part 1 5106 bytes content-type:text/plain; format=flowed; charset=Windows-1252; (decoded 7bit)
As the "GSR" has featured peripherally in discussions in the "is this going
to die" thread I decided to start a new thread related to it and others like
I'll make the following observation at the start. If people want to ask
about it or comment on it please do so offlist to avoid annoying people who
get annoyed by such things. While I am notionally the "inventor" of this
circuit the design was actually proposed to me by God. Add a smiley to that
if you want. The basic circuit fell fully formed into my mind essentially
instantaneously immediately after I had provided a detailed written
requirement. The design was to meet a very specific need, which it met and
continues to meet very well indeed. This doesn't mean that it will meet all
other needs as well, divinely inspired or not :-). GSR stands for 'God's
Switching Regulator". End of discourse. Comments offlist please.
There are about 10,000 of these circuits in use at present. If it is used in
a new project, which it probably will be, there will be 100,000+ of them in
the next year or two.
The GSR is a buck regulator (maybe it should have been GBR :-) ) - so output
voltage must be lower than input voltage (unless additional coils are added,
which I have done).
It is usually not the most efficient circuit available.
It's attributes include
Simplicity / low component count / low cost
Tolerance to wide range of inductors.
Good regulation for load and input variation, considering its
Operation over a VERY wide range of input voltages (20:1 in original)
Extremely strange mode of operation (which probably helps reduce EMI
If you want a high performance circuit and cost is not important then you
probably want to use one of the many many IC switching regulators available.
If you want an ultra low cost design that performs surprisingly well then it
may meet your need. As an amateur 9v battery to 5 volt converter it could do
The "Black Regulator", refined by Roman Black from a circuit originally
provided by Richard Prosser (if my memory serves) has less components and an
even lower cost. It has worse regulation, requires larger inductors, tends
not to like wide ranges of input voltages and it is much easier to
understand its operation.
The attached GIF is subset from a larger circuit. It uses a high side P
Channel FET as the main switch BUT most low power designs would use a small
PNP transistor as the main high side buck regulator switch. In such cases
zener ZBUK2 is not required. Resistors RBUK3 & RBUK4 would need to be chosen
to suit the circumstances/
The original circuit had an output voltage of about 9 volts AFAIR and
"switched" for input voltages from about 10 volts to 200 volts. The P
Channel FET was rated at 200 volts and this set the upper (un)safe limit of
operation. When Vin is below target output voltage the regulator acts as a
pass through circuit with about 1v or less drop. This is important in the
In operation the switching waveform of the FET (or high side PNP or
whatever) is "chaotic". No nice clean waveforms here. Switching can be made
far more regular and efficiency improved by adding a series
resistor-capacitor from the FET-Coil junction to the base of transistor
QBUK2. You can expect that EMI levels may rise when you do this.
It *probably* works like this ;-)
All off, Vout = 0, ZBUK1 doesn't conduct.
QBUK1 off so RBUK2 turns QBUK2 on which turns BUKFET on via RBUK3.
Voltage is applied to inductor via FET and current increases.
CBUK2 charges until ZBUK1 starts to conduct.
QBUK1 turns on (driven by ZBUK1) turning off QBUK2 turning off FET.
The following is standard buck regulator stuff:
Inductor 'rings" as current must continue to flow.
Right end of inductor now goes positive relative to left end.
DBUK2 conducts and CBUK2 voltage *RISES* as energy is transferred from
QBUK1 is held off as long as inductor has energy in it that drives output
and can transfer energy to load fast enough to maintain ZBUK1 in
When inductor can't keep up the fight, CBUK2 falls *slightly*, ZBUK1 stops
conducting and cycle repeats.
If there is an RC added as abovem, when inductor left end starts to fall,
QBUK2 is driven off more quickly adding regenerative turn off. The reverse
happens at turn on. This adds "formal" hysteresis and makes the circuit more
understandable. There is in fact formal hysteresis present due to the eenrgy
stored in the inductor, but some can't see this.
If the inductor is shorted out the design becomes a linear regulator. I have
NEVER seen this circuit drop into a linear regulation mode but i have heard
that reported by someone else who played with it.
Apart from the inductor there are a few tens of cents of parts used (if a
small TO92 bipolar transistor is used instead of the FET. Inductor cost
depends on application.
part 2 4235 bytes content-type:image/gif; name=gsr.gif (decode) part 3 186 bytes content-type:text/plain; x-avg=cert; charset=us-ascii (decoded quoted-printable)
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.296 / Virus Database: 265.6.2 - Release Date: 20/12/2004
part 4 79 bytes content-type:text/plain; charset="us-ascii" (decoded 7bit)
Got this response offlist - probably thought my offlist request re religious
stuff related to technical matters as well.
I told them I would respond on list but not identify them. They can self
identify if they wish,.
I am very interested of one of those regulators, for a power supply I want
to build, where the cost is most important, and after that the efficiency.
Around which value is the efficiency though ? I guess that's more than a
linear power supply...
I am beginner with electronics, so I have a limited capacity to understand
your circuit. I want to build a 13.8V power supply, that would be capable of
giving at maximum 3A of current. Could you indicate me the key parts that
need to be changed for that job ? And in the schematic, there is the
complete circuit ? Or just a part of it ?
Thank you for answering !
That is more or less the whole circuit - although there are always bits and
pieces that need adding on in a real circuit.
A filter capacitor on Vin is a very good idea. Depending on what the load
is it MAY be a good idea to add a small output resistor in series with the
output with a larger filter capacitor to ground after it.
You don't say whether 13.8v is input or output.
What are your desired input and output voltages?
What is the circuit to be used for?
The more you can tell about your application the easier it is to give good
Why is the cost important? - are you wanting to build many or just a few or
one? If a few or one then you MAY find that a commercial circuit or an IC is
still an OK solution.
The efficiency can probably reach 80%+ depending on many factors. I suspect
it would never get to 90%+. My friend who is using it to give 12v from 20v
to 35v in is getting 70%+.
For a 3A design using a FET may be a good idea. A P channel FET is needed -
which one depends on the required Vin .
NOTE - while this circuit COULD be used from rectified mains to 13.8 volts
DC I would not recommend doing so! The output is not isolated and the
circuit is "unusual" enough that you would want to have lots of confidence
that the design worked under all conditions. Mains can be very tricky.
Tell us more and see how it goes. If you don't want the list to know who you
are you can ask me offlist and I'll answer on list. For technical things
like this it's a very good idea to ask onlist so all can share with you.
What I said at the start about answering me offlist only applied to anything
applying to how God related to the design - some people get extremely
annoyed when such things are discussed onlist.
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.296 / Virus Database: 265.6.2 - Release Date: 20/12/2004
Ok, sorry for misunderstanding the part with offline reply, that's because
english is not my mother tongue :)
I don't need to hide, and I already stated the problem with the power supply
I want to build in another post, and I got some answers that were useful for
me, but when I saw your regulator, I thought it could be used in my circuit
also, reducing the costs of the power supply.
The input voltage would be between 14 and 22.4V, from a 16V transformer,
with a rectifier bridge and an 4700uF filter capacitor. Could the voltage be
applied directly to the schematic you provided ? Is it necessary another
filter capacitor ? I still need to calculate the ripple at 3A and see if I
can find a capacitor with so much ripple. 80% efficiency is good, and I can
bear with it, I don't need to obtain 99% out of it.
The output is to be 13.8V, and I don't know exactly which resistors I would
have to change for that. For 3A, I know I would have to use a FET which
resists at this load, I don't know exactly the amount of power to be
dissipated, and which would be the part to dissipate more heat - the FET ?
Tha application is a security system, and the load of 3A maximum would be
when the bell goes on - very rarely, and for a short time. The load is
composed of the sensors, the bell, other devices at around 12V and a
regulator at 5V for the other circuitry (not power consuming).
I hope there aren't too many questions to annoy you...
Thank you very much for your help.
Russell McMahon wrote:
> It *probably* works like this ;-)
C'mon Russell, this is silly and you should know better. Treating it like
magic doesn't do anyone a service. This circuit does what it does due to
real physics, no matter how you think it was inspired. In fact, it's not
that hard to understand at all.
> If the inductor is shorted out the design becomes a linear regulator.
Although it may not be that stable. This circuit depends a lot on parasitic
capacitances and other lags to oscillate with the inductor in place. For
the same reason it may not be stable with the inductor removed.
> I have NEVER seen this circuit drop into a linear regulation mode but
> i have heard that reported by someone else who played with it.
This is one of the dangers of this circuit. While very likely, oscillation
is not completely guaranteed depending on parameters that are hard to
The hardest part of a buck regulator is usually the high side switch driver.
Turning the switch on and off hard is key to good efficiency. Deciding when
to turn the switch on and off is usually the easy part. This circuit aims
for low cost and therefore uses a very simplistic high side switch driver.
The turn off time will be particularly slow, because is is just a 100Kohm
resistor working against the effective gate capacitance of BUKFET. This
will be slow, and is probably the single largest source of power loss in the
circuit. Unfortunately there is no free lunch. A faster high side switch
driver will cost a few 10s of cents in additional transistors and resistors.
The slow turn off of BUKFET also increases the danger of linear operation,
although I agree that it is still unlikely do to other lags in the circuit.
Other than the high side switch driver, two other issues occupy most of the
brain cycles in a professionally designed switching regulator. First,
inductor saturation must be dealt with. Either it must be tolerated, or
more commonly, arranged to never happen. This is where your circuit may
have a more serious problem.
A quick back of the envelope calculation shows that this is a problem with
this circuit during startup. C2 starts out at 0V with 100v applied accross
the 100uH inductor. Current therefore increases at 1A/uS. At this rate, C2
will take over 30uS to charge to 10.5V. If all were perfect components, the
inductor current would build to nearly 30A during this time. In practise
the inductor will saturate and become a resistor, stressing the inductor,
the FET, and C2. This will be much less of a problem if the input voltage
rises more slowly. It may not die right away even if switched on suddenly,
but the stresses to the FET and C2 will eventually cause premature failure.
Note that additional capacitance on Vout in the circuit being powered will
make this worse.
The other area that needs to be considered carefully is the reverse recovery
time of D2. In low voltage circuits this is usually dealt with by making D2
a Schottky diode, which has essentially instant reverse recovery time.
However, at higher voltages this is impractical. I don't know what a BYV26C
is, but if it's a silicon diode this needs to be thought about carefully.
If D2 is still conducting when the FET switches on, both the diode and the
FET will take a serious beating for a few 100nS at least. This will
eventually destroy one or the other. At the very least, D2 should be a fast
recovery diode. A better answer is to guarantee that the FET will never be
turned on when D2 is conducting. This is possible to do in a simple buck
regulator using just analog feedback as this one does, but needs to be
verified. You may be lucky that this has accidentally worked out, but
figuring out the details will be difficult due to the unpredictable lags
between Vout crossing the regulation threshold and the FET being turned
In short, this may be an effective low cost buck circuit. However, its
operation and limitations must be clearly understood before putting it into
production. The fact that 5000 are in the field operating isn't relevant.
At the least the special operating conditions of those units should be
noted. A small change in operating condition can lower the MTBF of this
> No virus found in this outgoing message.
> Checked by AVG Anti-Virus.
> Version: 7.0.296 / Virus Database: 265.6.2 - Release Date: 20/12/2004
Just like any clever virus would write. Do you really think anyone would
believe the *sender* that a message is virus free?
> C'mon Russell, this is silly and you should know better. Treating it like
> magic doesn't do anyone a service. This circuit does what it does due to
> real physics, no matter how you think it was inspired. In fact, it's not
> that hard to understand at all.
I was by no means attributing it to magic - I was attempting to forestall
nit picking over the fine detail or comments that it can't possibly work at
all. You may quite possibly recall that when I first introduced it some
years ago there were two people who proclaimed (very) long and (very) loud
that the circuit could not possibly work at all. No names mentioned of
course :-). Obviously it obeys the laws of physics - there was never meant
to be any suggestion to the contrary.
>> If the inductor is shorted out the design becomes a linear regulator.
> Although it may not be that stable. This circuit depends a lot on
> capacitances and other lags to oscillate with the inductor in place. For
> the same reason it may not be stable with the inductor removed.
I tried quite a few variants of this circuit with different transistors,
different inductors and different power and voltage levels. I never managed
to have it so anything other than operate correctly. While you can indeed
theortically posit a static case where the circuit operates in linear mode,
for me this hasn't ever happened. The Lorentz butterly wing flap may be
strong with this one :-)
Other people have reported it behaving in linear mode and being sensitive to
eg coil polarity!!!
The best I can say is that every attempt by me to make it fail failed.
>> I have NEVER seen this circuit drop into a linear regulation mode but
>> i have heard that reported by someone else who played with it.
> This is one of the dangers of this circuit. While very likely,
> is not completely guaranteed depending on parameters that are hard to
For the nervous, adding the series RC circuit that I mentioned adds non
inductive regenerative feedback. It also cleans up the switching waveform
and increases efficiency. It makes frequency of operation more stable which
would increase EMI at some frequencies.
> The hardest part of a buck regulator is usually the high side switch
> Turning the switch on and off hard is key to good efficiency.
I agree completely. As shown this is a starting circuit, and is entirely
adequate for basic operation where efficiency is not the driving factor. In
my final version I did in fact add a gate turn off circuit to improve
efficiency. Also a current limit on the output as it will enthusiastically
try to provide any load offered. In my application the extremely wide input
voltage range makes driver design more difficult as it must provide enough
drive at 10 volts or so in but not dissipate excessively at 200 volts in.
There are vanishingly few circuits around that handle this sort of input
> Deciding when
> to turn the switch on and off is usually the easy part. This circuit aims
> for low cost and therefore uses a very simplistic high side switch driver.
> The turn off time will be particularly slow, because is is just a 100Kohm
> resistor working against the effective gate capacitance of BUKFET. This
> will be slow, and is probably the single largest source of power loss in
Agree with all that.
> Unfortunately there is no free lunch.
Agree. But sometimes you can get a lunch that's almost as good for far less
cost. Or even a lunch that's better at far less cost.
> A faster high side switch
> driver will cost a few 10s of cents in additional transistors and
It also adds room and complexity and design effort. None of these are a
problem in my case, but I wanted to present the basic circuit so people
would get the idea. It does an excellent job as is, but can definitely be
improved by adding compleixty.
> ... First,
> inductor saturation must be dealt with. Either it must be tolerated, or
> more commonly, arranged to never happen. This is where your circuit may
> have a more serious problem.
The comments on startup in the circuit shown have merit but are not the full
story - largely my fault for not being even wordier ;-)
Part of my "mistake" here was to not draw a new circuit to remove some
confusing material. This was meant as an example circuit and all component
values would need to be designed for a particular application. The values
shown will not be approriate for most probable PICLister applications and in
fact some do not even represent my final values. I should have made that
- V100 in my case runs from 10 to 200 volts.
- MPSA42 is used only where its voltage rating is needed. in most circuits
with Vin of 12 to 30 volts something like a BC337 would do a good job.
- The FET would be replaced by eg a BC327 bipolar in low power applications
(as noted in original text).
- L1 would be chosen to suit the maximum Vin. In my case it is MUCH larger
than 100 uH due to the 200 volt maximum rating - it's actually somewhere in
the milliHenry range.
- Increasing C2 by adding more capacitance downsteam will indeed affect
operation. I noted in the last email that a small series resistor may be
required to partially isolate variations on C2 from the load capacitor.
(Roman's circuit also uses a series resistor in some cases). This resistor,
weher used, is an additional source of inefficiency, but is not a major
- Flyback diode shown is an excellent device but would be overkill in many
circuits (see Olin's comments on this diode)
> A quick back of the envelope calculation shows that this is a problem with
> this circuit during startup.
As shown yes - see above.
>It may not die right away even if switched on suddenly,
> but the stresses to the FET and C2 will eventually cause premature
Properly designed (see above) reliability is good. There have been some
failures in the 10,000 built but well within acceptable limits. I think this
circuit has been in use for about 3 years now.
> Note that additional capacitance on Vout in the circuit being powered will
> make this worse.
Yes - see above re series resistor.
This is a consequence of the transition triggering method which this corcuit
employs. Adding the series RC as mentioned previously would reduce this
> The other area that needs to be considered carefully is the reverse
> time of D2 ... I don't know what a BYV26C is, but if it's a silicon diode
> this needs to be thought about carefully.
It was. The BYV26C is an ultra fast recovery silicon diode designed for this
sort of application. In many cases it would be overkill. For very low power
low voltage applications a humble 1N4148 may suffice!!!
> If D2 is still conducting when the FET switches on, both the diode and the
> FET will take a serious beating for a few 100nS at least.
> ... You may be lucky that this has accidentally worked out, ...
It *could* have been luck :-)
> In short, this may be an effective low cost buck circuit. However, its
> operation and limitations must be clearly understood before putting it
> The fact that 5000 are in the field operating isn't relevant.
I consider it is relevant, as it shows that it can work well in a real world
But it shouldn't blind people to the fact that, like all circuits, it still
needs to be designed for a given application.
> At the least the special operating conditions of those units should be
> noted. A small change in operating condition can lower the MTBF of this
> circuit dramatically.
Agree. But note that many circuits have points where they suddeenly transit
from well behaved to disatrous. As you suggest, understanding the operation
as well as one may is always a good idea.
>> No virus found in this outgoing message. yada yada
> Just like any clever virus would write. Do you really think anyone would
> believe the *sender* that a message is virus free?
I hope not!
I don't know whether you can turn that reporting "feature" off.
The outgoing check is a good idea though (even though I hope you won't rely
on it :-) )
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.296 / Virus Database: 265.6.3 - Release Date: 21/12/2004