UCD 25 watts to 1200 watts using 2 mosfets

Jump to Latest
#1,561
CPX said:
..and i had lower efficiency🙁
Click to expand...
Yes, that is fitting to theory.

If 5% more efficacy and 1 Ohm capability are really a must, you might consider to increase the dead time in front of the level shifter (==> less idle consumption) and reduce the gate drive resistors to levels, which touch the limits of the IR chip... likely to be around 3-5 Ohms... and drive your IRFB4321 ...living with a hot driver IC would be the down side...
...and you could add a moderate adjustment of the output filter and snubbers.
Altogether, IMHO that's the limit which you can reach without complete redesign.

Can you link/post your current schematic including values?

P.S.
...might take until tomorrow evening before I can answer again...
Work is calling.
 
#1,562
Won't that increase distorsions?😕.

I do not have a schematic yet since prototipe is not finished...i am practicaly using lorylaci's schematic with some changes(lm317/337 stab.,snubbers,feedback resistors,added buffer..etc)..i will paint it when final version is ready🙂.

You answer when you can..you have no obligation to do this and i am thankfull for everything, i wish you a good day at work :cheers:
 
#1,565
CPX said:
I don't think so, i have Rloss=0.027*2*16^2=14W at 100 degreees and i lose over 68 W,even if i take out 12 W quiescent loss and 2 W snubber loss and 5W inductor and filtering loss te rest 35 W is switching loss🙁
Click to expand...

Your calculations must sure be bad about efficiency. The method for measureing efficiency, has some erros. For example due to bus-pumpong some of the energy is pumped back to the PSU.

I put the data into my Excel scheet, and got abouta max 20W losses/FET at 20 A output current.

If you say, switching losses dominate, then the easiest method to decrease switching losses, is to decrease switching frequency.

Then you should realise which type of switching losses dominate?
Is it the cross-condcution (or integarl of votlage times current FETs), or is it output capacitance, or is it the commutating current in the antiparralel diode?

More detail is need to make better decisions.
 
#1,566
lorylaci said:
If you say, switching losses dominate, then the easiest method to decrease switching losses, is to decrease switching frequency.
More detail is need to make better decisions.
Click to expand...

Amen. Full agreement from my side.

Regarding loss calculation. Let's assume that the formula of the application note would deliver a good approximation (also worth a discussion, but I think the formula is good enough to get a first idea).
There are two points in your Excel that can improved.
a) Psw1: Looks like a typo. The sheet calculates column E-F, but I think E+F would be correct.
b) Psw3: You assume the factor K to be 1. In fact K may differ massively from 1, because Qrr is strongly depending on di/dt and also somewhat from Tj and IF. Let's have a look to the data sheet of IRFB4321. Figure 19.
Qrr is shown as a function of di/dt and for two temperatures.
Even at moderate slow switching, we will have di/dt of 300A/us or more.
Figure 19 shows that at 300A/us and realistic temperatures the Qrr is around 1000nC instead the so called 'typical' 300nC at 25C and 100A/us.
Means factor K has to taken much higher than 1. IMHO any value below 2 is unrealistic.

CPX:
At your side, I tend to say that your thermal expectations are not reasonable.
Running the amp for 2 minutes at full power is already pretty tough. Nobody is listening to a beep of 500W/18kHz continously. At least I am glad that nobody forces me to do so. For thermal evaluation it is more reasonable to chose real music (feel free to search for heavy dub with high average power content). Run the power into a load resistor and in parallel feed it through a 1kOhm resistor to a normal speaker. So you can hear what's going on and if you wish you can even adjust for clipping levels that some bad boy DJs might apply.
 
#1,567
ChocoHolic said:
Amen. Full agreement from my side.

Regarding loss calculation. Let's assume that the formula of the application note would deliver a good approximation (also worth a discussion, but I think the formula is good enough to get a first idea).
There are two points in your Excel that can improved.
a) Psw1: Looks like a typo. The sheet calculates column E-F, but I think E+F would be correct.
b) Psw3: You assume the factor K to be 1. In fact K may differ massively from 1, because Qrr is strongly depending on di/dt and also somewhat from Tj and IF. Let's have a look to the data sheet of IRFB4321. Figure 19.
Qrr is shown as a function of di/dt and for two temperatures.
Even at moderate slow switching, we will have di/dt of 300A/us or more.
Figure 19 shows that at 300A/us and realistic temperatures the Qrr is around 1000nC instead the so called 'typical' 300nC at 25C and 100A/us.
Means factor K has to taken much higher than 1. IMHO any value below 2 is unrealistic.

CPX:
At your side, I tend to say that your thermal expectations are not reasonable.
Running the amp for 2 minutes at full power is already pretty tough. Nobody is listening to a beep of 500W/18kHz continously. At least I am glad that nobody forces me to do so. For thermal evaluation it is more reasonable to chose real music (feel free to search for heavy dub with high average power content). Run the power into a load resistor and in parallel feed it through a 1kOhm resistor to a normal speaker. So you can hear what's going on and if you wish you can even adjust for clipping levels that some bad boy DJs might apply.
Click to expand...

Thanks ChocoHolic. I never really used this spreadsheet, I made approximations usually.
After your reccomendations, it can be clearly seen that switching losses do overwhelm.
For IRFB4321 I would try an additional PNP transistor for gate discharge. Since this would be close, miller effect could have less chance.
I already told CPX that music has no continous power.

Parralleling more devices would decrease dI/dt per device, so parralel of two low Qrr (whichi usually has higher Rdson) would spare him some losses? (also total thermal resistance from die to heatsink would decrease, which combined with forced cooling would get him less thermal stress)
But of course then additional totem-poler per FETs would be needed (but if put next to FETs, then also would mean better switching)
 
#1,568
lorylaci said:
For IRFB4321 I would try an additional PNP transistor for gate discharge. Since this would be close, miller effect could have less chance.
Click to expand...

I also think that a PNP (fast high current type) for fast turn OFF and at the same time a low turn-ON-resistor of approx. 5 Ohms would be a reasonable starting point to drive the IRFB4321 with acceptable speed.
This combination should also keep shoot through idle losses in a reasonable range without any change in front of IR2110. Also the PNP would help move some heat out of the IR2110.


lorylaci said:
Parralleling more devices would decrease dI/dt per device, so parralel of two low Qrr (whichi usually has higher Rdson) would spare him some losses? (also total thermal resistance from die to heatsink would decrease, which combined with forced cooling would get him less thermal stress)
But of course then additional totem-poler per FETs would be needed (but if put next to FETs, then also would mean better switching.
Click to expand...

Here you are touching one point were I am not fully in line with the IR application note. The application note in combination with the growing Qrr for fast di/dt would indicate that fast di/dt would generally increase losses.
In fact there is one more aspect were high di/dt helps to reduce switching losses and this aspect is not shown in the application note.

Let's have a look to a switching event at hard switching condition.

Starting point:
Upper MosFet on, current running from load to upper MosFet.
(...situation during negative output signal and upper switch ON):
Now we turn OFF the upper switch, the current commutates from the N-chanel to the body diode, but remains running through the upper device.
After the dead time we start to turn ON the lower switch.
Before the half bridge can slope down, the lower switch has to take over the load current and has to remove the Qrr.
The time to take over the load current is can be calculated by I/(di/dt).
Here high di/dt is helpful to reduce losses.
After the load current is reached the current still increases to remove Qrr, here high di/dt is unfortunate regading losses.
As soon as also Qrr is removed, the reverse current of the diode steps back to zero pretty fast (snap off) and the output voltage slopes down to the negative rail. This sloping speed is depending on parasitic capacitances, snubbering, applied currents and applied di/dt.

With respect to one phase were high di/dt is increasing switching losses(removing Qrr) and other phases were high di/dt is lowering the switching losses (taking over load current and sloping speed) - we do get a certain optimum di/dt for smallest switching losses.
This optimum is also depending on load current, because the load current does directly influence the time which is needed to take over the load current, but does have just a minor influence on Qrr.

For really high currents in my application (up to 40A), I came to the conclusion that even the max di/dt that I can reach (limited by the inductance of the source leg to levels slightly above 1000A/us) is below the optimum of the IRFB4115.
I would also guess that in case of CPX amp, the achieved di/dt of is below the optimum of IRFB4321.

Does paralleling help?
...would need a detailed examination of the particular choice...
Up to now I would expect, as long as we do not exceed the limitations of the largest chip that can be put into a certain package, it is likely to find a single MosFet that can do it as good as two paralleled ones.
If we would need to change to the next larger package like TO247, then paralleling two TO-220 might become attractive, because snubbering TO247 is pain....
But for 600W from a bridge, the IRFB4321 should easily do the job without paralleling.

Sorry for the long and pretty dry text 😱
 
Last edited:
#1,569
lorylaci:
I measure efficiency corectly..bus pumping is neglijable at 500 hz,anyway
what i said was only an aproximation but switching losses are still very big.🙁.
My amplifier is mainly for myself..but i want to make it as good as posible..😛.

ChocoHolic:
I need good efficiency to get as much power as posible and the run the amplifier as quiet as possible(without a fan)

New efficiency data :

irfb4321 and lm311 320 khz

-500 hz real impedance 1.25 ohm
9A 74%; 14A 80%; 16A 82%; 17A 85%

-500 hz real impedance 3 ohm
1A 41%; 4A 77%; 8A 90%

I need 86% at 16A 1 ohm load for 500W output and i only have 82%,so i will probably have to accept 450W..😕

New distorsion tests(2 uF output capacitance, 10uH,lm319, irfb4321, 350khz switching)

100 hz real impedance: 1.18 ohm
1A 0.13%; 12A 0.11%; 16A 0.26%

1000 hz real impedance: 1.3 ohm
1A 0.12%; 13A 0.26%; 14.5 0.1%; 16A 0.6%

15000 hz real impedance: 1.6 ohm
1A 0.25%; 8A 0.17%; 10A 0.18%; 11.3A 0.08%

100 hz real impedance: 2.9 ohm
1A 0.05%; 4A 0.12%; 6A 0.10%; 8A 0.27%

1khz real impedance: 3.1 ohm
6A 0.085%; 7A 0.12%; 8A 0.6%

15khz real impedance:4.2 ohm
1A 0.07%; 3.3A 0.08%; 7.6A 0.46%

So distorsions are smaller with added output capacitance and smaller residual 😉.
 
#1,570
To squeeze a little more efficacy it might be worth to try lorelaci's proposal with the PNP for turning OFF and decreasing the value of the turn-on-resistor.

Overall your results are not bad already now.
450W vs 500W: Of course 500W is much easier for marketing.
But our ears won't notice much, it's just 0.5db .
Heat is a more important topic, but should be checked with realistic music program.
 
#1,573
I do not see so much changes in the PCB, mostly less jumpers at the neg rail.
Critical power paths seem to be unchanged, so I would not expect new headache in that area.

Higher distortion at lower f:
I am not convinced that this would be a general rule.
Did you keep ripple unchanged?
Or filter unchanged and may be at the same time increased feedback cap?
In any case, it seems that you have found a reasonable fit to your requirements. Enjoy!
 
#1,574
No..riple was bigger,and i saw that adding aditional capacitance reduced distorsions..i will make a few more tests when i will install lt1016... i hope i will have less noise with it..

I tested lorylaci's undervoltage modification but i used it as a standby control and seems to work good,practicaly it only blocks the level shifter,using ir2110 sd pin this can not be done.

Thanks again ChocoHolic,Lorylaci,Luka and others for your help:worship:...i will come back with new questions 😀 when the new board will be ready and i will try to bridge the amp.
 
#1,575
stewin said:
hi dzony pls can you post your revised schema on the relay part . because i build the schema but relay didnt work until i grounded emiter of bc337 next to r6 and connected R6 to base of 337 . the relay then switched, but did not delay on powering up . With the temp protection, the temp led lights but the relay does not switch off.
Click to expand...
 

Attachments

Last edited:
#1,577
AC conector needs negativ voltage to operate. But that negativ voltage must be independent of main negativ voltage. That whay AC conector goes directly on one of two secondary on main transformer. AC conection who has AC voltage with inverted 1N4007 diode and 1uF makes negativ voltage for protection.

Cheers
 
You must log in or register to reply here.
Top Bottom