Simplistic Pre-Regulator (Tracking, Mains Zero Crossing Detecting)

[robert67]

Rebuilding a vintage lab power supply (you can see how I gutted it here https://youtu.be/P1JIkpT76wY) I found myself in need of a tracking pre-regulator. Staying with the vintage theme I didn’t wanted a switch-mode type and I wanted it to be as simple as possible.
 
During my search I stumbled on the EEVblog forum upon a design that fascinated me because of its use of a thyristor for main zero crossing detection: https://www.eevblog.com/forum/projects/very-low-noise-preregulator-for-benchtop-power-supply/. So I took it and simplified it (maybe to a fault, but it’s my thing to use as few parts as possible) …

Advantages of the simplified design:

A) Of course much simpler (10 instead of 21 parts – not counting rectifier, capacitor and linear regulator)
B) Can pre-regulate down to lower voltages (using a BJT instead of MOSFETs)
C) Injects two orders of magnitude less current into the output of the linear regulator
 
Drawbacks of the simplified design:

A) Considerably less efficient (using a BJT instead of MOSFETs)
B) More expensive parts (I really maxed out on the parametric searches of the distributors)
C) Creates more noise (the pass transistor is switched off hard)

I haven’t ordered the parts for it yet, because some of them are quite expensive (e.g. $10 for a MJ11032G TO-3 BJT). Instead I wanted to put the design under your esteemed scrutiny first. So if you spot an error, please stick it to me. Or, even better, if you think it can be done with even fewer parts, please let me know. Here it goes …
 
DESIGN CONSTRAINTS

Not really technical constraints, but anyway …

1) The switching power transistor has to come in a TO-3 package (the vintage power supply has two external TO-3 heat sinks – one will be used for the pre-regulator and the other for the linear regulator – so a purely “aesthetical” constraint)

2) Can’t do anything about the AC input voltage and its wide swing (in Europe AC mains is specified as 230V +/-10%, and I’m reusing the transformer)

3) The input voltage range of the linear regulator (7V - 37V) and the maximum current (5A) is a given (OK, these are actually technical constraints)

CIRCUIT DIAGRAM

see attachment

PRINCIPLE OF OPERATION

Unless otherwise noted “current voltage” references the output of the bridge rectifier D1 at it cathodes (rectified sine wave), its anodes being the ground (0V).

The footnotes contain information in relation to the actual choice of values and parts.

A) Charging Operation

A constant current source (U1, R1) delivers a current to any point low enough below the current voltage (1).

Assuming D3 is “off” – see B) Voltage Feedback and C) Zero Crossing Detection – that current flows to the base of Q1, if the base is pulled low enough below the current voltage by the voltage drop across the collector and emitter of Q1 (2). This voltage drop includes the base-emitter forward voltage of Q1 (3).

As Q1 becomes conductive, current flows through D2 and charges C1, if the voltage drop across D2 (4) plus the voltage of C1 is below the current voltage minus the voltage drop across the collector and emitter of Q1 (3). 

(1) 2.7V
(2) 4.2V - 6.2V
(3) 1.5V - 3.5V
(4) 0.2V - 0.8V

B)  Voltage Feedback

A voltage divider (R2, R3, and R4) delivers a part of the voltage differential between C1 and the output of the linear regulator (1) to the base of Q2. If it is low enough below the voltage of C1 (2), a current flows through the base of Q2.

As Q2 becomes conductive, a current flows through its emitter and collector via R5 to the gate of D3, and switches it on.

The current from the constant current source (U1, R1) is then flowing through D3 to ground, instead to the base of Q1. The charging of C1 stops.

(1) nom. 7V
(2) 0.63V

C) Zero Crossing Detection

As long the current from the constant current source (U1, R1) flows through D3 to ground, D3 stays conductive, even if its gate no longer receives any current from Q2 through R5.

The constant current source (U1, R1) will be able to maintain that current as long as the current voltage doesn’t fall too low (2). So the charging of C1 can only resume after the current voltage has fallen to that low point.

That low point (2) is below the required voltage drop across the collector and emitter of Q2 and D2 to resume charging C1 (3). In effect the charging of C1 is further delayed until the current voltage has risen again beyond that required voltage drop, that is after its zero point.

(1) 2mA
(2) 3.6V
(3) 4.4V

CHOICE OF PARTS

U1: LM317HVT (http://www.ti.com/lit/ds/symlink/lm117hv.pdf) – At first glance a no brainer for a simple constant current source working at a maximum of 60V. I choose one in a TO220 package to keep it cool (I don’t like to burn my fingers on my breadboard). However, there are constant current LED drivers available that would do the job with just one part (e.g. AL5809-20P1-7), not needing a resistor. Alas, they are only available in SMD packages (doesn’t fit into the vintage theme at all). I also had a classical Zener (ZPY1)/transistor/resistors 4-part solution designed. Its minimum operating voltage was much lower, shaving off 12W from Q1’s worst case power loss. But its two parts more and “Winter ins Coming” (12W will help to fight the cold).

Q1: MJ11032G (http://www.onsemi.com/pub/Collateral/MJ11028-D.PDF) – Expecting high peak currents I choose the beefiest transistor in a TO-3 package I could find. It had to be a Darlington in order to keep the rest of the circuitry lean and simple.

D2: FERD40H100S (http://www.st.com/content/ccc/resource/technical/document/datasheet/group3/4e/b5/b8/47/b6/af/41/d6/DM00281757/files/DM00281757.pdf/jcr:content/translations/en.DM00281757.pdf) – A strong enough Schottky diode with an acceptable voltage drop. For various reasons I restricted myself to types coming in a TO220 package (mainly because I have about two dozen TO220 heatsinks lying around).

Q2: BC560C (https://www.fairchildsemi.com/datasheets/BC/BC556.pdf) – Nothing special here, but for the C models providing a high current amplification, thus enabling the voltage divider being high impedance. Its maximum collector-emitter voltage of 45V should be enough here, because it’s only seeing the already pre-regulated voltage.

D3: EC103D1 (http://www.ween-semi.com/documents/EC103D1.pdf) – I choose this one for its low gate trigger current. Again this allows the voltage divider in the voltage feedback part to be high impedance, because it enables Q2 to be driven with low base currents.

Thank you so much for your time!

Again, if I made some mistakes – stick it to me.
And if you think it can be done with fewer parts – please let me know.

[jaycee]

Linear Technology App Note 32 is worth a read.. theres examples of using SCR pre regulation there.
App Note 30 also covers the subject

[robert67]

Thanks for the pointers Jaycee! Both papers are very interesting reads and you can get a lot of ideas from them.

Unfortunately the BOM of the SCR pre-regulator (fig. 5 in app note 32 http://cds.linear.com/docs/en/application-note/an32f.pdf) doesn't fit my design goal of minimizing the parts count - it has about 25 And it requires an transformer with a middle tap which my vintage power supply doesn't have

In the old days SCR pre-regulators were quite common (http://powersupply.blogs.keysight.com/2013/05/more-on-early-power-supply-preregulator.html). But the simpler ones create a whole lot of noise, the SCRs being switched on somewhere within a half wave.

Linear Technology App Note 32 is worth a read.. theres examples of using SCR pre regulation there.
App Note 30 also covers the subject

[jaycee]

You can make that circuit work with the transformer you have, you would just require 4 SCR's instead of 2, and make a full wave bridge rectifier with them. Understandably though it's more complex.

I adapted the circuit in Figure 8a/b for use with an AC input - C1B and it's associated parts can go away if you have an AC input, because you can generate the higher Gate voltage using a Cockroft multiplier.

You could probably simplify it even more by using a P-channel MOSFET instead, as the concerns the designer had in 1989 dont so much apply now in 2017 - P MOSFETs with good power handling are readily available. This is exactly what Blackdog did with the pre-regulator design you linked to.

[jaycee]

heh just fully watched your video rather than skimming through it. The design seems pretty typical of the time. 10% resistors used in the current shunt is not too surprising - they would have struggled to do better than that in parts tolerance back then, at least for a sensible cost. These days, you can get 1% accurate metal strip resistors and there's no problem

It looks like they were using an RCA clone of the LM723, again, not too surprising. That chip does have its issues, though.

2x2N3055 seems a bit skimpy for 5A of output with the drop across the regulator you'd typically see. I bet it got hot.

Funnily enough I am working on a 0-30V 0-5A design myself. As a "starter for ten" I prototyped a 0-18V 0-1A design using all discrete parts, as going up in voltage/current is simply scaling this design up. I'm using nothing exotic so far - TL431 as the voltage reference, TL082's as the opamps, and TIP35C for the outputs on the small PSU. The two TIP35C's are being driven by an MJE15032, but to be honest a BD139 would probably do for just 1A.

The design is quite similar to how the Mastech power supplies work - I have another, small (probably 2-3VA) transformer providing +5/-5V supplies for the opamp. The centre tap of this floats around the main regulator's output, a neat trick. As an example of how that works, look up the ELV-22532 power supply schematic (you've probably already seen it if you've followed Blackdog's thread )

[schmitt trigger]

Pre-regulators for linear supplies have been around for a long time. People realized that sometimes the dissipated heat would be simply too much.

The first design I saw was in the late-1970s in a Popular Electronics Magazine article. I don't remember the details other than it used a pair of 723s as the controllers; one operating in its usual fashion, but the other as a comparator to trigger the SCR at a certain level.

Good luck on your project! I also love working on retro projects, and this certainly qualifies as one. Using TO3 power transistors is a must for authenticity. Hope you can get the metal-can 723s for the same reason.

[robert67]

You can make that circuit work with the transformer you have, you would just require 4 SCR's instead of 2, and make a full wave bridge rectifier with them. Understandably though it's more complex.

You're absolutely right. A transformer without a center tap together with a full bridge rectifier made of SCRs works just fine. In fact you can make the full bridge rectifier out of two diodes (from the AC inputs to one DC output) and two SCRs (from the two AC inputs to the other DC output).

[robert67]

heh just fully watched your video rather than skimming through it. The design seems pretty typical of the time. 10% resistors used in the current shunt is not too surprising - they would have struggled to do better than that in parts tolerance back then, at least for a sensible cost. These days, you can get 1% accurate metal strip resistors and there's no problem

Yeah, the current limiting was pretty awful, using a 3-position rotary switch to change between different shunts for a 0.5A, a 1.5A and a 5A limit 

2x2N3055 seems a bit skimpy for 5A of output with the drop across the regulator you'd typically see. I bet it got hot.

The transformer has actually two secondary windings: ~20Veff and ~15Veff. Their "pre-regulator" (the small board) switched between using only one winding (I guess it was the 20Veff) for lower output voltages and both windings in series for higher output voltages. So the maximum power loss must have been about 100W (20V*5A).

[jaycee]

Yeah, the current limiting was pretty awful, using a 3-position rotary switch to change between different shunts for a 0.5A, a 1.5A and a 5A limit 

Seems a bit odd. Would have been just as easy to use a pot!

The transformer has actually two secondary windings: ~20Veff and ~15Veff. Their "pre-regulator" (the small board) switched between using only one winding (I guess it was the 20Veff) for lower output voltages and both windings in series for higher output voltages. So the maximum power loss must have been about 100W (20V*5A).

Yeah, even allowing for that I'd say it was too skimpy. If you're looking for a good pass transistor to replace them, the MJ21194 is pretty hard to beat. Use an MJE15032 to drive them as darlingtons. Remember to add small (say 0.22 ohm) emitter resistors so that they current share - also do this with your 2N3055's actually!

[blackdog]

Hi Robert67,

It is good to see that you are using "my" prereg 

One "but", some of the items you left out, were not in the schematic for "fun"
There is know no limit on the "switch off" time, and you wil hate the transformer noise and high EMK pulses because of that.
Only trowing away components is almost never better, less components and stil have the same performance, thats OK

For everybody else "This is not a Thyristor pre regulator!!!!!"
Almost al Thyristor pre regulators are noisy as Hell!
This pre regulator, if build wel, with a controled "switch off" time, is low noise.

Kind regards,
Blackdog

[jaycee]

Hi Blackdog

Your articles on the other forum were great, even if reading them was a bit challenging (Google Translate can be quite imaginitive with technical terms)

I'm currently experimenting with using an LM2576 as a pre-reg, but with an extra stage of filtering. While it looks OK in LTSpice, i expect real life will be another matter!

[blackdog]

Hi jaycee,

Thank for the compliment!
It takes me a lot of effort to write those articles, i'am a dyslectic monkey. 
And if Google translates it into English, I can understand that it looks like it's written in Klingon ... 

Kind regards,
Bram

[robert67]

Hello Blackdog,

One "but", some of the items you left out, were not in the schematic for "fun"
There is know no limit on the "switch off" time, and you wil hate the transformer noise and high EMK pulses because of that.
Only trowing away components is almost never better, less components and stil have the same performance, thats OK

You're absolutely right. My design creates some nasty noise when the pass transistor switches off hard. Up to now I was totally focused on switching the pass transistor on during the zero point of a half wave. So thank you for the hint I will edit my topic and add that as drawback of my (current) design

Maybe I can somehow come up with a soft switch off with a few parts ...

For everybody else "This is not a Thyristor pre regulator!!!!!"
Almost al Thyristor pre regulators are noisy as Hell!

Yes, up to now the thread wasn't really about (flaws in) my design but SCR pre-regulators Which is also an interesting (vintage) topic though

Best Regards
Robert

[robert67]

Yeah, even allowing for that I'd say it was too skimpy. If you're looking for a good pass transistor to replace them, the MJ21194 is pretty hard to beat. Use an MJE15032 to drive them as darlingtons. Remember to add small (say 0.22 ohm) emitter resistors so that they current share - also do this with your 2N3055's actually!

In my pre-regulator design I'm using a MJ11032G: 300W, 50A (100A pulsed), NPN, darlington, TO-3. It's probably too big for the job, but I wanted to be on the save side. Anyway, it's going to replace one of the old 2N3055s on the back side.

For the linear regulator itself (which is not part of this thread) I'm currently using a single 2N5883: 200W, 25A (50A pulsed), PNP, TO-3. Yeah, it's a PNP, beefing up a reference L200 design. In between I used two of them with 0.1Ohm emitter resistors. But again, that belongs into another thread.

[not1xor1]

For everybody else "This is not a Thyristor pre regulator!!!!!"
Almost al Thyristor pre regulators are noisy as Hell!
This pre regulator, if build wel, with a controled "switch off" time, is low noise.

have you checked the RMS current through the transformer secondary?

several months ago I simulated your circuit via LTspice and (if I remember correctly) noticed that the transformer current is more than double the DC load current, so I suspect that the power (or at least great part of it) saved in the linear regulator is wasted by the transformer

of course even with a simple non linear load, like a diode bridge followed by a capacitor, 50-60% more (RMS) current goes through the transformer than through the load, but with your regulator I think that the transformer has to dissipate some 30-50% more power over a circuit with just a linear regulator

[robert67]

For everybody else "This is not a Thyristor pre regulator!!!!!"
Almost al Thyristor pre regulators are noisy as Hell!
This pre regulator, if build wel, with a controled "switch off" time, is low noise.

have you checked the RMS current through the transformer secondary?

several months ago I simulated your circuit via LTspice and (if I remember correctly) noticed that the transformer current is more than double the DC load current, so I suspect that the power (or at least great part of it) saved in the linear regulator is wasted by the transformer

of course even with a simple non linear load, like a diode bridge followed by a capacitor, 50-60% more (RMS) current goes through the transformer than through the load, but with your regulator I think that the transformer has to dissipate some 30-50% more power over a circuit with just a linear regulator

Interesting question (to Blackdog I guess). I have not given that any thought up to now. And I have to admit that I'm not a LTspice guy.

That said, your LTspice simulation results are a bit surprising. Using any kind of switching (at 50/60Hz in our case) pre-regulator instead of just a linear regulator REDUCES the active power (W) drawn from the transformer. Of the  course the apparent power (VA) will be higher, current and voltage being out of phase.

Anyway, your LTspice results suggests that the efficiency of the transformer decreases at a significantly higher rate than the active (W) power drawn from the transformer is decreased by the pre-regulator. And that is compared to the filter capacitor directly connected to the rectifier, where voltage and current are already considerably out of phase.

But the only difference here is that these pre-regulators (Blackdog's and mine) stop the current flow to the filter capacitor when it has reached the desired voltage. Blackdog's does that soft, avoiding any transients (@Blackdog: and thus noise  ). Mine does it hard, creating a transient that has somehow to be dissipated in the (magnetic field of the) transformer.

Could you tell us for which operating point of the regulator (max/min output current, max/min output voltage) you are getting those LTspice results?

[not1xor1]

For everybody else "This is not a Thyristor pre regulator!!!!!"
Almost al Thyristor pre regulators are noisy as Hell!
This pre regulator, if build wel, with a controled "switch off" time, is low noise.

have you checked the RMS current through the transformer secondary?

several months ago I simulated your circuit via LTspice and (if I remember correctly) noticed that the transformer current is more than double the DC load current, so I suspect that the power (or at least great part of it) saved in the linear regulator is wasted by the transformer

of course even with a simple non linear load, like a diode bridge followed by a capacitor, 50-60% more (RMS) current goes through the transformer than through the load, but with your regulator I think that the transformer has to dissipate some 30-50% more power over a circuit with just a linear regulator

Interesting question (to Blackdog I guess). I have not given that any thought up to now. And I have to admit that I'm not a LTspice guy.

yes, I was asking Blackdog...
I tested this kind of preregulator (but I used an opamp as a sort of flipflop) a few years ago and did not think then about checking the current through the transformer
so now I'm not sure what happens in a real circuit, i.e. I wonder if the spice model of the transformer I'm using is realistic

That said, your LTspice simulation results are a bit surprising. Using any kind of switching (at 50/60Hz in our case) pre-regulator instead of just a linear regulator REDUCES the active power (W) drawn from the transformer. Of the  course the apparent power (VA) will be higher, current and voltage being out of phase.

you lose something more in the transformer... probably just due to the Joule effect (i.e. caused by the current increase on the windings resistances)

Anyway, your LTspice results suggests that the efficiency of the transformer decreases at a significantly higher rate than the active (W) power drawn from the transformer is decreased by the pre-regulator. And that is compared to the filter capacitor directly connected to the rectifier, where voltage and current are already considerably out of phase.

But the only difference here is that these pre-regulators (Blackdog's and mine) stop the current flow to the filter capacitor when it has reached the desired voltage. Blackdog's does that soft, avoiding any transients (@Blackdog: and thus noise  ). Mine does it hard, creating a transient that has somehow to be dissipated in the (magnetic field of the) transformer.

Could you tell us for which operating point of the regulator (max/min output current, max/min output voltage) you are getting those LTspice results?

That is the same method I tested (real circuits) a few years ago... but a few months ago I made many simulations of different circuits... so now I have something more than 100 .asc files in that directory ... I just don't remember which is the right one 
but I think the problem is more noticeable at low output voltage and maximum load

in the next few days I'll see if I can find some more details

[not1xor1]

Hi, Robert67

OK, I read more carefully your introductory post
your circuit will be a sort of parody of the Vulcanian "live long and prosper"... something more like "live short and die in a puff of stinking smoke" 

just kidding... please do not take it as an offence...

the first part to die will probably be the LM317 used as constant current regulator
it may work for a short while like a zener dumping the huge voltage spikes caused by the transformer secondary winding acting like the inductor in a stepup switching supply

I ran a few simulations in LTspice comparing 3 circuits with as little as possible variations, implementing:
1) just a linear regulator
2) your schematic
3) Blackdog preregulator

of course no device runs the risk of exploding when running in software 
so I could realize that Blackdog preregulator, besides avoiding destruction of real hardware, at the cost of few more devices is even more efficient than your circuit

Regarding the Blackdog circuit, I could also check the RMS current through the secondary and noticed that while much higher than that of the plain linear regulator circuit, the transformer losses increase just about 15% (at least in that particular test: 1.3V/3.2A output)

on the other hand the preregulation is a bit lose and especially at low output voltage both the mosfets and the linear regulator waste a lot of power

Tomorrow I'll give more details with the shots of the circuits and of the measures and some more comments.
Later on I'll provide some more schematic I tested in past...

[robert67]

Hi Not1xor1 (1? Not (1 xor 1)?),

OK, I read more carefully your introductory post
your circuit will be a sort of parody of the Vulcanian "live long and prosper"... something more like "live short and die in a puff of stinking smoke" 

just kidding... please do not take it as an offence...

Good one, the Vulcan parody that is. And no offense taken! That's why I put the design out there ... to be shot down In some circles that process is called "review"  And I'm very grateful that you take your time to have a look at my Klingon abomination ("Today is a good day to die!")   

the first part to die will probably be the LM317 used as constant current regulator
it may work for a short while like a zener dumping the huge voltage spikes caused by the transformer secondary winding acting like the inductor in a stepup switching supply

Yes, I remember thinking at one point about that. But somehow I forgot about it in my quest to reduce the parts count. However, as Blackdog brought up the switch-off noise problem its was coming to me again.

I'm currently looking at the following solutions:

A) RC snubber (2 parts)
B) Zener diode (1 part) - not using the LM317 for that 
C) Soft switch-off like Blackdogs (at least a RC element and a diode, so 3 parts)
D) Any combination of A), B) and C)

I do not like the soft switch-off for two reasons: (1) It increases the power loss in the series transistor (not that this would matter in my circuit) and (2) it messes with the regulation (that's the killer for me: if +/- a few V is OK it's fine, but I want to stay within a few 100mV).

so I could realize that Blackdog preregulator, besides avoiding destruction of real hardware, at the cost of few more devices is even more efficient than your circuit

Regarding the Blackdog circuit, I could also check the RMS current through the secondary and noticed that while much higher than that of the plain linear regulator circuit, the transformer losses increase just about 15% (at least in that particular test: 1.3V/3.2A output)

on the other hand the preregulation is a bit lose and especially at low output voltage both the mosfets and the linear regulator waste a lot of power

I noticed that too about Blackdog's circuit: The MOSFETs are operated under certain (low voltage)  conditions in their linear regions. And they are always going through their linear regions during the soft switch-off. And the whole soft switch-off throws off the regulation a bit (see my dislike for it above). However, his design goal was a real quite pre-regulator, so the drawbacks in efficiency and regulation accuracy are acceptable.

My design goal was the simplest possible pre-regulator of this sort. So some noise and decreased efficiency is acceptable to me. Self-destruction of course not   

Tomorrow I'll give more details with the shots of the circuits and of the measures and some more comments.
Later on I'll provide some more schematic I tested in past...

I'm looking forward to it!

[not1xor1]

Hi Not1xor1 (1? Not (1 xor 1)?),

yes 

the first part to die will probably be the LM317 used as constant current regulator
it may work for a short while like a zener dumping the huge voltage spikes caused by the transformer secondary winding acting like the inductor in a stepup switching supply

Yes, I remember thinking at one point about that. But somehow I forgot about it in my quest to reduce the parts count. However, as Blackdog brought up the switch-off noise problem its was coming to me again.

I'm currently looking at the following solutions:

A) RC snubber (2 parts)
B) Zener diode (1 part) - not using the LM317 for that 
C) Soft switch-off like Blackdogs (at least a RC element and a diode, so 3 parts)
D) Any combination of A), B) and C)

I do not like the soft switch-off for two reasons: (1) It increases the power loss in the series transistor (not that this would matter in my circuit) and (2) it messes with the regulation (that's the killer for me: if +/- a few V is OK it's fine, but I want to stay within a few 100mV).

A) an RC snubber may help, but not so much, as far as I remember
B) a TVS diode is more appropriate (and usually powerful) than an ordinary zener, but its purpose should be that to protect from occasional spikes rather than dissipate them at all times
C) soft switch-off does work and greatly reduces the spikes and improve the overall efficiency... and you are right, you just need 3 components
D) so yes IMHO you need all of these solutions

in any case any reduction in the peak current reduces both output noise, spikes and transformer stress

if you reduce the current too much the PSU may fail to respond properly to transient load and you may decrease the overall efficiency, but there is a sweet spot where you can get the optimal result

possible solutions to reduce voltage spikes (and noise) and improve the overall efficiency while reducing the transformer stress are:
- a low value resistor (100-220m\$\Omega\$) between the bridge and Q1 collector
- rather than using a big capacitor, make a pi filter with 2 capacitors and a low value resistor with lower value capacitors

LTspice showed that these solutions do work with low voltage and a fixed load
I have to test more voltage variations and the behaviour with a fast changing load

I'll post more informations later
bye

[robert67]

Hi not1xor1,

A) an RC snubber may help, but not so much, as far as I remember
B) a TVS diode is more appropriate (and usually powerful) than an ordinary zener, but its purpose should be that to protect from occasional spikes rather than dissipate them at all times
C) soft switch-off does work and greatly reduces the spikes and improve the overall efficiency... and you are right, you just need 3 components
D) so yes IMHO you need all of these solutions

A) That's still my first choice so far. I would place the RC snubber directly across the transformer secondary, no need to let the switch-off spike travel through the rectifier first. As the pre-regulator switches off at max 37V the RC snubber needs to keep the spike just below 23V (at 60V the LM317HVT starts to fry).

B) I've ruled out zener and TVS diodes. There are simply no parts available that cut off the voltage safely at 60 volts without being "shorts" at 58.8V (made a parametric search at digikey). And MOVs would die a quick death being used 100 times per second.

C) Yes, soft switch-off works, but I still don't like that it also "softens" the regulation. Maybe I just have to get over that ...

D) So far I would go for a combination of A) and (if I can bring myself to it) C).

possible solutions to reduce voltage spikes (and noise) and improve the overall efficiency while reducing the transformer stress are:
- a low value resistor (100-220m\$\Omega\$) between the bridge and Q1 collector
- rather than using a big capacitor, make a pi filter with 2 capacitors and a low value resistor with lower value capacitors

Another three parts ...

This 100Hz switch pre-regulator is getting more complex by the minute. Kudos to Blackdog for his design!

Anyway, I'll try to calculate the values for the soft switch-off RC element.

I also explore a little the feasibility of a linear pre-regulator (something you mentioned in one of your earlier posts). But such a pre-regulator burns a lot of power in the series transistor(s) (worst case - drawing 5A at almost 0V - about 275W). Thankfully my transformer has two secondary windings (not the same voltage, that is not middle-tapped), so maybe the power loss can be reduced by "vintage" winding switching to an acceptable value.

I'll post more informations later

As always looking forward to it.

[not1xor1]

Hi not1xor1,

A) an RC snubber may help, but not so much, as far as I remember
B) a TVS diode is more appropriate (and usually powerful) than an ordinary zener, but its purpose should be that to protect from occasional spikes rather than dissipate them at all times
C) soft switch-off does work and greatly reduces the spikes and improve the overall efficiency... and you are right, you just need 3 components
D) so yes IMHO you need all of these solutions

A) That's still my first choice so far. I would place the RC snubber directly across the transformer secondary, no need to let the switch-off spike travel through the rectifier first. As the pre-regulator switches off at max 37V the RC snubber needs to keep the spike just below 23V (at 60V the LM317HVT starts to fry).

unfortunately you would need a large snubber capacitor to reduces the spikes to a reasonable value (I had to use 220µF in the simulation) and as a side effect that increases both transformer losses and overall power consumption

in the screenshots below you will notice that without snubber the spikes get to 600V which is unrealistic since the BJTs would reduce that to a much lower value (until they die)

these are the data I got (3A load):

snubber	| input power	| transf. RMS current	| transf. power loss (% rel. to lin. reg)
none	106W	7.5A	39.6W (+28%)
220µF+1ohm	117W	8.9A	47.5W (+60%)

B) I've ruled out zener and TVS diodes. There are simply no parts available that cut off the voltage safely at 60 volts without being "shorts" at 58.8V (made a parametric search at digikey). And MOVs would die a quick death being used 100 times per second.

you should use just a resistor or a 2 BJT constant current source (like I did in the simulation) to get an higher voltage tolerance

C) Yes, soft switch-off works, but I still don't like that it also "softens" the regulation. Maybe I just have to get over that ...

unfortunately with the emitter follower configuration you can't just use an RC + diode to slow down the switch-off.. just few mVs of base voltage decrease put the darlington off and a large capacitor value would take too long to be charged by the current source

D) So far I would go for a combination of A) and (if I can bring myself to it) C).

possible solutions to reduce voltage spikes (and noise) and improve the overall efficiency while reducing the transformer stress are:
- a low value resistor (100-220m\$\Omega\$) between the bridge and Q1 collector
- rather than using a big capacitor, make a pi filter with 2 capacitors and a low value resistor with lower value capacitors

Another three parts ...

This 100Hz switch pre-regulator is getting more complex by the minute. Kudos to Blackdog for his design!

Anyway, I'll try to calculate the values for the soft switch-off RC element.

I also explore a little the feasibility of a linear pre-regulator (something you mentioned in one of your earlier posts). But such a pre-regulator burns a lot of power in the series transistor(s) (worst case - drawing 5A at almost 0V - about 275W). Thankfully my transformer has two secondary windings (not the same voltage, that is not middle-tapped), so maybe the power loss can be reduced by "vintage" winding switching to an acceptable value.

I suggest you to use Blackdog's circuit
according to my simulations you just have to derate the transformer by 50%, i.e. if your transformer is rated for 5A on a resistive load you should limit it to 2.5A
BTW I noticed that the maximum stress on the transformer is at about 50% of max output voltage

Radj:                   1m      315    550    1k    1.5k    2k    2.7k
output voltage          1.2V    5.2    8.1   13.7   19.9   26.2   34.9
transf. current (RMS)   6.8A    =      =      6.7    6.5    6.3    5
input power           104.2W  112.7  120    133.6  147.5  159.2  162.4
transformer loss       32.8W   33.4   33.8   34.6   35.2    =     29.8
rect. bridge loss      12.3W    =     12.2   =      12     12.6   10.5
prereg. switch loss    30.8W   29.3   27.6   23.9   18.9   12.6    0.5
linear regulator loss  21.7W   19.5   19.3   18.9   18.4   17.9   15.6
load power              3.7W   15.5   24.3   41.1   59.8   78.5  104.7




the other circuit I mentioned in a previous post works like Blackdog's one, but keeps a tighter control on the delta Vprereg-out and manages to save a few Ws of power, but requires an additional transformer (or winding)

more efficiency may be achieved by a circuit which switches off on zero-cross, like the SCR ones, but it requires a large inductor and an awfully complicated circuit to properly manage output voltage, load and AC line variations...

the best solution IMHO would be to just use a buck converter IC as preregulator

[blackdog]

Hi,

Some remarks...
Normaly its about 60 to 65% from the transformer current you can use if there is a bridge and buffer capacitor and not 50% (but most of the time i use 50% to be on the save side)
In my design there is 2x 10000uF and i went back to 10000 a 12000 uF, this wil dubble the ripple but the active rectifier with the LT4320 keep my losses down so there wil be no drupout problems.
So there is more voltage on the buffer capacitor and of course extra ripple, maybe a 33V transformer is a better choice for 5-A at 30V...
This has al to do with the variation of the Power Line you are willing to accept.
A higher transformer voltage will give you a 5A-30V output by lower Powerline voltages, but also higher dissipation.

Because now the peak currents go down because the buffer elco is smaller, the dissipation in the transformer becomes lower, but also in the pre-regulator it becomes smaller.
The reference section is so good, this together with the high gain of the opamps means that you do not find the extra ripple on the buffer capacitor at the output of the power supply.

On the other forum (Circuitsonline.net) you cant find the latest version of the schematic, i'am stil working on it.
There is now a new candidate for the opamps, and dat is the OPA140 series.

This design of the preregulator is made to get the Power dissipation down, not to be the best in efficiency, and do it with low noise.

Hi not1xor1
the best solution IMHO would be to just use a buck converter IC as preregulator
You must be kidding...

I design a extreem Low Noise Lineair power supply an you start to talk about Buck converters.
Switching preregulator in low noise design is like Water and Fire... :-)
And yes a know, with high kost and a long design time, and a lot of testing you can do it.

But what i have seen from the professionals, no..., always high noise levels if they use switchers in front a analog regulator.
If you build my design wel, the 22KHz bandwith noise wil be smaler than 5uV and the Ri extreem low til at least 100KHz.

Everybody who thinks he can cut corners in my design will throw away performance, and spice is nice, but building and measuring is better.
You have to tune the preregulator like a stated for the transformer you are using.

Also the TVS diode for your transformer and line voltage and also the snubber, maybe you need to change the 2.2Ohm and 3.3uF...
If you are in a 60Hz country, you can use 20% smaler value for the buffer capacitors.
In a 50HZ country you can use this rule of thumb: 1A and 10000uF wil give you 1V TopTop Ripple.
Another rule of thumb is you need 2000 tot 3000uF per Ampere buffer capacitor, most designers use the 2000uF in there design.

But...
Even if you do the electronics the good way, and make a mess of the wiring, you wil trow the performance of this design out of the window 

Kind regards
Blackdog

PS
Most simpel solution is a 32V transformer with tabs, 8V, 16V, 24V and 32V.
Or more simpel, use 2 transformers of 2x9V, 150VA a 160VA and use some relais for switching or triacs

[robert67]

Hi not1xor1, Hi Blackdog,

please don't forget that my design goal is "as simple as possible" and that I'm willing to sacrifice some (a whole lot) of efficiency for it.

@Blackdog: Of course my (not obviously ) not working design sacrificed some performance compared to your original. I've stated that and I was willing to do so to save a few components. Anyway, I might go back to an older version of my design that was a lot closer to yours (still a few parts less  ).

@not1xor1: Thank you so much for the simulations! The results are most educating (if not a bit sobering). I was also busy myself and added kind of an "active" RC snubber and the soft-switch off:

See attachment Pre-Regulator 4.1.jpg (I really have to learn how to include attachments as images in the HTML)

The soft switch-off (D5, R8, R3) actually works as long as there is a load, that is current is drawn from the filter cap C1. And if there's no current drawn from C1 there's also no switch-off spike. In addition it doesn't interfere with the regulation. The soft switch-off can only supply base current when the voltage of C1 is decreasing, so no overshoot. However, if there is a load transient at switch off time from, e.g. 5A to 0A, the soft switch-off can not supply any current to the base and there is a spike 

The "active" RC snubber (D4, R6, R7, C2) works as follows: D4 just prevents the base of getting any current from the snubber. R7 discharges C2 while D3 is off, so it's nice and empty to absorb the switch-off spike. R6 is just a security measurement (that thyristor can take up to 16A pulses). Please ignore the values of the components, I had not the time to calculate them properly.

I also doodled together a simple linear pre-regulator (you suggested in a previous post it would be the best for the transformer - don't try to simulate that  ):

See attachment: Pre-regulator 6.jpg

That thing burns a whole lot of power. Probably more than the heat sink on my vintage power supply can take. Anyway, I mainly did it to compare the required number of parts.

Anyway, best regards to both of you

[not1xor1]

Hello Blackdog,

Normaly its about 60 to 65% from the transformer current you can use if there is a bridge and buffer capacitor and not 50% (but most of the time i use 50% to be on the save side)
In my design there is 2x 10000uF and i went back to 10000 a 12000 uF, this wil dubble the ripple but the active rectifier with the LT4320 keep my losses down so there wil be no drupout problems.
So there is more voltage on the buffer capacitor and of course extra ripple, maybe a 33V transformer is a better choice for 5-A at 30V...

Transformers are rated in VA, that is under a plain resistive load.
When you connect a rectifier bridge, a levelling capacitor and a resistive (or constant current load), both the phase and the RMS value of the transformer current change, so, in order to provide the nominal current to the load, the transformer has to withstand more power losses.

I made a rough model of a 30V 150VA transformer and simulated the behaviour with a pure resistive load compared to rect. bridge + capacitor and got the following results, that while might not have an absolute value,  IMHO have a meaningful comparative one:

load type   |        load values      |   transf.values         
            | V (RMS) I (RMS)  power  |  I (RMS)  power loss
AC resistor | 33.1V    5A     165.8W  |   5  A    27.3W
DC 3A       | 40.7V    3A     122  W  |   4.8A    30.3W
DC 5A       | 36.8V    5A     184  W  |   8.2A    55.7W

This shows that even derating the transformer to 60% of the nominal current the power losses increase by about 10%. That usually is not a problem, especially with large transformers in well ventilated cases.

But now, your preregulator circuit adds even more distortion to the transformer current and with a 3A constant current load and depending on the output voltage, the RMS current through the transformer secondary varies between 5A and 6.8A and the power losses go up to 35.2W (pls. notice that I tested just 7 different output voltages).

So due to the increase in transformer stresses, when using such a preregulator, I think it would be wise to limit the load current to 50% or less of the nominal transformer current.

Because now the peak currents go down because the buffer elco is smaller, the dissipation in the transformer becomes lower, but also in the pre-regulator it becomes smaller.

The capacitor value seems to have little effect on the transformer power losses, at least according to LTspice (transf. + brdige + cap + 5A constant current load, 10-20 seconds interval averages)

capacitor   transf. RMS current / power losses
 3300µF     7.6359A / 55.2  W
 6600µF     7.541 A / 55.585W
13300µF     7.4929A / 55.826W

That is: a 400% increase in capacitor value produces just  1% increase in transformer power losses.

Hi not1xor1
the best solution IMHO would be to just use a buck converter IC as preregulator
You must be kidding...

I design a extreem Low Noise Lineair power supply an you start to talk about Buck converters.
Switching preregulator in low noise design is like Water and Fire... :-)
And yes a know, with high kost and a long design time, and a lot of testing you can do it.

well... there are plenty of low noise IC switching regulators which may serve well the main goal stated by Robert67, i.e. minimum components count, and it is easy to filter out the high frequency noise with the aid of one or more LC filter stages between the preregulator and the postregulator

But what i have seen from the professionals, no..., always high noise levels if they use switchers in front a analog regulator.
If you build my design wel, the 22KHz bandwith noise wil be smaler than 5uV and the Ri extreem low til at least 100KHz.

Apart that the PSU ripple and noise is usually specified for a 20Hz-20MHz bandwidth, your preregulator design, while low noise, still is affected by two little defects:
- high frequency noise due to the charge of the 100µF input capacitor quickly transferred on each zero-cross
- on low load you also have very low frequency noise since the output capacitor is mainly charged via the input one and just once in a while by the transformer

See the the screenshots below with your preregulator low load noise compared to an SCR-style (using mosfets) preregulator at low load noise (which I left on the virtual world of LTspice due to the overcomplicated circuit and unrealistic value of some components)




bye

[jaycee]

I made a rough model of a 30V 150VA transformer

Would you be willing to share that model ? So far in my own power supply simulations, ive just used AC voltage sources, and added some Rser in a crude attempt to model the voltage drop under load. I'd very much like to see a real transformer model

[robert67]

Hi not1xor1, Hi Blackdog,

Concerning the use of a buck converter:

1. It's not vintage
2. A tracking one is also a dozen parts or more
3. Otherwise it is a highly efficient solution (but efficiency is not my design goal)

Concerning my vintage transformer:

1. It was good for a 30V/5A purely linear regulated lab power supply
2. Besides the output voltages without load I know nothing else about it
3. It's from a time when we had 220V instead of 230V AC in Europe

Concerning my new approach

1. 13 parts (without rectifier and filter cap)
2. Soft on and soft off (low noise, no spikes)
3. Of course wasting power in the series transistor 
4. Relatively nice to the transformer

Reasoning:

1. Why is a zero point detection needed?
2. Look at it as a linear regulator with a very ripply input voltage (Vripple = Vp)

Design:

1. The power part with constant current source stays the same
2. Instead of the thyristor a BJT can gracefully "steal" current from the series transistor's base
3. The voltage feed back stays almost the same, just introducing some resistance before the base of the PNP BJT to "soften" it's on/off
4. The zener plus resistor is just a stop-gag measure, because the regulation is in fact quite crappy/soft



Questions:

@Blackdog: Do I have your approval noise wise?
@not1xor1: Do I have your approval transformer stress wise?
Who knows this song: "This is my threaaa-aaad, and I cry if I want to, cry if I want to. You would cry too if this happens to you!" 

[not1xor1]

Hi,

See attachment Pre-Regulator 4.1.jpg (I really have to learn how to include attachments as images in the HTML)

you first have to post the message, then click on the Modify link opening a new tab window (with firefox I use the middle mouse button)

as soon as the new tab opens, go back to the previous one, where you have your original message, and copy to the clipboard the address of the image you want to embed

then return to the edit tab and click on the small Gioconda icon (first from the left on the second raw) to embed an image and paste the address between the tags

once you have finished just save your modified message

The soft switch-off (D5, R8, R3) actually works as long as there is a load, that is current is drawn from the filter cap C1. And if there's no current drawn from C1 there's also no switch-off spike. In addition it doesn't interfere with the regulation. The soft switch-off can only supply base current when the voltage of C1 is decreasing, so no overshoot. However, if there is a load transient at switch off time from, e.g. 5A to 0A, the soft switch-off can not supply any current to the base and there is a spike 

The "active" RC snubber (D4, R6, R7, C2) works as follows: D4 just prevents the base of getting any current from the snubber. R7 discharges C2 while D3 is off, so it's nice and empty to absorb the switch-off spike. R6 is just a security measurement (that thyristor can take up to 16A pulses). Please ignore the values of the components, I had not the time to calculate them properly.

according to LTspice the effect of the soft switch is very small as the base voltage drops too quickly
a darlington transconductance (deltaIc/deltaVbe) is just too high

the snubber takes most of the burden, but at the expense of sinking almost 200mA RMS at low output voltage

I also doodled together a simple linear pre-regulator (you suggested in a previous post it would be the best for the transformer - don't try to simulate that  ):

See attachment: Pre-regulator 6.jpg

I didn't mean that
I just stated that a preregulator increases the transformer losses compared to any linear regulator so you just have to reduce a bit the maximum allowable current of the PSU

[not1xor1]

I made a rough model of a 30V 150VA transformer

Would you be willing to share that model ? So far in my own power supply simulations, ive just used AC voltage sources, and added some Rser in a crude attempt to model the voltage drop under load. I'd very much like to see a real transformer model

I do not have neither the proper instrumentation, nor a real 30V transformer, so it is just a gamble 

I based the old model on the measurements (just inductance and resistance) of another 150VA 24V transformer.
Now I measured those data again with a better instrument, but had anyway to modify them a bit to get a reasonable behaviour...

The new model is just  a couple of inductances with the K spice directive:

* 230-30VAC-5A-transf
.subckt 230-30VAC-5A-transf 1 2 3 4
L1 1 2 .975 Rser=21
L2 3 4 22.3m Rser=0.25
R1 4 2 1G
K1 L1 L2 0.99
.end


Now I'm going to run a few tests to check if it may approximate fairly well a real transformer
Later will post an attachment with all the asc asy and sub files

[robert67]

Hi not1xor1,

you first have to post the message, then click on the Modify link opening a new tab window (with firefox I use the middle mouse button)

as soon as the new tab opens, go back to the previous one, where you have your original message, and copy to the clipboard the address of the image you want to embed

then return to the edit tab and click on the small Gioconda icon (first from the left on the second raw) to embed an image and paste the address between the tags

once you have finished just save your modified message

Thank you for that! The wiki help for the forum had nothing on that. I'll try that right away with my last post.

The soft switch-off (D5, R8, R3) actually works as long as there is a load, that is current is drawn from the filter cap C1. And if there's no current drawn from C1 there's also no switch-off spike. In addition it doesn't interfere with the regulation. The soft switch-off can only supply base current when the voltage of C1 is decreasing, so no overshoot. However, if there is a load transient at switch off time from, e.g. 5A to 0A, the soft switch-off can not supply any current to the base and there is a spike 

The "active" RC snubber (D4, R6, R7, C2) works as follows: D4 just prevents the base of getting any current from the snubber. R7 discharges C2 while D3 is off, so it's nice and empty to absorb the switch-off spike. R6 is just a security measurement (that thyristor can take up to 16A pulses). Please ignore the values of the components, I had not the time to calculate them properly.

according to LTspice the effect of the soft switch is very small as the base voltage drops too quickly
a darlington transconductance (deltaIc/deltaVbe) is just too high

the snubber takes most of the burden, but at the expense of sinking almost 200mA RMS at low output voltage

200mA RMS at 40Veff is just about 8W, I could live with that (burning more than 30W in my series BJT at high loads).

I also doodled together a simple linear pre-regulator (you suggested in a previous post it would be the best for the transformer - don't try to simulate that  ):

See attachment: Pre-regulator 6.jpg

I didn't mean that
I just stated that a preregulator increases the transformer losses compared to any linear regulator so you just have to reduce a bit the maximum allowable current of the PSU

Got it. But I wanted to explore that avenue anyway, just to see if it is in feasible any way (which it is not).

Any thoughts on my latest brain aneurysm (Reply #26 - now with picture in the text )?

[not1xor1]

Any thoughts on my latest brain aneurysm (Reply #26 - now with picture in the text )?

First thing, not related to your question, is that I always forgot to write you can just use a depletion mosfet and a resistor to get a current source withstanding 100s of Vs, rather than an LM317.

Now returning to your circuit... I've not simulated it but got the feeling that, with enough load current, it would be a great noise generator, a sort of low frequency inductor-less buck converter, but with unpredictable behaviour due to the current leakage through the zener diode which might be enough to switch on Q4 depending on the temperature... a resistor between BE would solve that problem, but not the other ones

You might reduce some of the spikes and get some more hysteresis through a capacitive divider, i.e. connecting for a short while a 100-470µF capacitor from the transformer to the 12.000µF one, it would need a series resistor to limit the peak current, a Pmosfet to open/close the circuit, a reverse polarized schottky diode to dissipate the capacitor charge when you switch off the Pmosfet and  some sort of timer or another higher threshold Vpre-Vpost comparator to keep it switched on just a while... so it would just became much more complicated with little benefit.

If a vintage solution is your main goal, if I were you, I would explore an SCR solution...

The clean way is to charge a capacitor through a resistor discharging it on each zero-cross to get a 100Hz sawtooth sync-ed with the transformer secondary.
Then you compare the difference of Vout - Vprereg with the sawtooth value and as soon as the sawtooth goes above the delta you switch the SCRs on.

A several mH inductor (like those used in passive PFC PSU) would avoid a sudden rise of current through the transformer and when the voltage gets down so that there is no current through the SCRs and they switch-off, a schottky diode would let to discharge the rest of the energy stored in the inductor to the levelling capacitor.

As far as I can remember (I ran the simulations about one year ago), at least in the magic world of LTspice  , that circuit can provide high efficiency and low noise if you use mosfets rather than SCRs and update constantly the Vout-Vprereg threshold according to the set output voltage, the output current and the transformer voltage variations (not instantaneous, but the RMS ones, i.e. those due the AC line) and put a large non polarized capacitor in parallel to the transformer secondary.

But you might just keep it simple, still saving some power, by using a high enough Vout-Vprereg threshold and skipping all those sophistication reducing so the required components.

That is:
- skip the sawtooth comparison circuit which charge the levelling capacitor as suitable each 10 milliseconds
- just use a larger threshold and switch on the SCRs as soon as the voltage gets below that

Sometimes it would get 2 cycles to return quite above the threshold or you might skip many cycles when you have low load and so would have some very low frequency noise depending on the deltaVout/deltaVin capability of the linear regulator, but it should work.

[not1xor1]

But you might just keep it simple, still saving some power, by using a high enough Vout-Vprereg threshold and skipping all those sophistication reducing so the required components.

That is:
- skip the sawtooth comparison circuit which charge the levelling capacitor as suitable each 10 milliseconds
- just use a larger threshold and switch on the SCRs as soon as the voltage gets below that

Sometimes it would get 2 cycles to return quite above the threshold or you might skip many cycles when you have low load and so would have some very low frequency noise depending on the deltaVout/deltaVin capability of the linear regulator, but it should work.

OK that seems to work... just 3 BJTs and one PNP darlington and 8 resistors... and much lower ripple than I expected.
But I still have to check for peak currents and power losses, transient load, low load, etc...


Epic fail... I made a big mistake... too good to be true 

BTW, Jaycee, I didn't forget that transformer model... later I'll post everything

[not1xor1]

I'm attaching a zip file with the transformer model and 3 different test circuits as ltspice .asc files.
.meas statement are included in the .asc files.

have fun

BTW tomorrow I'll post a few screenshots and the simulation files of a preregulator which may suite Robert67 needs. It is SCR-style but uses a PNP darlington as switch. Ripple is high (less than 10mV) but there are no spikes.

[not1xor1]

Concerning my new approach

1. 13 parts (without rectifier and filter cap)
2. Soft on and soft off (low noise, no spikes)
3. Of course wasting power in the series transistor 
4. Relatively nice to the transformer

Reasoning:

1. Why is a zero point detection needed?
2. Look at it as a linear regulator with a very ripply input voltage (Vripple = Vp)

Design:

1. The power part with constant current source stays the same
2. Instead of the thyristor a BJT can gracefully "steal" current from the series transistor's base
3. The voltage feed back stays almost the same, just introducing some resistance before the base of the PNP BJT to "soften" it's on/off
4. The zener plus resistor is just a stop-gag measure, because the regulation is in fact quite crappy/soft

I tested that kind of circuit years ago, but I was using MOSFETs then, while BJTs are much slower and your circuit, under some circumstances (regarding set voltage and load) behaves perfectly.
Unfortunately in some instances it shows the defects I verified with MOSFETs, although to a lesser extent. In the screenshot below you will see:
- V(base,N001): darlington Q1 is switched twice per half wave (i.e. at 200Hz)
  with mosfets that might be even 4-500Hz or more with huge spikes
- the red trace V(bridgeout) shows 80V spikes and some distortion the second time Q1 is switched on
- the sky-blue trace shows the current through D2
- output ripple with 3A load is anyway quite good at less than 1mV


And you cannot make it slower by just reducing the base current of Q4 since BJT parameters are highly dependent on temperature, besides that components age and resistor values drift and trimmer resistors age and drift even more (unless you want to invest 10ths of € in metal foil metrology grade trimmer  ).

The circuit may be anyway quite valuable if we manage to make it reliably (i.e. independently of component tolerances and value drift) slower.

I'm attaching the current .asc file in case somebody else wants to play with it.

BTW I suspect that sort of crowbar made by D3, R5, Q4 and Q1 being too slow to protect other circuits from damage.

P.S. it looks like files with .lib extensions can't be uploaded so I changed the extension of the darlington model to .txt. Now to make the simulation work either modify the include statement or change back the extension to .lib.

[oldway]

There is something we have not talked about for phase controlled pre regulators: the characteristics of the transformer.

Indeed, many make mistakes in choosing the transformer: they choose a toroidal transformer, which is the worst choice.

Why ?
Because a toroidal transformer has a very small short circuit impedance, which causes a very bad operation of the pre regulator with high peak current pulses, high rms current, strong heating of the transformer and diodes / scr's, high noise, low range of control angles, and so on....

A pre-regulator requires a high short-circuit impedance transformer (of the order of 20%) .... This is achieved by winding primary and secondary side by side, such as transformers for battery charger or MMA welders, or transformers for microwave ovens (MOT).
However, they have a magnetic shunt, which is not necessary for pre regulators.

[not1xor1]

The circuit may be anyway quite valuable if we manage to make it reliably (i.e. independently of component tolerances and value drift) slower.

Reducing by 10 times the current from the current source (not feasible with LM317 that's specified at 10mA minimum load) and placing a 2.2nF cap between C-B of Q4 seems to cure all problems.

It looks quite promising.

[oldway]

As for the original schematics, it is not a phase controlled pre regulator and it is a real disaster in terms of efficiency. 
Indeed, the drops in voltages are very high: 1.2V on the bridge rectifier, 1.2V on Q1, 0.4V on D2, 1.25V on LM317, ie in total 4.05V.

EDIT: you did not use a LM317 but current source by 2 transistors instead....then, 0.7V instead of 1.25V, that's 3.5V ....
EDIT2: using a thyristor as in original schematic, Q1 only will be full on or full off....this mean reduced power dissipation.
Without thyristor, Q1 will work in the linear caracteristics, it will limit current and dissipate a lot.
I already tried such a schematic for battery chargers, it is useless because it has too much loss.

[robert67]

Any thoughts on my latest brain aneurysm (Reply #26 - now with picture in the text )?

First thing, not related to your question, is that I always forgot to write you can just use a depletion mosfet and a resistor to get a current source withstanding 100s of Vs, rather than an LM317.

I'll keep that in mind for future designs.

Now returning to your circuit... I've not simulated it but got the feeling that, with enough load current, it would be a great noise generator, a sort of low frequency inductor-less buck converter, but with unpredictable behaviour due to the current leakage through the zener diode which might be enough to switch on Q4 depending on the temperature... a resistor between BE would solve that problem, but not the other ones

I did know the regulation is crappy (as I wrote), but I didn't thought about the zener leakage. Thanks!

You might reduce some of the spikes and get some more hysteresis through a capacitive divider, i.e. connecting for a short while a 100-470µF capacitor from the transformer to the 12.000µF one, it would need a series resistor to limit the peak current, a Pmosfet to open/close the circuit, a reverse polarized schottky diode to dissipate the capacitor charge when you switch off the Pmosfet and  some sort of timer or another higher threshold Vpre-Vpost comparator to keep it switched on just a while... so it would just became much more complicated with little benefit.

That seems to be the theme of this thread: Simple, but to get it actually working you have to make it much more complicated 

If a vintage solution is your main goal, if I were you, I would explore an SCR solution...

The clean way is to charge a capacitor through a resistor discharging it on each zero-cross to get a 100Hz sawtooth sync-ed with the transformer secondary.
Then you compare the difference of Vout - Vprereg with the sawtooth value and as soon as the sawtooth goes above the delta you switch the SCRs on.

A several mH inductor (like those used in passive PFC PSU) would avoid a sudden rise of current through the transformer and when the voltage gets down so that there is no current through the SCRs and they switch-off, a schottky diode would let to discharge the rest of the energy stored in the inductor to the levelling capacitor.

As far as I can remember (I ran the simulations about one year ago), at least in the magic world of LTspice  , that circuit can provide high efficiency and low noise if you use mosfets rather than SCRs and update constantly the Vout-Vprereg threshold according to the set output voltage, the output current and the transformer voltage variations (not instantaneous, but the RMS ones, i.e. those due the AC line) and put a large non polarized capacitor in parallel to the transformer secondary.

But you might just keep it simple, still saving some power, by using a high enough Vout-Vprereg threshold and skipping all those sophistication reducing so the required components.

That is:
- skip the sawtooth comparison circuit which charge the levelling capacitor as suitable each 10 milliseconds
- just use a larger threshold and switch on the SCRs as soon as the voltage gets below that

Sometimes it would get 2 cycles to return quite above the threshold or you might skip many cycles when you have low load and so would have some very low frequency noise depending on the deltaVout/deltaVin capability of the linear regulator, but it should work.

I tried to comprehend what you're writing. And it seems to work ... but I'm not sure. Guess I need to draw another circuit diagram 

[robert67]

A pre-regulator requires a high short-circuit impedance transformer (of the order of 20%) .... This is achieved by winding primary and secondary side by side, such as transformers for battery charger or MMA welders, or transformers for microwave ovens (MOT).
However, they have a magnetic shunt, which is not necessary for pre regulators.

Well, my vintage transformer is neither toroidal nor does it use side-by-side windings. Its a nice square shape hulk of laminated metal with under-over windings. So I guess its impedance is somewhere in the middle. 

[robert67]

But you might just keep it simple, still saving some power, by using a high enough Vout-Vprereg threshold and skipping all those sophistication reducing so the required components.

That is:
- skip the sawtooth comparison circuit which charge the levelling capacitor as suitable each 10 milliseconds
- just use a larger threshold and switch on the SCRs as soon as the voltage gets below that

Sometimes it would get 2 cycles to return quite above the threshold or you might skip many cycles when you have low load and so would have some very low frequency noise depending on the deltaVout/deltaVin capability of the linear regulator, but it should work.

OK that seems to work... just 3 BJTs and one PNP darlington and 8 resistors... and much lower ripple than I expected.
But I still have to check for peak currents and power losses, transient load, low load, etc...


Epic fail... I made a big mistake... too good to be true 

"If it's to good to be true, it's probably not true." 

I also struggled controlling those SCRs (with a "reasonable" number of parts ) but I completely failed 

[oldway]

...
Well, my vintage transformer is neither toroidal nor does it use side-by-side windings. Its a nice square shape hulk of laminated metal with under-over windings. So I guess its impedance is somewhere in the middle.

The transformer of your 1993 bench power supply use side by side windings, that's ok.

[robert67]

I tested that kind of circuit years ago, but I was using MOSFETs then, while BJTs are much slower and your circuit, under some circumstances (regarding set voltage and load) behaves perfectly.
Unfortunately in some instances it shows the defects I verified with MOSFETs, although to a lesser extent. In the screenshot below you will see:
- V(base,N001): darlington Q1 is switched twice per half wave (i.e. at 200Hz)
  with mosfets that might be even 4-500Hz or more with huge spikes
- the red trace V(bridgeout) shows 80V spikes and some distortion the second time Q1 is switched on
- the sky-blue trace shows the current through D2
- output ripple with 3A load is anyway quite good at less than 1mV

I was actually thinking about changing to a MOSFET as series transistor. There seems to be some old stock of TO-3 types (IRF140, IRF250, BUZ24, IRF150, 2N6764) available on eBay (European sellers - I wouldn't trust an "IRF150" for €3.99 from China).

The circuit may be anyway quite valuable if we manage to make it reliably (i.e. independently of component tolerances and value drift) slower.

Yes. I still like the basic idea of it.

However, making the regulation more reliable/precise with the current amplifications of three BJTs basically multiplied (one of them being an darlington) might be a mission impossible (and I'm not the Tom Cruise of electronics  ).

Anyway, I thought of using op amps for the regulation, getting rid of all the ever changing h(FE) and V(BE). Maybe that will work out somehow. But I'm skeptical if I can keep the parts count "reasonable" 

[robert67]

The circuit may be anyway quite valuable if we manage to make it reliably (i.e. independently of component tolerances and value drift) slower.

Reducing by 10 times the current from the current source (not feasible with LM317 that's specified at 10mA minimum load) and placing a 2.2nF cap between C-B of Q4 seems to cure all problems.

It looks quite promising.

I'll change the current source to 2mA with a depletion mode FET plus resistor and add the cap in my circuit diagram.

[robert67]

As for the original schematics, it is not a phase controlled pre regulator and it is a real disaster in terms of efficiency. 
Indeed, the drops in voltages are very high: 1.2V on the bridge rectifier, 1.2V on Q1, 0.4V on D2, 1.25V on LM317, ie in total 4.05V.

Yes it is The numbers are in my circuit diagram (over 30W are burned worst case in the series BJT). But then, efficiency was never one of my design goals. Simplicity (low parts count) and a the use of a TO-3 transistor are. And I will quite happily burn up to, say 70-80W, to achieve those 

EDIT: you did not use a LM317 but current source by 2 transistors instead....then, 0.7V instead of 1.25V, that's 3.5V ....

But that's two parts more  (see above)

EDIT2: using a thyristor as in original schematic, Q1 only will be full on or full off....this mean reduced power dissipation.
Without thyristor, Q1 will work in the linear caracteristics, it will limit current and dissipate a lot.
I already tried such a schematic for battery chargers, it is useless because it has too much loss.

The Vulcan science council (see the whole thread) has determined that a hard switch off via the thyristor causes (A) destructive voltage spikes due to the transformers inductivity [not1xor1] and (B) unacceptable noise [Blackdog] 

Basically you're buying "nice and quite" for selling out efficiency. If I wanted real high efficiency I could simply go for a buck converter (which has already been suggested [not1xor1]). But that would neither be "vintage" (so a no-go for me) nor low noise (for which that suggestion has already be damned  [Blackdog]).

[not1xor1]

The circuit may be anyway quite valuable if we manage to make it reliably (i.e. independently of component tolerances and value drift) slower.

Yes. I still like the basic idea of it.

However, making the regulation more reliable/precise with the current amplifications of three BJTs basically multiplied (one of them being an darlington) might be a mission impossible (and I'm not the Tom Cruise of electronics  ).

Anyway, I thought of using op amps for the regulation, getting rid of all the ever changing h(FE) and V(BE). Maybe that will work out somehow. But I'm skeptical if I can keep the parts count "reasonable" 

After many simulations I think that there are really no shortcuts.
You have to take into account many different parameters, i.e. power dissipation, spikes, output noise, etc...
Some circuit may work at a given output voltage, but may cause huge spikes at another one, or may respond quite badly to a transient load, or may burn your transformer, etc...

If your goals are save just a little of power, use not too many components and use a vintage circuit then it may be better to just make use of the 2 transformer taps and use a plain linear regulator like in this old oltronix circuit where a darlington and a couple of diodes are used to switch between the 2 different DC voltage levels:

[robert67]

The circuit may be anyway quite valuable if we manage to make it reliably (i.e. independently of component tolerances and value drift) slower.

Yes. I still like the basic idea of it.

However, making the regulation more reliable/precise with the current amplifications of three BJTs basically multiplied (one of them being an darlington) might be a mission impossible (and I'm not the Tom Cruise of electronics  ).

Anyway, I thought of using op amps for the regulation, getting rid of all the ever changing h(FE) and V(BE). Maybe that will work out somehow. But I'm skeptical if I can keep the parts count "reasonable" 

After many simulations I think that there are really no shortcuts.
You have to take into account many different parameters, i.e. power dissipation, spikes, output noise, etc...

Maybe you're right Anyway, I've added the additions you suggested (lower constant current, cap between collector and base) to the circuit diagram. However, now that the constant current is lower every thing else needs to be re-dimensioned too. And that leads to more problems ...



But I've mentioned in my last post that I thought about using opamps for the regulation ... So I'm given it one more try Instead of using two BJTs to do the regulation I went for a single opamp used as inverting summing amplifier and a MOSFET as series transistor. The opamp "ground" is the voltage of the main filter cap C1. It is summing the output voltage of the linear regulator (negative to the opamp "ground", that is V at C1) and a reference voltage (positive to the "opamp" ground, that is V at C1). If the output voltage of the linear regulator is lower than the reference voltage, the output of the opamp is positive, driving the MOSFET open, and vice versa. The gain of the opamp controls how "soft" the switching is. E.g. at a gain of 10 a voltage difference of 1V is needed to switch on the MOSFET fully. Every voltage difference below that will drive the MOSFET into its linear region.



That's my last try at a "simplistic" pre-regulator. If that doesn't work I will go either for an evil buck pre-regulator, or completely forget about pre-regulation and do as you suggested (using the two secondary windings - Incidentally I currently do exactly that while breadboarding my linear regulator ).

[not1xor1]

But I've mentioned in my last post that I thought about using opamps for the regulation ... So I'm given it one more try Instead of using two BJTs to do the regulation I went for a single opamp used as inverting summing amplifier and a MOSFET as series transistor. The opamp "ground" is the voltage of the main filter cap C1. It is summing the output voltage of the linear regulator (negative to the opamp "ground", that is V at C1) and a reference voltage (positive to the "opamp" ground, that is V at C1). If the output voltage of the linear regulator is lower than the reference voltage, the output of the opamp is positive, driving the MOSFET open, and vice versa. The gain of the opamp controls how "soft" the switching is. E.g. at a gain of 10 a voltage difference of 1V is needed to switch on the MOSFET fully. Every voltage difference below that will drive the MOSFET into its linear region.



That's my last try at a "simplistic" pre-regulator. If that doesn't work I will go either for an evil buck pre-regulator, or completely forget about pre-regulation and do as you suggested (using the two secondary windings - Incidentally I currently do exactly that while breadboarding my linear regulator ).

I think I already mentioned in a previous post an old schematic I devised years ago that uses a schmitt trigger as sort of SR flip flop.
Here it is:



Most of the spikes are eaten by the capacitor (3 100uF low ESR in parallel). The discharge resistor and the fill resistor are used to cut the input spikes when the mosfet is switched on.

The switch speed is easily tuned by changing C2 while the preregulator delta is almost independent of the output voltage and can be changed by adjusting the R3-R5 voltage divider.

Rfill dissipates quite a lot of power on high load and low output voltage, but the overall power loss is still low and both the mosfet and the linear regulator do not dissipate much power.

This is appropriate especially when you use a linear regulator bootstrapped by the positive voltage.

[robert67]

... The gain of the opamp controls how "soft" the switching is. E.g. at a gain of 10 a voltage difference of 1V is needed to switch on the MOSFET fully. Every voltage difference below that will drive the MOSFET into its linear region.

I think I already mentioned in a previous post an old schematic I devised years ago that uses a schmitt trigger as sort of SR flip flop.
Here it is:



Most of the spikes are eaten by the capacitor (3 100uF low ESR in parallel). The discharge resistor and the fill resistor are used to cut the input spikes when the mosfet is switched on.

Thanks for reminding me about that solution of yours. Going through all the alternatives I forgot about that one.

Anyway, the idea of my last design was to avoid switch-off spikes completely (by wasting power in the series transistor of course). The opamp works as an error amplifier or proportional regulator (voltage of C1 being the "process variable" and the desired voltage being the "setpoint"), not as a Schmitt trigger or comparator. It doesn't need a capacity at the gate of the MOSFET.

• As the voltage of C1 approaches the target voltage the MOSFET enters its linear region
• The current to C1 is decreased, the voltage of C1 rises slower
• As the voltage of C1 rises further the MOSFET is driven further into its linear region
• The current to C1 is further decreased, the voltage of C1 rises even slower

Anyway, after counting parts I decided it will be easier to switch my secondary transformer windings to keep the total power losses at bay, and to redesign my linear regulator to take 60V (switching from a L200 to a LM317 high voltage variant at the core).

Thank you so much for your help and your patients with my "brain farts".

[not1xor1]

Anyway, the idea of my last design was to avoid switch-off spikes completely (by wasting power in the series transistor of course). The opamp works as an error amplifier or proportional regulator (voltage of C1 being the "process variable" and the desired voltage being the "setpoint"), not as a Schmitt trigger or comparator. It doesn't need a capacity at the gate of the MOSFET.

• As the voltage of C1 approaches the target voltage the MOSFET enters its linear region
• The current to C1 is decreased, the voltage of C1 rises slower
• As the voltage of C1 rises further the MOSFET is driven further into its linear region
• The current to C1 is further decreased, the voltage of C1 rises even slower

I'll have a look at it... I'll see how it behaves in LTspice

Anyway, after counting parts I decided it will be easier to switch my secondary transformer windings to keep the total power losses at bay, and to redesign my linear regulator to take 60V (switching from a L200 to a LM317 high voltage variant at the core).

Thank you so much for your help and your patients with my "brain farts".

no problem
I think that using a multi-tap transformer, in most cases, may be even more efficient than using a pre-regulator.
Actually I'm quite interested in that subject and in my spare time I'm simulating many different kinds of preregulators.

In future I will probably start another thread to discuss about that.

[robert67]

Anyway, the idea of my last design was to avoid switch-off spikes completely (by wasting power in the series transistor of course). The opamp works as an error amplifier or proportional regulator (voltage of C1 being the "process variable" and the desired voltage being the "setpoint"), not as a Schmitt trigger or comparator. It doesn't need a capacity at the gate of the MOSFET.

• As the voltage of C1 approaches the target voltage the MOSFET enters its linear region
• The current to C1 is decreased, the voltage of C1 rises slower
• As the voltage of C1 rises further the MOSFET is driven further into its linear region
• The current to C1 is further decreased, the voltage of C1 rises even slower

I'll have a look at it... I'll see how it behaves in LTspice

That would be most appreciated 

Anyway, after counting parts I decided it will be easier to switch my secondary transformer windings to keep the total power losses at bay, and to redesign my linear regulator to take 60V (switching from a L200 to a LM317 high voltage variant at the core).

Thank you so much for your help and your patients with my "brain farts".

no problem
I think that using a multi-tap transformer, in most cases, may be even more efficient than using a pre-regulator.
Actually I'm quite interested in that subject and in my spare time I'm simulating many different kinds of preregulators.

In future I will probably start another thread to discuss about that.

Depending on the results of your simulation, I might even go back to the pre-regulator approach.

Turns out switching between the secondary windings is not that easy. The transformer has (now, after Europe went from 220V to 230V) some very "odd" output voltages (considering mains AC is +/-10%):

• 14.3Veff to 17.4Veff, 20.2Vp to 24.6Vp
• 20.5Veff to 25.0Veff, 29.0Vp to 35.4Vp

To keep worst case losses in the linear regulator at a reasonable level (below 150W - I have two TO-3 BJT on big heat sinks on the backside) I'll have to set the switch point depending on the current AC voltage. After trying to come up with a design using transistors I went very fast for an opamp in schmitt trigger configuration Unfortunately that opamp needs some input protection on the feedback from the linear regulator (max. 30V) when I just use the 29.0Vp (below 30V) to 35Vp winding to supply the switcher. That's two more parts and it messes with the nice high input impedance Or I run the opamp from the combined windings. But then I'll need 60V capable opamp and relay as well as a second rectifier/filter to measure the AC level on my "lower" winding

Anyway, maybe I attach my current design later to this message ...

[not1xor1]

Depending on the results of your simulation, I might even go back to the pre-regulator approach.

the requirement to keep the component list short is a serious limitation 
otherwise you might use both the center tap and a preregulator

regarding your new circuit proposal I'll have a look at it tomorrow

Turns out switching between the secondary windings is not that easy. The transformer has (now, after Europe went from 220V to 230V) some very "odd" output voltages (considering mains AC is +/-10%):

• 14.3Veff to 17.4Veff, 20.2Vp to 24.6Vp
• 20.5Veff to 25.0Veff, 29.0Vp to 35.4Vp

To keep worst case losses in the linear regulator at a reasonable level (below 150W - I have two TO-3 BJT on big heat sinks on the backside) I'll have to set the switch point depending on the current AC voltage. After trying to come up with a design using transistors I went very fast for an opamp in schmitt trigger configuration Unfortunately that opamp needs some input protection on the feedback from the linear regulator (max. 30V) when I just use the 29.0Vp (below 30V) to 35Vp winding to supply the switcher. That's two more parts and it messes with the nice high input impedance Or I run the opamp from the combined windings. But then I'll need 60V capable opamp and relay as well as a second rectifier/filter to measure the AC level on my "lower" winding

Anyway, maybe I attach my current design later to this message ...

I have an old transformer as well... I bought it when I was a kid... it is probably 45 years old now

I used it for a while with a 2N3055s audio amplifier I built when I was 15 (until it died in a cloud of stinking smoke  )
It is centre tapped with a big "150VA" printed on the case, but one winding i much thicker than the other one so I guess it is something like 24V 6.25A / 48V 3.12A. For that reason I was already thinking about possible centre tap switching solutions. (I wonder if I should rather spell it "centre"

Here is what I simulated so far, just a quick sketch which you might hopefully find useful.

The transformer used in the simulation is centre tapped (15+15V - I modified the previous 30V spice model).
An LT1013 is connected as Schmitt trigger and compares Vout/6 with a 2.5V reference.
As soon as the voltage gets above about 15V, the LT1013 output goes to about 18V (its supply voltage is just 20V in this simulation) switching M1 on via Q1 so the input of the linear regulator is supplied by an higher voltage.

This circuit might work even with much higher voltage by just changing Q1 and M1 and modifying the rate of the voltage divider at the input of the Schmitt trigger.

R9 and C5 reduce a bit the output spike when the input voltage is switched (I might try to slow down M1 commutation as well).
The adjust pin of the linear regulator is driven by a delayed ramp to test the input voltage switching.

BTW the green trace in the top pane is the power loss of the linear regulator.

[not1xor1]

Anyway, the idea of my last design was to avoid switch-off spikes completely (by wasting power in the series transistor of course). The opamp works as an error amplifier or proportional regulator (voltage of C1 being the "process variable" and the desired voltage being the "setpoint"), not as a Schmitt trigger or comparator. It doesn't need a capacity at the gate of the MOSFET.

• As the voltage of C1 approaches the target voltage the MOSFET enters its linear region
• The current to C1 is decreased, the voltage of C1 rises slower
• As the voltage of C1 rises further the MOSFET is driven further into its linear region
• The current to C1 is further decreased, the voltage of C1 rises even slower

Your circuit just provides a gate voltage proportional to the difference between the linear regulator input and its output (and so to ripple) so I do not think it can help in reducing power dissipation.
I tried by changing various component values and output voltages and could not see any difference in power usage between your circuit and a plain linear regulator.

Anyway I've here an old NSC application note, AN1, november1967, about the LM100 IC (the first linear regulator IC?).
Among the various examples there is a switching PSU.
So what about a switching preregulator? Isn't 1967 enough vintage for you? 
Of course I do not suggest to use an LM100 

[robert67]

the requirement to keep the component list short is a serious limitation 

Yes, it is

otherwise you might use both the center tap and a preregulator

I'm actually thinking about that too. Making the linear regulator itself 60V capable is not that easy.

BTW the green trace in the top pane is the power loss of the linear regulator.

Wow, just 70W power loss is not too shabby!

And here's the circuit I've come up to switch between the transformer windings (the original innards of the power supply used a 16A relay for that). The hysteresis of the Schmitt trigger needs to take the ripple voltage at the filter capacitor into account, otherwise you might end up switching that relay at 100Hz The relays I've been looking at just draw 17mA at 24V. So you don't need a big capacitor to keep the ripple down, and you don't need a high power opamp. If the output voltage of the linear regulator is below the minimum voltage of the capacitor you won't need that zener diode and resistor to protect the opamp input.

[robert67]

Anyway, the idea of my last design was to avoid switch-off spikes completely (by wasting power in the series transistor of course). The opamp works as an error amplifier or proportional regulator (voltage of C1 being the "process variable" and the desired voltage being the "setpoint"), not as a Schmitt trigger or comparator. It doesn't need a capacity at the gate of the MOSFET.

• As the voltage of C1 approaches the target voltage the MOSFET enters its linear region
• The current to C1 is decreased, the voltage of C1 rises slower
• As the voltage of C1 rises further the MOSFET is driven further into its linear region
• The current to C1 is further decreased, the voltage of C1 rises even slower

Your circuit just provides a gate voltage proportional to the difference between the linear regulator input and its output (and so to ripple) so I do not think it can help in reducing power dissipation.
I tried by changing various component values and output voltages and could not see any difference in power usage between your circuit and a plain linear regulator.

That's not what I've intended it to do I have to revisit that circuit ...

Anyway I've here an old NSC application note, AN1, november1967, about the LM100 IC (the first linear regulator IC?).
Among the various examples there is a switching PSU.
So what about a switching preregulator? Isn't 1967 enough vintage for you? 
Of course I do not suggest to use an LM100 

That's definitely vintage enough

[hli]

One question about the thyristor-based pre-regulator: when I understand this correctly then the series transistor charges the bulk capacitor until it reaches a certain voltage threshold above the output voltage. Then the thyristors gets triggered and turns off the series transistor until the next half-wave from the rectifier.
If so, then it would mean that the real regulator cannot suddenly increase the output voltage (which might be needed in a constant-current-scenario when the load-resistance is reduced), since the regulator just does not have enough input voltage for that, and the pre-regulator cannot charge the bulk capacitor until the next input half-wave (all depending on the exact timing).
Do I understand this right? Because this would kind of disqualify it for any serious lab power supply

[blackdog]

Hi hli,

How fast do you think a "normal" power Supply is?
Most power supply's have a 220uF to 1000uF capacitor over the output connectors.
Wat wil the dynamic output impedance be?
How much energy is in this output capacitor, how fast is the current loop, how much time before the next half sine is there to charge the buffer capacitor.
Normaly about 3V over the regulator transistor is enough for my linear regulator design for almost al dynamic situations.

Try the circuit, experiment!, and I do not mean with "Spice", but with Solder! 
It is not a complicated circuit.

Kind regards,
Blackdog

[hli]

How fast do you think a "normal" power Supply is?

I have never calculated that, but I would expect a transient response below 1ms (my LM317, although not qualifying as "proper" lab supply does follow 1A transient in about 10µs). But whatever it is - the time from the pre-regulator will _add_ to it. And on average, depending on the load situation, it might be like 5ms.

Try the circuit, experiment!, and I do not mean with "Spice", but with Solder! 

Will do! (when I find out which of all the many version of it I should try But thats why I was asking for experiences.

Thanks,
hli

[not1xor1]

One question about the thyristor-based pre-regulator: when I understand this correctly then the series transistor charges the bulk capacitor until it reaches a certain voltage threshold above the output voltage. Then the thyristors gets triggered and turns off the series transistor until the next half-wave from the rectifier.
If so, then it would mean that the real regulator cannot suddenly increase the output voltage (which might be needed in a constant-current-scenario when the load-resistance is reduced), since the regulator just does not have enough input voltage for that, and the pre-regulator cannot charge the bulk capacitor until the next input half-wave (all depending on the exact timing).
Do I understand this right? Because this would kind of disqualify it for any serious lab power supply

Do not forget that the same thing happens without a pre-regulator: the levelling capacitor is charged every 10ms (full wave rectifier).

So to make that kind of pre-regulator work properly you have to switch off the mosfet when the capacitor has enough charge to to provide an high enough voltage to the linear regulator for 10ms.

You have to consider the worst case, i.e. the maximum load. In that case the linear regulator, from the pre-regulator POV works like a constant current sink and we know that the voltage of a charged capacitor "C" discharged at a current "I" after time "t" decreases by I*t/C volts.

So suppose you have a 10,000uF capacitor and 5A as maximum load and a full wave rectifier: the capacitor voltage would drop by 5*10e-3/10e-3 = 5V.
Now if the linear regulator requires at least 2V (drop-out) to work properly at maximum load, to ensure the circuit works properly in every load condition, you have to switch off the mosfet as soon as the preregulator capacitor voltage gets 2+5=7V above the output voltage.

P.S. it would be misleading to call this kind of pre-regulator an SCR pre-regulator. Real SCR preregulator switches are off by default and are switched on just to provide enough charge to the capacitor before the rectified wave decrease switching off the SCRs (or a mosfet with a proper control circuit). Here the switch is on by default and is switched off as soon as the voltage gets above a given threshold.

[not1xor1]

Hi hli,

How fast do you think a "normal" power Supply is?
Most power supply's have a 220uF to 1000uF capacitor over the output connectors.
Wat wil the dynamic output impedance be?
How much energy is in this output capacitor, how fast is the current loop, how much time before the next half sine is there to charge the buffer capacitor.
Normaly about 3V over the regulator transistor is enough for my linear regulator design for almost al dynamic situations.

Try the circuit, experiment!, and I do not mean with "Spice", but with Solder! 
It is not a complicated circuit.

Kind regards,
Blackdog

But you can save a lot of  time and tin if you first understand how it works and why. 

And simulations let you see immediately the drawbacks of various kinds of circuits, when they do not go haywire and break the conservation of energy law.

Your pre-regulator is nothing new, it has been proposed in many different variations many times in various magazines during the last decades, but AFAIK no manufacturer uses that kind of circuit for their bench PSU, while there are various devices based on SCRs (or mosfet working like SCR switches) or switching pre-regulators.

Besides the very low frequency (fraction of Hz to few Hz) noise at low load, I already wrote about in past,, that kind of circuit has a really poor efficiency at low output voltages and high load.
The reason is that you switch off the current when the transformer secondary winding has already stored a lot of energy and that energy is then just wasted (apart a little stored in the input capacitor).

Even with a tighter threshold control (like that in reply 46) you can save just few tenths of watts when at low output voltage and with 3-5A load.

[hli]

Do not forget that the same thing happens without a pre-regulator: the levelling capacitor is charged every 10ms (full wave rectifier).

But for a 'normal' power supply we assume that the output voltage is always below (max. input voltage - ripple voltage - dropout voltage). So it should _never_ run into the situation where the regulator runs out of input voltage. Yes, the capacitor is charged only every 10ms, but we accommodate for that with high enough input voltage.

I was specifically referring to a low-output-voltage scenario, where the output voltage needs to rise suddenly (e.g. because you programmed the PSU that way, or because the load changes in CC setting. This should never be a problem without pre-regulator, it should be fine with a switching one (since they run at high frequency), but this one here has just no chance to react fast enough (depending on the timing of the required voltage change).

hli

[blackdog]

Hi not1xor1,

And how about experience...
Spice tells you its OK, and then you wil build it, "what te Fu.." a lot off time you put in Spice is going out off the window 

I can only say, do not use Spice as a shortcut to learn electronics, its a tool.
And yes, i use all kinds of electronic tools to make my live easier.

You have to learn bij real bulding that a resistor has also a inductens and also behave as a capacitor all depending on witch frequency you are using.
And then we also have all de paracitics of the way you build the circuit, yes! electronic is coplicated.
Try to see it as this, you can study all you want on say Youtube, about skateboarding...
The first time you wil step on the skateboard you realise that you didn't build/learn any muscle memory 

Take my power supply design, it is FAST!!!! and because of that, i have to follow the the rules of HF design for the wiring...
A lot of wiring is twisted to keep the induction and EMC down.
The same with probing the electronics, are you doing it the wright way, is the nasti spike you see realy in the circuit, or is it the wire to the power section that is radiating its field to the probing point.
I can go on the whole morning like this 

Iám absolutely NOT against Spice, but wath i see on this forum and the Dutch forum circuitsonline don't make's me happy.
Most user can't even draw a schematic and trust there Spice, because Computers don't lie...
The Spice user forgets they are lying to them self *grin*

Sorry for the lousy Ingles...

Kind regards,
Bram

 

[not1xor1]

Do not forget that the same thing happens without a pre-regulator: the levelling capacitor is charged every 10ms (full wave rectifier).

But for a 'normal' power supply we assume that the output voltage is always below (max. input voltage - ripple voltage - dropout voltage). So it should _never_ run into the situation where the regulator runs out of input voltage. Yes, the capacitor is charged only every 10ms, but we accommodate for that with high enough input voltage.

I was specifically referring to a low-output-voltage scenario, where the output voltage needs to rise suddenly (e.g. because you programmed the PSU that way, or because the load changes in CC setting. This should never be a problem without pre-regulator, it should be fine with a switching one (since they run at high frequency), but this one here has just no chance to react fast enough (depending on the timing of the required voltage change).

hli

if you read more carefully my reply you will understand that you have to adjust the switch threshold, independently from the current real load, to the worst case load. That is the pre-regulator switch has to stop charging the capacitor as soon as the capacitor voltage gets to Vout+maxVdropout+maxVripple, where as maxVripple I mean the voltage decrease of the capacitor when in the worst case load.

For instance if Vout is 1V, the worst case drop-out of the linear regulator is 2V, and you have a 20,000uF capacitor with a worst case load of 5A, you have to stop charging the capacitor as soon as the voltage gets to 1+2+2.5=5.5V.

That is a 20.000uF capacitor charged at 5.5V cannot get below 3V (1V + 2V drop-out) in 10ms even if you step load it from 1mA up to 5A (of course a real electrolytic capacitor would lose a really small bit due to self discharge).

[not1xor1]

Hi not1xor1,

And how about experience...
Spice tells you its OK, and then you wil build it, "what te Fu.." a lot off time you put in Spice is going out off the window 

well I built several variations of that kind of pre-regulators a few years ago, so I'm not referring just to spice simulations.
... and in any case I do think no real circuit can break the law of physics... 

[prasimix]

Power supply is at the end as slow, as the slowest section of it. At the beginning I'm used mosfet pre-regulator for charging bulk capacitor (thanks Blackdog once again) and it's reasonable fast but not faster then SMPS pre-regulator for simple reason that later works with few order of magnitude higher frequency and is powered by DC not low-freq AC. When mosfet (or lets call it mains frequency switcher) pre-regulator has to follow rapid changes on the output (i.e. something that is in range of mains frequency or faster) then I noticed a sort of "hiccup" phenomenon. Here is how it looks like then output is programmed with 50 Hz, 50% duty (10 ms ON, 10 ms OFF), cyan trace is pre-regulator output, yellow is post-regulator output, blue is programming signal from DAC:



Nothing suspicious on the screenshot above, but now over longer period of time:



... and zoomed in:



In practice such extreme programming is rarely existing in DIY environment and I didn't gave up from blackdog's solution due to that phenomenon but in that time for another more practical reason: it's practically very bad match with toroidal transformer (that someone already addressed in this thread). AFAIK there is at least one attempt to do something with such pre-regulator and toroidal transformer that is described here.

Regarding fast output programming, any output capacitance with affect speed. That could be especially problematic when lower programmed value followed a higher one (e.g. 5 V after 30 V). But for that situation many commercial power supplies are using so-called down-programmer circuit that is used for power sinking.

[hli]

if you read more carefully my reply you will understand that you have to adjust the switch threshold, independently from the current real load, to the worst case load. That is the pre-regulator switch has to stop charging the capacitor as soon as the capacitor voltage gets to Vout+maxVdropout+maxVripple, where as maxVripple I mean the voltage decrease of the capacitor when in the worst case load.

Well, in my scenario the worst-case load is a switch to full output voltage. Under that assumption, the pre-regulator neds to provide the full input voltage.
I'm not worrying about a constant load - I fully understand how the pre-regulator works for that, and how it needs to be configured. What I was looking at is how the pre-regulaotr affects the response to load transients that result in an increase of the output voltage

For instance if Vout is 1V, the worst case drop-out of the linear regulator is 2V, and you have a 20,000uF capacitor with a worst case load of 5A, you have to stop charging the capacitor as soon as the voltage gets to 1+2+2.5=5.5V.

What happens when, after the MOSFET has turned off, in the same 10ms cycle, the output voltage is request to increase to 10V (independent of the current that delivered)? AFAICS the output voltage would rise to (5.5V-Vdrop), so maybe 4V (depending on the regulator), and stay there until the next  charge cycle starts and the voltage at the capacitor rises above the 5.5V.

[not1xor1]

Power supply is at the end as slow, as the slowest section of it. At the beginning I'm used mosfet pre-regulator for charging bulk capacitor (thanks Blackdog once again) and it's reasonable fast but not faster then SMPS pre-regulator for simple reason that later works with few order of magnitude higher frequency and is powered by DC not low-freq AC. When mosfet (or lets call it mains frequency switcher) pre-regulator has to follow rapid changes on the output (i.e. something that is in range of mains frequency or faster) then I noticed a sort of "hiccup" phenomenon.

You are referring to sudden changes in programmed output voltage, don't you?
Well, probably an audio amplifier would be better  , anyway that kind of pre-regulator could still work in such cases if you provide a further mosfet connecting a capacitor at full voltage (via a low Q inductor + parallel resistor) just for those "emergency" situations... I mean class-G audio amplifiers style.

Regarding switching pre-regulators I think you can get quite low output noise (I've not built anything real yet) if you keep the frequency low, like in those old Jim Williams circuits.
If you have a look at those linear regulators IC datasheets, you may notice that even the humble LM317 has quite a good ripple rejection up to several kHz.

[not1xor1]

What happens when, after the MOSFET has turned off, in the same 10ms cycle, the output voltage is request to increase to 10V (independent of the current that delivered)? AFAICS the output voltage would rise to (5.5V-Vdrop), so maybe 4V (depending on the regulator), and stay there until the next  charge cycle starts and the voltage at the capacitor rises above the 5.5V.

such kind of pre-regulator is just not suitable for programmable PSUs (unless you add a capacitor backup like I hinted in the other reply).

On the other hand, if you just want to build an ordinary PSU you usually do not mind if it takes a few hundredths of seconds to get the output from 5V to 20V.

Apart from testing the response of a circuit to supply voltage transients (and you could find other methods to achieve that) I do not see any usefulness in fast variations of output voltage.

[prasimix]

What happens when, after the MOSFET has turned off, in the same 10ms cycle, the output voltage is request to increase to 10V (independent of the current that delivered)? AFAICS the output voltage would rise to (5.5V-Vdrop), so maybe 4V (depending on the regulator), and stay there until the next  charge cycle starts and the voltage at the capacitor rises above the 5.5V.

such kind of pre-regulator is just not suitable for programmable PSUs (unless you add a capacitor backup like I hinted in the other reply).

On the other hand, if you just want to build an ordinary PSU you usually do not mind if it takes a few hundredths of seconds to get the output from 5V to 20V.

Apart from testing the response of a circuit to supply voltage transients (and you could find other methods to achieve that) I do not see any usefulness in fast variations of output voltage.

You're right, good to write that down, just for the record.

[hli]

You are referring to sudden changes in programmed output voltage, don't you?

Thats one of the cases where this happens. I was more thinking of a lab supply in constant current mode. But there, the worst case is that the supplied current is too low for some milliseconds (not more than 10ms, obviously). Which is something one could live with.

[hli]

such kind of pre-regulator is just not suitable for programmable PSUs (unless you add a capacitor backup like I hinted in the other reply).

OK, so I understood the behavior correctly. Thanks for the explanations. For a programmable PSU, one might also sync these changes to the beginning of a half-wave, when the FET is turned on anyway.

hli

[blackdog]

Hi hli,

If you want a realy fast programmable power supply, you have to buy/design a 4 quadrant type.
Most of the programmable power supply you can buy are "slow" sometime they have a "down programmer" to push down the output if you change the output to a lower voltage.
This to make them more quickly respons in light load conditions.

If you want a quick reacting current source, buy or design one.
You wil never have a "one size fits all" power supply.

The design i showd is for "normal bench" use, but extreme low noise, fast  and stable, if i want a fast current source with a high compiance, i will use another device.

Kind regards,
Blackdog

[robert67]

First of all, sorry that I was not attending my thread for a while. I was under the weather and now I have to catch up with work. So my contributions will be sparsely in the near future. But it seems you're all quite OK without me 

A thought or two about regulation speed

Even some high priced linear lab power supplies use a relay to switch between different taps of the transformer secondary. Those relays, that is the good ones, have a switching times of 8ms (6ms release). Add to that a few ms of bounce and any considerations about your pre-regulator being able to charge up your filter cap in time when a rising voltage transient comes along is mute.

The only way to respond to rising voltage transients when using a linear regulator in considerably less than 10ms is not to pre-regulate and to keep your filter cap at maximum voltage the whole time. But then of course you're burning a whole lot of power at lower output voltages.

Update on my brain farts regarding the rebuild of my vintage power supply

• Transformer tap switch to avoid insanely high power losses since the old thing has two windings anyway
• Rectifier and filter cap
• Linear pre-regulator that will burn up to 150W just because I can I mean I have two massive TO-3 heat sinks at the back of my case and that pre-regulator will of course eliminate the ripple at the filter cap - so a "no noise" design (take that Blackdog )
• Linear regulator that just burns its own dropout voltage, so about 15W at 5A to dissipate inside the case (which has a lot of ventilation).

[not1xor1]

You are referring to sudden changes in programmed output voltage, don't you?

Thats one of the cases where this happens. I was more thinking of a lab supply in constant current mode. But there, the worst case is that the supplied current is too low for some milliseconds (not more than 10ms, obviously). Which is something one could live with.

What do you need constant current for? I can think about measuring very low resistances, or charging a battery.
In both cases there is no problem if the voltage doesn't change so fast.

Of course there are probably many other cases where a tight current regulation with rapid voltage changes might be essential which I just do not know and cannot even imagine about.

I will be glad to learn about some more usage examples, I'm just an hobbyist, not an engineer.