Super Regulator, collecting the facts

[Fred Dieckmann]

"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 in the > 1 MHz range also unless your living room is a measurement chamber."

Very good. Your rejection into 100 ohm LED load has fallen to 54 dB at 1 megahertz assuming no capacitance across the LED. This is still an very good rejection number
at 1 MHz. I agree that very good rejection of RF makes for better sounding audio equipment.

[peranders]

Originally posted by Fred Dieckmann
Very good. Your rejection into 100 ohm LED load has fallen to 54 dB at 1 megahertz assuming no capacitance across the LED. This is still an very good rejection number
at 1 MHz. I agree that very good rejection of RF makes for better sounding audio equipment.

54 dB for the LED alone but what does it mean in total, at the output?

[WaltJ]

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'll answer the comments above referencing also Fred D's sch. of the modified Isource, post # 390.

Yes, Jan, I can agree that this would work to some degree, and be both better and more simple (if not cheaper). I also agree with you that these JFET current regs. need lots of voltage to get beyond their knee, and flatten out to the highest Z. 4-5V maybe.

That said, they also may not be high enough in Z. I have found that a JFET cascode always works better for highest Z, albeit at the expense of even more voltage loss. Now we're up to 5-6V or more if another JFET is used as the cascoding device. But, given the high desirability of a cascode, modifying Fred's sch. further may be the very best alternative in terms of minimum component count, and also reasonable cost. Here's what to do with it:

1) Replace LED1 with a 6.8V zener, cathode to Vregin (dynamic Z will be a few ohms). This can be dropped down to 5.6V if headroom loss is a great worry.

2) Replace X2 with a 10k ohm resistor.

3) Replace Q2 with a PN2907A (lowest cost, lowest C part for the job, excellent Hfe).

4) Replace R1 with a 2N5459 JFET, selected for 4-5mA for the currents you describe. For lighter currents, a 2N5457 will suffice at around 2mA.

Most of the above parts are available from Digi-Key or Mouser for pocket change. High voltage parts can also be sub'd if desired for the 90V reg, with the 10k R adjusted upward in value.

This type of Isource has been tested both on the bench and via simulation, and the output Z is higher than that of the prior suggestion (which is limited by the non-cascoded xstr's sensitivity to Vcb variation).

A general point which should be understood about this type of regulator. With the inherent tight buffering of the op amp errors due to the rail bootstrapping, plus the reference diode being fed from the regulated output, the Isource errors constitute the major dynamic source of errors. Thus almost anything that can be done to improve Isource stability etc. is very much worthwhile (some might even read that to mean audible).

In other words, those posters asking about what the reg output does for various Isource topologies are on the right track. The answer is that it will change dynamically due to Isource imperfections, if you allow them to happen.

Now, after all that, maybe the prereg looks good?

wj

[Jocko Homo]

Not knowing which Q2 in which schematic, but..............

Do we really have to worry about saving 4 cents in a circuit that most likely has some high quality, and therefore, not cheap, film caps?

I bet Phred can find transistors with better h_fe linearity, and low capacitance, that don't cost $1 each.

With that $1 that they save, they can make their own current diode with a resistor and the right $0.40 JFET. And have enough left over to make a cascode version.

Unless, of course, there is no headroom. But this DIY, and I bet that they can find it if they want it bad enough.

Since this is DIY, I wonder how many guys trying to build a serious regulator are going to worry about a handful of inexpensive parts if it gives them another 10 dB or so of PSRR.

I bet most of them will do so, even with a pre-regulator. I would.....

Jocko

[WaltJ]

Not knowing which Q2 in which schematic, but..............

Just read the post, the sch. reference IS there.

FYI, a PN2907A costs a few pennies not $1.

Why *not* save the $, if one can have superior performance *and* low cost? The ckt. suggested does do that. The main expense is the loss of headroom, as cited.

[PRR]

I just knew someone would object to Walt's "lowest cost".

> Not knowing which Q2 in which schematic

The top one.


> find transistors with better h_fe linearity, and low capacitance, that don't cost $1 each.

Hfe linearity is a non-issue: this device works constant-current. Hfe is important; but 2907 has enough Hfe that its base current will be swamped by the R-LED or Zener current. The exact value of Hfe is not critical: we only need to hold the source current higher than Darlington base current, and constant, not any exact amount.

Capacitance is critical, and mentioned by Walt: "PN2907A.. lowest cost, lowest C part...". The 2907 is a pretty-bitty die with small junctions and thus low capacitance. If you go comparing capacitance, be sure you compare at the working condition: Cob varies a lot with voltage and in some devices Cob varies somewhat with current. 2907 Hob at 30V is a mere 2pFd; you need a good layout so your layout strays don't degrade this.

I'd also place great weight on Walt's claim: "This type of Isource has been tested both on the bench and via simulation, and the output Z is higher than that of the prior suggestion...". He probably has a very-nice bench, and this kind of extreme application can't be "proven" with talk/theory or just simulation.

Or are you objecting to the HIGH cost of PN2907A? Yes, DigiKey asks a whopping $0.26 each, a big budget-buster. Their labor costs to sort parts into baggies must be high. But PN2907A is a good choice in many-many chores, so Super-Size your order: price is half if you buy 100 ($13). Or go to Mouser: $0.11 each, $6.00 for 100. Even cheaper if you do Suface-Mount and buy a reel (and you may be able to get parasitic capacitance down another pFd).

If you must pay more: 2N2907A comes in a TO-18 metal-can package for $0.60. I doubt it is better in any way that matters here, and may add a tenth-pFd stray capacitance off that metal can.

The 2907 is so cheap because it is a real sweet-spot for size, doping, gain, etc. Millions or billions have been made. And foundry costs are mostly up-front: the more you make the less they cost. Don't be put-off because it is 10 for a buck.

[Jocko Homo]

25 years ago, '2907s were all that we had.

That was 25 years ago.

As for the $1 jab, Walt was objecting to those current diodes, that cost >$1.

You can make a cascode version, to keep Walt happy with high Z, with 2 JFEts, and one resistor. <$1. Ooops.......too much space. Never mind.

Jocko

[WaltJ]

25 years ago, '2907s were all that we had.

I said *PN*2907A, not 2N2907. There is an important difference between the two. There is also the principle of "if it ain't broke don't fix it". And PRR has stated quite well the merits of the 2907 family.

All of this is certainly not to make *me* happy. I thought the purpose of this forum was to discuss various means of execution, and the pros/cons. Is it somehow lost on the readers here just how/why high-Z in this application is critical?

wj

[Fred Dieckmann]

I a little confused as to the interpetations people are taking from my idea.

All advocated was the replacement of the 10K resistor that biases the green LED. This LED serves as the reference voltage for the bipolar current source. There is no change in headroom since the 10K resistor sees the input voltage minus the 1.8V drop across the Green LED. There is plenty of voltage to operate a jfet or two terminal jfet current source in its linear and high inpedance region.

Walt is correct in stating the the J502 is fairly expensive and it is avalible for around $ 1.60 from Mouser. Another choice would be the the Vishay J202 at about 42 cents.

http://www.mouser.com/catalog/616/306.pdf

I am not suggesting not using a preregulator, but just a simple change from a resistor to a jfet to increase the PSRR of that portion of the regulator. It
is a very simple change that looks worthwhile to me. I will hold off on how to get rid of the level shifting circuit while still retaining the bootstrap supply
for the op amp (mod to ALW 14 volt regulator) by using a certain transistor
as the buffer between the op amp and pass transistor. I am afraid a riot would
break out.

Later,
Fred

[Jocko Homo]

We are also here to discuss various means of execution. We, well, mostly me, hasn't kept anyone happy for >50 years. We specialize in using the Pease mode of "show me where it says that it can't be done that way".

But anyway......the "PN" version still is only good for 40 V, as compared to a 2SA970, which is good for over twice that amount. Which implies that the '970 has a smaller base region, and most likely low capacitance.

The '970 has half the capacitance of the '2907, when measuerd at 10 V, and 1 MHz, as both are in their data sheets. And it is much more linear.

No, the high-Z aspect is not lost on all of us.

Jocko