Super Regulator, collecting the facts

[WaltJ]

In figure 4 of the AE article from 4/00 there is no bypass on R102 -- 825R from adjust to the output of the SR -- should this also be a.c. coupled to the output?

Not necessarily, as it works fine just as it is shown. The proposed AC bypass you mention could be part of your stability problem re the 337. I can't see where it should be necessary at all.

wj

 

[peranders]

In figure 4 of the AE article from 4/00 there is no bypass on R102 -- 825R from adjust to the output of the SR -- should this also be a.c. coupled to the output?

The gain of LM337 is set to slightly under 2 so a cap will only reduce the gain < 6 dB, not much. It's more important if the gain would have been 5-10 or more.

 

[jan.didden]

I think it INCREASES the gain by 6dB as it increases the feedback for the error signal. I'm with Walt here, this extra cap could cause the 337 instability.

Jan Didden

 

[jackinnj]

Can you show us how you have made the grounding for all parts. This is very sensitive.

"Totae viae ad Romae duc*unt"

Edit: The positive regulator doesn't oscillate, the negative one with the pre-reg did -- both circuits are mirror images of each other with one central ground.

The pic I showed has connectors for ease of experimenting -- these have mucho milliohms of impedance.

I've re-read the 1995 articles and will incorporate Gary Gallo's current sensing scheme on the regulator board --

 

[peranders]

You could try to insert 47 ohms ( or near) in series with the power connection (pin 4) and the output. Add also 10-100 nF across pin 7 and 4. It might help.

 

[jan.didden]

[snip]Edit: The positive regulator doesn't oscillate, the negative one with the pre-reg did -- both circuits are mirror images of each other [snip]

Makes sense though - neg regs generally have more stability problems due to the (slower) PNP's used as series pass transistors. That's why some people use two pos regs with one pos output conneted to the gnd of the other to get bipolar supplies. But you need fully separated secondaries on the xformer for that trick.

Jan Didden

 

[jackinnj]

I can't make the elegant noise measurements of the 2/1995 TAA articles -- if you are going to do any measurements on this thing don't use a scope probe -- solder the measuring cable directly to the output pins.

Using my HP3581 (which closely corresponds to the Boonton 1120 and Tektronix AA501) I measure 600 nanoVolts of random energy at 10kHz using the 300 Hz bandwidth.

I use an Oreo cookie tin...wish I had an AP.

 

[peranders]

Makes sense though - neg regs generally have more stability problems due to the (slower) PNP's used as series pass transistors.

One other possible explanation is that the opamp is worse with you switch polarity.

 

[jan.didden]

One other possible explanation is that the opamp is worse with you switch polarity.

I wouldn't think so. You still use the opamp in inverting configuration with the same gain/ same feedback factor. I can't see any difference in that respect that could change the frequency behaviour.

Jan Didden

 

[peranders]

The PSRR is different at the positive vs. negative supply pin.

 

[jan.didden]

The PSRR is different at the positive vs. negative supply pin.

Hmmm. Yes. Never thought about that. You may have apoint there.

Jan Didden

 

[jackinnj]

In the "Improved" version with the LM317/337 pre-regulator there's only 2.3 Volts dropped across the Super-Regulator -- this should require a much smaller heat sink than 9 C/W.

 

[jan.didden]

I don't have an Audio Precision to make the noise measurements, but hooked up the HP3581A (which has 100 nanovolts lowest scale) to the Super Regulator with the Old Colony boards, zener instead of 1N4148 and an AD825AR on a high quality adapter and AD797AN. There is no pre-regulator for either setup:

An externally hosted image should be here but it was not working when we last tested it.

Interesting. Any idea why that curve shoots up at 40kHz or thereabouts? Is this a measurement artifact?

Yes, that 797 is pretty good. It's just that it often seems difficult to get it stable in these regs.

Jan Didden

 

[jackinnj]

Interesting. Any idea why that curve shoots up at 40kHz or thereabouts? Is this a measurement artifact?

Yes, that 797 is pretty good. It's just that it often seems difficult to get it stable in these regs.

Jan Didden

Jan I deleted the picture -- the WJ/JD articles used a constant Q filter which would should show a rising 3dB slope as WJ explained in 2/95 -- the HP3581 is constant bandwidth and I was trying to keep the BW as a % of the CF relatively constant -- the narrowest is 3 Hz -- but this is 30% of the lowest frequency and only 0.006% of 50kHz. In any case, I didn't want to mislead -- I will show the data in the graph below and it is really only useable as a relative assesment since the HP3581 has some nV of noise which can't be eliminated -- I thought there was an artifact as well, but the data is repeatable. Again, the Oreo cookie tin was used to good effect.

An externally hosted image should be here but it was not working when we last tested it.

 

[peranders]

Are you able to do some weighted measurements like DIN-A or something?

 

[jackinnj]

P-A -- I want to crank up the calorimetric true RMS meter first -- I just repeated the measurements again with the LT1055, and redid the AD797 -- I will have to look through my opamp box for more single JFET devices to test out.

 

[peranders]

Do you use LM329 + 1k+ 100 uF as voltage reference?

 

[jackinnj]

I use LM329DZ + 4.99K + 120uF/25V

 

[peranders]

Is your LP-filter 5k+100uF ?

 

[jackinnj]

There are a pair of 1K resistors which determine the error signal fed to the inverting input. The Voltage Reference is driven from the Positive Sense pin -- all of this is on a stock Old Colony board -- the only exception is that the filter capacitor which used to be on the positive supply pin is now across the diode to the collector of the 2N5087, and a zener is used instead of a 1N4148.

With respect to the noise -- I want to remake the calculations using some different equipment -- I have two filter setups which are not part of the graphs -- a 100 milliHz to 10 Hz filter which TI uses when characterizing the TL431, an a 10Hz to 100kHz filter which Linear used in Application Note 83"Performance Verification of LDO Regulators". I use these with the HP3403 true RMS meter which uses a thermopile, not a log-antilog RMS detector.