Assuming we want to keep the mean Z value down in the 20-25 milliohm range, the buffer R would either be .015 or .020 ohms. So let's look at those with the 330uF adjust cap and with output caps chosen to give the flattest Z in each case. This will also verify the Z division action I described earlier.
The yellow curve is with a .015 buffer R and 1000uF output cap.
The cyan curve is with a .020 buffer R and 470uF output cap.
Either one would probably do fine.
In order to flatten the Zphase all the way up to 20kHz, it looks like a buffer R value of around 40-50 milliOhms would be needed, and output C would probably come down to the 220uF range. I may play with that later.
Also I need to re-optimize the LM337 circuit with this same approach.
Impedance analysis is a sweet tool for fine-tuning power supply characteristics for audio.
The yellow curve is with a .015 buffer R and 1000uF output cap.
The cyan curve is with a .020 buffer R and 470uF output cap.
Either one would probably do fine.
In order to flatten the Zphase all the way up to 20kHz, it looks like a buffer R value of around 40-50 milliOhms would be needed, and output C would probably come down to the 220uF range. I may play with that later.
Also I need to re-optimize the LM337 circuit with this same approach.
Impedance analysis is a sweet tool for fine-tuning power supply characteristics for audio.
Attachments
where to locate the output series resistor?
Hi,
it could be before the tapping for the Radjust or after the Radjust.
What does the sym say about moving the resistor in front of the Radjust tapping point?
Or maybe experiment with splitting the resistor, 0r01 before and 0r01 located after.
How long does the PCB trace need to be to mimic these resistances?
There a new rub: My Stereo sounds far better when I change the default minimum trace width from 0.2mm to 0.15mm!!
Now that the reg output is isolated from the esr of the output cap, can the output cap be a low esr type or a film type with a parallel very large electrolytic?
Hi,
it could be before the tapping for the Radjust or after the Radjust.
What does the sym say about moving the resistor in front of the Radjust tapping point?
Or maybe experiment with splitting the resistor, 0r01 before and 0r01 located after.
How long does the PCB trace need to be to mimic these resistances?
There a new rub: My Stereo sounds far better when I change the default minimum trace width from 0.2mm to 0.15mm!!
Now that the reg output is isolated from the esr of the output cap, can the output cap be a low esr type or a film type with a parallel very large electrolytic?
Re: where to locate the output series resistor?
I have the Radjust right at the reg output. The data sheets are pretty explicit about doing that. But I agree, it might be worthwhile to see it's effects. Good idea.
Pretty long, I think. I thought of that too. I doubt if you want the output caps so far away from the regs.
I still have 10uF of film across the values I'm showing. I haven't kept repeating it, perhaps I should have.
This is not a sym, it's measurements.
I have the Radjust right at the reg output. The data sheets are pretty explicit about doing that. But I agree, it might be worthwhile to see it's effects. Good idea.
Pretty long, I think. I thought of that too. I doubt if you want the output caps so far away from the regs.
I still have 10uF of film across the values I'm showing. I haven't kept repeating it, perhaps I should have.
Re: where to locate the output series resistor?
http://circuitcalculator.com/wordpress/2006/01/24/trace-resistance-calculatorAndrewT said:
Re: where to locate the output series resistor?
You probably just stumbled across a trick well known in some circles:
http://www.edn.com/archives/1997/070397/14df_03.htm
AndrewT said:
You probably just stumbled across a trick well known in some circles:
http://www.edn.com/archives/1997/070397/14df_03.htm
Hi
C=L_out/R_out^2
and it needs to have just the same ESR as given by R_out. And a pretty low ESL of course.
When you don't have equal R's then this modifies to
C=4*L_out/(R_out+ESR)^2
Still zero ringing but a slight variation in effective Z_out in the crossover zone between the two conjugate impedances (which are not perfectly conjugate this time).
If there is signifcant ESL then the same process can be used to cancel its effect, with the proper parallel cap. Gets pretty tricky as the formula is very sensitive to small changes in the denominator because of the squaring. In doubt better go on the larger side of the C value and add known additional ESR. All this also applies to the selection of additional local bypass caps at the load -- don't build resonant tank circuits 😉
When one can live with a resulting impedance of 50..100mR then this all can give a ruler flat Zout way to 100kHz and above -- right at the cap. Layout/routing will have its share, though.
**) L_out would be in the ballpark of 1...10uH with most regs.
- Klaus
If you happen to know/measure the equivalent L_out**) of the reg (sans output cap) and likewise the R_out -- both of them overruled/dominated by external components or board trace properties to stabilize these values -- then the ideal C_out is given byjbau said:
C=L_out/R_out^2
and it needs to have just the same ESR as given by R_out. And a pretty low ESL of course.
When you don't have equal R's then this modifies to
C=4*L_out/(R_out+ESR)^2
Still zero ringing but a slight variation in effective Z_out in the crossover zone between the two conjugate impedances (which are not perfectly conjugate this time).
If there is signifcant ESL then the same process can be used to cancel its effect, with the proper parallel cap. Gets pretty tricky as the formula is very sensitive to small changes in the denominator because of the squaring. In doubt better go on the larger side of the C value and add known additional ESR. All this also applies to the selection of additional local bypass caps at the load -- don't build resonant tank circuits 😉
When one can live with a resulting impedance of 50..100mR then this all can give a ruler flat Zout way to 100kHz and above -- right at the cap. Layout/routing will have its share, though.
**) L_out would be in the ballpark of 1...10uH with most regs.
- Klaus
Andrew, taking Radjust after Rbuffer reverts it to Rbuffer=0 behavior, but with worse Z/phase above 10kHz. It might be effective if Rbuffer is an impedance-shaping RC like seen here:
http://www.diyaudio.com/forums/showthread.php?postid=1841178#post1841178
But the C values across a 20mOhm R would be huge.
I couldn't resist the temptation to see if this buffered output technique would bring any improvement to the Nat Semi LM317. The answer is, no, not yet. It just exposes some of it's weaknesses even more. Maybe with Rbuffer at .050 it would do OK.
The LM317 is looking more like a saddle without a horse to put on. It's unsuitable for audio, and an LM78xx will do just as well for digital circuits.
http://www.diyaudio.com/forums/showthread.php?postid=1841178#post1841178
But the C values across a 20mOhm R would be huge.
I couldn't resist the temptation to see if this buffered output technique would bring any improvement to the Nat Semi LM317. The answer is, no, not yet. It just exposes some of it's weaknesses even more. Maybe with Rbuffer at .050 it would do OK.
The LM317 is looking more like a saddle without a horse to put on. It's unsuitable for audio, and an LM78xx will do just as well for digital circuits.
Attachments
As I expected, the LT1085 and LM337 are very well matched with this buffered-output technique. Here is the plot overlay. The values used are the same on both regs: 0.020 Ohm Rbuffer, 330uF adjust capacitor, 1000uF output cap, and 330 Ohm load resistor. As before, a 470uF Cout will give a flatter Zout. Or you can use the 0.015 Rbuffer with 1000uF Cout. Your choice.
Another advantage: the output Z/phase doesn't change with different output voltage. Changing the adjust-to-ground resistor for 12VDC output gave the same Z/phase curves on both regs with no other component changes required. Very nice.
Who will be the first (besides me) to put one of these together and listen?
Another advantage: the output Z/phase doesn't change with different output voltage. Changing the adjust-to-ground resistor for 12VDC output gave the same Z/phase curves on both regs with no other component changes required. Very nice.
Who will be the first (besides me) to put one of these together and listen?
Attachments
who said the trace length would be long?
5mm of 1oz trace 0.1mm wide ~ 0r02 (170mA rating)
16mm of 2oz 0.2mm wide ~ 0r02 (half amp rating)
If traces this short are giving these values of resistance and your work is showing that this order of resistance measurably changes the performance of the regulators then that might explain why we all get different sounding results.
We forget to tell the simulator the real impedances in the leads and ground legs and inadvertently build a different circuit each time we etch/specify a PCB.
5mm of 1oz trace 0.1mm wide ~ 0r02 (170mA rating)
16mm of 2oz 0.2mm wide ~ 0r02 (half amp rating)
If traces this short are giving these values of resistance and your work is showing that this order of resistance measurably changes the performance of the regulators then that might explain why we all get different sounding results.
We forget to tell the simulator the real impedances in the leads and ground legs and inadvertently build a different circuit each time we etch/specify a PCB.
If we did know all the real impedances and component tolerances would not dwarf them by comparison, some of us would probably go to the lengths of hacking them all in.AndrewT said:
Many shops don't even specify, where an IC or other component comes from, but jbau found that the same regulator from another manufacturer makes a difference. How about the same components, when they come from different factories of the same manufacturer or were produced in different years, e. g. after a change in the production method.
Then again the quality of a solder joint might render all the effort useless anyhow, if we are talking about the difference between 15 mOhms and 20 mOhms.
that's the point.pacificblue said:
Jbau's experimentation is showing us that a very few ten's of milliohms makes a significant difference to the way 3pin regulators perform.
I certainly was not aware of how sensitive our regs are to PCB trace resistances.
I wonder how many others knew what Jbau is showing really happens?
Jbau,
thanks, keep the show running for as long as you can manage.
I believe you are doing us a service.
One comment I do wish to make.
Initially, I did not want to believe what you were saying about trying the very low capacitance on the adjust terminal. Now, I see a complete about turn with high adjust capacitance.
I see why and you have taken us along a long route of knowledge. We select what suits for each loading case. I had always assumed that high capacitance was right because I had tried it and it had worked for me. It now appears I was lucky. I must have inadvertently included trace resistance and my lack of knowledge led me to believe this was the only way.
You are showing there are many ways, if one really understands what is happening inside these ICs.
AndrewT said:
Totally agree. I guess I should say, my test fixture is not on a PC board, it's point-to-point wired but well laid out, so I can't take advantage of trace characteristics. Is a thin copper trace on substrate more thermally stable than a 3W resistor? I don't know the answer offhand. I think we'll see the same sensitivity with the wiring used for remote load sensing.
pacificblue said:
Agreed again. If I were designing for production as I did in the past, this would all have to be statistically accounted for from the start and QC procedures/measurements specified for production.
I do see patterns emerging over the years. There are exceptions (NEC was a noteable one), but generally, second sources of a particular device are usually inferior in one way or another. And in general, Nat Semi fares quite well in historical consistency. I've tested LM317's and 337's with date codes from 1984 to 2006 and they are quite consistent.
Andrew, thanks for your comments. I think part of the problem has been a focus on the LM317 in this device family. How many articles have been written about it, and the assumption has been, the LM337 is the same? When in fact, the 317 is the worst of them.
Also, the resistance is most critical between the reg output and the paralleled output caps/load resistor. After that, it's a different ballgame, and the same no matter what regs are used.
This also brings up the topic of "making measurements that matter" but we can leave that for another time.
I stayed up to the wee hours last night and installed the circuit of post #229 (but with 470uF output caps) into the Sony DAT. I never draw final conclusions listening at those hours, but initial impressions were mostly positive. No question, holography has suffered a bit. Image size is very large but not as convincing front-to-back. And some of the "electronic edge" is back. But the clarity and detail of the bass and mids was impressive. I'll add more impressions later this morning.
Have been listening for the last hour or so, and loving it. Previous comments generally hold true to this morning, though holography is better than I had described. It is overall much better but basically falls apart in the upper octaves, like a picture window that is leaning forward at the top.
But the true standout is power, control, and articulation of the bass and mids, it's really remarkable. It is simultaneously much more transparent AND visceral in the lower ranges. On one selection with a prominent kick drum, it always sounded good but now is better defined and pins me to the sofa in a way it never did. (I believe a formerly prominent audio reviewer used to call this the "slam factor.") On another, a live recording I made of a women's ensemble in a church, on one selection, there was a little kid sitting below the mics and to the right, who got fidgety during one piece and would kick the pew from time to time while they sang. How clearly this stands out now, more tonally correct and located properly in space at the bottom right. Good stuff.
Once again, what I am hearing best correlates not with the Z, which is quite linear up to 10kHz, but with the Zphase, which breaks upward above 2kHz. I'll work on extending that upward today. If we can get that tamed, we'll have a superb set of regulators.
Another thing I'm going to play with is the ground sense wiring. The Sony DAT power supply is on a separate board with about 12" of wiring connecting it to the main board. It strikes me that the inductance of the ground sense wires must be made as low as possible for it to be effective.
But the true standout is power, control, and articulation of the bass and mids, it's really remarkable. It is simultaneously much more transparent AND visceral in the lower ranges. On one selection with a prominent kick drum, it always sounded good but now is better defined and pins me to the sofa in a way it never did. (I believe a formerly prominent audio reviewer used to call this the "slam factor.") On another, a live recording I made of a women's ensemble in a church, on one selection, there was a little kid sitting below the mics and to the right, who got fidgety during one piece and would kick the pew from time to time while they sang. How clearly this stands out now, more tonally correct and located properly in space at the bottom right. Good stuff.
Once again, what I am hearing best correlates not with the Z, which is quite linear up to 10kHz, but with the Zphase, which breaks upward above 2kHz. I'll work on extending that upward today. If we can get that tamed, we'll have a superb set of regulators.
Another thing I'm going to play with is the ground sense wiring. The Sony DAT power supply is on a separate board with about 12" of wiring connecting it to the main board. It strikes me that the inductance of the ground sense wires must be made as low as possible for it to be effective.
Good stuff, thanks for sharing!
Two quick questions.
1) Exactly what kind of load was powered with the +/- regulators in your listening tests. Don't need name/brand, but qualitative description, pointer to schematic even better, if in public domain.
2) Sorry about the triviality of this one, but for my own edification, when you say Zphase, is this the phase angle between the load current and the voltage across the load?
Thanks!
Two quick questions.
1) Exactly what kind of load was powered with the +/- regulators in your listening tests. Don't need name/brand, but qualitative description, pointer to schematic even better, if in public domain.
2) Sorry about the triviality of this one, but for my own edification, when you say Zphase, is this the phase angle between the load current and the voltage across the load?
Thanks!
Hi iko,
Described it in a early post. It's a Sony 670 DAT recorder that I've modified heavily over the years. I have three others from the same era and the circuits are pretty much the same on them.
Just to be clear, I didn't physically install the test fixture supply into the Sony; I modified it's power supply circuits to match. The only exception is the load resistors. Quiescent current draw is already about 150mA on the (+) and 100mA on the (-) rail, so there's no need for any load resistors.
Here's the texbook definition:
"The total opposition that a circuit presents to an alternating current, equal to the complex ratio of the voltage to the current in complex notation. Also known as complex impedance."
In the context of a constant-voltage power supply, there is NO AC voltage being delivered, so the Zphase represents the timing relationship of the AC current delivery.
ikoflexer said:
Described it in a early post. It's a Sony 670 DAT recorder that I've modified heavily over the years. I have three others from the same era and the circuits are pretty much the same on them.
Just to be clear, I didn't physically install the test fixture supply into the Sony; I modified it's power supply circuits to match. The only exception is the load resistors. Quiescent current draw is already about 150mA on the (+) and 100mA on the (-) rail, so there's no need for any load resistors.
No problem. I know from past experience, if your main reference point is voltage waveforms on a scope, Zphase is like a Zen koan.
Here's the texbook definition:
"The total opposition that a circuit presents to an alternating current, equal to the complex ratio of the voltage to the current in complex notation. Also known as complex impedance."
In the context of a constant-voltage power supply, there is NO AC voltage being delivered, so the Zphase represents the timing relationship of the AC current delivery.
Measurement burnout
I'm hitting that wall again. Time to step away from the bench for a day or two.
I didn't work on extending the Zphase flatness today. But I did experiment with the ground sense leads, and I've decided to remove them and connect the adjust RC to the local power supply ground, for three reasons.
1. These regulators don't have sufficient bandwidth in the error amp to be effective over the entire audio band. So using it just accentuates an already compromised situation.
And I remember now, I ran into this very issue back when I was rebuilding analog recording consoles. Many of them used off-the-shelf modular power supplies. And I never saw one that used the remote rail or load sensing. Once I asked a factory tech why. He said, it doesn't have the bandwidth to be effective, and remote sensing only works at the sense point, and not if the load is distributed.
2. The self-inductance of even a 12" sense wire definitely affects the sound. Today I compared 18 ga., 22 ga., and 26 ga. sense wires. Obviously, the 18 ga. has much lower DCR but the Zphase shift at 10kHz in an 18 ga wire is much higher than the 26 ga., because the reactance is a greater proportion of it's total impedance up there, and is in series with a big capacitor. Another case of making things worse while trying to solve a problem.
3. The copper used to remote load sense would be better used in parallel with the existing supply or ground leads to lower their inductance.
It all makes sense to me. (pun intended)
I'm hitting that wall again. Time to step away from the bench for a day or two.
I didn't work on extending the Zphase flatness today. But I did experiment with the ground sense leads, and I've decided to remove them and connect the adjust RC to the local power supply ground, for three reasons.
1. These regulators don't have sufficient bandwidth in the error amp to be effective over the entire audio band. So using it just accentuates an already compromised situation.
And I remember now, I ran into this very issue back when I was rebuilding analog recording consoles. Many of them used off-the-shelf modular power supplies. And I never saw one that used the remote rail or load sensing. Once I asked a factory tech why. He said, it doesn't have the bandwidth to be effective, and remote sensing only works at the sense point, and not if the load is distributed.
2. The self-inductance of even a 12" sense wire definitely affects the sound. Today I compared 18 ga., 22 ga., and 26 ga. sense wires. Obviously, the 18 ga. has much lower DCR but the Zphase shift at 10kHz in an 18 ga wire is much higher than the 26 ga., because the reactance is a greater proportion of it's total impedance up there, and is in series with a big capacitor. Another case of making things worse while trying to solve a problem.
3. The copper used to remote load sense would be better used in parallel with the existing supply or ground leads to lower their inductance.
It all makes sense to me. (pun intended)
What music are you listening to? I'd like to listening to the same music to get some feeling. What speakers are you using? DIY or a certain brand/model?jbau said:
The womens ensemble CD is this one:soongsc said:
http://www.singers.com/7303c/Santa-Fe-Women's-Ensemble/Voices-Of-Light/
This was their first recording and some of the performances aren't the best. And of course I wish I had this power supply configuration in my mic preamp and DAT recorder back then... 🙂
The comment about visceral bass was about Sting's "Seven Days". I would never have classified it as a "slam factor" recording until now. There are many good ones for this. Annie Lennox' "Legend in My Living Room" is a good one for slam and power supply sag. And the kick drum on Shawn Colvin's "The Facts About Jimmy" is really deep and superbly recorded. Clearmountain's mixing work took a big step forward when he finally stopped mixing on Yamaha NS10's... ugh.
Almost my entire CD collection got ripped off a few years ago and it's only 1/4 of what it was then.
Spica TC-60 .
Yes.... 😀
- Status
- Not open for further replies.