By using some more components, I've been able to get some better results.
Output slew rate is between 2V/ns to 4V/ns with +/-36V supply. The transistors are rated to 5V/ns.
Worst rise time 38ns, best rise time 30ns
Worst fall time 64ns, best fall time 46ns
I believe that the difference between rise times or fall times are due to the output inductor helping or not with the switching. If this it right, the deadtime is somewhere arround 64ns-46ns=18ns. Looking at the waveforms I've spotted what looks like 30ns of deadtime.
Better results can probably be achived by fitting with the zenor voltages.
I've changed the comperator stage a bit, to improve the slew rate.
Schottky diodes has been added to the output stage, to improve current spikes while switching. These can probably be left out, since it dosn't matter that much.
Output slew rate is between 2V/ns to 4V/ns with +/-36V supply. The transistors are rated to 5V/ns.
Worst rise time 38ns, best rise time 30ns
Worst fall time 64ns, best fall time 46ns
I believe that the difference between rise times or fall times are due to the output inductor helping or not with the switching. If this it right, the deadtime is somewhere arround 64ns-46ns=18ns. Looking at the waveforms I've spotted what looks like 30ns of deadtime.
Better results can probably be achived by fitting with the zenor voltages.
I've changed the comperator stage a bit, to improve the slew rate.
Schottky diodes has been added to the output stage, to improve current spikes while switching. These can probably be left out, since it dosn't matter that much.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
I've also made a N-ch version of the output stage. It performs as well as the N-ch + P-ch stage with two IRF540 transistors. Switching times and dead times are about the same.
As simulation with two modern IRF6665 N-Ch transistors gave me switching times down to 10ns and deadtimes in the range of 20ns, without any optimizingn.
It seams like these fully descrete amps can be made to perform very well, at least in simulations.
As simulation with two modern IRF6665 N-Ch transistors gave me switching times down to 10ns and deadtimes in the range of 20ns, without any optimizingn.
It seams like these fully descrete amps can be made to perform very well, at least in simulations.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Heres a version using two n-ch IRF6665 mosfets in the output stage. I've replaced the mosfet in the driver logic with a bjt. It actually performs better.
By looking at the waveforms, Deadtime + switching time seems to be arround 20ns for both rising and falling edges.
I've very surprised that this is possible with a fully descrete design. Now does anybody know where mortals can get these IRF6665 transistors? Fairchildsemi makes the FDD3682 which performs very well (low gate charges and Ron), but IRF6665 has half the gate charge.
By looking at the waveforms, Deadtime + switching time seems to be arround 20ns for both rising and falling edges.
I've very surprised that this is possible with a fully descrete design. Now does anybody know where mortals can get these IRF6665 transistors? Fairchildsemi makes the FDD3682 which performs very well (low gate charges and Ron), but IRF6665 has half the gate charge.
The 6665's should be available. They were pitching them to me last year. What was still not released was the reference design---there was still some work being done on the protection circuitry associated with the gate driver IC, in which the lower FET's voltage drop is being sensed on a cycle-by-cycle basis.
They do have astonishingly low parasitic inductances as well as the optimized gate charge vs. on R. A bit tricky to solder for DIY use however. They showed some data that dramatized the improvement in reduction of EMI between what had been considered low-L packages and the new package.
The claim was also made that a lot of the improvement in a prototype amplifier was due to the pin-strappable deadtime gate driver ICs they have come out with, and the role that feature plays in reducing jitter/noise in the gate drive.
I saw bench performance that was exemplary out of a relatively simple hysteretic converter using an LM318 in the front end. I speculated that a quieter front end, perhaps a hybrid discrete/IC design, could bring their noise floor down even further, and they agreed.
They do have astonishingly low parasitic inductances as well as the optimized gate charge vs. on R. A bit tricky to solder for DIY use however. They showed some data that dramatized the improvement in reduction of EMI between what had been considered low-L packages and the new package.
The claim was also made that a lot of the improvement in a prototype amplifier was due to the pin-strappable deadtime gate driver ICs they have come out with, and the role that feature plays in reducing jitter/noise in the gate drive.
I saw bench performance that was exemplary out of a relatively simple hysteretic converter using an LM318 in the front end. I speculated that a quieter front end, perhaps a hybrid discrete/IC design, could bring their noise floor down even further, and they agreed.
I have been meaning to give the rep a call anyway, so I will ask him. They ought to be in distribution by now, although I don't recall offhand who is on top of the new parts. Of course Digikey will have them eventually, but I don't think they are high on the priority list for relatively new parts.
Maybe I could cadge a few samples out of them.
Maybe I could cadge a few samples out of them.
After I added C25 and Q5 the difference wasn't that big. I've also tryed to add a two feedback 33pF capasitors to the comperator to give it a hysteresis behavior, but it didnt make any difference.
I could imagine though, that hysteresis would be a good thing IRL.
Anyway I do have a scope (200Mhz), but I don't have the time to build this the next few days.
I could imagine though, that hysteresis would be a good thing IRL.
Anyway I do have a scope (200Mhz), but I don't have the time to build this the next few days.
ONsemi's NTP3055 and NTP2955 are N channel and P channel MOSFETs optimized for bridge use. They are immune to (so-called) reverse recovery dv/dt problems, and have betted RDS(on) matching than 540/9540. The price is slightly higher than 540/9540 in China (3.5 Yuan v.s. 2.5 Yuan).
Have any person tried that in a class D ?
Have any person tried that in a class D ?
NTP2955 is only rated to Vbr=60V, which should never be reached. Also the Ron i high. Cg is low however, which gives faster switching = lees distortion and less f*C loss.
NTP3055 I can't find.
Anywhy why do you want to match the fets and what do you mean by that?
At a given switching frequency the trade off between Ron and Cg is not the same for P-ch and N-ch Fets. Matching for equal loss in both transistors is difficult. P-ch Fets have 3 times worse carateristics than N-ch. Aming for a fully N-ch design is best. Modern N-fets like FDD3682 and IRF6665 perform outstanding.
I've always thought that it would be impossible to design a fast descrete high side N-ch driver and make it work in a Half-bridge, but it turns out that it can be done. Actually it's not tant difficult.
Fuji also knows how to make good semiconductors: 2SK3598-01 100Vbr 20A 22nC 62mOhm. Also their SMD TFP package is very nice.
TPCA8006-H (100Vbr 67mOhm 12nC) or TPCA8007-H (100V 47mOhm 15nC) from Toshiba are also a wise choice.
NTP3055 I can't find.
Anywhy why do you want to match the fets and what do you mean by that?
At a given switching frequency the trade off between Ron and Cg is not the same for P-ch and N-ch Fets. Matching for equal loss in both transistors is difficult. P-ch Fets have 3 times worse carateristics than N-ch. Aming for a fully N-ch design is best. Modern N-fets like FDD3682 and IRF6665 perform outstanding.
I've always thought that it would be impossible to design a fast descrete high side N-ch driver and make it work in a Half-bridge, but it turns out that it can be done. Actually it's not tant difficult.
Fuji also knows how to make good semiconductors: 2SK3598-01 100Vbr 20A 22nC 62mOhm. Also their SMD TFP package is very nice.
TPCA8006-H (100Vbr 67mOhm 12nC) or TPCA8007-H (100V 47mOhm 15nC) from Toshiba are also a wise choice.
IMO the tricky part is having the motion of the upper FET "aid" the turnon/turnoff rather than fight it.
Apparently there are some fairly significant patents on some of the schemes for high side drive, embodied in integrated driver parts.
It is appealing though to do the drive in discrete form if the spurious coupling can be kept under control, since it should be possible to reduce jitter, noise, and signal self-heating shifts and thereby reduce open-loop noise and distortion.
I built a tight-layout discrete driver, but using medium-sized leaded parts like TO92 case transistors it became clear that unwanted couplings just couldn't be adequately controlled. The project was about to migrate to an SMD carefully thought-out layout when I was diverted to other activities.
Apparently there are some fairly significant patents on some of the schemes for high side drive, embodied in integrated driver parts.
It is appealing though to do the drive in discrete form if the spurious coupling can be kept under control, since it should be possible to reduce jitter, noise, and signal self-heating shifts and thereby reduce open-loop noise and distortion.
I built a tight-layout discrete driver, but using medium-sized leaded parts like TO92 case transistors it became clear that unwanted couplings just couldn't be adequately controlled. The project was about to migrate to an SMD carefully thought-out layout when I was diverted to other activities.
Very good project... but i'm thinking.... may be is better use two N devices instead of complementary power devices... I think it should solve a lot of problems.
Another way.... we know that there are a lot of diff in capacitance between the same complimentary couple, so why don' t try to use a smaller P channel device like 9520 that have smallest capacitance?
This is only an idea, I think to remember I saw this solution in some commercial amps, but I' m not sure... I will try to make some searches.
By the way...... B R A V O (good boy) !!!!!
Another way.... we know that there are a lot of diff in capacitance between the same complimentary couple, so why don' t try to use a smaller P channel device like 9520 that have smallest capacitance?
This is only an idea, I think to remember I saw this solution in some commercial amps, but I' m not sure... I will try to make some searches.
By the way...... B R A V O (good boy) !!!!!
Hello, I have recently recieved a project in which I have to build a fully discrete class D amp. Here are the specs that I need:
50 watts into an 8 ohm load
Input impedance to be 50 Kohms
Frequency response - DC -- 100 Khz +- 2 db
Output impedance such that the damping factor is 100 or more.
1 volt into the input will give full output
I'm thinking the schematic mentioned earlier would be a good place to start but if anyone knows if it can do those things, please let me know. I won't have access to PSpice until monday so I can't test it yet.
Also any tips on how to proceed would be greatly appriciated.
50 watts into an 8 ohm load
Input impedance to be 50 Kohms
Frequency response - DC -- 100 Khz +- 2 db
Output impedance such that the damping factor is 100 or more.
1 volt into the input will give full output
I'm thinking the schematic mentioned earlier would be a good place to start but if anyone knows if it can do those things, please let me know. I won't have access to PSpice until monday so I can't test it yet.
Also any tips on how to proceed would be greatly appriciated.
It's not a problem to reduce the order of the outputfilter (which i 2 for a LC filter). In UCD it's done by adding a differentiated version of the outputfilter back to the summing point (done with the feedback capasitor) and making the filter a part of the modulator. UCD has a 1. order responce.
You can take contol of the LC filter and make it load independent by makin PD feedback.
If you want to reduce the order of the filter by one, you can do what I've done in the schematic below.
I haven't added any values, but the concept is right. This dosn't only apply to switching amplifiers.
You can take contol of the LC filter and make it load independent by makin PD feedback.
If you want to reduce the order of the filter by one, you can do what I've done in the schematic below.
I haven't added any values, but the concept is right. This dosn't only apply to switching amplifiers.
An externally hosted image should be here but it was not working when we last tested it.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.