Re: Help improving eff of a 24v -> 350v flyback converter

Mike wrote:

ye gods, DIY ferrite grinding - you are keen. 


Naa, It was actually very easy. Much easier than winding the bobbin.



1) input filter: 


It's gone now, Only 2-220uf caps in parallel on the input,
but it made no difference in the output noise or efficiency.


OK, then its not oscillating.


an input filter can make a smps oscillate. a smps is a constant power device - as Vin increases, Iin DECREASES. so it looks like a negative resistance, |R| = Pin/Vin^2.

bung a negative resistor acros the C of an LC LPF, and voila - oscillator!


you could read middlebrooks seminal paper, or just do this:
- ensure the impedance looking up the filter (2200uF C5 then 100uH L2, with C13 pretty much shorted by the "ideal" power supply) is less than |R|. for a low power supply this is usually automatic, as |R| is very high. for higher power supplies, check. spice is your friend here. get a decent model of L (include R and use its SRF to caluclate a shunt C), and also the cap (inc. ESR). pop these in parallel, excite with 1A current source frequency sweep. plot voltage, its actually impedance.

you will pretty soon see that filter resonance can be a real pain. 100uH, 2200uF has Fo = 340Hz, Zo = 0.213R so the cap ESR probably damps the filter (its a HUGE cap), but maybe not - ESR could be as low as 0.02R. in which case use a pair of 2200uF's in parallel, with an added 0.213R resistor, as an R-C damper to be applied across C5. or pick a C5 with enough ESR to do the job. Actually, Cdamp = 3Cout, Rdamp = Zo is best - see unitrode app note on snubber design, or just take my proof by blatant assertion :)

the general ROT is to use lots of C and bugger all L, that helps decrease Zo (making the filter more likely to damp itself with ESR), and greatly reduces the filter impedance seen by the smps.


2) output filter: 


I shorted across the output filter inductor. All that did was double
the noise at the switching frequency. Hard to tell what it did to the
low frequency noise since it got buried in the other noise.


if it only doubled the Fsw noise, its a pretty crappy filter. with a 500Hz cutoff and a 75kHz Fsw (call it 50kHz) you should have about 80dB of reduction - IOW a lot. that you dont suggests:

1) very low self-resonant frequency of coil.

with 3.5mH, 1.2nF will give an SRF of 75kHz.... depending how its wound this wont be hard to do.


2) piss-poor output cap - lots of ESR, ESL (about 30nH is usually a good guess).

I downloaded the KMX spec, its 3 Ohms ESR at 100kHz. ouch. with a DCM flyback if you know the duty cycle and DC output current, trig gives you the secondary currents. 220V, 2.2k = 100mA DC. assuming 50% duty cycle, the avg sec. current during (1-D) = Idc/(1-D) = 200mA, its a triangle so peak current = 2*Iavg = 400mA peak. that'll give you 1.2Vpp ripple for a single cap....

with so much ESR, you will get a significant benefit from paralleling a big fat film cap - I happen to have a whole bunch of 3u3 400V MKPs, I'd use one of them :) do an impedance plot of your KMF cap, it'll look like 3 Ohms for a fairly wide frequency range (ignoring ESR increasing with decreasing F :), so a 1uF cap will have a lower Z for F > 50kHz or so.



again, beware resonances. in this case Zo = 10.8R, Fo = 491Hz. betcha that rings like an SOB :(

3) R1, C13 (the other C13) 


Ok, ok, I fixed the part numbering. :)


- thats an LPF, Fo = 946Hz. congratulations, you just added a 2nd pole to your single-pole smps transfer function, thereby making it possible to build an oscillator.

Oh, just looked at the LM3578 data***. error amp? I dont see no stinking error amp.... C11 couples a ramp into the comparator input, which your voltage feedback signal gets compared against. yuck. but the comments re. R1,C13 still apply.

A UC3842, a UC3842, my kingdom for a UC3842 :)


WOW! You have your very own kingdom! 

yep. All I need now is Serfs.



measure the LF ringing - if its about 300Hz, then its an input filter oscillation - short L2 and it will go away. if its about 500Hz then its just the output LC filter oscillating all by itself - short L1 and it will go away. If its about 1kHz, its a closed-loop oscillation caused by C13 (the other one). remove the 330p and again, it will go away.


It looks like it's at about 1KHz.

that makes it a closed-loop oscillation. that chip is kinda bodgy really, no proper error amp.



I see what you are attmpting with C13, just use a smaller cap (and/or split R1) to keep the frequency well above your closed-loop frequency. 33pF would be a good start, 10pF even better.


I swapped the 330pf with a 10pf and again no difference. 

:( worth a try.


I have had some success. I used the pot core with it's new found gap
and wound half the 130T sec with 7/37 Litz wire(maybe 7/36) then wound
the 10T pri with #22 solid magnet wire, then finished off the sec.
With no other changes from the original circuit the eff jumped to
76%(from 68%) and the low freq output noise dropped to around 20mV and
is much closer to sine wave. I donno, but this may be because I
grounded the transformer core. The noise at the switching frequency
dropped to around 60mv. The core temperature increased by 18deg C
after an hour of operation.

60mV is pretty damn good. improved efficiency mostly due to reduced Fr (interleaving + litz). grounding the core certainly helps, as there is capacitive coupling to the core from (to) both windings. you may also find electrostatic screens helpful - a wide foil turn (not shorted), with a wire in the centre to 0V. one between wdg & core, one between windings.


At this point I tried the things above, but no further improvments.
All testing was done at 200v out and a 2.2k load, or about half the
power I wanted to get from it.
I then adjusted the output voltage up to 220v with the same 2.2k load.
The input current was 1.18A, but after just a couple minutes the
current had increased to 1.32A and the longer it ran the faster the
current increased. The core temperature seemed to follow the input
current. Core starting to saturate?

Mike


do a splat test. charge up a big cap. then short it across the inductor and measure the current (I have 10 1/4W 1R resistors I use as a current sense resistor). if Ecap >> Eind, then the cap voltage wont change much, so use V = LdI/dt, knowing V, dI and dt (digital scope very handy here) to calculate L. you should get a straight line (constant L) until the L saturates, at which point the slope will increase quite a bit (such that L becomes the air-cored value).


Cheers
Terry
.

Quantcast