<>

[diyaudio]

  Good progress. Your waveform looks good. try to take a waveform snapshots of the mosfet Vgs, Vds.

[T3sl4co1l]

You need to check if the ferrite will saturate.

Peak flux density Bmax = V*dt / Ae
dt is maximum switch on-time (for DCM), V is applied voltage, and Ae is the effective cross sectional area (core datasheet).

For CCM, the total V*dt is integrated over many cycles, so it's easier to reference it by peak current.  (Current works equally well for DCM, too.)  In that case, the expected flux is:
Phi_max = L * Ipk = V * dt

Note that H has units of Vs/A, i.e., inductance converts flux (applied volts * time) to the current flowing through the inductor, or vice versa.  It's Ohm's law, time-dependent.  (As well it should be, because a linear inductor is linear in the same way a resistor is.)

Ferrites saturate in the > 0.2T range, with the most common MnZn types going up to 0.4T or so.  They get hot when operated at this flux density, cyclically, though.  It may be more practical to choose 0.1-0.2T for a core of this size, at this frequency.

This also allows you to calculate the gap.

The minimum gap, and therefore the minimum number of turns as well, is given by:
N = V * dt / (Bmax * Ae)
Bmax is whatever peak flux density you choose.
Alternately, V * dt = V / (2 * F) for a 50% duty cycle unipolar circuit (like this flyback), or change the 2 to 4 for a bipolar circuit (e.g., the typical half bridge forward converter), or to 4.44 for a sine wave of Vrms.

Flux is determined completely independently* of inductance.  Inductance is determined by the ratio of Ae to l_e and the core properties.

L = A_L * N^2
A_L = mu_r * mu_0 * Ae / l_e

*To the first order.  Second order effects, like fringing flux and leakage, have a small effect (~10%), so "complete" isn't very complete at all.  But those effects are additive, so the results from this assumption are conservative.

We can now take this in two directions: when we introduce an air gap, we can say we're reducing the effective permeability.  This holds l_e constant, and varies mu_r.  In that case,
mu_eff = l_e / (l_g + l_e / mu_r)
And we use this value for mu_r in the usual A_L formula.

Alternately, we can say the effective length is all air gap, in which case mu_r = 1 (because we're taking air equivalent), and the core's contribution is l_e / mu_r.  (In effect, the fact that the core has high permeability, means its length is short-circuited by the same factor.  This is why cores are so helpful!)  This way, we get:
A_L = mu_0 * Ae / (l_g + l_e / mu_r)

One final alternative, that's somewhat separate from these: we can calculate the amount of air gap we need, in the first place.  Air gap stores energy, and the energy density in that gap is given by:
e = Bmax^2 / (2*mu_0)
The cross section of the gap is Ae (usually slightly more, for the same exception above*), and the thickness is the gap (or, the gap plus the core's effective air gap length l_e / mu_r, since we now know about that).  The energy in an inductor is 0.5 * Ipk^2 * L.  Smooshing all these together allows us to solve for the required gap, without knowing anything more about magnetism and cores.


I would strongly suggest avoiding electrical tape (the vinyl stuff), because it's squishy and melty.  Polyester and polyimide tapes and films are fine, or you can use paper in a pinch, if you don't mind that it's not exactly UL94V-0 self extinguishing.

Tim

[MagicSmoker]

Waveforms look pretty good to me, especially for being on a veroboard. I annotated one of the scope shots:

[diyaudio]

Waveforms look pretty good to me, especially for being on a veroboard. I annotated one of the scope shots:

I don't agree with your annotation on the Vgs waveform.

That's not typical for a Vgs signal. He's waveform needs to have the typical miller charge characteristics, with minimal overshot (with correct snubbing in place) see attached example for Vgs (Voltage Drain to Source) notice in the example the typical steps during the different miller charge for turn on and turn off times. You need to learn how to tame your waveform, you might also have to look at Soft Switching vs Hard Switching. (fast turn on interval is NOT important for this project) You also need to use a different ground clip when probing Vgs signals, loose that clamp clip and use a solid wire to ground for probing. Dave has done a video on this subject already.

You are still using thin Ebay wire to supply power to the breadboard and then to the mosfet board, that is probably one of reasons why you have such a large stray inductance, also your transformer has leakage inductance causing the ringing, you still need to inspect and measure this and work on your transformer winding, measure and snub where it applies.   

[T3sl4co1l]

I don't agree with your annotation on the Vgs waveform.

Yabbut he didn't annotate Vgs?

That's not typical for a Vgs signal. He's waveform needs to have the typical miller charge characteristics, with minimal overshot (with correct snubbing in place) see attached example for Vgs (Voltage Drain to Source) notice in the example the typical steps during the different miller charge for turn on and turn off times. You need to learn how to tame your waveform, you might also have to look at Soft Switching vs Hard Switching. (fast turn on interval is NOT important for this project) You also need to use a different ground clip when probing Vgs signals, loose that clamp clip and use a solid wire to ground for probing. Dave has done a video on this subject already.

The Miller step is visible on the rising edge.  At this scale, it shows as merely a brightening of the beam, but that's one plus of analog scopes (or sufficiently accurate and oversampled DPOs).  The falling edge step kind of gets lost in the ringing junk, and may be fairly weak due to the nature of the driver circuit.  Later (during the free induction ringdown), the baseline ringing is due to Cgd and unclamped Vgs.  The driver has a +/- 0.7V deadband, which is fine.

Tim

[MagicSmoker]

Waveforms look pretty good to me, especially for being on a veroboard. I annotated one of the scope shots:

I don't agree with your annotation on the Vgs waveform.

Probably because I was commenting about the Vds waveform... 

[diyaudio]

Waveforms look pretty good to me, especially for being on a veroboard. I annotated one of the scope shots:

I don't agree with your annotation on the Vgs waveform.

Probably because I was commenting about the Vds waveform... 

sorry. hahaha

[diyaudio]

Hey Tim, so I've been trying to understand your previous post and have a lot of questions, but I'd like to get through it one question at a time.

"You need to check if the ferrite will saturate.

IMHO

The best way to uncover the mystery's of known OR unknown magnetic specifications for SMPS  (and you don't have to build this now) is build a saturation test rig with variable Frequency and Duty Cycle adjustment, that's what I showed a few posts ago.I'm able test and verify core saturation levels with optimal frequency (for any material SMPS Transformer OR Inductor) and trace the numbers back to the datasheet (if its available), this is my preferred technique, I've have had good success thus far and tested custom wound transformers on a ETD49 EPCOS core and commercial grade magnetic components with supplied datasheet figures and the numbers are always spot on. Using brute force esoteric math calculations is cumbersome.

[diyaudio]

Diyaudio, I should really do that because this is getting quite tiring.
I've been at it continuously for 4 hours a day (and it's a good thing I'm on vacation now), working on the FBT calculations and researching the net, to very little gain...

Does your test rig look something like this? (but replaced with the TL494 of course, and with both variable frequency and duty cycle):


If you have a schematic for your test rig, I would appreciate if you uploaded it here.

I don't actually have one for good reason (cause its dead simple to make) using a TL494 replace the oscillator section with a pot for variable frequency adjust and apply a voltage from 0 ~ 3.3V to the DT PIN for duty cycle adjustment now you have variable frequency and dead time adjust, use a power fet I used a (50n06 I had 10 laying around) and a non inductive resistor in series with the drain to obtain and sample the current ramp. There are so many variants of this circuit using a TL494 is over kill, you can even use a one 555 timer as shown here.
https://hackaday.com/2013/10/28/making-a-power-inductor-checker/

     

[T3sl4co1l]

You've seen half a dozen different formulas, but realize they're all different rearrangements of the same fundamental equations.

All of which go entirely by definition, so dimensional analysis is 100% accurate here!

So, flux density?  That's density by area, so, flux per area.  In mks units, use V.s / m^2 == T.  Or replace m --> mm and s --> us, which is more handy for SMPS.  (This also uses MHz, uF and uH.)

There's a run-down of basic definitions here:
https://www.seventransistorlabs.com/tmoranwms/Elec_Magnetics.html

Saturation flux density is in the datasheet.  It's a bulk material property, so if it's not in the core datasheet, check the material datasheet (e.g. 3C90).  The flux density values they give in the inline table is kind of a recommended operating point: they specify the power dissipation under that condition, which as you can see, is pretty modest (under a watt).  A core that size would be sweating at 1W, so you'd want to use less power than, whatever Bmax gives 1W... (to figure that out, check the material datasheet for core loss vs. flux density).  The values they give are usually somewhat conservative, so you can run, say, 200mT at 100kHz without much worry.

Or you can always run at lower Bmax, as long as you don't mind needing the extra turns, and reap the rewards of lower losses.

Tim

[MagicSmoker]

There are many different approaches to designing a flyback converter depending on whether you are using an off-the-shelf transformer, a random core from the junk pile that looks about right, or a known core purchased from a reliable distributor. For the beginner in SMPS design the 2nd option is the worst because of too many unknowns.

The first thing to keep in mind about the flyback transformer is that it is really a multi-winding coupled inductor, and not a true transformer per se (but everyone calls it a transformer anyway, including me). This is because current is only flowing in either the primary or the secondary at any given time, never both, so each winding acts like an inductor. The second important point is that ALL of the energy delivered to the output has to be stored in the flyback transformer first. The third point is that the windings are all linked together by conservation of amp*turns (AT or A*T)and so the primary acts like a current sink (storing energy in its inductance) while the secondary acts like a current source. The upshot of this is that the relative voltages across each winding will be fixed by the turns ratio - as in a true transformer - but the absolute voltages are not fixed, just as is expected of a current source.

For example, if current ramps up from 0A to 10A in a 10t primary while the switch is on, then the current exiting the secondary immediately upon switch turn off will ramp down from a level of (10A * 10t)/Nsec. If the secondary is 1t then the current will attempt to ramp down from 100A; if the secondary is 20t then current will ramp down from 5A, etc. Of course, for current to flow in the secondary there needs to be a load across it, and it is the load resistance which sets the output voltage! So if the secondary winding has 10t - same as the primary - and we put a 1R resistor across it then the peak secondary voltage will be 10V. How about a 100R resistor? Well, to get 10A to flow through 100R you need 1kV... Hence one of the reasons why the flyback is so popular for high output voltage applications. It's also why you never want to run a flyback unloaded if the controller IC can't do pulse-skipping or otherwise shutdown the switch drive signal if the output voltage rises too high.

There Ain't No Such Thing As A Free Lunch, however, and in this case the voltage across the secondary when its diode is conducting is reflected back across the primary through the turns ratio, and this reflected voltage, plus the input voltage, is what the switch must withstand when it is off. Similarly, when the switch is on the voltage across the primary is reflected across the secondary through the turns ratio and adds to the voltage the secondary diode must withstand (on top of the output voltage). Hence why I said the relative voltages across each winding are always enforced through the turns-ratio, even if the absolute voltage transformation ratio is not fixed.

All this means that we use the inductor equations to design the primary of the flyback transformer, and then select an appropriate turns ratio to trade off peak currents and voltages in both the primary switch and the secondary diode. If we use a step-down ratio then we reduce peak primary current (when the switch is on) but increase peak primary voltage (when the switch is off off), and vice versa for the secondary diode.

The relevant equations for an inductor are:

E in Joules = 0.5LI² (where E is in uJ if L is in uH and I is in amps)
L * dI = V * dt (where L is uH if dt is in us and dI is in amps)

However, the wide range of flexibility in duty cycle, frequency, turns ratio and even inductance in a flyback can make it somewhat maddening to design, so the best approach to take will depend on initial limiting conditions. For example, if you are using a salvaged core that you don't have full specifications on - and especially if you are manually introducing an air gap - then you might need to make frequency a dependent variable to achieve the necessary power throughput (a word of advice: your life will be much easier if you use pre-gapped core sets from DigiKey, Mouser, etc.).

Duty ratio is also a bit of a wild card in that the maximum allowed by the controller IC (e.g. - the TL494 is 45% per output) is not necessarily the limit when a flyback is in DCM. For example, duty Cycle in a transformer isolated flyback with a Npri:Nsec turns ratio specified by n is:

D = (n * Vout) / ((n * Vout) + Vin)

So if the transformer has a 2:1 step down ratio (ie, n = 2) and Vin[min] is 10V then the calculated duty cycle is 0.50, which means the switch on time and diode on time will be equal. In discontinuous conduction mode (DCM), however, all of the windings go to 0A for some portion of each switching period, so it would be perfectly reasonable for switch on time to be, say, 7us and diode on time to also be 7us, but total period at 50kHz is 20us, so that means for 6us each switching period all of the windings have stopped conducting (but then you get the infamous DCM ringing, as I pointed out previously). The way you ensure DCM is by *lowering* the primary inductance (which also causes peak current to rise).

Use the website below to play around with turns ratio, frequency, inductance, etc., in a flyback:

http://schmidt-walter-schaltnetzteile.de/smps_e/spw_smps_e.html

Just ensure that Ton doesn't exceed 9us at Vin[min] and 50kHz fsw, of course.

Faraday's equation - the one you have found in a wide variety of formats - is only needed to check that the flux swing isn't too wide, leading either to saturation (which is disastrous) or just excessive core loss (which in most ferrites is proportional to around the 2.5 power of total flux swing).

Finally, there are pre-gapped ferrite cores available commercially which are usually specified by A_L in nH/t², and this usually assumes the use of one unground half and one ground half. Needless to say, pre-gapped cores are going to be a lot more predictable and consistent than attempting to insert spaces between ungapped cores you salvage. Also beware that magnetic amplifiers are commonly used in ATX power supplies to derive lower voltage rails without requiring another winding on the transformer and the cores that are used in this application are extremely unsuitable for use as flyback transformers (or regular chokes).

EDIT: Changed N -> n in formula above

[T3sl4co1l]

To myself: "You're going to be disappointed with that windup..."

Then I read "60V peaks".

Well, yeah.

Better?  Well, there's that at least.

As long as you're going to use doubled up wire, you can do it in single layers each.  One layer of primary, tape, one layer of secondary, tape, primary, tape, secondary.  Connecting layers in parallel reduces the impedance further.

As shown, impedance is probably something like 200 ohms, but system impedance is around 5 ohms.  The huge mismatch manifests as leakage inductance, hence the peak.  Basically, the leakage is (200/5) times worse than it needs to be.

What does the current waveform look like?

Is the winding phase correct?

Tim

[jbb]

As T3sl4co1l said - leakage inductance is responsible for those big spikes.

Leakage inductance is very dependent on winding geometry, which is why an interleaved winding can help.

Here's my suggestions:

• In many ways an ideal winding exactly fills the width of the bobbin in one layer. This gives a nice flat surface to wind the next layer on and also reduces intra-winding capacitance.
• A layer of mylar tape between layers makes it much easier to keep the next one flat.
• Where leakage inductance is critical, interleave the windings (see below).
• Note windings often take a little more space than you expect.  This can be due to things like the copper distorting as you bend it over a 90 degree angle.
• You'll see things written about fill factor.  Some handbook will tell you that 0.7 is possible.  I suggest you don't try to go higher than 0.5 without some practice.
• Make 'em look all pretty and nice. Scrappy windings can have unpredictable parameters or other problems.

For 2A RMS maximum, you might be OK with a single strand of 0.5mm diameter
Winding option A:

10	Tape
9	S2	13 turns 0.5mm dia

8	Tape
7	P2	14 turns 0.5mm dia

6	Tape
5	S1	13 turns 0.5mm dia

4	Tape
3	P1	14 turns 0.5mm dia

2	Bobbin
1	Core

Then wire P1 and P2 in parallel, S1 and S2 in parallel.  This might give you some troubles with circulating currents, though, because (P1 and P2) and (S1 and S2) follow different paths through the magnetic fields inside the winding pack and therefore could develop different Electro Motive Forces (EMFs). Different EMFs can lead to currents circulating through the windings.

Winding option B:

10	Tape
9	S2	6 turns 2x0.5mm dia

8	Tape
7	P2	7 turns 2x0.5mm dia

6	Tape
5	S1	7 turns 2x0.5mm dia

4	Tape
3	P1	7 turns 2x0.5mm dia

2	Bobbin
1	Core

Then wire P1 and P2 in series, S1 and S2 in series. Because the parallel windings are right next to each other (similar to what you've wound already) you can be pretty confident you won't have circulating current problems.

Caution: you cannot rely on single-insulated enamel wire (aka magnet wire) for safety.  This is because the enamel has little pinprick holes in it, which can rub through into short circuits (rare but possible).  If you need safety insulation (e.g. to get from mains to a human-touchable circuit) the transformer design needs a lot of careful attention.

[T3sl4co1l]

Yeah, series windings are usually better.  You can get away with many strands that way, 2 or 3 or even 4 strands, laying flat together (make sure they don't overlap).  Or copper foil / tape (but you'll need to add insulation), which is fantastic for high currents.

Also, don't use vinyl "electrical" tape: it's very stretchy, and melts at a low temperature.  I would rather use masking tape -- it's paper, which is bad, but paper was used back in the day, and paper doesn't ignite until higher temperatures (class A insulation).  Ideally, polyester (class B I think) or polyimide (class F? good for >200C!) should be used.

Without varnish filling the gaps, it's also a good idea to derate the wire size further.  Not a big deal on small transformers like this, but you'll notice things getting much hotter as you go up in size.

Tim

[diyaudio]

What's up with strange PI filter and the feedback node before the output inductor? Surely this isn't correct. ? as a measure I would use a pot and adjust the duty cycle to the max (0.4)  to confirm your transformer is working correctly at maximum load regulated so the process of elimination can begin. You have two problems remaining.. transformer efficiency verification  and voltage mode feedback verification (with compensation) both can be tested independently.

[T3sl4co1l]

Feedback before the LC assists with compensation.

Tim

[jbb]

Whoa, so many replies haha. Didn't expect such a low-scale, nooby project to be so watched.

Well, I can't speak for others on the forum, but it's nice to see someone deliberately working through something challenging.  I'm sure you've learnt a lot! Flyback supplies are common, but actually surprisingly tricky.

EDIT: Ok, performed the testing procedure again. And holy moly, do my eyes deceive me? 83% mean efficiency! Awesome. Notice that my 12V initial input voltage dropped massively to about 10V! Ugh! This is why I don't like breadboards! But I don't have a choice at the moment...

That efficiency's not bad at all for a not-particularly optimised converter.  Moving to a proper PCB might help a little - but is it worth it to you?  Once you get the TL494 feedback behaving properly (which should be fine on the prototype board) you've gotten most of the learning out of it, and it looks pretty tidy.  However, laying out a proper PCB could be good practice too.

2. Tim, I got masking tape...somewhere...just need to find it. The core does get rather warm. 56C when drawing the full 1.3A load.

No worries there - that's just 'warm' rather than hot.

[T3sl4co1l]

That's not a zener, that's a Vbe multiplier.  The voltage might be stable enough, but the tempco, hoo boy...

You aren't going to improve on the TL431; you can mimic it, but you'll end up with whatever problems you had in the first place, but worse because it's discrete instead of integrated.    In short: fix your problems first.

The spikes come from dI/dt at the rectifier (and dV/dt at the transformer, if it's common mode), feeding through because of capacitor ESL.

Solution?  Add ferrite beads in series with the line(s), and bypass with a ceramic cap.

Tim

[T3sl4co1l]

2. From opening up a lot of old SMPS bricks, I noticed they often have zener diodes, usually 1Watt, reverse biased with the output. Or sometimes, there might be a hefty bleeder resistor. For example, for a typical 5V 2A charger, there might be a 5.1V zener reverse biased with the output. So, at worst, the power loss would be somewhere around ~(5.7V - 5.1V) * 0.25A load_min = 0.15Watts. Does that sound about right?

This calculation is for a 0.6V drop at 0.25A, but where is the 0.25A going?  It can't just dead end.

It has to go on to ground, sooner or later.  So you have another 5.1V * 0.25A = 1.25W somewhere, in order for this to be valid.

But at 0.25A, the supply isn't at 5.7V anymore, so the 0.6V drop doesn't actually exist anywhere.  (Classically, it would be dissipated within the Thevenin source -- but with switching and reactive power flowing around, there's no particular reason to believe the drop is due to an internal resistance that dissipates real power!  AC is fun like that, eh?)

So you just have / need the 5.1V * 0.25A load.

You might look into sharpening the gate drive, see if that improves the minimum pulse width.  Also, what it's doing near cutoff, if it is oscillating or going chaotic or what.

Also, make sure the TL431 and opto aren't saturated.  Note that TL431 can't pull down below about 1.9V (can you figure out why? ).

As for the zener: that's usually for clamping, or TVS protection.  It could also be for minimum load.  It's worth noting, adapters usually have crappy regulation -- no TL431 + opto, just aux winding feedback -- so the output can be even more wild than what you've got here.

Better supplies often have a crowbar: a zener or separate TL431 triggering an SCR that shorts the output if it rises over so-and-so.

3. Instead of having a zener or bleeder resistor, is it possible to use capacitive reactance, before the rectifier, for loading the secondary side output?
R = 1/(2pi*f*C). Set R equal to some value that will draw, let's say 0.25A @ 5V, current on the secondary. Frequency is switching frequency. Solve for C.

Mmmh, maybe, but it's a brute force way, and isn't friendly to the transformer and transistor.

Tim