Super Regulator, collecting the facts

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Bts

I agree it's normal to have the LM317 ref decoupled, my point was that if this makes such a big difference to the sound quality for the whole regulator then something downstream is sensitive to this change - and I think this is where Fred is coming from.

Dave

BTS - Bye, Tata, Seeyou.
 
Isource improvement(s)

I replaced the 10K resistor bias for the current source LED with a jfet current source. This should result in about a 45dB improvement in LED voltage change with respect to input voltage change. Replacing the 10K with two 5K resistors with a cap to ground might be an even better approach.

These comments are most germane to the Audio Electronics 4/2000 article, and also indirectly to Andy Weekes' (ALW) PCB implementation upon which Fred D. is commenting above.

Shortly after the AE 4/2000 article was published, I received an email from Russell Twining, who suggested an alternative of essentially what Fred D. tried to say above (but Fred got it got wrong). Russell suggested that instead of an LED a 2.5V LM336 could be used. I agree with this point entirely, it would be a better solution that making the 10k feed resistor a JFET Isource (expensive!). Note that the ED 6/23/1997 regulator design referenced in post #350 uses this approach, with a 1.2V diode.

Yes, one could say that the current source circuits as published in AE 4/2000 have a weakness. I did in fact allude to this in the article, saying it is difficult to improve with a simple fix which does *not* mod the PCB. The alternate step of the 2.5V or 1.25V reference diode is quite worthwhile, and can be implemented without any PCB surgery to the Didden PCB, or the ALW PCB. It does leave the Early effect of the current source transistor(s) as a remaining error.

Russell also suggested that bypassing of a split 10k feed resistor (into two equal values), with a capacitive bypass from the split node to the Vin rail (not ground) would be another improvement. I agree with this as another possibility, unfortunately one not compatible with the Didden PCB which was the environment of the AE 4/2000 article.

Because of all of this, and the fact that I was attempting to make the new electrical design as close to a "zero PCB change" as would be possible, I took the approach of the prereg as a next-best system solution. Use of the prereg renders any imperfections in the Isource such as Fred D. has targeted rather moot.

If anyone implements the prereg, I think they'll agree it sidesteps all of the above issues. Plus, it also fixes the Vcb of the pass transistors, which is also worthwhile.

wj
 
AX tech editor
Joined 2002
Paid Member
Re: Isource improvement(s)

WaltJ said:


These comments are most germane to the Audio Electronics 4/2000 article, and also indirectly to Andy Weekes' (ALW) PCB implementation upon which Fred D. is commenting above.

Shortly after the AE 4/2000 article was published, I received an email from Russell Twining, who suggested an alternative of essentially what Fred D. tried to say above (but Fred got it got wrong). Russell suggested that instead of an LED a 2.5V LM336 could be used. I agree with this point entirely, it would be a better solution that making the 10k feed resistor a JFET Isource (expensive!). Note that the ED 6/23/1997 regulator design referenced in post #350 uses this approach, with a 1.2V diode.

Yes, one could say that the current source circuits as published in AE 4/2000 have a weakness. I did in fact allude to this in the article, saying it is difficult to improve with a simple fix which does *not* mod the PCB. The alternate step of the 2.5V or 1.25V reference diode is quite worthwhile, and can be implemented without any PCB surgery to the Didden PCB, or the ALW PCB. It does leave the Early effect of the current source transistor(s) as a remaining error.

Russell also suggested that bypassing of a split 10k feed resistor (into two equal values), with a capacitive bypass from the split node to the Vin rail (not ground) would be another improvement. I agree with this as another possibility, unfortunately one not compatible with the Didden PCB which was the environment of the AE 4/2000 article.

Because of all of this, and the fact that I was attempting to make the new electrical design as close to a "zero PCB change" as would be possible, I took the approach of the prereg as a next-best system solution. Use of the prereg renders any imperfections in the Isource such as Fred D. has targeted rather moot.

If anyone implements the prereg, I think they'll agree it sidesteps all of the above issues. Plus, it also fixes the Vcb of the pass transistors, which is also worthwhile.

wj

Walt, hi,

One change I have tried and also described in a post somewhere up north was the replacement of the entire input current source by a JFET current source. In other words, get rid of the LED, the PNP and related resistors and connect a FET CS directly from raw input (or after the prereg if implemented) to the pass transistor base. This improves the input ripple/noise rejection manyfold. But I found that a 1 or 2 mA CS was not enough to obtain good regulation above 100mA load or so. I had to look for a CS spec'ed at 5.6mA, that worked fine. As I noted earlier, these CS'es need some 5+ V across them so they ruin your dropout voltage. Still, it is an elegant and worthwhile mod if you have the extra input voltage. And it wouldn't need a PCB change I think.

I thought that this was also what Fred meant, but I now see that this was no so. Anyway, any comments from your side?

Jan Didden
 
Ain't no mountain high enough

A cap to Vcc for a voltage for an RC filter for the current through The LED is of course correct, which will be quickly apparent when one draws the schematic. I do think the different approaches with the jfet current source and LM336-2.5 are worth examining. A green LED biased at 0.4 mA has an AC impedance or about 100 ohms and a voltage drop of about 1.8 volts. The J502 has a typical impedance of 7 megohms. the ratio 7e6 / 100 is about 97 dB of rejection.

The 0.4 ohms impedance of the LM336 with a 10 K resistor gives about 88 dB of rejection. This is further compromised by fact the impedance of the LM336 starts to rise above 1 KHz as it's feedback falls off. The rejection is then decreasing by 20 dB per decade. I believe the voltage noise is also a little greater with the LM336 than with the LED. And of course the drop out voltage becomes slightly higher. These are still outstanding numbers and certainly a very valid approach that may be cheaper and easier to find parts for. No PCB surgery should be necessary for either approach. Differences in approach mainly those of preference for design with ICs or with discrete transistors, and there is certainly room for both in audio design.

There are many paths up the mountain. Walt has led 95 per cent of the way for the development of this very sophisticated regulator design. We are all extremely thankful for his fine work. I truly appreciate his input. Post like this indicate why he is such a welcome addition to the forum. We eagerly look forward to his further insight into circuit design. I feel sure he would not begrudge anyone the satisfaction of choosing his own path for the last few hundred yards to the summit.

Respectfully,

Fred
 

Attachments

  • led bias.jpg
    led bias.jpg
    24.5 KB · Views: 1,177
Re: Ain't no mountain high enough

Fred Dieckmann said:
A cap to Vcc for a voltage for an RC filter for the current through The LED is of course correct, which will be quickly apparent when one draws the schematic. I do think the different approaches with the jfet current source and LM336-2.5 are worth examining. A green LED biased at 0.4 mA has an AC impedance or about 100 ohms and a voltage drop of about 1.8 volts. The J502 has a typical impedance of 7 megohms. the ratio 7e6 / 100 is about 97 dB of rejection.

The 0.4 ohms impedance of the LM336 with a 10 K resistor gives about 88 dB of rejection. This is further compromised by fact the impedance of the LM336 starts to rise above 1 KHz as it's feedback falls off. The rejection is then decreasing by 20 dB per decade. I believe the voltage noise is also a little greater with the LM336 than with the LED. And of course the drop out voltage becomes slightly higher. These are still outstanding numbers and certainly a very valid approach that may be cheaper and easier to find parts for. No PCB surgery should be necessary for either approach. Differences in approach mainly those of preference for design with ICs or with discrete transistors, and there is certainly room for both in audio design.

There are many paths up the mountain. Walt has led 95 per cent of the way for the development of this very sophisticated regulator design. We are all extremely thankful for his fine work. I truly appreciate his input. Post like this indicate why he is such a welcome addition to the forum. We eagerly look forward to his further insight into circuit design. I feel sure he would not begrudge anyone the satisfaction of choosing his own path for the last few hundred yards to the summit.

Respectfully,

Fred


Fred, what about a LM336 and the JFet CCS?
 
Folks, sorry to break in, without checking up on all previous inputs. First of all, what Fred just put here is a typical current source used by designers for at least the last 25 years. Yes, it works. It is also very good.
When 'optimizing' a regulator, several factors must be thought through.
1. What do you want it to do?
2. How noisy is it?
3. What is its practical output impedance? Over frequency?
4.What is its transient response, both to load changes, and source changes?
There may be many other factors that I cannot think of just now.
When you change op amps, you change many of these factors.
Please keep this in mind.
 

PRR

Member
Joined 2003
Paid Member
JC> What is its transient response... to load changes

One thing that "bothers" me about many published regulators:

Where is the pull-down?

If you hang a resistor or light-bulb on them, they only have to source current. If the load turns off, they just have to stop sourcing very quick.

But if you hang a choke-loaded stage on a source-only regulator, and clip it, the load "kicks-up" and the regulator can't do a thing about it.

That's rare (though on another forum everybody is hanging transformers on mike-amps). But very short "kick-ups" happen in high-speed circuits. The kick alone may not be a big deal, but if the regulator freaks-out and does not recover quickly, it turns a short kick into a longer bobble.

Sure, we can suppress such kicks with a capacitor. But then why not suppress everything with a cap? Call me obsolete: I still like 4,700µFd caps. I use regulators only to work close to chip limits; but it is amazing how many of my chips have run on UNregulated +/-17V for years.
 
diyAudio Senior Member
Joined 2002
Hi Paul,

The kick alone may not be a big deal, but if the regulator freaks-out and does not recover quickly, it turns a short kick into a longer bobble.

That kind of behaviour is to be expected but:

Do you think this would actually happen in a real life condition?

Not too often IME...

Call me obsolete: I still like 4,700µFd caps.

Believe me, you're not the only one.


Cheers,;)
 
john curl said:
(...)4.What is its transient response, both to load changes, and source changes?
There may be many other factors that I cannot think of just now.
When you change op amps, you change many of these factors.
Please keep this in mind.

Understood. When I mentioned the op-amp idea, it was mostly offered to Fred for consideration, and I know he's well aware of these issues too. I've done quite a bit of SPICE simulation of a 90 Volt version of the Jung super regulator and would be glad to post the results if anybody's interested. This includes transients of load current and unregulated input voltage as well as stability analysis. However, I suspect that the interest in the 90 Volt version would be much less than for the classic version of the regulator. For the 90 Volt version, I had to use a slower pass element, which was also combined with an additional bipolar in a Darlington configuration for increased output current. Because of the extra phase lag from this, I wasn't able to achieve stability with the feedback resistor bypassed.

I simulated the performance with a number of op-amps, and indeed the AD825 had the best transient response for pulses of load current. The AD797 was a complete disaster in this application both for stability and transient response. However, I found that the OP37 gave transient response almost as good as the AD825, but with about a 40 dB improvement in line rejection at 120 Hz. Since this is for a power amp application, the low-frequency ripple rejection is important to me. Those results were what prompted me to mention the idea of op-amp substitution.
 
Here's the schematic from the simulator. Bypass capacitors on the unregulated input are not shown. The Zobel network of R17 and C5, together with R7 killed a 50 MHz parasitic in the pass transistor that showed up in transient. All the 100 uF electrolytics have 30 nH and 0.25 Ohms in series internal to the model. I've got multiple zeners in series because the tolerance on the lower voltage ones is tighter than the high voltage devices. The dissipation in the TL431A is such that the DIP version is necessary. It's too much for the TO-92 unfortunately. Based on Walt's post earlier, I'll probably replace the green LED with a voltage reference. I may use a MOSFET for Q2 based on Fred's input. The dissipation in some of the resistors is much higher than I'd like.
 

Attachments

  • junghv_reg.gif
    junghv_reg.gif
    9.8 KB · Views: 1,324
Thanks Fred. I was looking at the Rubycon YXF series, 160 Volt. It shows 0.24 Ohms at 100 kHz. This seems close to series resonance, so I used this for the Rs. Now, for the .01 uF, I'm not sure what you mean by the 1 Ohm thing. Do you mean this is the parasitic series resistance? Or are you saying the Zobel network should have 1 Ohm instead of 10 Ohms?
 
A Frank Capa production .......

The Zobel network should have 1 Ohm
or could. What ever you find works well.

For the Panasonic FC 35V 100uF the ESR is about 0.12 ohms and the FC series 100 V 100uF is about 0.18 ohms. These are very low ESR caps and your figures sound about right for some other electrolytic capacitor series. The Rubicon ZA series 35 V 100 uf would have an ESR of about 0.05 ohms. You might even want to look for some low ESR 50V 220 uF caps and change the value for the low voltage part of the circuit and stack two in series with resistor voltage dividers to put 45 Volts across each one for the output . i will double the ESR numbers shown here but still give lower total ESR


Some 50 V 220uF caps:

NIC NRSJ 0.023

Rubicon ZL 0.042

UCC LXY 0.073

Elna RSL 0.08

Panasonic FC 0.09

A sort of fun toy:
http://www.kemet.com/kemet/web/homepage/kechome.nsf/vapubfiles/KSpice205_32/$file/KSpice205_32.EXE
 
Maybe I have forgotten more than you know at present

"When Fred talks here about JFET current sources he forgets that 1 MOhms isn't true at high frequencies, much less or at least less. "

The Jfet current sources has a capacitance between 2 and 3 pF. Impress us and do the math to find out where to the rejection starts to roll off with frequency for one megohm. It is above the audio range BTW.
 
peranders said:
3 pF => 53 kohms at 1 MOhms but don't forget stray capacitance also. I think it's equally important how the regulator performs in the > 1 MHz range also unless your living room is a measurement chamber.

Excuse me, Per-Anders, maybe I am mixing you up with
somebody else, but I thought it was only about two or three
days ago when you recommended someone to remove the
input RFI filter of a phono amp, which none of us others seemed
to agree with. Have you attend a weekend course on RFI? ;)

Edit: Yes, we are nice today. :)
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.