http://www.genomerics.org/selfosc/selfosc.html
I got bored but I'm getting better apart from it throwing up other questions so I may get worse.
Anyway, if it floats your boat there may be more to be added.
Cheers
DNA
I got bored but I'm getting better apart from it throwing up other questions so I may get worse.
Anyway, if it floats your boat there may be more to be added.
Cheers
DNA
Great work. I can only encourage you to dig deeper 
What would also be very interesting is a reliable model in the frequency domain, that would enable the design of the right feedback loop (phase compensation etc.)
Have tried to make some models in Mathcad .... but I'm not sure at all if it's correct
Keep it up
What would also be very interesting is a reliable model in the frequency domain, that would enable the design of the right feedback loop (phase compensation etc.)
Have tried to make some models in Mathcad .... but I'm not sure at all if it's correct
Keep it up
IVX said:Wrong link? Check this out.
Yes, sorry, I went to bed early and switched it off. Sort of between 8AM and 10PM GMT would be good.
I suppose I could claim I need my beauty sleep because I am pug ugly but then, since I am so ugly beauty sleep won't help too much.
It's up at the moment.
I'll drink lots of cider and leave it on overnight.
DNA
Baldin said:
What would also be very interesting is a reliable model in the frequency domain, that would enable the design of the right feedback loop (phase compensation etc.)
Keep it up
Just got on to that bit, thought I knew a little. The result is almost alarming, I didn't expect it. Probably done something wrong but it certainly looks cute....
DNA
Self-oscillating class-d amplifiers always have enough loop gain to saturate the output stage such that effective gain forward through the amplifier and back again around the feedback loop self-adjusts to drop to unity (i.e., zero dB) exactly at the frequency where phase shift around the entire loop just reaches 360 degrees (positive feedback, unity gain). The frequency of oscillation is controlled entirely by phase loss - gain plays no roll. In this regard, a self-oscillating class-d amplifier is nothing more than a high level, high power phase-shift oscillator. Setting operating frequency (and limiting its smear when producing large signal audio) is all about designing and controlling the phase shift around the loop.
An earlier diyAudio thread covered most of the pertinent details: Non clocked (e.g. free running, self oscillating) class d
Regards -- analogspiceman
An earlier diyAudio thread covered most of the pertinent details: Non clocked (e.g. free running, self oscillating) class d
Regards -- analogspiceman
Thanks for the link....
I'm sort of moving in a different direction, could be the same though.
Karls last comment about the delay was the same place I started from when I tried to do this last time but I fell over
More stuff added. Have another look.
I might be babbling but it seems to be going somewhere.
Cheers
DNA
I'm sort of moving in a different direction, could be the same though.
Karls last comment about the delay was the same place I started from when I tried to do this last time but I fell over
More stuff added. Have another look.
I might be babbling but it seems to be going somewhere.
Cheers
DNA
Yes..., that's what I thought or expected it would be and on it's own....
I've done something similar with flyback converters and they didn't throw up that resonant peak in linear models of the loop when I put the delay line in to simulate it.
Unless I didn't plot high enough frequencies to see it. Including it did make transient response in the linear model agree with the switching model.
It's all too tempting to think that placing a delay line in a feedback loop would give the result I'm seeing
DNA
I've done something similar with flyback converters and they didn't throw up that resonant peak in linear models of the loop when I put the delay line in to simulate it.
Unless I didn't plot high enough frequencies to see it. Including it did make transient response in the linear model agree with the switching model.
It's all too tempting to think that placing a delay line in a feedback loop would give the result I'm seeing
DNA
AS---- You were right. I hadn't terminated the delay line properly. 
The resistor was there but I stuck it in the ground leg. Now, with the resistor in the right place, it behaves the way you might expect it to. Flat amplitude, frequency dependent phase.
The amplitude response still crosses over below the expected oscillation frequency but the phase margin hits 0 degrees at about the correct frequency.
Must try to be less silly.
DNA
The resistor was there but I stuck it in the ground leg. Now, with the resistor in the right place, it behaves the way you might expect it to. Flat amplitude, frequency dependent phase.
The amplitude response still crosses over below the expected oscillation frequency but the phase margin hits 0 degrees at about the correct frequency.
Must try to be less silly.
DNA
Third page is up.
It's more burble, but the full LC output filter is in there and I've realised how to combine my other ramblings to work out how the oscillation frequency varies with output voltage..... sort of.
Comes with a spreadsheet.....
It's a bit of a wimp out because I've just done it for the single inductor filter bit but, if I can get my head around the complex number stuff, then I'll do one for the LC filter one.
NTL decided to mess about with their network today so I'm not sure if the computer is really living where the DNS servers say it is. So, things might not work.....
Anyway
http://www.genomerics.org/selfosc/selfoscc.html
Cheers
DNA
It's more burble, but the full LC output filter is in there and I've realised how to combine my other ramblings to work out how the oscillation frequency varies with output voltage..... sort of.
Comes with a spreadsheet.....
It's a bit of a wimp out because I've just done it for the single inductor filter bit but, if I can get my head around the complex number stuff, then I'll do one for the LC filter one.
NTL decided to mess about with their network today so I'm not sure if the computer is really living where the DNS servers say it is. So, things might not work.....
Anyway
http://www.genomerics.org/selfosc/selfoscc.html
Cheers
DNA
Hi DNA !
That's cool. ..and sometimes slightly complicated for my old brain, but I am digging through....
Hm, you are concerned about the instable frequency?
In an older thread Bruno gave me a hint for a hysteresys resonator, see this thread
http://www.diyaudio.com/forums/showthread.php?threadid=36349&perpage=10&pagenumber=11
page 11 posting 105 (Bruno) and posting 110 (myself).
From my understanding this hysteresis resonators have less frequency modulation . I am going to try this this principle. In reality I plan to use a LM160 as comparator instead of the 2-transistor-comp in my posting. LM160 is a fast type with just 20ns delay and inverting/noninverting output (WOW, I got it without any difficulties in the component heaven building in Guangzhou! Amazing store, this component shopping building ..stuffed with small stores... And NO! These LM160 are no fakes, at least they really offer high gain, almost no influence of input signal level and really slightly less than 20ns. If they are copies, then it also fine for me. They are simply good.)
Is there any reason why you prefer the self oscilating circuit with feedback from behind the filter rather than a hysteresis oscilator? ...I mean... I am just at the beginning and could still change my plans, if I understand some other simple circuitry has mayor advantages..... I have to learn about Eagle first anyway... must get rid of my oldfashioned handdrawings and human wire router....
That's cool. ..and sometimes slightly complicated for my old brain, but I am digging through....
Hm, you are concerned about the instable frequency?
In an older thread Bruno gave me a hint for a hysteresys resonator, see this thread
http://www.diyaudio.com/forums/showthread.php?threadid=36349&perpage=10&pagenumber=11
page 11 posting 105 (Bruno) and posting 110 (myself).
From my understanding this hysteresis resonators have less frequency modulation . I am going to try this this principle. In reality I plan to use a LM160 as comparator instead of the 2-transistor-comp in my posting. LM160 is a fast type with just 20ns delay and inverting/noninverting output (WOW, I got it without any difficulties in the component heaven building in Guangzhou! Amazing store, this component shopping building ..stuffed with small stores... And NO! These LM160 are no fakes, at least they really offer high gain, almost no influence of input signal level and really slightly less than 20ns. If they are copies, then it also fine for me. They are simply good.)
Is there any reason why you prefer the self oscilating circuit with feedback from behind the filter rather than a hysteresis oscilator? ...I mean... I am just at the beginning and could still change my plans, if I understand some other simple circuitry has mayor advantages..... I have to learn about Eagle first anyway... must get rid of my oldfashioned handdrawings and human wire router....
ChocoHolic said:
Yes, it probably looks complicated because of the way I've gone about doing it. I don't really know what I'm talking about or where it's going. I should tidy it up sometime, but I probably won't. I thought I'd figured it out and then found I hadn't but things seem to be coming together.....
If you mean finding out what the self oscillation frequency is and how it changes then yes that's part of it. If I'm going to play then I need to work it out. I was also worried about the linear side of things going unstable but that doesn't seem to be as important, if it does happen.
I had started out by thinking about hysteresis control and how the method is kind of equivalent, or not. It does introduce a 'delay' function but that might end up being more complex than the 'time delay' that UCD relies on.
You might try thinking about the case with a single inductor as your filter and do some sums to see what happens just because it's simpler to analyse. If you're cool about calculating ripple currents and turning them ito voltages across the load and then working that back to the input to your comparator then it should give you an idea about what is happening.
Or it might work the other way around. Say you've got 100mV of hysteresis at your comparator input then you transform that to the output of your amplifier according to your feedback network... if it's 40:1 then you are looking for a change of 4V at the output.
A change of 4V across 4Ohms needs a change in inductor current of 1A.
At that point you are sticking one of the supply rails, VS, across the inductor with the other end of it at VOUT so you can work out how long its going to take before it flips over. dI=VL.T/L and other such stuff.
In this case it's sort of just simple algebra but it looks like when you try and extend it to a 'proper' output filter things get much more complicated. Then you're looking at how the filter itself affects the feedback/output voltage.
In the UCD case I'm beginning to think it's much simpler because a pure time delay creates an error base...... Oh, head explodes!!!
Anyway, that's the basic idea if you want to have a go at analysing things. Just don't believe I'm right about the method but give it a go and then see what your next problem might be.
I shouldn't be starting a web page here.
Since I haven't had a proper look at the hysteretic method I sort of don't know but it looks like it's going to be more complex than the simple UCD time delay. Fast comparators are nice and but it almost feels that once you've added the other delays into the circuit, level shifters and drivers, then, even if your goal was to produce an hysteretic amplifier, you've probably got something that will behave as a hybrid.....
It's not a preference. I'm just looking at things and trying to understand them. I have spent (perhaps wasted) a lot of time trying to get clocked amplifiers to behave themselves under all conditions. You can see some of the pain involved on the Titanchen section.
A lot of that will be down to poor choices and inexperience but, comparatively speaking, this UCD method is just too sexy for it's shirt. That's why I'm trying to analyse it. The more you figure out the more you might be able to make choices about the design.
I'm not going to say hysteretic control is a non-starter because I don't know. However I've just got a gut feeling that you might build one and then get really annoyed with it and, in a fit of disgust, rip the hysteresis out and then it will become what it wanted to be in the first place..... UCD.
Even using software you will be a Human Wire Router
I'll shut up now.
Cheers
DNA
yes off course my brain will nor be allowed to go on vacation when using a layout software... I will probably still have to self route and restrict many critical parts of the PCB.Hm...
For you UCD is easier to understand than the hysteresys self oscillation? Actually I prefer to try the hysteresys rather than the UCD, because I do not fully understand the UCD...
Please note, I would pick the feedback (pos & neg) from the halfbridge, not from behind the filter. So the filter behaviour, filter phase shifts and di/dt + dv/dt from the load circuit do not influence the flipping.
...grey theory so far, curious about reality, but I still feel that this easier to understand and debug rather than a UCD, which causes quite some struggles in my head. For UCD I already start to struggle if I want calculate the no signal steady state resonating frequency...
'The more you figure out the more you might be able to make choices about the design.'
Yes ! 100% agreement. ..and I am starting to become more and curious about the UCD. I think I have to dig for a detailed description, which does explain the way of UCD operation in a chocobrain-friendly way. Promising links welcome.
Go on with your way of work and documentation, I like it !!!!
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.