Bipolar discrete shunt regulators

[Onvinyl]

Hi lineup,
it might be my personal mistake, but every time I tried a output transitor cascade similar to you U5, U9, U6 I ran into instability troubles.

Haven't tried you exact circui, but very close.

I would cascode U2, that is very efficient in increasing the stiffness of the CCS.

Did you build it?
Rüdiger

 

[apassgear]

Hi All

As I have explained on the GB thread related to the Toolereg we have asked Jan Didden who is one of the most experienced member of DIYA regarding voltage regulators to review the technical aspect of this regulator and he has kindly accepted to do this on his own lab. To this end, Steen has kindly sent to Jan the compact prototypes for him to do this work and as far as we know Jan has already started to do so and will make some observations so we can all discuss his findings and decide on corrections if needed

We greatly appreciate Jan Didden’s kind offer to help us refine this excellent shunt regulator designed by Colin Toole.

 

[Lumba Ogir]

Lineup,
I would try to clean that mess up making a stable circuit with one CFP compound and without any ugly compensation capacitors.

 

[jan.didden]

Tony was so kind to send me one of the prototypes. I appreciate this, I don't have a lot of experience with shunt regs so was looking forward to getting one on the test bench.

The prototype I got was set for 30V and I ran it with 35V input.
The Zout measured quite high for such a circuit - more than an ohm.

After some measurements I found that R10 (100) is too small. It carries only 2.7mA or so, so the Vbq4-Vbq7=275mV. Since Veq7-Vbq7 needs to be some 600mV, Q4 was heavily saturated, Vcq4 was 300mV above Vbq4. Consequently, the whole current source above Q3 didn't work too well. I changed R10 to a diode to get Q4 some breathing room. (Actually I should use two diodes there in series). Better.

Then I shorted R11, because it worsens the reference for Q3. With R11, if Vbq3 drops to because Vout drops and it wants to drop the current through Q2 by lowering Icq3, it also drops the current through R11 and the zener. That means Veq3 drops as well and that works against dropping Vbq3, so worsens the regulation. Better.

The other issue is that Vb1 is only 1.8V or so below Vout, so Vcq3 is also that high, again limiting the available voltage for the whole current source stuff to work linearly. I put a 12V zener in series with R8 so Vbq1=Vcq3 is now around 14V below Vout. Now we're getting somewhere. Zout at 10kHz dropped to 17 milli-ohms. Not bad for a first try.

I also see some very low level 2.5Mhz hash but that's for later.

This is measured with 30mA load current (DC) and 20mA AC current pushed into the output terminal to simulate load signal current. Since the signal ripple is now so low (around 350uV incl noise) my AP has problems to measure it so I need a low noise preamp. I have one, but I overloaded it some time ago with DC and now it is broken. So, before I continue I have to repair it, and that's what I am going to do now first.

Jan Didden

 

[lineup]

Hi lineup,
it might be my personal mistake, but every time I tried a output transitor cascade similar to you U5, U9, U6 I ran into instability troubles.
Haven't tried you exact circui, but very close.
I would cascode U2, that is very efficient in increasing the stiffness of the CCS.

Did you build it?

Here is my circuit again:
http://www.diyaudio.com/forums/attachment.php?postid=1679826&stamp=1228822503

No Idid not build it.
But I spent several hours trying out all those small and bigger capacitors. And trimmming for many conditions.

Seems the real trick to get a good shunt is to adjust those caps to the load to be used.

Even those C2, C7 100nF are important.
As well as one suitable cap across the output. And of course across the load.
Here I use C4 = 1 uF.
This will keep circuit stable, when you remove all load + load bypass caps.

 

[AndrewT]

Hi Jan,
thanks for coming in as our resident expert.
Most of your concerns/observations have already been raised in the threads, but the members chose to ignore the warnings.

Can I suggest that R10 and R3 be changed to red LEDs. The extra voltage drop across R10 location is not an issue. I think we can afford the extra voltage drop across the input CCS. With 100r it was very low bordering on LDO performance. Using LEDs increases the min Vdrop across the input CCS to ~ 2.7V, if Q6 needs to run near saturation. Your 5V Vdrop keeps it well clear of saturation (Vceq6~2.5V).

What about changing Q5 from a med power device to a To92 small signal device?

 

[jan.didden]

Note: this was send to me by email by Colin but as we agreed to discuss here, I copied it so we can keep track of the issues.

For R10, it would seem simplest to just change it to another LED, as this gives about the same vdrop as two series diodes with probably lower impedance.

I can see how R11 causes problems for the reference, and I think I see two possible solutions. The first is what you suggest, eliminate R11, and the second, feed D2 with it's own current source. For my own purposes, I have actually changed this part to a reference diode, subsurface zener IIRC.

Do you really feel we need a 12v drop in series with R8? It would seem to me that even just a standard 1.7v LED would give the current source sufficient headroom, as it would put Vcq3 around 3-3.5v. Also, I plan to use this for some lower voltage regulators, 3.3v and 5v, so do you think that it would be prudent to remove the cascode on the CCS feeding q3 because of saturation concerns?

17mohm is quite nice for this, IMO. It will go down significantly with a larger C1-3. May I ask how you are measuring this? Real world impedance measurement like this is something I have been wondering how to do for quite some time now. [/B]

Colin,

Yes, a LED for R10 seems an even better solution.
As to the reference zener, replacing it with a better ref doesn't make that much of a difference (possibly less noise though). Look at R11 as a local (emitter) feedback resistor, decreasing the gain of that stage, and less gain in the control loop means less performance. There's no reason for R11 IMHO anyway.

Yes, 12V in series with R8 is a bit more than necessary; it's what I had handy. Especially if you want to use it with lower Vout. Another LED here would probably be good enough. Anyway, I'm not sure you need Q7 at all.

I also played with R8 itself, the lower this can be, the better the performance. (R7/R8 set the gain of Q1 and you want to maximize that). 100 ohms seems OK, but shorting leads to oscillations. So that needs some attention.

I also removed the very high-Q capacitor C2 and replaced it with a regular 100uF/35V electrolytic with ESR, for better damping. It cut the low-level oscillations I saw (around 2.5MHz IIRC) in half.

So, all in all, a pretty good regulator. If you want to measure Zout, connect a tone generator or other sine wave source (an amp with a CD with a steady tone will also do) and feed the signal into the output of the reg via a large coupling cap and resistor (like 100uF with + on the pos output of the supply, and the neg pin via 1k or so to the source). Then you set the source to say 20V, which inserts 20mA into the reg. You can load the reg with a DC load. Then measure Vac at the regulator output, and I mean the output directly at the pcb. Measuring after some lenght of wire means you're measuring the wire which may swamp the reg Zout. I measured around 350uV with 10kHz so you really need something better than your multimeter .
So, measuring 350uV with 20mA gives you Zout as 350u/20m = 17.5 milliohms.

Let us know what you find.

Jan Didden

 

[Lumba Ogir]

Lineup,

But I spent several hours trying out all those small and bigger capacitors

Wasted hours.
Seems the real trick to get a good shunt is to adjust those caps to the load to be used.
The real trick is to make a stable circuit. Yours is not stable.

 

[jan.didden]

Here is my circuit again:
http://www.diyaudio.com/forums/attachment.php?postid=1679826&stamp=1228822503

No Idid not build it.
But I spent several hours trying out all those small and bigger capacitors. And trimmming for many conditions.

Seems the real trick to get a good shunt is to adjust those caps to the load to be used.

Even those C2, C7 100nF are important.
As well as one suitable cap across the output. And of course across the load.
Here I use C4 = 1 uF.
This will keep circuit stable, when you remove all load + load bypass caps.

Nice reference. I would increase R4, delete R6 for even better clean reference.
Why don't you sim the Zout over freq?

Jan Didden

 

[AndrewT]

Hi Jan,
re your output impedance measurement method.
If the applied test voltage is 20Vac then the peak currents being injected and extracted from the PSU are ~+-28mApk.
Does that mean the resistive load applied to the PSU must draw >=28mA? i.e. 30Vdc across 1k0

If this applies, then the CCS must be set to pass ~>= [2 * Ipk] + Iq.
For the Cetoole reg, this is approximately >=57mA +Ir5+Ir2+Ir1+Ir15+Ir7>=57 +3.3+2.8+2.3+2.6+5 ~>=73mA. i.e. R6<=13r

 

[jan.didden]

Hi Jan,
re your output impedance measurement method.
If the applied test voltage is 20Vac then the peak currents being injected and extracted from the PSU are ~+-28mApk.
Does that mean the resistive load applied to the PSU must draw >=28mA? i.e. 30Vdc across 1k0

If this applies, then the CCS must be set to pass ~>= [2 * Ipk] + Iq.
For the Cetoole reg, this is approximately >=57mA +Ir5+Ir2+Ir1+Ir15+Ir7>=57 +3.3+2.8+2.3+2.6+5 ~>=73mA. i.e. R6<=13r

Andrew,

The pos peak is no problem, the shunt can always sink more current (up to the max dissipation and current of course).

For the neg current, the shunt can only go down to zero so the max you can pull out is the DC shunt setting. The version I got had R6=25 ohms and on the current meter it drew about 80mA IIRC.

Jan Didden

 

[jan.didden]

Hi Jan,
thanks for coming in as our resident expert.
Most of your concerns/observations have already been raised in the threads, but the members chose to ignore the warnings.

Can I suggest that R10 and R3 be changed to red LEDs. The extra voltage drop across R10 location is not an issue. I think we can afford the extra voltage drop across the input CCS. With 100r it was very low bordering on LDO performance. Using LEDs increases the min Vdrop across the input CCS to ~ 2.7V, if Q6 needs to run near saturation. Your 5V Vdrop keeps it well clear of saturation (Vceq6~2.5V).

What about changing Q5 from a med power device to a To92 small signal device?

R3 yes a LED is better, especially with low CCS currents. BTW I assume that low-drop-out is not a key requirement here.

On changing Q5 to TO92, I wouldn't do that, it carries the full current so a BD seems a good choice.

Jan Didden

 

[AndrewT]

..For the neg current, the shunt can only go down to zero so the max you can pull out is the DC shunt setting. The version I got had R6=25 ohms and on the current meter it drew about 80mA IIRC.

I think you are confirming that the CCS must supply both the resistor load and the test AC load.
If this is correct, then R6=25r cannot supply 80mA, only ~38mA. Subtract the Iq of ~16mA, leaves the maximum Rload + test current<=22mApk.

 

[jan.didden]

Well, suppose you have a no-load situation where the CCS delivers 80mA, and the shunt eats up these 80mA. Now you are going to send +28mA into the output. The result will be that the CCS still supplies 80mA and the shunt sinks 107mA, no?

Edit: Kirchoff's law: the total sum of currents into/out of a node is zero. So, the output node: in comes 80mA from the CCS, 28mA from the measuring source, so out must go 107 (eaten up by the shunt). There's nothing so practical as a really good theory

Jan Didden

 

[AndrewT]

On changing Q5 to TO92, I wouldn't do that, it carries the full current so a BD seems a good choice.

but the first transistor of the CCS never sees a Vce >2V.
A 50mA CCS would only dissipate <100mW
Any 500mW 100mA To92 can cope with this.
I would be very tempted to use BC5xx series.
If one requires between 50mA and 100mA then any of the 200mA To92 would do.
For between 100mA and 200mA, any 500mA To92 or Eline would do.
ZTX can dissipate 1000mW and with 200mA @ <1.5Vce passing is <300mW. there are also the 2n5551 and MPSa42 that could take upto 150mA.

I cannot see why a medium power device is necessary. Yes, it's an easy, no calculation device, but cheaper and better are available for this duty.

 

[jan.didden]

but the first transistor of the CCS never sees a Vce >2V.
A 50mA CCS would only dissipate <100mW
Any 500mW 100mA T092 can cope with this.
I would be very tempeted to use BC5xx series.
if one require between 50mA and 100mA then any of the 200mA To92 would do.
for between 100mA and 200mA, any 500mA To92 or Eline would do.
Ztx can dissipate 1000mW and with 200mA @ <1.5Vce passing is <300mW.

I cannot see why a medium power device is necessary.

Yes sure, no problem. Just my personal preference to run parts way below rating.

Jan Didden

 

[AndrewT]

Well, suppose you have a no-load situation where the CCS delivers 80mA, and the shunt eats up these 80mA. Now you are going to send +28mA into the output. The result will be that the CCS still supplies 80mA and the shunt sinks 107mA, no?

Edit: Kirchoff's law: the total sum of currents into/out of a node is zero. So, the output node: in comes 80mA from the CCS, 28mA from the measuring source, so out must go 107 (eaten up by the shunt). There's nothing so practical as a really good theory

Jan Didden

I think you are missing the point.
R6=25r cannot supply 80mA. It only has ~[1.55V-0.6V] across it.
Where does 80mA come from to allow the test AC voltage to be applied?
You originally said that in the test method an Rload can be applied.

You can load the reg with a DC load.

or does your recent reply mean that the Rload must be removed?

 

[AndrewT]

Edit: Kirchoff's law:................There's nothing so practical as a really good theory .../B]

This is a put down that is disrespectful and unwarranted.
Keep personalities out of a technical discussion. There are far too many others that drop to this level in their discussion. I don't think we need to take this thread in that direction.

 

[Ed LaFontaine]

I''d like to express a feeling that I hope others share:

I'd like to see the greatest benefit come out of this thread, along with the increase in understanding to be gained from looking at things in a new way.

I'm happy Janneman is taking on this effort. I think Andrew will continue to ask questions I couldn't pose.

 

[jan.didden]

I think you are missing the point.
R6=25r cannot supply 80mA. It only has ~[1.55V-0.6V] across it.
Where does 80mA come from to allow the test AC voltage to be applied?
You originally said that in the test method an Rload can be applied.
or does your recent reply mean that the Rload must be removed?

Andrew,

The schematic I have says: 25ohms - 89mA. I didn't check the actual resistor on the board, but noticed that my lab supply indicated 80mA. That's all there is to it.

As I said, a pos current into the output is only limited by the current and dissipation limit of the shunt. The max neg current you can pull out is the standing current, which I believe is 80mA or thereabouts. Do you disagree with that?

Jan Didden