My take on a discrete shunt voltage regulator

[HiFiNutNut]

They are waves repeating. I set the voltage to 0.01V (lowest) and the time at 0.5uSec (shortest).

Anyway, it was quite late so I did not do any further. I put the 22uF back in the output without local bypass (this gave a thin flat line) and put on a LM4562 to have a listen. The LM4562 had never sounded so good, nearly as good as the OPA627. The downside was at the very high frequency, the LM4562 sounded bright.

So I think I will give it a harder test tonight or tomorrow - play music when monitoring the rails and see if I find any ripples or "TAILS".

I think my earlier thought that "the output cap must be at the output" was pre-matured. I did not have sufficient evidence to support that statement so more tests are to be done.

In whichever case, things are improving and this is closer to my expected result.

With all these experiments, it points to low impedance and low inductance to be the key to success. That is never easy to achieve.

[ikoflexer]

Originally Posted by HiFiNutNut

They are waves repeating. I set the voltage to 0.01V (lowest) and the time at 0.5uSec (shortest).

Good job hifinutnut. Do you remember by any chance how many vertical lines distance between the peaks of the waves?

[keantoken]

I just realized that the MPSH81 is an RF transistor... I was sort of puzzled by the incredible bandwidth of the simulated circuit. Hmmm. And the 2N5088 is a high-gain transistor, to keep from loading it's high source impedance.

Does R6 have a large affect on stability?

And why use different FETs for J1 and J2? Wouldn't one type work for both?

The circuit is simple enough that it could be SMD'd and put right at the voltage input of the subject PCB, rather than on a whole other PSU dedicated board and associated lengths of wire.

Thanks,
- keantoken

[HiFiNutNut]

The waves were different from each different set of value of output cap and its location. They were 2 to 10 of (sometimes superimposed) sinewaves within each grid so the frequencies were between 4mHz to 20mHz with magnitude between 3mV to 50mV, also depending on where the plobe was.

But the simpliest of putting a 20uF (ESR=0.34R) at the output, without local bypass, with remote sensing, gives a thin flat line with no resonances. However, this is when there is no music signal and the load is purely from the bias of the opamp, a constant load. I will need to test when music is playing and see if it can trigger any ringings on the rails.

[ikoflexer]

Originally Posted by keantoken

I just realized that the MPSH81 is an RF transistor... I was sort of puzzled by the incredible bandwidth of the simulated circuit. Hmmm. And the 2N5088 is a high-gain transistor, to keep from loading it's high source impedance.

Does R6 have a large affect on stability?

No, I use R6 to measure the shunt current in the real circuit.

And why use different FETs for J1 and J2? Wouldn't one type work for both?

It would work, but not the same. I'm keen on using j201 for J2, because of its properties in the Vref, but I prefer a higher current for J1.

The circuit is simple enough that it could be SMD'd and put right at the voltage input of the subject PCB, rather than on a whole other PSU dedicated board and associated lengths of wire.

I'm planning to design one, yes, for low current needs.

[ikoflexer]

In post #164 C1 got moved to the base of Q2. Tonight I tested this mod and it was as stable as ever. Also, tested a 4uF film cap on the output and you can see the attached resulting trace images. The vertical distance between the faint lines in the image is 100 microvolts. The trace shows very little noise, mostly a flat trace about 10 to 15 microvolts thick.

The rest of the psu consisted of a transformer, followed by a rectifier, followed by a 10000uF capacitor. Don't take this as a suggestion, I just happened to have the rig setup this way from another project so I just used it. The DC voltage (if you can call those crests DC) at the regulator input was about 38V.

If anyone's wondering, I think this is it, v2 should stay pretty much as is. I've tested it even as a somewhat sloppy built and it was stable. For more current I've even tried two mosfets in the CCS and two mosfets in the shunt. The limit power dissipation will be set by the safe operating area of the mosfet and the size of the heat sink. Just tonight my screwdriver slipped a bit and the current went up to about 7A for about 20 seconds, but nothing got fried. The output voltage was set to 22V. Yep, the regulator dissipated over 100W for a short time.

As I said before, there are some build details described in post #1, but if anyone's considering building it and has questions, don't hesitate, ask away.

[Salas]

OK then no problems with the C1 move. Did you manage to see any wee difference for noise uV with that move on the scope?

[ikoflexer]

Not sure, the total noise is below what I can measure in microvolts. The C1 mod makes sense though theoretically and if the rest of the regulator is built properly it should be beneficial.

At this point even moving the oscilloscope probe puts more noise on the screen.

The whole thing is so sensitive. I tried measuring the noise with the HP3581A wave analizer (also a microvolt RMS voltmeter) but as soon as the probes touched the output it infused about a 100uV noise from creating a ground loop, because now both the scope and the HP3851A were grounded.

I built a battery powered noise preamp last week or so which I used for testing the noise of different Vrefs but I forgot to hook it up to have a look tonight. Honestly, it's hard for me to worry about the regulator noise after in theory and in practice it show as you see.

BTW, since I don't think I posted yet the simulated total noise at the output of the regulator, here it is, up to 50kHz.

[Salas]

I know its academic, that is why I asked if you could see a hair thiner noise line. Maybe it does something when the ref isn't a big enough Norton, and the cap does see a difference between across a low impedance and with RC constant. Maybe with an LM ref Z diode for example.

[HiFiNutNut]

I did some tests an hour ago. I connected my CD player to my preamp (opamp buffer) and played some music while using the oscilloscope to measure the preamp input and the regulator rails.

When the scope probe was at the preamp input, the screen displayed all those wonderful dancing lines.

When the scope probe was at the rails, whether at the regulator output end or at the load end, the screen displayed a thin, flat, straight, static line.

I thought I had to be careful so I checked very carefully and did it many times. The result was confirmed.