFT.
jbau said:
If it is only to make schematics, you will probably need less time to learn that in Spice than it takes to make a schematic from characters.
Now to run some symmetry tests with the LT1085/LM337 pair, which, as we saw back in post #147, are very well matched (the best I've seen yet) and free of the 600Hz resonance that the LM317 has. I ran the impedance curves again and you can find them below. Vset component values for ±15VDC are the same for both regs:
Vin-Vout = 6V
out to adjust : 240 ohms
adjust to gnd : 2.7k ohms || 10uF tantalum
output cap: 1500uF lytic w/ 10uF film
load resistance: 330 Ohms
In the theory of "low, linear, and matched" power supply Z/phase being desirable, this is two of the three. Linear it is not, with a 4-times rise in Z from 100Hz to 20kHz, and corresponding inductive phase from 100Hz on up. It will be interesting to see how this reflects into the symmetry error.
Vin-Vout = 6V
out to adjust : 240 ohms
adjust to gnd : 2.7k ohms || 10uF tantalum
output cap: 1500uF lytic w/ 10uF film
load resistance: 330 Ohms
In the theory of "low, linear, and matched" power supply Z/phase being desirable, this is two of the three. Linear it is not, with a 4-times rise in Z from 100Hz to 20kHz, and corresponding inductive phase from 100Hz on up. It will be interesting to see how this reflects into the symmetry error.
Attachments
And here is the symmetry curve of the 5532A inverter, the exact same setup as before, with grounded + input. The measurement of this exact configuration with the optimized LM317/337 regs is also plotted for easy comparison.
And we see that, despite the tight matching and lower Z values everywhere (and lower noise, too), the symmetry error is WORSE across the board. The only area of improvement is in the 500-1000 Hz region where the LM317 resonates.
So obviously lower Z and tight Z matching is not enough. But since I'm not ready to toss out the "lower, linear, and matched" theory yet, I have to conclude that Zphase IS very important, and maybe the most important of the three. To the point that it's worthwhile (i.e. audibly beneficial) to give up some ultimate lower Z values in order to linearize the Zphase.
So that is what I'll do next. Optimize the adjust circuits for lowest Z and flattest Zphase. And compare that to a simple RC filter that might accomplish the same thing but with a little higher Z values.
It would be really nice if someone who has what they feel to be a superior bipolar power supply would post the Z/phase curves for it under load conditions. There are several guys on this forum who have the equipment to do it. It would be really useful for everyone to see what these other, highly-touted supplies look like impedance-wise.
And we see that, despite the tight matching and lower Z values everywhere (and lower noise, too), the symmetry error is WORSE across the board. The only area of improvement is in the 500-1000 Hz region where the LM317 resonates.
So obviously lower Z and tight Z matching is not enough. But since I'm not ready to toss out the "lower, linear, and matched" theory yet, I have to conclude that Zphase IS very important, and maybe the most important of the three. To the point that it's worthwhile (i.e. audibly beneficial) to give up some ultimate lower Z values in order to linearize the Zphase.
So that is what I'll do next. Optimize the adjust circuits for lowest Z and flattest Zphase. And compare that to a simple RC filter that might accomplish the same thing but with a little higher Z values.
It would be really nice if someone who has what they feel to be a superior bipolar power supply would post the Z/phase curves for it under load conditions. There are several guys on this forum who have the equipment to do it. It would be really useful for everyone to see what these other, highly-touted supplies look like impedance-wise.
Attachments
I was getting burnt out doing nothing but measurements and needed some musical relief. So I linearized the Z/phase on the ±15V regulators that supply the preamp section of my amplifier and then spent a couple days listening. Once again, absolutely stunning results. What a treat.
As I was listening, I thought of the thousands of hours I have spent over decades doing listening tests of various electronic and acoustic devices, without having a bloody clue what the Z/phase behavior of the power supplies behind them was.
Not knowing the Z/phase behavior of the power supplies inserts a HUGE uncertainty into any listening test. Period.
OK, back to measurements. Here's another interesting one. It's the Z/phase curves of the LT1085 regulator at 15V output with various values of adjust capacitors, from 24nF to 10uF. (There was no significant difference with adjust caps larger than 10uF so I didn't plot those.) Since the LT1085 isn't plagued by the resonance that the LM317 is, I lowered the Vset resistances down to the factory recommended value. The Vset components:
Vin-Vout : 6V
out to adjust R : 120 ohms
adjust to ground R : 1.3k ohms
Output load : 330 ohms, 1000uF
If you're not familiar with phase plots in conjunction with their respective magnitude spectra, these plots are a good example of how the phase changes as a function of the impedance distribution. Even though the values don't change that much in an absolute sense (they're all below 30 milliOhms), merely changing the impedance trend and distribution changes the phase (i.e. alters the timing relationships) before the actual impedance changes. That's why calculating things like PSRR and ripple at one frequency doesn't tell the whole story. And it's also why judgments of inductive vs capacitive behavior can't be taken as absolute reality when dealing with regulator outputs. By making some tradeoffs (i.e. optimizing), it can be easily changed. And there is great audible benefit in doing so.
Anyway, the sweet spot looks like it will be a little below the 124nF curve so I'll dial in on that shortly, and then optimize the LM337 to match it as best as I can.
As I was listening, I thought of the thousands of hours I have spent over decades doing listening tests of various electronic and acoustic devices, without having a bloody clue what the Z/phase behavior of the power supplies behind them was.
Not knowing the Z/phase behavior of the power supplies inserts a HUGE uncertainty into any listening test. Period.
OK, back to measurements. Here's another interesting one. It's the Z/phase curves of the LT1085 regulator at 15V output with various values of adjust capacitors, from 24nF to 10uF. (There was no significant difference with adjust caps larger than 10uF so I didn't plot those.) Since the LT1085 isn't plagued by the resonance that the LM317 is, I lowered the Vset resistances down to the factory recommended value. The Vset components:
Vin-Vout : 6V
out to adjust R : 120 ohms
adjust to ground R : 1.3k ohms
Output load : 330 ohms, 1000uF
If you're not familiar with phase plots in conjunction with their respective magnitude spectra, these plots are a good example of how the phase changes as a function of the impedance distribution. Even though the values don't change that much in an absolute sense (they're all below 30 milliOhms), merely changing the impedance trend and distribution changes the phase (i.e. alters the timing relationships) before the actual impedance changes. That's why calculating things like PSRR and ripple at one frequency doesn't tell the whole story. And it's also why judgments of inductive vs capacitive behavior can't be taken as absolute reality when dealing with regulator outputs. By making some tradeoffs (i.e. optimizing), it can be easily changed. And there is great audible benefit in doing so.
Anyway, the sweet spot looks like it will be a little below the 124nF curve so I'll dial in on that shortly, and then optimize the LM337 to match it as best as I can.
Attachments
I am following your results with great interest; thank you so much for your effort. I am quite puzzled though, why, when so many people have reported bad sound with these prepackaged regulators, you limit your measurements exactly to these. I'd love to see what results you might get measuring some of the discrete regulators that have been developed and reportedly have much better sound than the 317 and the like.
Have you looked into the LT3080? It's a five pin, but can be modified for three pin compatibility with the 317.Its claim is that since it uses a completely independent precision current source for programming separate from the voltage follower opamp output stage, allows increased linearity of the regulators output impedance and phase due to isolation of the load from the voltage reference setting stage. I have some extra ones from a project, and would be happy to send one for your testing. I used them in conjunction with the 337 for a +/- 15v supply.
Tom
Tom
ikoflexer said:
I think the point of this thread is that how you implement the 3x7 regs can have a significant effect and this could explain the reports of "bad sound". A more cynical interpretation is that the reports of bad sound from LM317 regs are biased and therefore not worthy of much attention.
One thing my experiments have shown me is that because the 3x7 regs do not have true external sensing they are very sensitive to the connection between the load and the regulator. Using wire that is too narrow a gauge or too long can easily make Zout an order of magnitude higher at the load.
okapi said:
Well then, sorry about the diversion. Cynicism or conspiracy aside, there may be something to that myth. I'm not pushing, just thought it would have been interesting.
True of any regulator, unless remote sensing is an option, and furthermore, is done properly.
Thanks for your interest, ikoflexer.ikoflexer said:
There are many answers to this. The most basic was given in my first post. I have four devices that I want to upgrade the power supplies for. There is not physical room in any of them for anything larger than what's there. So I'm constrained by that.
But beyond that, yes, there are many opinions on this site. Some to be taken more seriously than others. Very few are backed up by measurements of any sort, let alone measurements that might reveal something relevant. I have yet to see any work of this type by anyone. (If I've missed something, please point me to it.) So, bottom line, I trust my own ears and measurements more than I trust the opinions I've read.
I guess it should be said, I haven't been completely transparent. So in case it hasn't been obvious, this isn't the first time I've done this sort of work. It's just the first time I've applied these concepts to bipolar power supplies.
A long-time friend of mine who owns a local recording studio was by today and heard this for the first time, and was blown away. "Tube smoothness with solid state definition" was his comment. He's bringing in a pair of his outboard mic pre's to mod this weekend.
Me too. And I've made several requests for exactly that. Go back a couple posts to read the latest one. But to my surprise noone has responded. If you have one and want to see how it measures, send it to me and I'll gladly run the curves on it.
I have about 4,000 sq ft filled with electronic equipment of all sorts. I have measured the Z/phase curves on 70-80 bipolar supplies now, in audio equipment and in test equipment, some discrete, some not. Some switchers, some linear. They are all over the map performance-wise. The ones with the lowest, flattest Z/phase I have seen are the ones I've optimized from the lowly LM317/337 combo.
Hi Tom, no I haven't. Sounds interesting. What current source do you use with it? I'd be happy to test it.tailspn said:
One thing to consider is the matching of the two regulators. It's not going to be possible to lower the 337's Z much further and keep the phase linear. You can see in the curves that something in the 20milliOhm range is about as low as it can be taken and maintain the good phase behavior.
So if we find a better positive reg, we'd need to find a negative that can match it...
And... it IS possible that we might find that there are good reasons to not have infinitely small impedance on the rails. For example, the lower it is, the smaller the variation is needed to create large phase shifts. I can easily imagine there being a point of diminishing benefit, and the "sweet spot" may not be the lowest one.
jbau said:
Sorry, I must have missed your call to arms. Kept an eye on your thread for more than a week, but missed some of your in depth look at certain issues. I'll have to go back and do a comprehensive read. It would be my pleasure to send you one of my prototypes. I'll contact you offline.
That would be good news for me, as I got a whole bunch of the lmx17 family. I try to keep an open mind and take into account both measurements and subjective experience.
I'm not sure I follow. If the AC component is so small, as to be completely inaudible, would its phase shift still matter?
Perhaps that's where the misunderstanding lies. It's not the voltage AC component that we're worried about, it's the current waveform. And that's what the phase measurement is describing; the altering of the timing relationships of current drawn through the load being measured. And it's relevant whether the Z stands 2 microOhms or 2 Ohms above ground.
This will be clearer if you see how the Z/phase is measured. Back in post #58 there is a link to the Agilent Impedance Measurement Handbook. Take a look at chapter 2, which shows the various impedance measurement techniques, the strengths and shortcomings of each, and a basic block diagram of how it's implemented. The "Auto Balancing Bridge" method is the one most useful for audio. It measures drive voltage and output current across the load by Kelvin connection at each frequency, and calculates the Z and phase from that. I don't remember the formula, that's what computers are for... I coded this stuff many years ago.
Not shown in the book (they DO want you to buy one of their instruments, after all...) is: you can replace the I/V opamp and necessary range switching with a simple current probe, and do the same measurement with a network analyzer.
Like you, I'm keeping my mind (and ears) open to developments as I encounter them. I've seen lots of models that show supplies with vanishingly low impedance values over a wide freq range. But I've never seen those values confirmed in an actual measurement of one that has been built. Models are like opinions. It's easy to cook one up. But I am not convinced until it's built and working and I can measure its performance.
Another day of scorching outdoor temps gives a good excuse to stay inside and work on circuits... 🙂
Here are the Z/phase curves for the optimized LT1085 / LM337 pair. Average impedance level isn't as low as I'd hoped to achieve, but the Z/phase match is still the best I've seen yet. The phase is ±10º up to 10kHz. The component values:
LT1085
out to adjust : 100 ohms
adjust to gnd : 1.12k ohms || 100nF film
output cap/load: 1000uF lytic w/ 10uF film || 330 Ohms
LM337
out to adjust : 120 ohms
adjust to gnd : 1.32k ohms || 39nF film
output cap/load: 1000uF lytic w/ 10uF film || 330 Ohms
The LT1085/LM337 combo is a definite improvement over the LM317/337's, and I see no reason to use the latter unless you have a bunch of 317's laying around that you need to use. From what I've seen so far, this may be as good as it will get for the 3-terminal adjustable regs. I'll report on the sound quality as soon as I get them installed.
Next up for me is a look at the fixed 5V regs that feed the ADC and DAC. There are millions of devices out there that use fixed ±5V regs fed from the 12 or 15 volt rails. The analog portion of these devices will benefit greatly from linearizing the phase, especially the A/D converter. I'll start a new thread for that one.
Here are the Z/phase curves for the optimized LT1085 / LM337 pair. Average impedance level isn't as low as I'd hoped to achieve, but the Z/phase match is still the best I've seen yet. The phase is ±10º up to 10kHz. The component values:
LT1085
out to adjust : 100 ohms
adjust to gnd : 1.12k ohms || 100nF film
output cap/load: 1000uF lytic w/ 10uF film || 330 Ohms
LM337
out to adjust : 120 ohms
adjust to gnd : 1.32k ohms || 39nF film
output cap/load: 1000uF lytic w/ 10uF film || 330 Ohms
The LT1085/LM337 combo is a definite improvement over the LM317/337's, and I see no reason to use the latter unless you have a bunch of 317's laying around that you need to use. From what I've seen so far, this may be as good as it will get for the 3-terminal adjustable regs. I'll report on the sound quality as soon as I get them installed.
Next up for me is a look at the fixed 5V regs that feed the ADC and DAC. There are millions of devices out there that use fixed ±5V regs fed from the 12 or 15 volt rails. The analog portion of these devices will benefit greatly from linearizing the phase, especially the A/D converter. I'll start a new thread for that one.
Attachments
jbau - since you haven't changed the fundamental internal circuit design of the LT1085 and LM337 in these experiments, can you please explain how the impedance remains flat up to 20kHz (and actually decreases at higher frequencies) when the gain of the error amplifier in both these regulators starts falling after about 10Hz?
Also, why are the values of the support ciruit components in the optimised LM337 in post 176 above i.e.:
out to adjust - 240 ohms
adjust to gnd : 2.7k ohms || 33nF
output cap: 1500uF w/ 10uF film
load resistance: 330 Ohms
different to these latest ones i.e.:
out to adjust : 120 ohms
adjust to gnd : 1.32k ohms || 39nF film
output cap/load: 1000uF lytic w/ 10uF film || 330 Ohms
Are they both optimised??
out to adjust - 240 ohms
adjust to gnd : 2.7k ohms || 33nF
output cap: 1500uF w/ 10uF film
load resistance: 330 Ohms
different to these latest ones i.e.:
out to adjust : 120 ohms
adjust to gnd : 1.32k ohms || 39nF film
output cap/load: 1000uF lytic w/ 10uF film || 330 Ohms
Are they both optimised??
Gopher said:
because the output capacitor(s) is in parallel with the regulator. I see essentially the same thing. You don't have to go to much higher frequencies before Zout starts to rise as you would expect.
Gopher said:
My guess is: YES they are both optimized- for the given resistor values. I think that the efficacy of the adjust capacitor is related to the adjust resistor, therefore, a higher capacitance is required when the resistance is lowered.
- Status
- Not open for further replies.