Class D amps with ext clock over 1MHz?

[Genome]

| In article <3SPgd.399$9t2.354@newsfe5-win.ntli.net>,
| [...]
| >IGBTs are a different class of beasty.
|
| They are not really that different. They look a lot like a N-MOSFET
| driving a PNP.
| --
| --
| kensmith@rahul.net forging knowledge

Ken Smith compares the basic mosfet to an IGBT and realises that they are
'not really that different'.

A mosfet looks a lot like a mosfet.

An IGBT looks a lot like a mosfet driving a bipolar transistor.


Ken Smiths Wife

'Why did you shag me up the arse last night?'

Ken Smith

'That's why I had shit on my knob this morning!'

Don't worry, it's subtle.

DNA

 

[legg]

| Pooh Bear wrote...
| >
| > Joerg wrote:
| >
| >> Is there a fairly comprehensive list on the web about class D audio
| >> amps that run above 1MHz, synchronized or externally clocked?
| >
| > None that I know off offhand. Tripath's 'spread spectrum' design
| > maxes out at about 600kHz IIRC.
|
| One serious issue that's not often talked about is an asymmetric
| power MOSFET turn-on and turn-off time. Moreover, FET turnoff
| time has a slow recovery tail, and is memory dependent for short
| time intervals. The distortions from this issue are exacerbated
| at high PWM frequencies, and lead to a degraded performance.
|
|
| --
| Thanks,
| - Win
|
| (email: use hill_at_rowland-dotties-org for now)

I'd like to know the basis for that one?

I think it's more the evident practical effects of load and drive
methods on switching speed, than the characteristic of the fet alone
that he may be refering to. The digital source can't tell whether a
switching event is assisted or degraded by load conditions.

On the other hand, he could be talking about IGBTs and not switching
over to his 'write' brain for correct keyboard control. IGBT tails are
effectively stored, reappearing at turn-off in subsequent,
externally-forced, 'ZVS' events. IGBTs are not popular in typical low
voltage 'feedback-free' digital-to-driver coil applications, to my
knowledge.

RL

 

[Pooh Bear]

|
|
| Winfield Hill wrote:
|
| > Pooh Bear wrote...
| > >
| > > Joerg wrote:
| > >
| > >> Is there a fairly comprehensive list on the web about class D audio
| > >> amps that run above 1MHz, synchronized or externally clocked?
| > >
| > > None that I know off offhand. Tripath's 'spread spectrum' design
| > > maxes out at about 600kHz IIRC.
| >
| > One serious issue that's not often talked about is an asymmetric
| > power MOSFET turn-on and turn-off time. Moreover, FET turnoff
| > time has a slow recovery tail, and is memory dependent for short
| > time intervals. The distortions from this issue are exacerbated
| > at high PWM frequencies, and lead to a degraded performance.
|
| Absolutely. Just noticed the same today looking at IGBTs.
|
| Tripath get 'better than most' results by taking feedback from the output
| to counter this. They have a programmable dead-time too. Needless to say -
| the shortest dead-time gives the highest performance with the greatest
| risk of cross-conduction.
|
| Driving gates at high frequencies takes some current too !
|
|
| Graham
|

IGBTs are a different class of beasty.

Oh indeed. The effect ( switching time ) is more pronounced. Just commenting,
since I was looking at some 'Fairchild' parts yesterday ( for an SMPS ).

Graham

 

[Ken Smith]

| In article <3SPgd.399$9t2.354@newsfe5-win.ntli.net>,
| [...]
| >IGBTs are a different class of beasty.
|
| They are not really that different. They look a lot like a N-MOSFET
| driving a PNP.
| --
| --
| kensmith@rahul.net forging knowledge

Ken Smith compares the basic mosfet to an IGBT and realises that they are
'not really that different'.

A mosfet looks a lot like a mosfet.

An IGBT looks a lot like a mosfet driving a bipolar transistor.

What are you on about?

I think what I said is perfectly clear.

 

[Joerg]

Hi Chris,

What is it that we don't like about fully analog, self oscillating
designs? There's a number of them out there, some of which are of the
highest quality out of any class of amp, considering the power levels
that is. Non switch near 1Mhz either.

Actually the MAX9712 does switch there. Even higher since it is spec'd
up to 2MHz. It can be synchronized so it clocks exactly where you want
it to. So far I haven't found anything similar from the large companies
though.

Regards, Joerg

 

[Terry Given]

yeah but...
temp is slow changing compared to the 600kHz etc switching speed, so
thats not a problem
load current ditto, though its faster moving than theta.

NT

its trivial to build fairly accurate thermal models in software, and in
general you ought to know what the system thermal model is anyway
(temperature is important for reliability). Thermal sensors can also be
used, along with sufficient experimentation to ascertain the correlation
between the measured temperature and the temperature(s) of interest.

You can also design the control loop to improve its rejection of
dead-time related effects - google internal model control. Better yet,
do all of the above. Non-linear modulator gain can take care of
distortion due to storage time etc.

"all-digital" is kind of a meaningless thing - virtually every digital
controller has some form of analogue input. It might be more meaningful
to talk about the requirement for extra circuitry. The feed-forward
approaches probably dont need any extra hardware to implement, whereas
feedback methods perhaps do. Most of this stuff is a lot easier to do in
software, if you can live with the few tens of microseconds computation
time (modern DSPs kick ass)

Cheers
TErry

 

[Pooh Bear]

yeah but...
temp is slow changing compared to the 600kHz etc switching speed, so
thats not a problem
load current ditto, though its faster moving than theta.

Compare die temp at switch-on vs long term idle vs full load vs 'average' load. They're all different. So no
single dead time adjustment fixes the issue.

Load current can change radically. Again - compare no load to full load.

Graham

 

[Pooh Bear]

its trivial to build fairly accurate thermal models in software, and in
general you ought to know what the system thermal model is anyway
(temperature is important for reliability). Thermal sensors can also be
used, along with sufficient experimentation to ascertain the correlation
between the measured temperature and the temperature(s) of interest.

It's rather less than trivial to model the die temp accurately on a dynamic basis..

You can also design the control loop to improve its rejection of
dead-time related effects - google internal model control. Better yet,
do all of the above. Non-linear modulator gain can take care of
distortion due to storage time etc.

If you monitored the switching speed in 'real time' then it would be possible to make a good adjustment for it.
Maybe this is what Tripath do ?

"all-digital" is kind of a meaningless thing - virtually every digital
controller has some form of analogue input. It might be more meaningful
to talk about the requirement for extra circuitry. The feed-forward
approaches probably dont need any extra hardware to implement, whereas
feedback methods perhaps do. Most of this stuff is a lot easier to do in
software, if you can live with the few tens of microseconds computation
time (modern DSPs kick ass)

Don't forget we're talking about nanoseconds of switching speed. From memory I think that Tripath offer a
selection of 25 / 50 / 100 ns dead time.

Oh - and the lower the load impedance - the longer the required dead time too. Note that audio amps don't drive
resistive loads - so that impedance isn't fixed, nor purely real ( just to make things more interesting ).

Don't forget to factor in the power supply droop and ripple to the PWM equation while you're at it.


Graham

 

[Terry Given]

Terry Given wrote:




It's rather less than trivial to model the die temp accurately on a dynamic basis..

not really, all you need are the relevant dimensions and materials. And
there are sneaky ways of measuring your models too - I have seen some
scarily accurate dynamic thermal models of IGBT and mosfet assemblies
(Semikrons mathcad worksheets are an edifying read, but they wont give
them to you

If you monitored the switching speed in 'real time' then it would be possible to make a good adjustment for it.
Maybe this is what Tripath do ?

pass. I have seen some gate drive circuits that force dead-time to a
(measured) minimum, but they cost extra, and gate-drive circuits
typically are built 4(6) at a time, so are cost sensitive.

Don't forget we're talking about nanoseconds of switching speed. From memory I think that Tripath offer a
selection of 25 / 50 / 100 ns dead time.

Oh - and the lower the load impedance - the longer the required dead time too. Note that audio amps don't drive
resistive loads - so that impedance isn't fixed, nor purely real ( just to make things more interesting ).

Don't forget to factor in the power supply droop and ripple to the PWM equation while you're at it.

why not measure actual power supply voltage, and use that to calculate
the required duty cycle. but dont forget switch forward voltage drop
(which is line, load and sock-colour dependant)

Graham

Cheers
TErry

 

[Pooh Bear]

not really, all you need are the relevant dimensions and materials. And
there are sneaky ways of measuring your models too - I have seen some
scarily accurate dynamic thermal models of IGBT and mosfet assemblies
(Semikrons mathcad worksheets are an edifying read, but they wont give
them to you

LOL - I bet !

I guess we simply disagree on the meaning of 'trivial' in this instance.

pass. I have seen some gate drive circuits that force dead-time to a
(measured) minimum, but they cost extra, and gate-drive circuits
typically are built 4(6) at a time, so are cost sensitive.

Gate drive is - of course yet another factor in the overall performance. You can't charge ( or discharge ) the gate
capacitance instantly - and in the meantime the mosfet is running in 'linear operation'.

why not measure actual power supply voltage, and use that to calculate
the required duty cycle. but dont forget switch forward voltage drop
(which is line, load and sock-colour dependant)

On a cycle by cycle basis - that would be most accurate. But it'll be a wicked DSP that can do all these calcs in
say 2-5us.


Graham

 

[Terry Given]

Terry Given wrote:




LOL - I bet !

I guess we simply disagree on the meaning of 'trivial' in this instance.

lol! Once you sort out how to do it, transient thermal impedance-type
calculations arent really that complex, and of course using spice makes
the modelling very simple - once you have meaningful (ah, thats the
tricky bit) values of the relevant thermal resistances and capacitances.

many FET models have a temperature input, so you can measure the power
thru the FET (yay 4 spice function blocks), run it through an
electro-thermal model (ie r's and c's) of the heatsink and feed back the
actual junction temperature. It is fun to show positive thermal feedback
(ie thermal runaway) in mosfets - something I have seen in practice, and
had some great arguments over (amazing how many people dont read the
RTdson-vs-Tj graph

Gate drive is - of course yet another factor in the overall performance. You can't charge ( or discharge ) the gate
capacitance instantly - and in the meantime the mosfet is running in 'linear operation'.

yep. I'm analysing a self-oscillating flyback design at the moment, and
I suspect their rotten efficiency (78% for 25W?!) is due to piss-poor
gatedrive - ie exactly what you are talking about.

On a cycle by cycle basis - that would be most accurate. But it'll be a wicked DSP that can do all these calcs in
say 2-5us.

technically you only need to go as fast as the supply ripple - and if
you know something of the supply ripple characteristics, you dont even
have to go that fast.

I am looking at a 150MHz 30MIPS dsp for a job at the moment. 30MIPS =
150 instructions in 5us - thats more than enough for quite a complex
controller, if you're sneaky about it. My PC programmer mates laugh at
the puny throughput I get from my DSPs compared to their pentiums -
several instructions per ns!

Graham

Cheers
Terry

 

[classd101]

Maybe on newer versions. It topped out at about 600-800 kHz a few yrs back.


It takes feedback pre the output filter IIRC.



Which were you thinking of ?

Agreed that none switch at 1 MHz or near that.


Graham

Hi Graham,

I think I agree that Triphath's feedback is pre filter, seems to be
some confusion online about that.

Even 800 KHz is high, but yeah I can't recall what version it may have
been I was reading about when I saw 1 MHz so.. its really not a
"feature" anyway.

Which was I thinking of?

Well there's a good handful of them out there that I've seen, even TI
I believe has a dual channel 500W reference design which is self
oscillating.

There are a few which are certainly noteworthy, but I won't mention
because they aren't available for anything other than mass production,
and have little to no literature available for them.

So with that I'll mention:
Bang & Olufsen's "IcePower" modules, takes feedback pre & post filter.

LC Audio's ZapPulse modules: feedback pre filter

UcD modules: feedback post filter, fully discrete, very simple
circuit. Widely regarded as the top dog, ultimate available today, for
too many reasons to bother mentioning, basically if you haven't yet
heard of it, you're behind the times.

There's also another which is self oscillating and uses a single pulse
error correction scheme but since there is no litterature on it, and
only available in large quantities, I won't bother to dig the name up,
but it is out there

I'll finish up my little wish list with the Mueta, some interesting
reading material is available on it, though it remains mythical at
this point, the technology certainly sounds promising. Personally I
don't think it will dethrone the UcD.

I don't mention these to discourage anyone from trying something new,
but if you want to see what is cutting the edge these day's, there you
have it, might save ya re-inventing the wheel.

Also these modules (UcD specifically) are said to have better sound
than pure digital systems in direct comparison, sorry.

Regards,
Chris

 

[classd101]

| Pooh Bear wrote...
| >
| > Joerg wrote:
| >
| >> Is there a fairly comprehensive list on the web about class D audio
| >> amps that run above 1MHz, synchronized or externally clocked?
| >
| > None that I know off offhand. Tripath's 'spread spectrum' design
| > maxes out at about 600kHz IIRC.
|
| One serious issue that's not often talked about is an asymmetric
| power MOSFET turn-on and turn-off time. Moreover, FET turnoff
| time has a slow recovery tail, and is memory dependent for short
| time intervals. The distortions from this issue are exacerbated
| at high PWM frequencies, and lead to a degraded performance.
|
|
| --
| Thanks,
| - Win
|
| (email: use hill_at_rowland-dotties-org for now)

I'd like to know the basis for that one?

DNA

Hi,

Proper fet selection for a high speed class D purpose still seems like
a black art to many, help? I know of the gate charge ratio trick to
help reduce gate step turn induced turn on, what else is there?

Also, more and more are incorporating adaptive delay these days, it
"looks" simple, is it?

Thanks,
Chris

 

[Pooh Bear]

Hi Graham,

I think I agree that Triphath's feedback is pre filter, seems to be
some confusion online about that.

Even 800 KHz is high, but yeah I can't recall what version it may have
been I was reading about when I saw 1 MHz so.. its really not a
"feature" anyway.

Which was I thinking of?

Well there's a good handful of them out there that I've seen, even TI
I believe has a dual channel 500W reference design which is self
oscillating.

There are a few which are certainly noteworthy, but I won't mention
because they aren't available for anything other than mass production,
and have little to no literature available for them.

So with that I'll mention:
Bang & Olufsen's "IcePower" modules, takes feedback pre & post filter.

LC Audio's ZapPulse modules: feedback pre filter

UcD modules: feedback post filter, fully discrete, very simple
circuit. Widely regarded as the top dog, ultimate available today, for
too many reasons to bother mentioning, basically if you haven't yet
heard of it, you're behind the times.

There's also another which is self oscillating and uses a single pulse
error correction scheme but since there is no litterature on it, and
only available in large quantities, I won't bother to dig the name up,
but it is out there

I'll finish up my little wish list with the Mueta, some interesting
reading material is available on it, though it remains mythical at
this point, the technology certainly sounds promising. Personally I
don't think it will dethrone the UcD.

I don't mention these to discourage anyone from trying something new,
but if you want to see what is cutting the edge these day's, there you
have it, might save ya re-inventing the wheel.

Also these modules (UcD specifically) are said to have better sound
than pure digital systems in direct comparison, sorry.

Thanks for the input. I haven't heard of some of these - maybe being in the Uk may be the reason ?

You, in turn, might care to look at the Audio amplifier products being offered by Powersoft. www.powersoft.it

They are highly regarded in the pro-audio sound reinforcement market.

I met their guys at PLASA ( UK pro-audio trade show ) last month. Very impressive figures generally and a nice
new OEM module. 1 kg provides 1kW of audio - that's everything including the PSU ! Connectors at one end take
AC line - at the other end, audio line in and spkr out. Switching is - surprisingly - < 200kHz but doesn't seem
to impair performance. DSP based modulator btw.


Graham