DIYHiFi.org

For the sake of audio
It is currently Thu Nov 23, 2017 4:39 pm

All times are UTC




Post new topic Reply to topic  [ 150 posts ]  Go to page 1, 2, 3, 4, 5 ... 10  Next
Author Message
 Post subject: 1kW Gen2
PostPosted: Sun May 13, 2007 3:49 am 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
Here I go again. In my first baby steps I learned that I was not really happy with the driver IC. But also not with selfoscillating in general.
The disadvantages of self oscilating are generally the possibility of lock ups and no chance to synchronize multiple chanels. This missing synchronization can lead to audiable beating frequencies if multiple chanels are used at the same time.

My new concept is avoiding this. Furtheron it is using standard components. And the synchronization is tranformer coupled in order tp get rid of HF errors in larger systems and may be when synchronizing with SMPS. In such systems GND at one end is never the same like GND at the other end...

The heart of the my new concept is a level shifting comparator which delivers the drive signal for a standard Halfbridge driver, i.e. RI2110.
Furtheron I am planning to split the feedback, which allows to have some db feedback even across the output filter without instability. If I look to some first simulations then I can achieve 5db...10db feedback from behind a 2nd order output filter.
Another clue is the logic supply for the IR2110. It is derived from the comparator output signals. This ensures that the IR signal treshholds are directly linked to the IR input signals. This is giving a very good stability of the dead time vs. temperature and supply voltage changes.

The following schematics are without output filters, without supply for the drivers and without rail caps.
This is keeping things simple and shows the principle.
Up to now still theory, but I am starting with the PCB these days.....

Any comments are welcome. Also hints for better half bridge drivers that can handle at least 150V.


You do not have the required permissions to view the files attached to this post.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Sun May 13, 2007 3:52 am 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
The level shifting comparator has a propagation delay, below my measurement limit (10ns).
The falling/rising edges do slope within 50ns.
The attached picture shows the comparator outputs when operating the system at 1MHz.

Please note I used 1:10 probes. So the scale is 1V/grid.
Time base: 200ns/grid.
This speed and slew rates are already at the limit of my 30MHz bandwith scope. So the real sloping might be faster....


You do not have the required permissions to view the files attached to this post.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Sun May 13, 2007 8:58 pm 
Offline
Account disabled on owner's request

Joined: Wed Dec 29, 2004 8:31 pm
Posts: 899
Looks like a great start for a breadboard, and as you say demonstrates the concept... nice work.

Hints for the drivers, it's important to have as low an output impedance as possible, helps to absorb transients, but to some extent they've already had an effect by the time they appear at the gate terminal. In terms of this issue, that cap from gate-source may cause more problems than it addresses.

I guess you're using the additional cap from gate to source to help slow turn on further, but it must also slow turn off. I'd look at other ways.

At say 150V you wont' want to be relying on the body diodes either, as they'll still be going through charge recombination while the other mosfet turns on, it quickly becomes destructive. You could use 1 diode // to them to take up most of that current flow, or even more robust, yet less efficient, two diodes per mosfet, one in // and the other anti series, which effectively takes the body diodes completely out of conduction.

These IC drivers aren't ideal either, I believe they're high impedance mid swing! There's better switching to be had with complimentary emitter followers to buffer them, for which your delay circuitry could still be placed in front of, then you get the lowest output impedance possible. The transistors used in such a buffer may be disproportionate, in order to provide the asymmetrical gate drive signals you desire, what JohnW once coined as "soft deadtime" and/or you could take care of that in their bias network. In the very least though, consider a local PNP turn off. There's no limite at all to how fast you should turn off a mosfet.

In keeping with that, I don't really think you need the additional resistor in series with the turn off speed up diode.

Another trick worth looking at is the use of a snubber from gate-drain to slow the rate of rise at turn on, but has little effect at slowing turn off. If I remember right such a thing emplys a zener and only effects the beginning rate of rise, and can drop out after. You start to see there's many options and tradeoffs, just have to find the right mix for your application that gives the performance and reliability you're after.

One last tip is that you may want to leave provisions on your PCB to allow for RC snubbers across the mosfets, which would help with ringing and the very high EMI it causes.

What you said about self oscillating amps may be true, but doesn't have to be. They can usually be synched to a clock but if you do, you then lose alot of the advantages of them, at least without further complication.

Similarly, a clock based amp needs to be a little more complex in order to account for the usual delays as well, which is intrinsic to self oscillation.

Still, heterodyning can be beaten through careful design towards low EMI, both radiated and conducted, and both in the design and implementation phases. If all is done right, you can use several amps placed right next to one another and all from the same supply with no audible beating. Thanks to Hypex for proving it.

I hope those tips help, thanks for sharing, and welcome to the dark side :mrgreen:


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Mon May 14, 2007 1:33 pm 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
Hi Ho!
...yes that's definitely more interesting feedback than in DIYaudio...

But oooohps.... I was of the opinion that my gate drive is not so bad.
It is the result of my examninations with my self resonant topology where I used the IRS21954. The cap at the gate has two functions: Slowing down the turn ON in order to limit the di/dt to values that I can practically handle without generating to bad voltage peaks already by the few nH of some mm track routing and T0220 geometry.... Trouble seems to start around 1kA/us. Furtheron this cap does provide a low impedance gate drive at any time, a property that I really like. I also tried zero Ohms for turn OFF and it was OK. Also I tried PNP to speed up the turn OFF, also fine and that really fast turn OFF allowed me to speed up the turn ON again without to much trouble. But stop... there is another limitation that you already mentioned. The body diodes. The max alowed dv/dt is 5.5kV/us during diode recovery.

Up to now I definitely wanted to avoid external diodes, because the double diode approach is completely killing my will to design a simple amp. The variation with parallel diodes is usually regarded as not helpful, because the MOSFET body diode offers lower forward voltage drop than the a suitable external type and the resulting effect would be very small. So far the grey theory, I never did a real life comparison and I guess that also the parasitic inductances of the layout and the diode itself might influence the behaviour. Do you have trustworthy examinations on hand? If yes please send to me. Parallel diodes would be still in the range of acceptable circuit complexity.

Back to the gate drive. I use Rg101 (etc) to adjust the dead time after settling a proper turn ON speed with Rg111 (etc).
...well dead time... I would not really call it dead time. The current commutation times are around 20ns..30ns. The voltage sloping time of the half bridge is around 50ns. The gate drive signals are sloping from LOW to high within about 200ns, (showing a Miller plateau of about 80ns when hard switching, of course no Miller plateau when running idle with inductive filter load, +/-5A). And gate drive sloping from HIGH to LOW takes place within 70ns (almost no Miller plaeteau).
...so this is not like switching OFF the first switch then wait some nice dead time and then switch ON the second switch... no, no... I view at it like a analogue transition event. And I found that you can allow quite some overlap, the circuit behaviour is unexpected forgiving and shows just minor temperature drift. I came to the conclusion that I can adjust an optimized transition on a piece by piece basis. Of course that's no solution for mass production, but for DIY it is perfectly fine. It offers a wide allowed window for adjusting according to your personal taste for the trade off between losses and low distorsions.

I also tried drain-source-snubbers (..hey.. there are not so many caps out there, which are specified for 5kV/us...):
-Caps only: I tried the value range between 220pF...2nF. Result quite poor. The related currents peaks caused shut downs already much earlier than with caps.
- RC: Simply no noticable effect.

JohnW is suspecting HF ringing which I cannot see with my slow scope (...yes I know, I am using quite poor tools compared to my design requirements. But I do not want to use our scopes from work. I am trying to be a positive role model and this does also mean not to misuse company resources for personal advantage. Silly me... )
In order to reduce possible HF ringing I am considering to put a small series resistor to each gate cap, may be in the range between 1 Ohms and 5 Ohms.

Half Bridge Drivers: I am also suspecting that they are not the best choice, but up to now nobody proposed a better one that could handle at least 150V. So I picked these traditional 'workers'.

A big :partyman: to the 'dark side' :wave:
Now everybody GO and :jump: and help the newbie on the block.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Mon May 14, 2007 1:45 pm 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
...hm here is one of my older files for the halfbridge dead time adjustment, unfortunately it does not contain all signal screen shots and also not the zoomed views... Furtheron it does refer to another schematic so forget the component names... It is just to visualize the transitions and show that it is reasonable to adjust to some crossconduction at zero load condition and the behaviour if the inductive acting filter is connected.


You do not have the required permissions to view the files attached to this post.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Mon May 14, 2007 1:48 pm 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
correction:
-Caps only: I tried the value range between 220pF...2nF. Result quite poor. The related currents peaks caused shut downs already much earlier than without caps.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Mon May 14, 2007 2:38 pm 
Offline
Account disabled on owner's request

Joined: Wed Dec 29, 2004 8:31 pm
Posts: 899
Hi,

There's some other forum?? diyaudio?? Oh... aren't they the ones with that one thread dedicated towards class d design related issues? (fools). :Hangman:

Caps only in terms of snubbers is a really bad move and yah, won't work at all. The series resistor is required to damp the ringing.

Without a better scope you'll just be guessing at the values, the snubber really should be designed according to the ringing that's actually there, it is extremely likely there is a lot of ringing going on too, especially on a breadboarded prototype.... I wouldnt' worry about it until you take it to PCB in fact, and attempt an optimized layout. Once you do, you might try something like 10n + 10 ohm as a standard snubber. It may slow the transistion of the switching node some, but very little. If the layout is reasonable it should effectively damp the ringing.

It really is dead time, as it still amounts to a delay in the time domain, and it should be minimal. But yeah, dead time is usually considered some set digitally implemented delay, which is not the case at all as you've seen. I like the term "soft deadtime" as it's somewhat descriptive of what goes on.

To keep it minimal the rising and falling gate signals should intersect as close to ~Vth as possible, dead on Vth would be equal to zero deadtime, but Vth does vary with temperature and output so it's a real balancing act what becomes optimal without greatly complicating the drivers, for which the effectiveness of is also arguable. Once you concentrate on Vth as the point of reference you'll perhaps see it isn't entirely forgiving.

I personally seriously dislike that cap from gate-source. I consider it a bandaide measure. I think if you add any series resistance with it you'll be making any gate step even more prominant, and the added inductance + possible resonance that comes with that cap is already perhaps problematic. There's alot of options to use in place of it, and I think most of them are probably better. Some supposedly think that cap is a good measure against gate step in the first place, Cdv/dt induced turn on, but as mentioned by the time it see's that cap the induced voltage has already taken effect, so it's pretty much useless other than to trick you inton thinking it's helping.

I dont' know much about IC gate drivers, I prefer discrete, so I have no recommendations there at all. You never know what the driver du-jour will be or how long it will be available for, and I find they are always limited in some fashion or other, so you end up having to design the circuit around them, and I hate that kind of restriction.

In terms of the body diode it's worth taking care of because it is unreliable unless you do so, and allows you to further speed up the switching transition, end result it's more robust, more efficient, and better distortion figures.

There is a great deal of misinformation regarding that parallel body diode though, as you mentioned, most just throw it in without having half a clue why, or even that it's probably doing next to nothing for them. As far as I know there's no definitive study, I've seen some interesting papers but, you can consider it marketing. You're right that the usual parasitics limite the effectiveness of it, and the forward drop as well, even with shottkey's, is another limiting factor.

The key is in layout as always, and selecting the right diode for the current required. It's a real bandaide though. Even some mosfets that include the // diode internally are only going to be so effective as well, due to the inductance of the internal bonde wires alone.

However it seems to still be of some use if a proper diode is chosen and tightly implemented. They do not completely take over for the body diode, it will still conduct, but the stored charge is cumulative, and if it conducts less current, there is less stored charge, and so recombination is sped up to some degree even if the // diode only takes some of the load off it.

Depending on the level of insanity (power) you're after, it'll become manditory to take care of it properly if you want a full range amp that has to switch efficiently and with low distortion without blowing apart.

Thinking about it like that, the extra few components it takes becomes trivial considering what they do. It's when you start to push the envelope of higher power that necessitates the extra measures, and the utmost in simplicity is no longer an option!

Best,
Chris


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Tue May 15, 2007 4:10 pm 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
I am thinking about Chris' words that the dead time (no matter if soft or not) is also seen in the time domain. ...this moved my brain again to another direction... towards propagation delay time.
Up to now I was looking to this as a uncritical topic, which just has to be taken into consideration for stability of feedback loop.
But now I am wondering about possible audible effects. The halfbridge drivers are far from perfect here. And I somehow feel that I could design a discrete level shifting comparator, which could serve the gate drive of the lower MOSFET and the floating gate drive of the upper MOSFET as well.
Not just a working one, I hope I can solve it better than my approach above, which is suffering from some 120ns..150ns delay of the IR2110 + 1/2 of the sloping time of the comparator outputs + dead time circuit, ending up in about 300ns delay between input and output of the power bridge .
Give me five weeks... (have to travel).
Nevertheless comments and hints are welcome. I will have internet access most of the time.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Wed May 16, 2007 2:43 am 
Offline
Account disabled on owner's request

Joined: Wed Dec 29, 2004 8:31 pm
Posts: 899
Aha, there you go.

It may not be so directly audible as it is an added source of increasing distortion.

It's also one of the great advantages of any self oscillating design. The better clocked amps take this into account as well but not without added complexity, while the poor ones do not, maybe only trying to minimize it, and then pretending it doesn't matter.

"And I somehow feel that I could design a discrete level shifting comparator, which could serve the gate drive of the lower MOSFET and the floating gate drive of the upper MOSFET as well. "

UCD territory now :) Even doing that though won't actually solve the problem if you keep it clock based, you'll need to correct for the time related errors throughout the loop.

Maybe Fred from D-amps would like to chime in here, he's done it rather simply.

Cheers


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Wed May 16, 2007 12:24 pm 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
??? Why is propagation delay by principle no issue in self resonant topologies?

And how do 'sophisticated' clocked designs get around propagation delay times? Do they have a time machine and know some ns in advance that they have to flip next....? :shock:

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject:
PostPosted: Wed May 16, 2007 2:15 pm 
Offline
Account disabled on owner's request

Joined: Wed Dec 29, 2004 8:31 pm
Posts: 899
The time delays in self oscillating amps are an intrinsic part of the freedback/oscillation loop. In essance, they're taking of for free!

In a clock based amp, you'll typically find them ignored, except for the more respectable ones, which I believe consists of nothing more than local integrating loops, and then summing the total error with the clock signal.

So you also then see what happens if you go and synch a self oscillating amp to a clock signal, advantage lost!


Top
 Profile  
Reply with quote  
 Post subject: Re: 1kW Gen2
PostPosted: Sun Jun 17, 2007 6:21 am 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
... strolled around and got inspiration from Bruno's level shifters and drivers.
For my application I upgraded them a little bit.
1.) I cascoded the high side tail.
2.) ...down, down, down with the impedances!! This greatly improves speed, of course on the expense of increased driver losses. Especially in the cascoding transistor there is some heat.
3.) I integrated a LED in the antisaturation circuit and now can immediately see if the driver is active or not. Blue is nice.

Overall the propagation delay now is going to be dominated by my gate drive dead time adjustment only.
Saving about 200ns vs. the approach with the IR chip. Very nice.


You do not have the required permissions to view the files attached to this post.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject: Re: 1kW Gen2
PostPosted: Sun Jun 17, 2007 12:43 pm 
Offline
Account disabled on owner's request

Joined: Wed Dec 29, 2004 8:31 pm
Posts: 899
Hi,

I posted a good number of circuits back in the day very similar to this. The dual diode clamp technique allows the use of a regular small signal diode, and apparently is a bit more reliable at higher temperatures (that it should never see to begin with). Mainly I used it since it works, and I don't have a supply of schottky's in my junk box. Also it takes a very carefully selected schotty in order to do the job properly, where the small signal diodes already are decent.

I would cascode the lower driver as well to help match delays further.

I would save the money on your blue LED, the cool light really tells you nothing useful as to the performance of it. Besides, you pretty well know almost instantly by the smoke whether or not it's working, at least to the extent of what the light tells you.

I think your lowered values and such is a move in the wrong direction though. It's causing increased dissipation in the cascode transistor, and also further loading the regulator for no good reason. It is very tricky to optimize this properly, but you need to think in terms of how much current is really needed, and where it really needs to be.

Also there's really only one place where you need the low impedance, and there are much more efficient ways to get it.

Regards,
Chris


Top
 Profile  
Reply with quote  
 Post subject: Re: 1kW Gen2
PostPosted: Sun Jun 17, 2007 1:35 pm 
Offline
Cow

Joined: Sun Jun 11, 2006 3:36 am
Posts: 112
Hi Chris,
yes the blue diodes are just show. I could use simply any voltage drop thing.. But I love the blue... And it is good to indicate if the driver is active... This does not mean if switching is fine, it means showing running or protect mode.

I do not think that I need to cascode the lower leg. Cascoding is avoiding that the charge of the parasitic C-B capacitor has to be delivered from the base drive circuit. For the lower leg the C-B voltage is only sloping 2V up and down. Means the required C-B charge does not cause a noticeable delay and also no noticeable reduction of sloping speed.
For the high side leg the C-B voltage is sloping more than 100V. Means there would be a natural asymetry that would require 50 times more charge to drive the high side leg without cascoding. Already with cascoding the high side leg only...- the C-B charge is neglectible for both legs.

Hm, why do you say resitor values are unnecessary low?
I simply went the approach of measuring the real thing. With 330 Ohms between B-E of the BC817 the turn OFF-sloping was quite lame, means in the hundrets of ns. With values between 20 Ohms ...33 Ohms it slopes down within 40ns. Which values do you reach with which resistors+transistors?
Second please consider that I am going to drive quite heavy MosFets. So the required driver output currents are probably by factor 3 higher than what most designs around 200W would need. Just consider the turn ON, it should be dominated by the 15 Ohms. Means BC817 has to deliver 800mA peak. For such fast and high transients you can be happy if you get a current gain 50, resulting in 16mA required base current.

_________________
Best overall performance is derived by the right trade off.


Top
 Profile  
Reply with quote  
 Post subject: Re: 1kW Gen2
PostPosted: Sun Jun 17, 2007 3:14 pm 
Offline
Account disabled on owner's request

Joined: Wed Dec 29, 2004 8:31 pm
Posts: 899
ChocoHolic wrote:
Hi Chris,
yes the blue diodes are just show. I could use simply any voltage drop thing.. But I love the blue... And it is good to indicate if the driver is active... This does not mean if switching is fine, it means showing running or protect mode.

I do not think that I need to cascode the lower leg. Cascoding is avoiding that the charge of the parasitic C-B capacitor has to be delivered from the base drive circuit. For the lower leg the C-B voltage is only sloping 2V up and down. Means the required C-B charge does not cause a noticeable delay and also no noticeable reduction of sloping speed.
For the high side leg the C-B voltage is sloping more than 100V. Means there would be a natural asymetry that would require 50 times more charge to drive the high side leg without cascoding. Already with cascoding the high side leg only...- the C-B charge is neglectible for both legs.

Hm, why do you say resitor values are unnecessary low?
I simply went the approach of measuring the real thing. With 330 Ohms between B-E of the BC817 the turn OFF-sloping was quite lame, means in the hundrets of ns. With values between 20 Ohms ...33 Ohms it slopes down within 40ns. Which values do you reach with which resistors+transistors?
Second please consider that I am going to drive quite heavy MosFets. So the required driver output currents are probably by factor 3 higher than what most designs around 200W would need. Just consider the turn ON, it should be dominated by the 15 Ohms. Means BC817 has to deliver 800mA peak. For such fast and high transients you can be happy if you get a current gain 50, resulting in 16mA required base current.


Hi,

I just figured it's an expensive show at ~3$ a blue diode or whatever. I only commented as you made use of the choice term "upgrade", and while it would be funny to see all the commercial knock off amps using blue LED's this way, I have to take pity on the poor DIY'ers who've only got a small box of junk parts to work with.

You're right, you don't "need" to cascode the lower leg, it doesn't need speeding up, but it might benefit from slowing down to maintain symmetry. The additional delay imposed by cascoding it could help to match the delay of the upper driver's level shifter. It is backwards thinking, but it works.

Your voltages are not so high, extremely efficient mosfets are readily available in this range and even greater. Using a heavier one is a big hit on efficiency, but if you're tailering the circuit to what's in your parts box.... OK. Otherwise, consider that there's a great need for a good mosfet already, not only for efficiency but for THD as well. Using one of that nature will greatly relax the demand on the driver as well, and so you won't need much more than a few mA at the base to easily drive the mosfet with the required "soft" turn on slope very efficiently, therefore your cascode transistor will at most only get warm, and your driver's regulator will also run cooler.

You can also select a driver transistor with a higher transistion frequency, I made very good use of a 2N3906 here for example. I found a high FT to be a key parameter to good performance here.

Then if you're still stuck, you might also make use of the usual techniques required of even driver IC's for decent performance at high speed and high load... buffer it. This simple two transistor version is not suitable by itself for power levels greater than a few hundred watts.

Cheers,
Chris


Top
 Profile  
Reply with quote  
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 150 posts ]  Go to page 1, 2, 3, 4, 5 ... 10  Next

All times are UTC


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Group