Hi guys, I am working on a new class D amp and I am having some questions/problems that I can't quite explain. Please refer to the attachements which will be divided up to show the different waveforms:
The circuit is a full bridge with only switched feedback [for now] using the LM5104 gate driver and IRLIB4343 FETs. I am using the TL3116 complementary comparator for now without trouble, but I know it can ring on negative pulses which I haven't seen. I don't know what the topology I used is but it's similar to what is in the JBL EP250 where clock/audio/switched feedback summing and integration is done across the comparator pins. The inductors are T130 toroids mounted right next to each other, which could be a problem since stray flux could couple?? I was thinking that flipping one inductor upside down might cancel any stray coupling?
The circuit is a full bridge with only switched feedback [for now] using the LM5104 gate driver and IRLIB4343 FETs. I am using the TL3116 complementary comparator for now without trouble, but I know it can ring on negative pulses which I haven't seen. I don't know what the topology I used is but it's similar to what is in the JBL EP250 where clock/audio/switched feedback summing and integration is done across the comparator pins. The inductors are T130 toroids mounted right next to each other, which could be a problem since stray flux could couple?? I was thinking that flipping one inductor upside down might cancel any stray coupling?
Attachments
The left screenshot is of the switching outputs - the A side on top and B side on bottom. Middle is zoomed in rise times and right is zoomed in fall times. Modulation was at 0 in all screenshots. The deadtime resistor is set to 20k here, which corresponds to about 100ns.
First, I was unable to snub the FETs well enough. I am using the IRLIB4343 which has a very small Qg and is generally very easy to drive - I think that may be part of my problem. I started with a 4.99ohm gate resistor and that gave about 6ns rise times and a ripple about 70MHz. A 680pF cap and 6.8ohm snubber gave the best results [on most of them, some needed a bit of tweaking], but it still wasn't as clean as I thought it could be. I have the dead time on the LM5104 turned up, as you can see, so I don't think overshoot is the problem. The board layout is very tight, which explains the high ringing freq, but that didn't clear the ringing up so I tried to enlarge the gate resistor and slow down the gate. I took the gate resistor all the way out to 15ohm without any real change in the rise time. I then went down to 3.3ohm and still there was little change in rise times.
Also, adding the snubbing sped up the rise time and slowed down the fall time from drain to source. I would have though adding capacitance across the FET would have slowed both edges down. Am I missing something? It is clear that my fall times are very long.
Can anyone help explain what might be going on here, and/or suggest anything to try?
First, I was unable to snub the FETs well enough. I am using the IRLIB4343 which has a very small Qg and is generally very easy to drive - I think that may be part of my problem. I started with a 4.99ohm gate resistor and that gave about 6ns rise times and a ripple about 70MHz. A 680pF cap and 6.8ohm snubber gave the best results [on most of them, some needed a bit of tweaking], but it still wasn't as clean as I thought it could be. I have the dead time on the LM5104 turned up, as you can see, so I don't think overshoot is the problem. The board layout is very tight, which explains the high ringing freq, but that didn't clear the ringing up so I tried to enlarge the gate resistor and slow down the gate. I took the gate resistor all the way out to 15ohm without any real change in the rise time. I then went down to 3.3ohm and still there was little change in rise times.
Also, adding the snubbing sped up the rise time and slowed down the fall time from drain to source. I would have though adding capacitance across the FET would have slowed both edges down. Am I missing something? It is clear that my fall times are very long.
Can anyone help explain what might be going on here, and/or suggest anything to try?
Attachments
And here are the switching nodes of the amp with the dead time turned all the way down. You can see larger spikes which I am attributing to crossconduction. The tops of the waveforms aren't very straight, and there are those weird 'corners'. They seem to be related to when switching occurs but I'm not sure what might be going on here. On that subject, my switching doesn't occur on top of each other which doesn't really make sense to me. The comparators are configured the same and very, very close to each other. I know I want my switching on top of each other but I'm not sure how. Adjusting gate resistors didn't make much difference, as I stated above. Any help?
Attachments
I haven't yet connected any feedback. THD, with an AUX-00025 filter and the AP monitor filter at 22kHz, is about 8.25% at 1W into 12ohm. Turning the dead time down brings THD down to about 7.75%. After 1W THD is basically constant until clipping. The distortion signature is a lot of third harmonic and the output shows considerable crossover distortion. Distortion drops to about 1% around 500mW output and then begins rising until clipping.
Hi,
to mee it looks that:
- either your dead time is different for L to H and H to L transition
- or you have relatively large DC component in filter inductor current.
Second case looks more probable to me, especially when I look at unequal fall times in first picture of your third post. With such a large dead time, both rise and fall time would be equal if inductor current would have no DC component.
Do you have DC coupled current probe? Try to measure inductor current.
Best regards,
Jaka Racman
to mee it looks that:
- either your dead time is different for L to H and H to L transition
- or you have relatively large DC component in filter inductor current.
Second case looks more probable to me, especially when I look at unequal fall times in first picture of your third post. With such a large dead time, both rise and fall time would be equal if inductor current would have no DC component.
Do you have DC coupled current probe? Try to measure inductor current.
Best regards,
Jaka Racman
Hi,
Your wavforms look normal to me. You need to keep your fall time fast overwise you will destroy your outputs, but on the third wavforms they look normal. I have a half bridge proto and the waveforms are identical. I don't think you will clean them up anymore. You say you have snubbered the output, that will have removed most of the ringing from the wavform. The spike on the switching corner is just the inductor trying to keep the current flowing when you quickly clamp high or low. The little kink you see on the slower edge is the point where the next fet turns on. The other fet has already turned off, your signal should ramp toward the ampsbias, usally if left it will total overshoot this point due to inductor. The signal moves toward the bias, then your other fet turns on and speeds up the movment as it trys to clamp so hence the slight kink. You can prob speed up your deadtimes to remove the slower portion of the rise/fall. be carefull not to go to far! But all in all your wavform looks reasonablly normal to me. What is the rise/fall time of the bridge? You should aim for around a max of 50n/s from start to finish and keep an eye on the propogation delay. If you keep these two beasts tame there is no reason why you should not be able to cut deadtimes down to 25n/s. Don't be paranoid though, your waveform is a lot better than some of my attempts.
Regards
Mad.P
Your wavforms look normal to me. You need to keep your fall time fast overwise you will destroy your outputs, but on the third wavforms they look normal. I have a half bridge proto and the waveforms are identical. I don't think you will clean them up anymore. You say you have snubbered the output, that will have removed most of the ringing from the wavform. The spike on the switching corner is just the inductor trying to keep the current flowing when you quickly clamp high or low. The little kink you see on the slower edge is the point where the next fet turns on. The other fet has already turned off, your signal should ramp toward the ampsbias, usally if left it will total overshoot this point due to inductor. The signal moves toward the bias, then your other fet turns on and speeds up the movment as it trys to clamp so hence the slight kink. You can prob speed up your deadtimes to remove the slower portion of the rise/fall. be carefull not to go to far! But all in all your wavform looks reasonablly normal to me. What is the rise/fall time of the bridge? You should aim for around a max of 50n/s from start to finish and keep an eye on the propogation delay. If you keep these two beasts tame there is no reason why you should not be able to cut deadtimes down to 25n/s. Don't be paranoid though, your waveform is a lot better than some of my attempts.
Regards
Mad.P
- Status
- Not open for further replies.