I think we are all aware that the triangle cannot a perfect triangle.
On the other hand most likely also everybody will agree that the deviation with the chosen values (time constant multiple orders of magnitude longer than period of switching frequency) will not be a show stopper.
Let's try to exclude one or the other locking issue.
Measuring the phase reversal in locked situation will most likely not work.
In case phase reversal happens and causes the amp to stop, then this will automatically have the consequence to activate the locking mechanism described by Pafi.
The simplest method to exclude phase reversal by design is to put the two antiparallel diodes from the inverting input of U2 to GND. If this does not change the behavior you should still leave the diodes in the circuit in order to exclude this locking mechanism in any case.
Overcoming the locking mechanism, which has the integrator at one or the other rail and both inputs of the 2110 LOW should be possible by two means.
1) Putting a resistor in parallel to C6. Chose a time constant of 20ms or longer to ensure that it has minor impact on loop gain over the entire audio band.
2) Ensure AC coupled input.
Both together should ensure an automatic unlocking behavior as follows.
In case of being locked both MosFets are OFF.
Feedback will feed more or less nothing to the integrator.
AC coupled input will feed no DC to the integrator.
The resistor parallel to C6 will discharge the cap and as soon as the output voltage of the integrator reaches levels smaller than the triangle peaks - the comparator and the phase splitters will start up again.
On the other hand most likely also everybody will agree that the deviation with the chosen values (time constant multiple orders of magnitude longer than period of switching frequency) will not be a show stopper.
Let's try to exclude one or the other locking issue.
Measuring the phase reversal in locked situation will most likely not work.
In case phase reversal happens and causes the amp to stop, then this will automatically have the consequence to activate the locking mechanism described by Pafi.
The simplest method to exclude phase reversal by design is to put the two antiparallel diodes from the inverting input of U2 to GND. If this does not change the behavior you should still leave the diodes in the circuit in order to exclude this locking mechanism in any case.
Overcoming the locking mechanism, which has the integrator at one or the other rail and both inputs of the 2110 LOW should be possible by two means.
1) Putting a resistor in parallel to C6. Chose a time constant of 20ms or longer to ensure that it has minor impact on loop gain over the entire audio band.
2) Ensure AC coupled input.
Both together should ensure an automatic unlocking behavior as follows.
In case of being locked both MosFets are OFF.
Feedback will feed more or less nothing to the integrator.
AC coupled input will feed no DC to the integrator.
The resistor parallel to C6 will discharge the cap and as soon as the output voltage of the integrator reaches levels smaller than the triangle peaks - the comparator and the phase splitters will start up again.
This was my Idea.
Pafi: You can still post this forum, but take a few second to think less offensive words.I personally don´t care them, but other people may be more sensitive. It isn´t a good idea to go out by the back door. Stay here, my friend.
RGRDS.
okay - no more about the triangle 
Here's a bunch of stuff to consider:
1) I just don't see any spec for HIN and LIN on the IRS2110 data sheet. They show a "functional Block Diagram" that shows a resistor to ground - but that's it - maybe I missed it?. In fact, they show it going to some ground, but they also show a different ground on same picture for the com pin. Go figure. Probably a separate analog and digital ground, but it's not obvious nor stated any where. Don't count on that "resistor".
Since the data sheet shows maximum values of 0.3V over VCC and under VSS, you can assume they use Schottky diodes to protect the inputs. Again, there is nothing in your circuit to keep the XOR gates from driving unlimited current into IRS2110 - even it's just on the edges since you are AC coupled there. I wouldn't trust this circuit at all. Driving current from the XOR +6V supply into the IRS -12V rail may cause unexpected results.
2) BootStrap : C9 is 1uF? That might not be big enough. You want to have plenty of reserve to drive the upper mosfet. I think bigger is better here.
You hinted that the bootstrap supply is +12V, I trust you - it's just not drawn.
3) Consider external, fast recovery diodes across the output mosfets. Not sure if this is biting your design or not - just that the intrinsic diodes are usually slow.
4) I still object to feeding back full carrier levels - but understand how it should work, just the same. Having said that - U2 is going to behave differently depending upon the source impedance of the audio (as others have pointed out). You'd be much better off adding a buffer amp at the input port. That will make the time-constant of U2 the same at all times, (R9 || R10) * C6.
I noticed that the phase response of this, U2, depends heavily upon the source impedance of audio. If no input, i.e. open circuit, the time constant is R10*C6 ; pretty long. But check out the gain and phase plots. You've got roughly 90 degrees phase shift throughout the audio spectrum. I suspect you'd prefer 180 degrees.
If the input is low impedance, the phase response is changing from around 170 at DC to 90 at 20 kHz. Somewhat better.
If you use an input buffer, *and* add some resistance in parallel to C6, you move the pole frequency out a whole lot (wider bandwidth). I tried it with 47k, arbitrarily, and the phase shift is nearly 180 degrees throughout the whole audio band. Pretty much the same as Choco and you have already said, but with a slightly different explanation
This arrangement looks more predictable and repeatable.
Here's a bunch of stuff to consider:
1) I just don't see any spec for HIN and LIN on the IRS2110 data sheet. They show a "functional Block Diagram" that shows a resistor to ground - but that's it - maybe I missed it?. In fact, they show it going to some ground, but they also show a different ground on same picture for the com pin. Go figure. Probably a separate analog and digital ground, but it's not obvious nor stated any where. Don't count on that "resistor".
Since the data sheet shows maximum values of 0.3V over VCC and under VSS, you can assume they use Schottky diodes to protect the inputs. Again, there is nothing in your circuit to keep the XOR gates from driving unlimited current into IRS2110 - even it's just on the edges since you are AC coupled there. I wouldn't trust this circuit at all. Driving current from the XOR +6V supply into the IRS -12V rail may cause unexpected results.
2) BootStrap : C9 is 1uF? That might not be big enough. You want to have plenty of reserve to drive the upper mosfet. I think bigger is better here.
You hinted that the bootstrap supply is +12V, I trust you - it's just not drawn.
3) Consider external, fast recovery diodes across the output mosfets. Not sure if this is biting your design or not - just that the intrinsic diodes are usually slow.
4) I still object to feeding back full carrier levels - but understand how it should work, just the same. Having said that - U2 is going to behave differently depending upon the source impedance of the audio (as others have pointed out). You'd be much better off adding a buffer amp at the input port. That will make the time-constant of U2 the same at all times, (R9 || R10) * C6.
I noticed that the phase response of this, U2, depends heavily upon the source impedance of audio. If no input, i.e. open circuit, the time constant is R10*C6 ; pretty long. But check out the gain and phase plots. You've got roughly 90 degrees phase shift throughout the audio spectrum. I suspect you'd prefer 180 degrees.
If the input is low impedance, the phase response is changing from around 170 at DC to 90 at 20 kHz. Somewhat better.
If you use an input buffer, *and* add some resistance in parallel to C6, you move the pole frequency out a whole lot (wider bandwidth). I tried it with 47k, arbitrarily, and the phase shift is nearly 180 degrees throughout the whole audio band. Pretty much the same as Choco and you have already said, but with a slightly different explanation
This arrangement looks more predictable and repeatable.
Attachments
Last edited:
The inputs of the IR2110 are CMOS and they are therefore swithcing at half of the supply voltage. In order to make them work better with a wide range of sources you may use a lower supply voltage for the input section than for the driver section in order to adapt to the source.
There is also the possibility to have small voltage differences between the ground of the input section and the grounf of the driver section. As far as I understand this is not meant to be used as a continuous situation but to ease EMC issues a bit.
Regards
Charles
There is also the possibility to have small voltage differences between the ground of the input section and the grounf of the driver section. As far as I understand this is not meant to be used as a continuous situation but to ease EMC issues a bit.
Regards
Charles
okay - no more about the triangle
Here's a bunch of stuff to consider:
1) I just don't see any spec for HIN and LIN on the IRS2110 data sheet. They show a "functional Block Diagram" that shows a resistor to ground - but that's it - maybe I missed it?. In fact, they show it going to some ground, but they also show a different ground on same picture for the com pin. Go figure. Probably a separate analog and digital ground, but it's not obvious nor stated any where. Don't count on that "resistor".
I could measure an internal 100K resistor, and diodes to + and to - as the majority of CMOS devices.
Since the data sheet shows maximum values of 0.3V over VCC and under VSS, you can assume they use Schottky diodes to protect the inputs. Again, there is nothing in your circuit to keep the XOR gates from driving unlimited current into IRS2110 - even it's just on the edges since you are AC coupled there. I wouldn't trust this circuit at all. Driving current from the XOR +6V supply into the IRS -12V rail may cause unexpected results.
CMOS usually has a couple of CMOS transistors with a RDSon of about 125R, so, at 12V is about of 100mA maximum. But, I will measure DC levels under working conditions.
2) BootStrap : C9 is 1uF? That might not be big enough. You want to have plenty of reserve to drive the upper mosfet. I think bigger is better here.
Apparently, it is sufficient. In fact, IR says .22, and lower as frequenciy is increased, which appears to natural as Xc goes down.
You hinted that the bootstrap supply is +12V, I trust you - it's just not drawn.
Yes, +12 with respect to the negative power rail, not to chassis ground, and isolated from it.
3) Consider external, fast recovery diodes across the output mosfets. Not sure if this is biting your design or not - just that the intrinsic diodes are usually slow.
OK, good idea. I didn´t use it considering internal MOS diodes, but I´ll try your suggestion.
4) I still object to feeding back full carrier levels - but understand how it should work, just the same. Having said that - U2 is going to behave differently depending upon the source impedance of the audio (as others have pointed out). You'd be much better off adding a buffer amp at the input port. That will make the time-constant of U2 the same at all times, (R9 || R10) * C6.
I noticed that the phase response of this, U2, depends heavily upon the source impedance of audio. If no input, i.e. open circuit, the time constant is R10*C6 ; pretty long. But check out the gain and phase plots. You've got roughly 90 degrees phase shift throughout the audio spectrum. I suspect you'd prefer 180 degrees.
If the input is low impedance, the phase response is changing from around 170 at DC to 90 at 20 kHz. Somewhat better.
If you use an input buffer, *and* add some resistance in parallel to C6, you move the pole frequency out a whole lot (wider bandwidth). I tried it with 47k, arbitrarily, and the phase shift is nearly 180 degrees throughout the whole audio band. Pretty much the same as Choco and you have already said, but with a slightly different explanation
Thank you for the time to the simulations. But in fact, full voltage appears in 3 15K resistors in place of 1 47K, and I use the circuit as the author made the article I posted. In fact, normally -I insist- the circuit works fine. It only goes blocked SOMETIMES it is ran without load. Inserting a 1K as output, it continues working fine and doesn´t block.
This arrangement looks more predictable and repeatable.
Again, thanks for the posts, simulations. and your time.
Osvaldo.
What about the startup of your circuit? You need to get that bootstrap voltage up before the output can switch.
With a loaded output, there's a DC path to ground - maybe enough to get your circuit started. Without it, the output is floating.
Maybe you need to guarantee at startup, that the lower mosfet is on first, so that the bootstrap voltage charges up.
Do you have any power sequencing problems also? The power amp voltage should be full, before you start the comparator running. Anyway you can guarantee the startup condition? Oh - wait, the SD pin! Can you add circuitry that keeps the driver in shutdown, until the supplies come up?
With a loaded output, there's a DC path to ground - maybe enough to get your circuit started. Without it, the output is floating.
Maybe you need to guarantee at startup, that the lower mosfet is on first, so that the bootstrap voltage charges up.
Do you have any power sequencing problems also? The power amp voltage should be full, before you start the comparator running. Anyway you can guarantee the startup condition? Oh - wait, the SD pin! Can you add circuitry that keeps the driver in shutdown, until the supplies come up?
Yes, but that's not what I was getting at.
That bootstrap has to get energized somehow, before this thing will work correctly, yes? That means, either it must startup with the lower mosfet on, or you could also ensure that the driver makes a couple of on-off transitions before the feedback is engaged.
I was thinking that a DC load forces the the mosfet junction ot be zero at startup - before the mosfets are switching. That allows the integrator to work, initially, maybe enough that the driver does switch the mosfets enough to charge up the bootstrap.
That resistor across c6 should also help here too. It'll keep the integrator output initially at zero volts, instead of undefined state. Again, the circuit may startup more reliably.
I believe that may be two circumstances, first turn on upper fet, and first tun on down fet.:
1) the #1, the cap is empty, and will not turn on, then it goes to the #2
2) the lower fet will turn on because it´s driver is always powered from supply. Then, the BS cap will be charged via BS diode, and lower fet, the at the following upper turn on this fet is well drive because BS cap is full, ar at least, say, 2/3 charged (.66 *15V is 10V), more than needed.
I agree with you at this point.
Sorry, I couldn´t understand how the hell do paragraph quoting.
Fixed quote issue
Maybe I am missing something here - when you say "locked", the C6 is stuck with some voltage across it. What else is going on? Is the the triangle still making triangles? Is the IR2110 still switching?
I'm under the impression that the IR2110 has stopped working when locked up. Is this not right?
I'm under the impression that the IR2110 has stopped working when locked up. Is this not right?
The trouble is that the charge accumulated in C6 gets a value near the supply rail, and how the triangle is about 2Vpp, the comparator gets hooked, and the IR2110 simply "sees" nothing in its inputs. If I manually discharge the caps shortcircuiting it with a screwdriver, the the amp continues working pretty fine till a new load removal. In fact, not always I remove the load, and with no DC path to gnd, the amp is blocked. This fact is what confuses me.
Thank to "cop" for the re-work in the post
Oooooh - that's a horse of a different color! Yes -the comparator stops, either stuck high or low because the math makes it so. No comparator change, then no XOR change, then - because AC coupling, HIN and LIN both settle to zero - and both mosfet's open up. No output, no feedback, no U2 action - nothing. Yep.
The problem lies with the triangle, either HCF4047 or TLC274 - or something is loading down the TLC274.
1) Check HCF4047 frequency when locked up. Still 500 kHz?
2) Latchup condition on TS3702 comparator. Remember how I keep harping on that XOR turning on the protection diode in IR2110? Just suppose, that the +/- 6 volt rail droops low, even for an instant - simultaneously triangle output drives U3 input above the rail. That can cause latchup. The triangle opamp doesn't have enough strength to damage anything - but the U3 looks like a dead load.
3) Similar to (2), check the +/- 6 Volt power rails when the problem occurs. Still 6? Possibly U2 is trying to power up all the 6 volt logic.
Yeah - I like that cop icon
Yes, I designed it to be used with no more than 1V rms input, or 1.414Vpp, so it will no overmodulate the PWM signal.
´till the opamp input, all the rest to the RC oscillator, all this continues working pretty fine. The trouble is form opamp to output. No frequency nor amplitude varies when working fine, or blocked by C6.
Yeah - I like that cop icon
Your thesis looks pretty, I will check it.
Thankyou.
cool. You might consider using a buffer at the input - very high input impedance, and it makes the feedback network a constant impedance.
Yes, I did.
Have you a opinion about CMOS opamp (like TS274, TMS274, TLC084, the older CA3130, etc.) as audio amplifiers? I´m not musician, but Electronic Engineer.
No - I couldn't say for sure. I use NE5532's a lot. If you scroll around the forum you will find a lot of opinions on this very topic. At my job, we use OPA2227's and OPA2132'2 a bunch - both nice little opamps but we don't use 'em for audio. OPA2132 is fet input, but bipolar elsewhere. I think the OPA2227 is all bipolar.
I guess it all depends on what you want. Seems like you want all cmos - any particular reason? Anyway, if you pick parts with same footprint, then you can test them all.
I guess it all depends on what you want. Seems like you want all cmos - any particular reason? Anyway, if you pick parts with same footprint, then you can test them all.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.