DIYHiFi.org

So, what we have is essentially this, as shown a couple of years ago. People who built this regulator during this time usually, just as I did, preferred it to the conventional classic purely capacitive classic supply, as well as to the 3-pin integrated regulators (LM338) and, as per rule, reported improvements were important and not the subtle ones.



Some of you who have been building it asked for PCB or kit, to make this building endeavor easier. So, here is the chance. With minor changes it will be now released as a PCB and kit, with or without mains transformer, as Audial product. You may expect it to see it released till the end of the June.

Here is the PCB top side preview.

OK, I'll try explaining here a couple of things about this doohickey. The first important thing: it is out of the scope of this thread to discuss whether this is a "regulator" or not. Call it a "stabilizer" or any way you like, it is fine with me. The task it performs is obvious and semantics doesn’t really matter in that regard.

So, the circuit doesn't use a feedback to maintain constant voltage of its output, "comparing" it to the reference voltage and using an error amp to compensate for the change of voltage caused by the variations of the load current. Its output simply "follows" the reference voltage. So, there is no feedback to "regulate" the output voltage.

In other words, the output impedance is higher than what one may expect to see. The output impedance is basically given by Shockley's equation, which defines the impedance of the emitter of ideal transistor as Z = 0.026 / Ic, and which is for practical transistors rather closer to 0.030 / Ic. So we have 3 Ohm @ 10mA. Way too high? Methinks it is relative. The same equation means also that when the current gets higher the impedance will decrease, to 0.3 Ohm @ 100mA. I agree that you will play completely safe only when you have a constant current consumption, so class A amp and possibly balanced topology. Nevertheless, the circuit like this proved to work practically and sonically very good also with class AB amps, and in fact it was mostly used with gainclone amps.

The good thing, inductive raise of the impedance of such a regulator is smooth (1st order), and the absolute value of this impedance, as said, is not too low, and this fact is the good one when it comes to the integration of this regulator with output capacitor. A classic f/b based regulators, when combined with output capacitor, usually exhibit peaking in the output impedance module, whereas a regulator like this doesn't do, and its transient response is consequently clean. Another good thing is that such a regulator has a wide bandwidth and output bypass cap value can be relatively small.

Hello Pedja,
Do you find useful to include a 'minimal load' resistor on board? With about 1.5k/1W all the transistors in the triplet will be turned on. It is unlikely that someone will use this 'stabilizer' (well put!) for a very light load, less than say 13mA, but in such event the output Z and V will vary in little steps.

Nice work!
P.S. I always use this type of circuit for headphone amps; much more relaxed sound than feedback types.

Hello sidiy,

Yes, more relaxed, also more natural or simply more "normal" sound is what it brings in my experience.

As for the load, I'll actually always suggest testing supply before plugging it to the actual circuit, and for this reason the load resistors shouldn’t be higher than say 1k with 22V output, providing 22mA. I also suggested testing with dummy load in the originally posted article [ http://www.pedjarogic.com/gc/supplies.htm ]. A 6-6.5mA of this current will flow through the second, and 12-13mA will flow through the first Darlington transistor. Currently the first transistor also runs at some 6mA, since its emitter is fed via 220R instead via 100R. Schematics will be posted later.

Also, I rather wouldn't recommend a Darlington like this for light loads. With small currents you can do better if you strip the circuit down, readjusting Darlington and using better transistors. You can also strip the regulator even further, totally omitting Darlington and staying with Zener as most simple shunt regulator, which is also very interesting solution (we can discuss this later).

Pedja

I'm glad you made those points for me. Saved me some typing.

Anyway. As for output Z of one of these doo-hickies (have you been hanging around me that long now?):

While several ohms may seem high to some of you, you have to take into account the amount of current variation that you think will occur. If you are only drawing 13 mA, what is the chance that the circuit will suddenly swing up to 50 mA or? Not likely. If you only see a 10-20% variation, from that you can come up with a number for the AC signal that you will see under normal operation.

Odds are that you will come to the conclusion that really, really low output impedances aren't always needed.

But, if the thought of a "high"output impedance still bothers you, consider this:

Same 13 mA draw, single follower, and we come up with 2-3 ohm output Z.

A 100 uF cap will have more than that for frequencies below 500 Hz or so.

Are you really going to stick a 4700 uF cap everywhere that you have such a low current draw?

Yeah, some of you may.

Now you have to deal with how the cap sounds, etc., blah, blah.

Compare that to a simple circuit that has a flat impedance over quite a large frequency range.

Think about that some more, and Pedja's approach may start to make sense.

Maybe not. Your choice.

Jocko

Doesn't the DC current gain of the transistors also have impact on the output impedance of the reg?



No, not at all.
You have there a very good PSU. You have all the issues sorted out (and now you have added CRC before the reg), I can't figure out how to make better, for high current loads (like a power amp). It's simple and very elegant.

You are right, relation from above is valid when transistor is driven by zero impedance sources. The source is here the previous transistor, and its output impedance, divided by the output transistor current gain, is added to the final output impedance.

So, if this previous transistor runs at 6mA its output impedance is (supposedly) 5 Ohm. With current gain of BD911/912 equal to 100, this will be divided to 0.05 Ohm. This implies that the higher current gain helps lowering the output impedance, but the output impedance total value can not be lower than the value given by the equation above.

If you do make this PCB, I might ask that the output transistors be located closer to the edge of the board, so they can be bolted to a big heatsink if someone wants to use this for a really high-current application. Maybe you can put the output pads between the transistors.

I've been considering the best place to put the output transistors so their locations suit the most of possible applications, and I've come across the option shown above... It makes possible to use PCB TO-220 heatsinks and pieces effective up to 2.7°C/W are normally available, and with a bit of luck you will even stumble across 2.1°C/W.

If you choose transformer voltage wisely, a voltage drop across the regulator won’t be higher than 7-8V. Even with small class A amp you won’t have more than 1.5A of quiescent current. This means that the output transistor will have to dissipate 11-12W. So, the heatsinks as above are sufficient.

Now, you may want to use one pair of regulators for both channels of such an amplifier, or you may have any other reason to pull more current... in that case you can use one bigger heatsink for both transistors. (That's also why the output pads are not between the output transistors.) You may even cut the PCB a bit... But I'd rather recommend staying within 2-3A of quiescent current per one board anyhow, it is not really meant to be used for more than that.

Having said all this... I'll still have your suggestion on mind, and I may consider adding another pair of output transistors footprints. Existing ones will still remain as they are, as I don’t want to discard the use of PCB heatsinks. They are handy, available and quite effective in most cases.

And, to clarify where I am at currently... A couple of rev0 boards are already finished. This means that I don't have much of work until they are actually released.

There is one other possible, key word, possible, benefit. With a couple ohms output z this can help block noise getting in from the rectifiers and before. The output z along with the 100u cap = RC filter. IMO this is a more elegant way to filter than using a RC after the regulator. Yes, this breakpoint moves with the varying z. So, as the current goes up, output z goes down. But, the voltage drop across the output of the regulator stays the same. First approximation not taking delta temp, source z, etc. into account.

And, IMO, I think that this type filtering is a larger part of the reg being easy to listen to vs fb/no fb.

I've come to the conclusion that for a regulator I want either the lowest ouput z with the lowest noise I can get or, something like what you've done. It's the "in between" items I don't care for.

Hi Mike,

I am not sure I understood this... where do you propose an RC? The schematic included in the first post doesn't display this, but the board includes the resistors between two 4700uF caps at both sides (of both regs), forming common mode RC filters. The R value will depend on the acceptable voltage drop and associated power dissipation, but in most cases it may be exactly a couple of Ohms. (Five watts wire/ceramic body resistors can be mounted and one may use two in parallel, mounting on and another beneath the board.)

As for the noise, as long as we don't speak about MC or RIAA stage I normally wouldn't consider noise as issue. Funnily enough, speaking about regulators, my experience is similar... Many times a sonic performance of solutions applied relate fairly well with their noise performance.

Pedja, read Mike's post again (I had to read it again too) because he is talking about your current reg, no RC added. He means the output impedance of the reg, along with the output cap, forms an RC filter.

Anyway, what do you think about this?
I guess that for a power amp this is almost paranoia - your reg should be low noise as is.

That is extremely interesting idea, as I do can confirm that this is (soundwise) very critical part of the reg. But let's see if Mike can confirm if this was what he actually suggested.

A few ohms output z of the regulator (the emitter impedance, .03/Ie) and the 100u output cap = RC filter.

Remember the old days when we had to calculate the output impedance of each transistor stage, and calculate the input impedance of the next Q stage? Remember too, we also had to look at the output and input capacitance of each Q stage.

There have been similar circuits posted in the past but I'm not certain who digs into the math anymore though (other than Fred, Jocko, JC, more?)

True, this is a low noise design.

Sometimes I think we forget to look at these circuits like we did 40 years ago. I'm guilty of it...

My other comments; I don't mind a regulator that uses feedback, if it's done correctly. (IMO only please). Like the Jung regulator, a few others. I also like circuits like you posted Pedja. The in between stuff I mentioned that I don't like would be things like a straight LM317 circuit and it's brothers and cousins. Carlos has done some pretty amazing things with the LM317 but I'll bet he's sweated many hours to get it to sound right.

Most of what I've done the last couple years has dealt with regulators. I've become convinced that the power supply is 90% of the sound. Literally.

I took a OPA2134, bypassed with Pana FMs close to the chip, added RC snubbers, tuned the RC values to what I like the sound of, added a C from the + rail to the - rail and powered it with my shunt regs. The 2134 is even run at unity gain. ( I can see and hear a few people cringing and thinking I'm nuts) This amp is used as a DI input in a studio. I've been told it's one of the best sounding DIs out there. Why? It's all about the V and I delivered at the supply pins. Disclaimer; This is for a DI and the sound of it may not be liked or appropriate for a home stereo.

When I tune a circuit for a certain sound I do it via the regulator and bypassing, not by changing the amp stage. Caveat; the amp stage has to be decent to begin with.

Sorry I wasn't clear, at work and typing in a hurry. Will step down from the soap box now.