Another look at the LM317 and LM337 regulators - Page 6 - diyAudio
Go Back   Home > Forums > Amplifiers > Power Supplies

Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 29th May 2009, 07:13 PM   #126
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Quote:
SY: I'm still personally unclear on why a flat impedance is more desirable than an impedance which is just low everywhere.
What is "just low everywhere"? That's a pretty imprecise descriptor. Part of the point is, you don't know if your supplies are "just low everywhere" or not. Or if they're equally "just low" or not. It's blind faith that they are. So I'm putting numbers to it. And listening. And hearing a huge difference. And waiting patiently for someone else to do it instead of saying what they think.

But until then, is there a power delivery situation in audio where a flat and low source impedance/phase isn't a good thing? Every one I've ever encountered, nonlinearities in either the source/load were reflected in the output. How can it be otherwise.

Wavetek thinks flat supply impedance/phase is important. Their 288 function generator uses 317/337 with lots of support circuitry to get the system impedance flat. But the way they do it, the 317 is 0.2 ohms higher than the 337. I can post a schematic if you wish.

Quote:
If I could verifiably hear a difference between a circuit being fed from a power supply with a source resistance of 1 milliohm versus 20 milliohms, I'd be more concerned with fixing the circuit.
Well SY, you can test that one out easily. Build the circuit, and listen. It's almost trivially simple.

I am still the only person who has built this and heard what it does. mjf built one that is close, but too many things are different to say if it performs the same or not. One of these days someone will stop just throwing their opinion at this and actually do something. I can tell you that one audio manufacturer will soon do just that. They have been intrigued enough by this project to buy the necessary equipment to test it for themselves.
  Reply With Quote
Old 29th May 2009, 07:26 PM   #127
SY is offline SY  United States
diyAudio Moderator
 
SY's Avatar
 
Join Date: Oct 2002
Location: Chicagoland
Blog Entries: 1
Why would I want to build a circuit which is so poorly designed that a few milliohms in the supply line is audible?
__________________
And while they may not be as strong as apes, don't lock eyes with 'em, don't do it. Puts 'em on edge. They might go into berzerker mode; come at you like a whirling dervish, all fists and elbows.
  Reply With Quote
Old 29th May 2009, 07:49 PM   #128
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
That's called posturing, SY. The fact remains, you don't know. Ignorance is bliss, I suppose.
  Reply With Quote
Old 29th May 2009, 07:52 PM   #129
SY is offline SY  United States
diyAudio Moderator
 
SY's Avatar
 
Join Date: Oct 2002
Location: Chicagoland
Blog Entries: 1
I've built one or two amps and preamps with 317 and 337 based regulators. Just not one that had such poor supply rejection or whatever is causing your issue.
__________________
And while they may not be as strong as apes, don't lock eyes with 'em, don't do it. Puts 'em on edge. They might go into berzerker mode; come at you like a whirling dervish, all fists and elbows.
  Reply With Quote
Old 29th May 2009, 09:04 PM   #130
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
And now even a moderator is trashing the project, so quick to condemn something he has never investigated. Apparently it's not cool to bring a different perspective around here.

How does one put a mod on ignore?
  Reply With Quote
Old 29th May 2009, 09:32 PM   #131
Mr Evil is offline Mr Evil  United Kingdom
diyAudio Member
 
Join Date: Aug 2004
Location: Behind you
Well you haven't done a very good job of explaining why a flat output impedance is desirable. Surely once it is low enough then it ceases to be a problem no matter how much it varies?
__________________
https://mrevil.asvachin.eu/
  Reply With Quote
Old 29th May 2009, 10:22 PM   #132
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
Mr. Evil,

this is academic. A 10mR peak at 5kHz within an overall Zout of 100uR surely doesn't matter at all. It's flat enough because it's low enough.

--------:--------

jbau's point (and mine, too) seems to be that the textbook 317 circuit is far from optimum (and IMHO the datasheet is full of very misleading stuff on component values selection) and there is room for big improvement with little effort and low cost.

I, like maybe many others, have both built and simulated a bunch of 317 circuits (including the EDA test circuit I just simmed now, which gives pretty consistent results with any of the three 317 models I have).

A "typical circuit" with Cadj=0 and Cout=10uF has chances to exhibit a sharp impedance peak at ~5kHz which can go up to tens of ohms. The load step reponse looks correspondigly horrible.

With a bit of tweaking (a few passive parts) one can get a really flat Zout of ~100mR from 10Hz to 100kHz. Here Gopher is right on track, the point is to compensate the 1uH (or thereabouts) output inductance of the basic circuit. Which is easy when you know the LF output resistance... and still easier if you override both with passives to get stable, predictable conditions for this ESL/ESR. It's essentially the same problem as with paralleling different caps, the parameters of the smaller one have to be choosen such as to compensate the ESL of the bigger one. And Cadj basically lowers Zout on the LF side, that's why we need a rather big time constant there. Its ESR/ESL doesn't matter much. Er, this is only my 2cts, YMMV.


Of course all depends on the circuit it powers. Single ended, supply referenced stuff is very likely to have poor PSRR.

-------:---------

jbau, I'd really like to see the Wavetek 3x7 circuits if you are willing to share. There is always something to learn.

- Klaus
  Reply With Quote
Old 29th May 2009, 10:36 PM   #133
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Evil, I thought I'd explained this in bits here and there, but we'll try again. There are many reasons and I probably won't hit them all. Gopher explained one a couple posts back, where multiple devices share the same rails. Add to that, power supplies are ideally constant-voltage, current on demand. But that only exists in theory. They all have finite source impedances which define their ability to deliver current. And it isn't just one number. It's multiple curves that change as a function of current level, frequency, probably other things. The devices drawing current from it look back at the supply through a transmission system that also has non-flat impedance. Current drawn through a source that isn't a flat impedance will modulate the voltage where the impedance is higher. With solid state devices, this happens as frequency increases. Unfortunately, most circuits have poorer PSRR as freqs increase. Add to that, phase matters. Reactance alters timing relationships, so current doesn't arrive when it's demanded at all freqs. That is why I generalize, the tonal response in most systems ends up resembling the impedance curves of the power systems in it.

Add to that, in a bipolar supply system, if the negative supply has different impedance curve than the positive, that's an even greater mess.

I agree with the conviction that, AT SOME POINT, a low-enough impedance will render it all a moot issue.

But where is this point? What is "low enough"? And is it "low enough" across the audio band and maybe a bit beyond? Do you know? If so, how do you know? Noone has answered that yet. And the reason why is simple: they can't, because it's work they have never done.

I'm coming into these questions open-minded because I don't have the answers. I see issues that I've had to grapple with in other systems (speakers, transformers) with great success, and I think I'm approaching it with tools and techniques appropriate to the task. What I have discovered so far, is that flat matters. Low matters. And equal matters. Just how flat, how low, how equal, I don't know.

For my speakers, I have freq response, phase, impedance, harmonic distortion and dynamic linearity data for them so I know how they behave. This is the first time I've ever listened to a program source of any kind where I knew the Z characteristics of it's analog supplies. Up until now it's been an undefined variable.

I'm going to give a little more time to this project, and then I need to install these into the pieces that I'm upgrading and move on to the next project. Time management also matters. And sitting here fighting off the crap being slung at me by people who should know better feels like a huge waste of time. It's exhausting.
  Reply With Quote
Old 29th May 2009, 11:07 PM   #134
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
Quote:
Originally posted by jbau
But where is this point? What is "low enough"? And is it "low enough" across the audio band and maybe a bit beyond? Do you know? If so, how do you know? Noone has answered that yet. And the reason why is simple: they can't, because it's work they have never done.
Now this depends so much on the powered circuit that any generalizations seem impossible. Say we have a standard opamp circuit which will operate in class A/B. Then we have rectified load currents in the rails, which can have pretty sharp edges depending on how sharp the A/B-transition is. Hence the bandwidth of the supply should be significantly higher than the audio range, I normally try to achieve a x10 factor, 200kHz. And the region above that (up to a bit beyond the GBW of the opamp) must be handled by local decoupling at any rate (preferably L-C or Ferrite-C, not a C alone -- the parallel cap issue again).

The +/-rail asymmetry can be handled by using identical circuits for both rails, requiring individual secondarys and bridges/filters of course and only feasible with a single global supply.

- Klaus
  Reply With Quote
Old 29th May 2009, 11:26 PM   #135
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
KSTR,
I don't have the luxury of using a discrete supply in these equipments. I'm stuck with a 3-term reg, that's why I'm doing this.

I've attached the +-22V supply drawing from the Wavetek 288, the 12V circuits are the same except some different resistor values.
Attached Images
File Type: jpg 288ps.jpg (46.1 KB, 1741 views)
  Reply With Quote
Old 29th May 2009, 11:39 PM   #136
KSTR is offline KSTR  Germany
diyAudio Member
 
KSTR's Avatar
 
Join Date: Jul 2007
Location: Central Berlin, Germany
jbau,
Thanks for the schematic, I'll dig into this, looks interesting, a feedback tracking reg....
  Reply With Quote
Old 30th May 2009, 01:01 AM   #137
SY is offline SY  United States
diyAudio Moderator
 
SY's Avatar
 
Join Date: Oct 2002
Location: Chicagoland
Blog Entries: 1
jbau, just a procedural point- when someone is posting as a moderator, the cop icon is shown. No cop icon, it's just another member.

I think I asked a pretty reasonable question- why would you want to use a circuit with terrible PSR? If you want the output impedance to be flat, stick a 0R5 resistor on the output.

And seriously, do get the Dietz paper I referenced earlier. He covers very thoroughly the points you're trying to explore here. Pease elaborates on it in his superb book (which you should also have, and which has the Dietz paper as an appendix).
__________________
And while they may not be as strong as apes, don't lock eyes with 'em, don't do it. Puts 'em on edge. They might go into berzerker mode; come at you like a whirling dervish, all fists and elbows.
  Reply With Quote
Old 30th May 2009, 01:07 AM   #138
Account disabled at member's request
 
MJL21193's Avatar
 
Join Date: Mar 2007
Quote:
Originally posted by jbau


How does one put a mod on ignore?

I've added SY to my buddy list...
Attached Images
File Type: png 1116.png (29.2 KB, 1606 views)
  Reply With Quote
Old 30th May 2009, 01:17 AM   #139
SY is offline SY  United States
diyAudio Moderator
 
SY's Avatar
 
Join Date: Oct 2002
Location: Chicagoland
Blog Entries: 1
I never noticed there was such a thing. Cool. I'll eat a pancake in your honor.
__________________
And while they may not be as strong as apes, don't lock eyes with 'em, don't do it. Puts 'em on edge. They might go into berzerker mode; come at you like a whirling dervish, all fists and elbows.
  Reply With Quote
Old 30th May 2009, 01:28 AM   #140
okapi is offline okapi  United States
diyAudio Member
 
okapi's Avatar
 
Join Date: Aug 2004
Location: Buffalo
Quote:
Originally posted by SY
jbau, just a procedural point- when someone is posting as a moderator, the cop icon is shown. No cop icon, it's just another member.

I think I asked a pretty reasonable question- why would you want to use a circuit with terrible PSR? If you want the output impedance to be flat, stick a 0R5 resistor on the output.

And seriously, do get the Dietz paper I referenced earlier. He covers very thoroughly the points you're trying to explore here. Pease elaborates on it in his superb book (which you should also have, and which has the Dietz paper as an appendix).
Hi Sy, i don't want to put words in jbau's mouth but i think one of the important aspects of his argument is that the impedance not only has to be low enough but also benefits from being matched on the positive and negative rails.

An alternative idea is that when it Zout is not low enough matching becomes important.

Earlier, i suggested putting a resistor in series with one of the legs of the PS to deliberately unmatch the positive and negative sides and then repeat the THD test or a different test. I still think this is a good idea.

Would you also mind commenting on what you consider "low enough". If this is too general a question could you comment on some of the design parameters that allow you to determine what is low enough?

One final point for this post, in this thread i have commented several times about the effect of ESR (gopher has repeated this point) mainly because i have read the dietz article. My understanding is that the changes jbau has made are not that sensitive to ESR. Jbau, please correct me if i am wrong.
  Reply With Quote
Old 30th May 2009, 01:40 AM   #141
SY is offline SY  United States
diyAudio Moderator
 
SY's Avatar
 
Join Date: Oct 2002
Location: Chicagoland
Blog Entries: 1
"Low enough" depends on the circuit, the draw, the expected variation in current, and (most of all) the PSR. One can't say, "It needs to be 116 milliohms" as a universal.

My preamp (ImPasse) which uses a 317 regulator, draws 36mA, with maximum signal current swing of perhaps 1.5mA at full output (20V). If there were a 0.5R source impedance, that would represent a 0.75mV signal on the rail, which is 89dB down from the signal, assuming zero PSR. If the output stage were configured with a CCS in the cathode, the PSR would be on the order of 90dB. I think power supply contribution to the output of -180dB is probably acceptable.
__________________
And while they may not be as strong as apes, don't lock eyes with 'em, don't do it. Puts 'em on edge. They might go into berzerker mode; come at you like a whirling dervish, all fists and elbows.
  Reply With Quote
Old 31st May 2009, 08:18 AM   #142
diyAudio Member
 
Join Date: May 2008
Very interesting thread.

Quote:
Originally posted by jbau
And I want to see the + and - supply be as equal as possible.
Did you consider to look at the combined impedance of power supply and amplifier? Maybe a certain imbalance of the power supply could mirror-image a possible imbalance of the amplifier's + and - circuit's impedances and improve the complete system even further.

Quote:
Originally posted by jbau
I'd like to find a positive regulator that was a better match for the good LM337.
How about using the LM337 for the positive rail? With high-current regulators a positive regulator is often used for the negative rail. The opposite will work as well and could give you the perfect match.
__________________
If you've always done it like that, then it's probably wrong. (Henry Ford)
  Reply With Quote
Old 31st May 2009, 10:07 AM   #143
Gopher is offline Gopher  United Kingdom
diyAudio Member
 
Join Date: Nov 2005
Location: UK
Can someone please post a scan of the Errol Dietz paper. Can't find it anywhere on the net.
  Reply With Quote
Old 31st May 2009, 10:48 AM   #144
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Send me an email, I'll send you a PDF of it.
  Reply With Quote
Old 31st May 2009, 11:05 AM   #145
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Quote:
by pacificblue
Very interesting thread.
Yeah, if we can get those who aren't really interested to back off, we may get somewhere...

Quote:
Did you consider to look at the combined impedance of power supply and amplifier? Maybe a certain imbalance of the power supply could mirror-image a possible imbalance of the amplifier's + and - circuit's impedances and improve the complete system even further.
Sure, that would be a logical followup, but I think we need to nail down this one first. Gotta know the ground you stand on before you walk on it... (literally)
Quote:
How about using the LM337 for the positive rail? With high-current regulators a positive regulator is often used for the negative rail. The opposite will work as well and could give you the perfect match.
Okapi sent me a LT1085 to test, and it turns out to be a very good match with the LM337. I'll post some graphs later, they're on a different computer.
  Reply With Quote
Old 2nd June 2009, 01:56 AM   #146
diyAudio Member
 
Join Date: Jul 2003
Location: Columbus, Mississippi
Quote:
Originally posted by jbau


<SNIP>
Okapi sent me a LT1085 to test...
<SNIP>
And I FINALLY sent the LT1083 for you to test... sorry for the delay.

Greg in Mississippi
__________________
Everything matters!
  Reply With Quote
Old 2nd June 2009, 02:11 AM   #147
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
As mentioned above, the LT1085 and LM337 mate up really well impedance-wise, see the curves below. I've run curves on some 30 power supplies now, mostly in situ, and this is the best match I've seen yet. This combo should sound really good, though the phase is only flat to 400Hz. Both have the same 240 ohm / 2.7k ohm Vset resistors with 10uF adjust caps, into a 1500uF and 330 ohm loads.

I wish I had two identical of something so I could do careful listening comparisons with the optimized 317/337 pair.

General observations on the LT1085:
The 1085 only needs about 45mA of load to get it into it's linear zone.

The adjust cap affects a much larger range than it does on the LM317, from say 120Hz up to 5-6kHz. Caps larger than 10uF showed no improvement.

Because of the tight matchup, I plan on using this pairing for future tests on the effects of supply imbalance. But I'm inclined to install the LM317/337 into my gear due to the better phase characteristics of that pair.
Attached Images
File Type: gif 1085337.gif (8.1 KB, 1462 views)
  Reply With Quote
Old 12th June 2009, 05:44 AM   #148
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Here's a progress report on work I'm doing to develop a measurement that might show the effects of Z / phase imbalance on signals processed by devices on the supply, hopefully something that might correlate with what we can hear.

My favorite stumulus for impulse and polarity-sensitive testing is the haversine pulse. (Google 'haversine pulse' and you'll see it's widely used.) Unlike a square-edged pulse, it has an inherent lowpass characteristic that generates less aliasing errors. Subtle symmetry tweaks give a notch-free passband of 2 times the pulse frequency that is approx. 6dB down at the pulse frequency. So a 10kHz haversine pulse gives useful info up to 20kHz. Place it in the center of a hanning-weighted sampling window for best results.

My idea is to pulse a circuit, first with a positive haversine, then with a negative. Add them together. If they are equal, then they would cancel each other out. If not, then the signal that remains shows their difference, including phase differences. FFT the difference signal and see the difference spectrum in the freq domain. Then compare that to the impedance overlays of the power supply and look for correlations.

I have digital function generator, the Exact 625, that does haversine AND invert AND symmetry offsets in the digital domain. And it can be externally triggered for synchronous sampling with averaging. Perfect for the job, I thought. This would give excellent repeatability and no change in the analog circuits for the + and - pulse.

Since the control for this experiment would be the output of the generator itself, I fine-tuned the difference measurements on the 625 until I got stable, repeatable results. I wasn't expecting to see anything special. But what I saw was surprising. See the curve below (sorry for the reflections...) (Oh, it's a linear freq scale, as noted in the pic.)

The two pulses are far from equal. Only -66dB difference at around 6kHz, and only going down as low as -73dB at lower frequencies. Then I saw the ripples around 300-600Hz. Can it be? A difference spectrum with features that reminded me of what I had seen in the LM317/337 Z / phase curves? Hmmm...

So I removed the top to the generator, looked inside, and there they are - LM317 and LM337 regs for the 15V analog circuits, with Vset values on both that are close to the factory values: 247 ohms / 2.5k ohms, with a 6.8uF adjust cap, and a 6.8uF output cap. It also had some kind of tracking arrangement using opamps but I don't have schematics and do not want to strip it down to see more.

At first, I was disappointed. I thought I'd found the perfect signal source for my test. But then, I realized; this unit can be the control AND the DUT in one unit. How much better a control can you get than a DSP generated signal? The unit's analog signal path IS the device under test. All that is needed is the Z / phase curves of the 625's supplies to correlate with.
Attached Images
File Type: jpg ex25a.jpg (68.1 KB, 1284 views)
  Reply With Quote
Old 12th June 2009, 05:48 AM   #149
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Below is the overlaid Z / phase curves of the 625 regs outputs, looking very much like the other 317/337 curves I've measured. (The spike up at the far left in the blue curve on the Z graph was a data collection error, I had a wrong "goto" branch in my program that didn't catch it.)

Compare this to the difference spectrum in the previous post. Looking for correlations, I've marked off two obvious ones.
: The bump at 325Hz <-> the middle of the phase gap at 325Hz.
: The ripple at 590Hz <-> the Z null and phase departure point at 590Hz.
: As the Z and phase is rising, the difference spectrum becomes worse.

It's late and I have to wrap up for now. For a first attempt, the technique and the results are encouraging. I hope to do more on it this weekend.
Attached Images
File Type: gif exact625.gif (8.7 KB, 1184 views)
  Reply With Quote
Old 17th June 2009, 12:00 PM   #150
jbau is offline jbau  United States
diyAudio Member
 
Join Date: Apr 2009
Coming up for are long enough to give another status report. Well this is proving to be quite the challenge. Several things have become apparent since my last post. First, the tools are bent. The instruments I'm using to isolate this phenomena are contaminated by the same condition that I'm trying to measure. It turns out the plot two posts back does not just show the errors of the haver pulser, it also has the errors of the waveform analyser.

Even high-end analysers/digitizers are showing significant asymmetry errors when sampling positive and negative pulses. By significant I mean + to - pulse differential error of 70-ish db down across broad parts of the audio spectrum. That's worse than 12-bit performance. And (surprise, surprise...) their bipolar power supplies are not impedance balanced.

Another major issue is trigger jitter. Getting accurate phase data from time domain measurements is totally dependent on consistent trigger response. When the digitizer gets the trigger edge from the pulser, the time delay from receipt of trigger must be always the same to accurately identify the start of the data record in the sampling stream. Inside the digitizer, if the trigger input is being monitored by a cpu in a software loop, the trigger uncertainty is amplified by a host of factors regarding the execution time of the software loop. (That eliminates sound cards from being useful here.) And even interrupt-driven triggers have uncertainty greater than one sample interval. HP is the only company I know of that has dealt with this issue in it's digitizers, even as far back as the HP 3582A. Unfortunately, most of their analysers are only 12 bits and that's probably not enough for this job. Their current, higher-res units would require a second mortgage on the house and that ain't gonna happen...

I've come up with ways to deal with the asymmetry issue, and I'll be posting those later. Whether or not they're sufficient hasn't been determined yet.

This website needs a separate forum for measurement techniques. What do you think, mods? This measurement isn't limited to power supply stuff. It feels stupid to be tacking it on to a thread about 3-terminal regulators. I'd like stop posting it here and start it up in another, more appropriate area of the board. Any suggestions?
  Reply With Quote

Reply


Hide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off



New To Site? Need Help?

All times are GMT. The time now is 10:26 AM.


vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2014 DragonByte Technologies Ltd.
Copyright 1999-2014 diyAudio

Content Relevant URLs by vBSEO 3.3.2