Issues with transformer/inductor cores

I'd like to open this topic on transformer cores and calculating them. I've had some issues with my SMPS projects that I guess are related to the transformer core, mainly too much primary current, big inductive spikes on the primary winding, audible noises coming from the transformer or not being able to sustain the load (the secondary voltage would collapse even with a small load). I'd also like to isolate transformer issues from the issues with the output inductor.
So I've built myself a small test rig that can be powered from a bench power supply with current limiting. Most of the transformer cores I tested behave badly, drawing a lot of current and making audible noises. Some of them are salvaged from electronics but I also have store bought ETD29/39 cores.
Some waveforms are pictured below. The one that looks better is from a green toroid, the only one that behaves well. I've also had the same issues with buck/boost converters that use toroid inductors from computer power supplies - the only core that behaved well was the green one. Frequencies are around 50-100kHz.
So what am I doing wrong? I use the formula below for calculating the primary turns, with a Bmax value of 1500.
Also, can a transformer be tested with no output circuitry (just the two windings), no feedback or even without a secondary winding, just the primary?
 

Attachments

  • DS1Z_QuickPrint4.png
    DS1Z_QuickPrint4.png
    58.4 KB · Views: 2
  • DS1Z_QuickPrint1.png
    DS1Z_QuickPrint1.png
    44.6 KB · Views: 2
  • turns calculation smps transformator.png
    turns calculation smps transformator.png
    6.1 KB · Views: 2
The topology of the test circuit is two switch forward. The schematic is similar to this. But the issues I have are regardless of the power supplies I've built (buck, boost, flyback, half bridge etc)
 

Attachments

  • TSFW_Amp_PSU.pdf
    36.2 KB · Views: 4

Silvio

Member
Hello Bogdan, all I can say seeing your post is that the green core has the highest permeability, that means that it offers the most inductance per turn.
You should measure the inductance more than the number of turns involved. Each ferrite material offers different features and permeability. Some are suitable for high frequency and others may not cope more than a 100khz. We tend to salvage a lot of transformer cores from ATX computer power supplies, usually these are used in the region of 20khz or so. These can be used for up to around 100khz but then trying to work with higher frequencies the cores tend to heat up due to core losses.

The green round cores are usually found in the input common mode chokes before the bridge rectifier. These cores will usually give an inductance of 1mH or more for a few turns. I use green cores for GDT for example. Compared to a yellow core which is usually found as the output inductor will give and inductance in the uH range for say 20 -30 turns you will get 60-80uH for a core of say 30-35mm diameter. The brown tan cores are found for example in audio output filters in class D amplifiers. These cores lend themselves well at high frequencies but need quite a few turns to get the right inductance.
I was experimenting myself with different cores to make an input inductor for my resonant smps. This was placed in series with the primary winding of the main transformer. Trying different cores it was found that these cores (brown tan) gave the best waveform with minimum spikes generated. I only needed 2uH and I used 2 stacked cores of 30mm diameter and winding 8 turns to get this small inductance needed. Doing the same on a yellow core for example I would have ended up with 2 turns and only covering a small part of the ring core.
Too much wire on a core may make it saturate early, too little will not lend itself well either.

The right magnetic field is required for every purpose. A transformer having too much turns in the primary will have less magnetic lines and will not transfer enough energy in the secondary winding. Too little will make it saturate early and tend to heat up the core quickly. The way you wind transformers plays also an important role keeping in mind the amount of coupling and also the leakage inductance which the latter is to be kept to a minimum as possible, these practices are the key for a good transformer.
If you came across Excellent IT software on this website you can download it and you can play with different core materials for different topologies you will find out how the number of turns will vary with different core materials and frequency.
Further knowledge can be acquired by following Robert Balanos on youtube. This man is an engineer and design smps for NASA. There is a lot of mathematics involved in designing an smps, it is not so simple as it seems.

Regards Silvio.
 
Thank you Silvio, I have watched Robert Balanos' videos and found them full of great information. I know building an SMPS is not an easy task but it seemed weird that I'm getting the same bad results with different topologies and different cores, and I thought maybe I'm doing the calculations wrong. I was expecting to at least get the power supplies in a stable state without oscillations and the switching transistors heating up, and then tackle the output, regulation and such.
Apart from TNY series that run at over 100kHz, I stayed around 50kHz which seemed reasonable for most cores. That's why I built the test circuit so that I can alter the frequency and see how the core behaves.
I would, for example, get the two core halves without the bobbin, wind a few turns of wire directly on the core (for 12V only 2-3 turns are needed) and then hold the halves pressed together while running the circuit and observe the primary waveform. Sometimes even increasing the number of turns did nothing, and even winding the transformer "properly" gave the same results.
I also tried adding a secondary and loading it, but still no success.
The formula I posted above was put into a spreadsheet and I would play with Bmax and the frequency to see how it affects the number of turns, but the differences are usually small.
 

Silvio

Member
Thank you Silvio, I have watched Robert Balanos' videos and found them full of great information. I know building an SMPS is not an easy task but it seemed weird that I'm getting the same bad results with different topologies and different cores, and I thought maybe I'm doing the calculations wrong. I was expecting to at least get the power supplies in a stable state without oscillations and the switching transistors heating up, and then tackle the output, regulation and such.
Apart from TNY series that run at over 100kHz, I stayed around 50kHz which seemed reasonable for most cores. That's why I built the test circuit so that I can alter the frequency and see how the core behaves.
I would, for example, get the two core halves without the bobbin, wind a few turns of wire directly on the core (for 12V only 2-3 turns are needed) and then hold the halves pressed together while running the circuit and observe the primary waveform. Sometimes even increasing the number of turns did nothing, and even winding the transformer "properly" gave the same results.
I also tried adding a secondary and loading it, but still no success.
The formula I posted above was put into a spreadsheet and I would play with Bmax and the frequency to see how it affects the number of turns, but the differences are usually small.
Well seeing the waveforms tells me something might not be quite right. First of all you have to start with a good square wave when feeding the gate, secondly adequate dead time is necessary. Snubbers are also needed on the primary especially if the pulse width is not 50%. Another thing you must consider is having adequate gate voltage to switch on the gates of the mosfets. This is very important as during short pulse width the mosfet tend to switch in the linear region which in turn heats up rapidly. Lastly if you are trying to make a variable smps with a wide voltage range then care must be taken due to this. Adjusting the feedback may not be an easy task especially when the the load is not constant.

I have been playing myself lately with a similar issue. I was using a simple resistive divider in the output to feed the opamp inside the sg3525. My setup included a gate drive trafo so that I do not have to use an opto coupler to isolate the primary from the secondary. When one sees the output waveform generated from the SG chip will notice a slight overshoot on the risetime then falls back to continue the square wave. This will represent itself on the GDT which in my case my GDT was very fateful in copying this on the output winding. I had also a totem pole after the SG so that I can have some more current through the GDT primary. Another thing that I noted was when the GDT is loaded it produced a little spike during switch off, here I included a soft snubber to reduce it (47r- 10nF) With all this the smps worked beautifully with a heavy load around the mid voltage output. (my topology was HB)

The maximum voltage calculated was 40v at full pulse width keeping in mind that headroom was needed for regulation as my final output was 30v @ 5A. Having such a wide range of voltage 3v-30v was not easy to keep the feedback stable on all voltages I had a glitch somewhere on the mid voltage like 22v which the pulse width was gittering up and down and not deciding where to stay. The mosfets did not like this and I blew a few until I learned what is really happening. I tried to alter the value of the capacitor on the compensating pin and finally I settled with only 10nF in parallel with 470K resistor. This gave me the best result. I also had to include a 22nf capacitor across the potentiometer which was a 50k ohm. It is also important to use a snubber across the output diodes as spikes are generated from the output inductor. These will reflect back on the primary.

All this took hours of experimenting and trying to find the best values for good operation. Keep in mind that every setup has its own issues and have to be dealt with separately. All this depends on the type of typology, frequency, transformer construction, voltage range and current. A fixed voltage smps will be much easier to do than a variable one having two rails with such a wide voltage range and having dual rails.
Here are a few pics of my work
 

Attachments

  • Variable smps 5.PNG
    Variable smps 5.PNG
    1.1 MB · Views: 5
  • Variable smps 4.PNG
    Variable smps 4.PNG
    2.3 MB · Views: 5
  • Variable smps 1.PNG
    Variable smps 1.PNG
    1.2 MB · Views: 5
I did a test with an EE16 transformer. A simple gate drive transformer with 20T on each winding, 470uH inductance. Signal is coming from SG3525. As you can see, the scope can't even trigger properly, and the transformer is hissing. By comparison, the waveform that looks good is the "reference" green toroid transformer.
 

Attachments

  • DS1Z_QuickPrint6.png
    DS1Z_QuickPrint6.png
    53.7 KB · Views: 0
  • DS1Z_QuickPrint9.png
    DS1Z_QuickPrint9.png
    119.7 KB · Views: 0

Silvio

Member
I did a test with an EE16 transformer. A simple gate drive transformer with 20T on each winding, 470uH inductance. Signal is coming from SG3525. As you can see, the scope can't even trigger properly, and the transformer is hissing. By comparison, the waveform that looks good is the "reference" green toroid transformer.
I can see from the pictures you posted that indeed the waveform in the second picture is rather distorted and full of oscillations. I understand that the bad waveforms are coming from the EE16 transformer. You did not show how you set this up. I don't know if you have a current amplifier after the SG chip or you are feeding it directly to the trafo.

The overshoot is rather common coming out of the SG chip and I usually always see this especially if the chip is fed with more than 12v. This is not an issue to my believe as this helps in charging the gates and makes the rise time quicker.

Regarding the inductance I think you must have at least 1mH for 12v on the GDT for a frequency of 50-80Khz. A toroidial transformer offers much better symmetry and that is why you are getting a better waveform from it. I have seen GDT on pro equipment even made from transformers like EE16 or smaller, however when checking the waveform these produce are not so fateful and the waveform produced is rather a bit distorted. To sum it all up a toroid core gives the best waveform. Green cores seem to have the most inductance per turn and these will offer the least turns to get to the required inductance. I myself used a small snubber in the primary of the GDT as explained in the last post. Where windings are present they will always produce a spike on switch off and that is why I included the soft snubber. I could do this as I had a totem pole driving the GDT with adequate current to spare.

Lastly if you post a small schematic of your setup I would understand better in trying to help you out. Please also include the place you are probing with your scope.

Regards Silvio.
 
No, it's just the bare SG chip and the transformer connected directly to the outputs. The current is not that high either (about 100-200mA). I'm not in the lab right now, but last time I added a capacitor on the VCC rail and a snubber as you mentioned and got a decent waveform (given that the whole circuit is on a breadboard). But yes, the green core is unbeatable, so apart from isolation I don't see any reason to use the EE16 transformer. In my setup the GDT is only used as a driver, not for primary-secondary isolation.
The schematic is attached below. It's really only a small daughter board that I can attach to the main power supply or play with on a breadboard. I soldered the GDT "primary" between OUT A and OUT B and the probes on the two secondaries.
 

Attachments

  • SG3525_board.pdf
    17.5 KB · Views: 1
So I tested the configuration below (half bridge) with some cores (connected to TR_PRI). The windings are bifilar.
First picture is with a green toroid unloaded
Second picture is with the same core loaded with a 20W bulb. The current draw is only around 600mA, where usually the bulb draws about 1.5-2A at full power.
Picture no. 3 is with an EI core unloaded and no. 4 with the core loaded.
Picture no. 5 is with an ETD29 core unloaded and no. 6 with the same core loaded.
The number of turns are calculated with the usual formula.
 

Attachments

  • DS1Z_QuickPrint1.png
    DS1Z_QuickPrint1.png
    40.8 KB · Views: 1
  • DS1Z_QuickPrint2.png
    DS1Z_QuickPrint2.png
    56.2 KB · Views: 0
  • DS1Z_QuickPrint4.png
    DS1Z_QuickPrint4.png
    42.7 KB · Views: 0
  • DS1Z_QuickPrint3.png
    DS1Z_QuickPrint3.png
    50.6 KB · Views: 0
  • DS1Z_QuickPrint5.png
    DS1Z_QuickPrint5.png
    51 KB · Views: 0
  • DS1Z_QuickPrint6.png
    DS1Z_QuickPrint6.png
    50.2 KB · Views: 1
  • Screenshot 2021-03-02 120125.png
    Screenshot 2021-03-02 120125.png
    55.2 KB · Views: 2

Silvio

Member
So I tested the configuration below (half bridge) with some cores (connected to TR_PRI). The windings are bifilar.
First picture is with a green toroid unloaded
Second picture is with the same core loaded with a 20W bulb. The current draw is only around 600mA, where usually the bulb draws about 1.5-2A at full power.
Picture no. 3 is with an EI core unloaded and no. 4 with the core loaded.
Picture no. 5 is with an ETD29 core unloaded and no. 6 with the same core loaded.
The number of turns are calculated with the usual formula.
I can see that the green core is producing a nice square wave. The waveform is getting distorted because the SG chip is getting overloaded. You cannot load 600mA when the SG chip can only give 100mA continuous and around 400mA peak and momentarily. All you need is a current booster after the chip ( a totem pole made with 4 transistors) If you try loading around 100mA without the totem pole I guess you will get a much better waveform. If you need more current than 100mA then a totem pole to boost things up is surely needed. Remember that a good waveform is the success of a SMPS.

Pictures 3-6 show that you have poor charging to the gates and slow turn off. You need to boost the current up to charge the mosfet gates better.

Do not take notice of the component numbers as I was still sorting them out. You can see the totem pole configuration before the gate drive trafo
The second pic shows my GDT in the center using a green toroid core
 

Attachments

  • totem pole 2.PNG
    totem pole 2.PNG
    1.7 MB · Views: 5
  • GDT.PNG
    GDT.PNG
    1.2 MB · Views: 5
Last edited:
I will try that, but the green core was supplied through the transistors directly from V+, not through the chip, and the bulb was connected on its secondary. Probably the gate drive current is proportional with the drain current.
I did however notice a difference between transistors (I tested some IRF530, 50N60, IRF840). The 50N60 performed the worst.
 
So this is the updated schematic. LOAD is a 20W bulb. With the bulb connected directly, it draws 2A, but as connected in the schematic, through the transistors, only 700mA.
 

Attachments

  • sg3525_hb_gdt.pdf
    16.9 KB · Views: 1
  • DS1Z_QuickPrint7.png
    DS1Z_QuickPrint7.png
    40 KB · Views: 2
Last edited:

Silvio

Member
So this is the updated schematic. LOAD is a 20W bulb. With the bulb connected directly, it draws 2A, but as connected in the schematic, through the transistors, only 700mA.
Well as you can see now the waveform is much better. I do not know where you are probing but if you are probing the input of the GDT then You should try a different capacitor like 103 (10nF can be ceramic type) with the input snubber or maybe try playing a little with the value of the 47 ohm resistor. Be careful not to load much as this will waste power to the GDT. BC547 and 557 can only handle 500mA of current and are general purpose. The base resistors are to be left as they are as there is ample gain in the totem pole transistors. If you have say BD149/140 will give you a better choice as they can handle more current and are more suitable for switching purposes. I have also seen 2n5551 and 2n5401 used for this purpose and they where switching a 3Kw smps having IGBTs driven by these transistors.
I can also see some ringing in the waveform and this has to be kept to a minimum as possible. You should resonate the snubber values with the ringing frequency to dump it as much as possible. I will post a pdf in switching power supply information how to calculate snubber. Another snubber should be added to the transformer primary. This should be either directly across the primary winding or you can add a snubber across each switching transistor instead. This will be your choice and see what works best.
The risetime now seem much better and the waveform is rather square as it should be. This will give a better gate charge and switching the fets hard on and not let them work in the linear region as this will heat them up very quickly. Ringing may be also due to long wires feeding the fets and maybe when a pcb is made this would be less. For experimenting you should twist the wires going to the gates to minimize noise pickup and stray inductances as much as possible.

Here are a few pics with my results
You can see that the amplitude of the waveform will decrease with load. Check the voltages on the oscilloscope readings. Ringing was very minimal here. The load was around 10W. The totem pole made with BD139 and BD140.
 

Attachments

  • GDT input wave no load.PNG
    GDT input wave no load.PNG
    913.4 KB · Views: 2
  • GDT input wave loaded.PNG
    GDT input wave loaded.PNG
    946.1 KB · Views: 2
  • Setup of GDT and load.PNG
    Setup of GDT and load.PNG
    1.1 MB · Views: 2
This setup is currently on a breadboard with a lot of wires in the air so some resonance is to be expected. The probe is placed directly at the bulb terminals.
I realized that I forgot to place the series capacitor, this is the waveform with it. Also I noticed that the waveform actually swings between +5V and -5V, which makes sense because it's VCC/2 minus losses. But the current draw is still small, if the bulb can draw as much as 2A why does it only draw around 500mA in this circuit? In the end it sees almost the full VCC minus losses, so it should also be able to draw as much current as possible.

EDIT: double checked with an IR2110 circuit, current draw is identical and the waveforms look about the same.

 

Attachments

  • DS1Z_QuickPrint8.png
    DS1Z_QuickPrint8.png
    49 KB · Views: 1
Last edited:

Silvio

Member
From the screen shot with the series capacitor it seems that the value of the capacitor is small and the voltage is dropping quite a bit. I suggest you remove the series capacitor as it is not needed now. With that EE16 your input inductance seems a little bit small at 500uH you should use more turns to it so that you have at least 1mH in the input. The way you are winding the trafo is also important. Very tight coupling is important here and stray capacitances are to be kept to a minimum. Safety is also important and adequate insulation between the windings is important. You will be having 320v and 12v so each winding should be isolated properly with minimum tape between layers and 2mm margin clearance on each end of the bobbin. A toroidial core will get you a better result. I used CAT 5 wire from a data cable to wind mine. The GDT is for a full bridge smps and contain 5 twisted wires.
Your lamp seems a bit too much of a load, in real life you will not use that much. It is important that at least with 8-10w load you must have an amplitude of 10v or more otherwise the mosfets will take more time to switch on and off. With poor switching voltage the mosfets will enter the linear mode and heat up quickly and eventually blow.
You asked me why the lamp is drawing little current. It is because there is a severe voltage drop arriving at the filament and only around 7v rms are going to it. That lamp is too much of a load. The mosfets will not draw that much for sure. If you tested the output with an IR2110 and got the same result then it proofs that the load is excessive.
PS try putting a series resistor (10-20 ohm) with the primary instead of the capacitor. See if the ringing gets any better.
 
Top