View Full Version : Sg3525 low power psu for dac

11-22-2017, 12:19 PM
Hi Alls !
I'm freshly registered here, but I read you since many years.

Now is my day to take the plunge and to show you my smps project.
I don't have engineer knowledge, only school memories.
Since the time I read the forum trying to understand the smps rules, I have now finished the schematics of a low power switchmode psu for preamp use.

The PSU is based on the SG3525 chip and don't use regulation since I will use some regulators at the outputs.
Push-pull design with TN23/3C90 core transformer with 3 secondary (2 for analog +/GND/- and another for digital +/GND) 2x5 PRIM / 2x9+1x5 turns SEC
I use a mosfet driver TC427 with 2x IRFS7440 MOSFET.
I'm planning to use at maximum SMD components for compact design and because I prefer soldering SMD than through hole (my eyes are still good for the moment :D)
For sure some parts cannot be put in SMD (snubbers essentially).

In attache the schematic of the entire PSU.
Many "sub-block" are inspired from well-know others PSU like BCAE1 or Elliot Sound or valveaudio. I'm not able to design entirely this circuit. So maybe I have done some beginners mitakes.

I will have some questions if someone can look the schematic.
Since the secondary winding ( 8- 9 for digital ) is isolated from the analog windings (4 - 5 - 6 -7), I have 2 dif GND, one digital and one analog. It is good for the DAC.
With this configuration does I need to use 2 common mode inductor (one for each output) or does I need to wind all of the secondary in the same inductor ? If I do thos does I need to connect absolutely the 2 gnd ?
I prefer to use 2 different inductor since for this power, I can probably buy them ready to use.
Since the PSU is unregulated, does I really need the inductors ? I need the cleanest possible psu, and in my head, running unregulated is better for noise.

Any comments or suggestion will be appreciated,

Thanks in advance to those who take time to read the post,


11-22-2017, 06:11 PM
Sorry, missing the parts numbers on the schematics.
I'm playing with Excellent IT, and I think the TN23 core I planned to use is not good. A TN26 will be better.
Print screen in attache.



11-22-2017, 11:08 PM
i think you have to reverse diode direction like this

11-23-2017, 06:43 AM
hi ,pichou nice project . please post photos of your smps while testing it.

11-23-2017, 08:14 AM
Thanks for your answer :)

Ouuuupsss 4 diodes in my nose :ambivalence:

I will start routing the board first and I will show you the result.

11-23-2017, 08:16 AM
Through my experience so far common mode inductors on the output tend to be more noisy than single bead inductors. I do not know what load you have on the output but as the smps is unregulated then noise is not so prominent.
You have enough capacitance on the output to filter out. I also suggest that you make a smaller capacitor such as 1uf film cap instead of the 100uf in the input of the regulators this will filter better any residual noise. You can also try ring inductors but separately wound and not in common mode fashion.

Your input and output snubbers have to be trimmed on the actual setup as ringing if any has to be suppressed on the actual ringing frequency. This tend to vary with different transformers and winding technique.

Use tight coupling and symmetry in your windings to eliminate leakage inductance as much as possible.

Regards Silvio

11-23-2017, 09:37 AM
Silvio :cool:
Thanks for your answer and advices.
Load on the output will be approx 0.5A on analog symmetric output and something like 1A on digital.

Ok for the capacitors and the inductors. To be sure about the inductors you advise me to use, you talk about one single ring inductor for each transformer output ? Or a common mode for analog and a separate single for digital ?
I have an old amp at home to make some tests (Esoteric 7056, the amp side is burnt put the psu is ok). If i remove the regulation to go 50% duty and the secondary common mode inductor I experiment some ripple. If I leave the inductor in place but still without the regulation, ripple is better.
Something like this can be used for digital side (rated at twice the current needed) ? https://www.mouser.fr/ProductDetail/Bourns/SRR1206-120ML/?qs=sGAEpiMZZMsg%252by3WlYCkU%252bb%2fp1ujdgwWzfYJ DYvjIr8%3d

The snubbers in the schematics are "just to put a value". I will try use this jig to found the correct value when the transformers will work : http://www.diyaudio.com/forums/power-supplies/243100-simple-math-transformer-snubber-using-quasimodo-test-jig.html
Good idea ?
Anyway, I will leave a place free in the board and populate this only when I will found the good value.

Another question, is, what about the manner I have plug the IN-/IN+/COMP/VREF to run unregulated, do you think it is correct ?

Cheers !

11-23-2017, 02:56 PM
About the output inductors I was talking about the dual rail output. I was experimenting with smps of higher current than that you are using (8-10A) when I encountered this problem with noise using the common mode inductor wound on the same core for the 2 rails. I got rid of it when I used separate inductor for each rail.

Usually they use a common inductor for several outputs ( Like the ones used in ATX power supplies) These all have a single rail though and not a double output. They do it like this usually for economy and also to get better regulation across all outputs these using the same ferrite ring with separate windings for each rail.
The inductor helps in both ways as it chokes a little the output and also help with better smoothing. It also causes some voltage drop so some headroom in the output for regulation is needed. It will also help out for surges in the output and limits the surge current.

Regarding snubbers well you can also check the blog post I posted called Calculation of snubber components and driver gate resistors.

About the IN+ and IN- I guess you are talking of the sg3525 chip. All I can see is that R15 is too low and this should be around 470K the duty cycle will suffer if it is too low. I also recommend that you make a small preset of around 500R instead of the discharge resistor R9 and then you can set the dead time to around 1-1.5uS. You can put a fixed resistor instead when the correct value is found.

I hope that helps Silvio

11-23-2017, 03:33 PM
That helps :cool:

Ok for the inductors !
That I can do is to put some dual footprint on the pcb to run common mode or separate inductors (with some headroom between them, I think is necessary) depending results...

I will look your entry blog about snubbers !

Thanks for the correction about the sg3525. I will put 500R for R9 in the schematics and move the value after some tests. I think I need first to have a good square wave signal before moving deadtime (would be difficult to measure DT if the signal have some spikes or other distorded form), so first I will found the good snubbers.

Now, i'm in the routing !

02-23-2018, 03:30 PM
Hi !
The PSU seems to work properly !
I have play with the snubbers, but i think I don't need them, at the primary and secondary side, there is no overshoot or something strange, it's even the opposite. In attache a view of the mosfet command. Smooth square signal and little undershoot at the end of the waveform.

Silvio, I have try to use your method to found the correct value for the gate resistor, I found 16R. I still use the smallest resistors I have for the moment, 22R....I don't understand the second part of your how to. :(



02-24-2018, 11:29 AM
I was just wondering why you did use the sg3525 for a non regulated power supply for low power. You could have readily used the IR2153. This would have made your circuit easier and having the oscillator and mosfet drive in one package. This chip could handle medium power fets such as IRF840 or IRF740 and a power between 100 and 400 watts can be achieved quite easily.

Regarding the gate resistor, well this calculation allows one to find the correct value for the gate resistor.
This is a combination according to the mosfet driver drive and source current capabilities. The gate resistor is chosen so that the maximum current of the mosfet driver are not exceeded. Hence the mosfet is charged and discharged quickly enough according to the frequency of operation.

Lets consider an IR2153 chip which have a drive and source current of around 200mA and an IR2110 which has a drive and source current of 2A. As one can see the both chips have different current capabilities and the selection of the gate resistor for both chips will be different.
I admit the IR2110 is designed to handle larger mosfets with higher gate charge requirements. The explanation is made so that one can calculate the gate resistor according to the driver used.

It is also important that one must observe the data sheets for both driver and mosfets used. These tend to vary between different manufacturers.

I hope that helps

Regards Silvio

02-24-2018, 12:00 PM
Silvio, because I have many sg3525 chip in stock bought at very low price. The idea is to start with this small psu, unregulated, and to continue to learn about smps with some sort of regulation and with bigger psu. One chip to learn is enough for me. Maybe it is not the easiest, but for future use I think it's a good choice. I'm not feel good for the moment with alls the different specifications of this sort of chip but I will look the IR2153 for my knowledge.
For the moment the psu works very good on alls the outputs. I can use twice the current I need on outputs without any heat for a total of 3.5A at primary side. Idle current is 0.15A. With my dsp/dac plugged, I draw a total of 1.7A on primary side.
I will try to mount a second psu with just the needed components for the SG to experiment with gate resistor/freq/snubbers/transformer. Maybe I can correct the waveform for a more neat square. The issue seems to be an over-damped form (correct terms ?). I know snubbers can correct overshoot, but here I need the opposite. Right ?

02-24-2018, 04:19 PM
I don't know where are you measuring that wave form is it from the gates of the mosfets or is it from the output of the SG chip?
It seems that you have a slow rise time and also a slow turn off time. The curves on the wave form clearly shows that. What seems to be the under shoot is the slow turn off and the dead time.

03-12-2018, 03:31 PM

Thanks for your answer Silvio.
I have done some tests and play with CT/RT/RD to obtain a 1ÁS dead time and the cleanest square signal I can with just the SG and driver. Ct is 1nF, RT is 15K, RD is 470R. The freq is around 47.5Khz. If I go lower for RD with 250R, dead time is 860ns. With the old CT and RT value charge and discharge time are longer.
I have done some screenshots of the results, hope they are goods for comprehension:
SG3525 output alone, the 2 outputs are identical :

Close view of start and end of the waveform :

SG3525 + driver TC427 :

Close view of start and end of the waveform :

A view of the 2 waveforms with some markers showing 944ns dead time :

I can see some ringing in the outputs. What can be the cause ? Routing of the board ?

03-12-2018, 09:18 PM
Hello Pichou,

The output from the mosfet driver is most important as this is driving your fets. From what I can see I understand that you have a very nice square wave form coming out to the gates.

Regarding the ringing well this may be caused:-

long runs from driver to gate or from SG to driver.

long leads from driver to mosfet these need to be as short and as thick as possible

High current traces near signal traces thus coupling with them

lastly transformer primary kicks back when switched off from the mosfet. This is only suppressed by a proper snubber across the winding.

usually this ringing is observed only when smps is loaded. The more load the more it rings

One other thing that might be causing ringing is the output inductor. This I have experienced myself and may be the case that you have to make another snubber just after the secondary to suppress any feedback to the trafo

Here is link with some helpful tips for building car audio power supplies


Hope that helps

03-13-2018, 12:45 PM
Your comments are every time useful and welcome ! Thanks for your time !

On the screenshots, there is only the primary side mounted on the board. So ringing is certainly due to my board layout. I think is due to the poor Rigol quality probes too. If I move them vertically to the pcb, ringing is....different...

I will continue to experiment first by calculate and placing gate resistors and mosfets and play with that, make new screenshots and wind a new transformer. When the transformer will be ok, I will continue with new screenshots and by placing rectification and see what happens with a load, add the caps and output inductor etc etc. Step by step approach to learn.

I have a question, does I need to load each output of the transformer to define the primary snubber ? If I understand, this snubber correct only the winding of the primary ? But can pass some strange effects from the secondary like a mirror.
I can put snubbers on secondary side, I have let the place. I can put a snubbers for each secondary output. But for the symmetric analog output, does the best way is to put the snubbers across the 2 legs of the secondary or between each leg and middle point ?

Thanks for the bcae1 link ! I know it and more I read it more I understand the "rules" for smps.

03-13-2018, 06:03 PM
Some new screenshot of the gate signal direct with the probe on the pin of the mosfet.
I think I have a good (the best) result without gate resistor. A little better slope if I put a 10K in pulldown. I have done some test still without gate resistor but with different value of the pull-down, and if I decrease the value I have very little dead time added and no visible enhancement on the slope of the curve. Does it's a problem to run without gate or pull-down resistor ?

Gate without anything :

Gate with 10K pull-down :

Gate with 3K pull-down :

Gate with 1K pull-down :

03-13-2018, 09:22 PM
Hi Pichou,

It is advisable that you use a gate resistor. This will limit the gate current as your diver might not last long without it.
Your driver can deliver around 1.5 amps.

You can take a look at the blog post regarding this subject.If you are using the same fets described in the schematic you posted earlier then use the specified one (22R) You can use a reverse diode to enhance switch off of the fet (1n4148 diode will do) see diagram I posted

The pull down resistor is there to prevent false switching of the fet. The driver actually switches off the fet, 10K ohms is enough in this case.

You can take some scope shots across the primary windings. You will notice that the rise and fall times will be much better there.

.I would like to see some pics of your project this will help me with better suggestions


regards Silvio

03-14-2018, 07:34 AM
Regarding the snubber you will try first without it and see how things perform.

If you will be needing to load the smps it does not really matter which winding you load on the secondary it still has the same effect as long as your loads are resistive. (Car bulbs in series will do) I don't know your output voltage/s.

Lastly use symmetry in your windings and tight coupling especially with the main secondary winding which will carry most of the load.
A good transformer is the key for a clean waveform.

04-11-2018, 02:24 PM
Here is a photo of the psu. Don't know if you can see it in large dimensions...
Transfo is a PRIM 6+6T made with 0.7x1 / SEC1 11+11T with 0.5x1 / SEC2 7T with 0.6x1
RD 470R / CT 1nf / RT 15k / F=48khz
R gate 11R + pull down 10K + Diode like your schematic.
Fall and rise time are better !
Snubber for PRIM is 100nF / 100R. Can't get lower for the resistor or higher for the cap because resistors heat quite a lot...Calculated value is near then 12R....impossible to put that value without huge current consumption and heat...Have to work on this more I think I don't understand or I made mistake somewhere....
No snubber for secondary for the moment.
Iddle current is 0.54A
Secondary voltage with no load at the regulators inputs are : 13.87v for bridge secondary, and +/-22.2v for dual rail output.

With no load here is some screenshots, strange spike, not seen before....::SD
Gate with no load :
Transformer primary with no load. What cause this "step" ?
Now some screenshots with load on the output (after the regulators).
1.35A at 7.8V on Digi output, 0.55Ax2 at 12v on analog output, total current drained on primary is 2.52A.
Gate is ok I think, but at the primary of the transformer the "step" is clearly visible. Does it's a problem to have this waveform ?
Because I found the same strange form on secondary....for sure...
Here is a view of the bridge secondary : same "step"
And a view just after the diodes. This problem is visible even after the diode.6621
The same problem is visible on the analog dual rail output.
I know I have some parasitics on the gate trace, but all my problems seems to show a bad transformer construction I think...

04-11-2018, 09:17 PM
Hi Pichou

I still do not know your input voltage but I am assuming it is 12v (correct me if I am wrong)

It will be better to put 2 snubbers across each primary winding (Your topology is push pull from what I can see)

You can also fit an 18v zener across drain source so that any voltage generated during switch off will be shorted out with the zener

Try lower capacitance with your snubber like 10nf and work out again the resistor value

Through experience I found that when the pwm is working in full duty not much noise is generated and usually the output inductor is not needed, this can generate a spike back to the primary winding. This is why the secondary snubber is needed

Having spikes in your wave forms can disturb the operation of the pwm signal to the gates

You need to use better symmetry in your trafo try to interleave the windings together if possible Primary and secondary side by side and wound together to the best possible way due to different turns in each winding.

04-12-2018, 12:07 PM
one last comment on the winding of the transformer

The symmetry in the primary windings need to be equal as possible. Make this winding in bifilar method so that the outcome will be as close as possible. Any difference between the two windings tend to generate a different wave shape or spike.

04-12-2018, 01:17 PM
Hi Silvio,

Yes, it's a car psu. During tests I use have 12.35v.
Yes it's pushpull with a snubber on each prim side.
I will try with lower cap. I've read some articles and see a bell curve for the snubber action. So, bigger is not better.

For the 18v zener, i assume I need to put it in the opposite direction of the internal diode of the mosfet ?

OK, I will try to make a new transformer.
Probably with 0.5x3 on primary and 2x0.4 and 1x0.6 for secondary.

04-13-2018, 05:53 AM
The zener should be placed the same direction of the mosfet diode and also parallel with the snubber so that only when there is over voltage will it conduct.

Regarding your snubber well this should be enough to dump excessive ringing and not dumping it all together otherwise you will be loading your fets for nothing.

04-13-2018, 10:38 AM
Hi Silvo,
Tried 2 news transformers following your instructions with the same visible step on primary.
Here is a photo of one transformer with only the primary and the 7 turns secondary. There is one winding of the 7 turns alone in the middle of the primary entry.
Primary is made with 3x0.5 mm wire and are winding parallel on the core with the other primary inside it (to made a 6 wire parallel profile = sec1 wire1/sec2 wire 1/sec1 wire 2/sec 2 wire 2 and so..).
Better is to twist each 3 stands first and twist the 2 primary together ?

For the moment impossible to measure the snubbers. With no load, I can see something like ringing, but with little load this first part of the waveform goes down to make the "step".

No load :

Little load (0.3A@15v) :

Little more load :

Photo of the first transformer :
Same behaviour

Photo of the second transformer :
Same behaviour

04-13-2018, 11:39 PM
Hi Pichou,

A few tips to help you out solve the problem

from what I can see the rise time is not fast enough and this may be the cause of the step you are seeing. Try lowering a bit more the gate resistor like 10 ohms and see the outcome

Please check also what is coming out of the SG chip as it could be the signal is not coming out as it should and thus disturbing the driver output stage. The duty cycle should be around 47-48%

The trafo winding looks good and I do not think is the problem now.

check again your dead time and see that it is around 1us, I can see you have a lot of ringing try to play a bit with the dead time and widen it a bit like 1,5 or 2uS and see the outcome

Push pull windings tend to ring a bit but that can be suppressed quite considerably with a good tailored snubber,

bridge out your output inductors and check again without them.

You can measure one side at a time to see more clearly

If I am not mistaken this problem came out since you mounted the smps on the pcb. If so check your tracks and see that you do not have a mistake somewhere especially around the sg chip. decouple the supply rail to the sg with a 100nf ceramic cap as closely as possible to the chip or directly with its pins

good luck

04-17-2018, 05:25 PM
Hi Silvio !

Gate resistor is already 11R (2x22r). Duty cycle is ok, dead time 1Ás from 3Ás without resolving my problem.
Decoupling is 100/10/0.1 on each power device.

All the problems disappear when I remove the inductor on output ! You are right with them ! Over....engineering from my head...thinking making better with the self...not in all case I understand...

Returning to step "charge psu and calculate snubber" !

Another time, thanks for you help Silvio !

04-17-2018, 07:02 PM
You're welcome

04-19-2018, 01:21 PM

Trying to define the snubber at primary side, but I don't understand something on your how to Silvio.
First, I measure the ringing freq with the psu loaded on each output (total of 2.6A on primary at 12.35v).
I found something like 14.7Mhz, it's very high....but the ringing is not a proper sine wave, there is some "non linear" behaviour.


When I put a 10nf cap on the snubber, freq is reduced to 3.48 mhz.
I don't have the same behaviour your are talking on your how to (the freq is shifted up).

04-19-2018, 08:43 PM
Hi Pichou, well I guess your capacitor is too large you may try something smaller like 1nF or 470pf and see the outcome. It also seems that you already have a high ringing frequency.
Regarding behavior of the ringing wave that is quite normal that it starts diminishing and lowering in frequency at the end.

The practice I used was read in a book and if I am not mistaken it was called power supply cookbook by Marty Brown. I tried it myself and it worked. The only difference between us here is that I was working on an off line half bridge smps and my voltage at the primary was around 160v. Well yours now is only 12v so that may be the different behavior.

Do not give up on this you can experiment with different value caps and different resistors. The ringing may not be all dumped but reducing the amplitude to 1/3 of what it was originally will be acceptable. A lot of dumping may cause unnecessary load on the switching transistors and this will bring more losses to the smps. It will also may affect the rise and fall times. Try to find a good compromise.
I would like to point out that snubber resistor wattage may be 2 and even 3 watts at times.

Regards Silvio