Super Regulator, collecting the facts

[jewilson]

Peranders,

As long as you can grade the AD797 to your circuit it will be ok, however a 50% yield is to expensive for me.

Andy,
I have a bunch of OP37 AZ's but their not stable either. You have a gain of about four or more. So that why I use the OP27's but I will populate the board with AD825's.

Now all I have to do is get rid of conducted radiation from my PS audio CD transport

[peranders]

Originally posted by jewilson
Peranders,

As long as you can grade the AD797 to your circuit it will be ok, however a 50% yield is to expensive for me.

If I'll get 50% yield something is very wrong in the design and I'm going to see to this will never happen. 99.7% yeild is my goal.

[rbroer]

While reading here http://tangentsoft.net/audio/opamp-linreg.html, I noticed the 2000 Jung revisited version is different from the EDN one.
One notices the extra PNP (for positive version) driven by the opamp.
Any reading material on this development; why was this extra bjt included (error is multiplied with even more gain) ?

[rbroer]

Ok, after looking carefully, I now see the extra PNP seems to just buffer the opamp output...

[jewilson]

Q2 allows the error amp to bias the pass transistor very fast, improving the ripple and dynamic performance. As the load varies at the error amp inputs it will change the current flow for the base of Q1.

[mlloyd1]

I changed to AD817 as suggested by Jung. I wasn't seeing any issues with the AD797, but I wanted to use them in a different application for which they were much more suited and I had a bunch of the AD817. Works great.

Listen to Walt. He knows what he's saying. Trust me ...

And no, I'm not with the government :-)

mlloyd1

PS - And I never tried the neutralizing cap with the AD797. I wish I had, Scott's papers on the AD797 were interesting reading.

Originally posted by Nicke
... Why not use the AD825 or AD817 as Walt suggested in the articles ...
Nicke

[Fred Dieckmann]

"Q2 allows the error amp to bias the pass transistor very fast, improving the ripple and dynamic performance. As the load varies at the error amp inputs it will change the current flow for the base of Q1."

The additional follower does drive the output with a lower impedance than the open loop output impedance of the op amp alone helping to move the additional phase shift from the output transistor to a higher frequency. It also increases the linearity of the op amp since the impedance that output of the op amp sees is higher by a factor of the Hfe of the buffer transistor. "In loop"buffers have been used for audio op amp circuits for quite some time now to decrease distortion by greatly increasing the impedance the op omp
output stage has to drive.

Interesting the second Sulzer article, "Regulators Revisited" TAA 1981, used a follower buffered pass transistor for higher current applications. It was not recommended unless needed for high current applications due to the stability concerns caused by the slow pass transistors used. This is not the case for the output transistors used in the newer regulator designs, which have an Ft of 10 to 30 times higher than transistors like the 2N3055. This article also suggest the use of a three terminal preregulator. Is a tribute to Mr. Sulzers work that the state of the art designs for audio regulators so closely resemble topologies developed nearly a quarter of a century ago.

[peranders]

Originally posted by andy_c
So what does this overly optimistic analysis tell us? Well here's an amplitude and phase plot of the simple RLC circuit showing the roughly 18 dB of attenuation at 10 MHz. Now if you have the AD797 data sheet handy, look at figure 14, open loop gain and phase vs. frequency. We have unity feedback because of the capacitor bypassing the feedback resistor. If we look somewhat below 10 MHz, we see in the plot below that there's about 50 degrees additional phase lag. This is also right where the phase of the AD797 starts to look squirrely too. So eyeballing it a bit, it looks like the unity loop gain frequency will be about 7 MHz and the phase margin will be about 10-20 degrees. Not good.

Andy, you are almost right on target here. I get an oscillation from about 90 mA load at 6 MHz! A small LP-filter is necessary for the supply voltage of the AD797. I have tested 4.7 ohms + 100 nF. Without this filter I have a steady oscillation at no load except for the feedback and the reference.

With this filter AND without 100 nF at the output the regulator works very nicely.

[andy_c]

Originally posted by peranders
(...)I get an oscillation from about 90 mA load at 6 MHz! A small LP-filter is necessary for the supply voltage of the AD797. I have tested 4.7 ohms + 100 nF. Without this filter I have a steady oscillation at no load except for the feedback and the reference.

With this filter AND without 100 nF at the output the regulator works very nicely.

Interesting, Per-Anders. Thanks for posting that. The super reg topology is a tricky one for sure. Much trickier than it looks like on the surface. If you need a high Q capacitor to bypass the op amp power supply to ground for stability of the op amp itself, then without the series resistor, the darn capacitor loads the regulator output directly, making it want to oscillate. Looks like you've found the fix with the series resistor.

I often wonder after diving into a lot of theoretical BS whether the conclusion will agree with reality. But it looks like if a good model of the load capacitor including ESL and ESR is available, then it's possible to predict stability or lack of it with pretty good accuracy.

[Fred Dieckmann]

I really want to get my measurements done and this is getting really old. Maybe it is time to get Jocko over here to solder for me. There might be some other tricks for optimizing the regulator for high speed op amps. I would use a very fast pass transistor for one thing, with a 5 to 10 ohm base resistor with it. Something with an Ft of 150Mhz or more if possible.

Andy........ "high Q " is the catch. It is interesting that the data sheet recommends using a series resistance for the larger cap when using a low ESR cap of several microfadads in conjunction with a 0.1 uF ceramic for decoupling the AD797. As for modeling something like this....... I would not get to hopeful in that department. Op amp Spice macromodels are very crude and most probably don't model PSRR. Even if they did, things like capacitive and inductive coupling in the PCB board would require an incredible amount of work and many more measurements than those required to try an RC filter.