Hi CPX,
most likely Lorylaci and the other guys in this thread know this family of amps by far better than I do. I never build this one.
So my comments are based on theory and observations in different designs only.
I think the proposal of lorylaci to move to smaller inductance values goes to the right direction, if you want to have 18kHz without attentuation.
Well, you said, you need a certain amount of ripple.
Now you have 10uH and 1uF.
You will get the same ripple with 5uH and 2uF and at 1 Ohm load the attentuation of a 18kHz signal will be just 0.35db (instead of 3db with 10uH and 1uF).
The downside of such low inductors is that you are chasing high triangular currents through the power stage.
When you start playing with such values, then keep an eye on the self heating of the e-caps on your PCB.
But frankly speaking, I do not see the need to achieve max output power at 18kHz. There is no music program with such massive content of 18kHz and also no tweeter that would survive such attacks.
Efficacy:
Your values appear lower than stated values from many others, but you cannot compare your values with other designs that run at significantly lower frequencies.
If you want to stay with a simple gate drive directly from the IR2110 and achieve high efficacy at the same time - I agree to the opinion that you probably have to move to switching frequencies around 200kHz-250kHz.
Driving an IRFB4321 in a fast manner demands pretty some driving current.
most likely Lorylaci and the other guys in this thread know this family of amps by far better than I do. I never build this one.
So my comments are based on theory and observations in different designs only.
I think the proposal of lorylaci to move to smaller inductance values goes to the right direction, if you want to have 18kHz without attentuation.
Well, you said, you need a certain amount of ripple.
Now you have 10uH and 1uF.
You will get the same ripple with 5uH and 2uF and at 1 Ohm load the attentuation of a 18kHz signal will be just 0.35db (instead of 3db with 10uH and 1uF).
The downside of such low inductors is that you are chasing high triangular currents through the power stage.
When you start playing with such values, then keep an eye on the self heating of the e-caps on your PCB.
But frankly speaking, I do not see the need to achieve max output power at 18kHz. There is no music program with such massive content of 18kHz and also no tweeter that would survive such attacks.
Efficacy:
Your values appear lower than stated values from many others, but you cannot compare your values with other designs that run at significantly lower frequencies.
If you want to stay with a simple gate drive directly from the IR2110 and achieve high efficacy at the same time - I agree to the opinion that you probably have to move to switching frequencies around 200kHz-250kHz.
Driving an IRFB4321 in a fast manner demands pretty some driving current.
Yes 18 khz is to high..<0.3 db at 16 khz should be ok.
I am not happy to decrease switching frequency because i am loseing frequency response,music sounds much better at >350 khz than at 200 khz
, so i would need an external booster to compensate and also noise(hiss) is bigger with more feedback capacitance.
It is a compromise situation and do not know what to chose..
I am not happy to decrease switching frequency because i am loseing frequency response,music sounds much better at >350 khz than at 200 khz
, so i would need an external booster to compensate and also noise(hiss) is bigger with more feedback capacitance.It is a compromise situation and do not know what to chose..
Last edited:
Yes 18 khz is to high..<0.3 db at 16 khz should be ok.
I am not happy to decrease switching frequency because i am loseing frequency response,music sounds much better at >350 khz than at 200 khz
It is a compromise situation and do not know what to chose..
Not increasing swithcing freq, but the decreas of prop delay that makes music better.
Not only the feedback capacitance that changes frequency, try to experiment with the resistors too.
The whole idea is that these phase lead network should set the frequency.
I would also tend to lorylaci's view.
I would tend to this view because of multiple reasons.
- Lower switching frequency doe not necessarily mean poor frequency response. Allowing larger output ripple may also help for tuning.
- Why do you think your amp would have a slow step response? The early clipping at 18kHz does not automatically mean slow step response, please note the filter is inside the loop. How does the step response look?
- Benefits of high switching frequencies can only be derived, when we achieve almost perfect gate drive. And in any case it will lower efficacy.
- Adding gate drivers like proposed by nigel is possible but naturally also increases the chance to get further unexpected effects. (Nevertheless for high switching frequencies with IRFB4321 likely to be necessary.)
===================
Perfect gate drive is always a relative thing...
The screen shot that I asked for would give us a first impression of dead time adjustment and gate drive control/speed.
Next step would be to check the same signals, but during neg signal half wave. There we will find the hard switching situation, which is more critical for undesired HF-ringing.
In both examinations it may happen that we need two screen shots.
One for turning ON the MosFet and one for turning OFF the MosFet.
Depending on duty cycle both events together might simply not fit on the screen with a resolution of 50ns/grid.
Take care during probing.
If you connect the grounded scope to the negative rail, then you must ensure to have a isolated floating power supply and also floating signal generator.
I would tend to this view because of multiple reasons.
- Lower switching frequency doe not necessarily mean poor frequency response. Allowing larger output ripple may also help for tuning.
- Why do you think your amp would have a slow step response? The early clipping at 18kHz does not automatically mean slow step response, please note the filter is inside the loop. How does the step response look?
- Benefits of high switching frequencies can only be derived, when we achieve almost perfect gate drive. And in any case it will lower efficacy.
- Adding gate drivers like proposed by nigel is possible but naturally also increases the chance to get further unexpected effects. (Nevertheless for high switching frequencies with IRFB4321 likely to be necessary.)
===================
Perfect gate drive is always a relative thing...
The screen shot that I asked for would give us a first impression of dead time adjustment and gate drive control/speed.
Next step would be to check the same signals, but during neg signal half wave. There we will find the hard switching situation, which is more critical for undesired HF-ringing.
In both examinations it may happen that we need two screen shots.
One for turning ON the MosFet and one for turning OFF the MosFet.
Depending on duty cycle both events together might simply not fit on the screen with a resolution of 50ns/grid.
Take care during probing.
If you connect the grounded scope to the negative rail, then you must ensure to have a isolated floating power supply and also floating signal generator.
I use a pair of heel-pinchers to drive my cattle through multiple gates.
[sorry for the blatant off-topic; just couldn't resist it
]I had a look to the PCB in posting #1440.
There is one critical path which encloses a pretty large loop area.
Description of the path:
GND ==> rail caps ==> upper MosFet ==> lower MosFet ==> rail caps ==> GND
If this large loop area causes trouble in your particular case can be tested by putting a ceramic cap (i.e. X7R/1uF/200V) directly from the drain of the upper MosFet to the source of the lower MosFet, with shortest possible wires.
SMD, ceramic, X7R is preferable, because it combines low parasitic inductance with inherent HF losses ==> shifting rail ringing to very frequencies and bringing a pretty nice damping.
If you do not have a such a ceramic cap, you can of course start this trial by using a reasonable MKT that you might have on hand.
There is one critical path which encloses a pretty large loop area.
Description of the path:
GND ==> rail caps ==> upper MosFet ==> lower MosFet ==> rail caps ==> GND
If this large loop area causes trouble in your particular case can be tested by putting a ceramic cap (i.e. X7R/1uF/200V) directly from the drain of the upper MosFet to the source of the lower MosFet, with shortest possible wires.
SMD, ceramic, X7R is preferable, because it combines low parasitic inductance with inherent HF losses ==> shifting rail ringing to very frequencies and bringing a pretty nice damping.
If you do not have a such a ceramic cap, you can of course start this trial by using a reasonable MKT that you might have on hand.
The specified path is connected to the ground plane on the other side(red points).
Requested wave forms(i hope):
50 hz - 15 A load positive half wave lower mosfert gate drive
50 hz - 15 A load negative half wave lower mosfert gate drive
50 hz - 15 A load switching bus versus mosfert gate drive
positive half wave
negative half wave
I also made some distorsion tests using sound card:
50 hz - 3.5 ohm
1A 0.08%; 2A 0.11%; 6A 0.24%; 7.5A 0.3%
100 hz - 3.5 ohm
1A 0.075%; 6A 0.2%; 8A 0.26%
1khz - 3.5 ohm
1A 0.08%; 3A 0.14%; 6A 0.26%; 8A 0.28%
15 khz 3.5 ohm
1A 0.08%; 3A 0.15%; 5A 0.238%; 6.7A 0.28%; 8A 2.5%
50 hz - 1 ohm
1A 0.14%; 4A 0.18%; 6.5A 0.12%; 11.5A 0.17%; 16A 0.4%
100 hz - 1 ohm
1A 0.14%; 7A 0.09%; 11.2A 0.13%; 16A 0.39%
1khz - 1 ohm
1A 0.17%; 7.5 A 0.45%; 12A 0.67%; 13.5 A 0.7%; 16A 3%
15 khZ - 1ohm
10A 0.5%; 12A 2%;13A 3.5%
An externally hosted image should be here but it was not working when we last tested it.
Requested wave forms(i hope
):50 hz - 15 A load positive half wave lower mosfert gate drive
An externally hosted image should be here but it was not working when we last tested it.
50 hz - 15 A load negative half wave lower mosfert gate drive
An externally hosted image should be here but it was not working when we last tested it.
50 hz - 15 A load switching bus versus mosfert gate drive
positive half wave
An externally hosted image should be here but it was not working when we last tested it.
negative half wave
An externally hosted image should be here but it was not working when we last tested it.
I also made some distorsion tests using sound card:
50 hz - 3.5 ohm
1A 0.08%; 2A 0.11%; 6A 0.24%; 7.5A 0.3%
100 hz - 3.5 ohm
1A 0.075%; 6A 0.2%; 8A 0.26%
1khz - 3.5 ohm
1A 0.08%; 3A 0.14%; 6A 0.26%; 8A 0.28%
15 khz 3.5 ohm
1A 0.08%; 3A 0.15%; 5A 0.238%; 6.7A 0.28%; 8A 2.5%
50 hz - 1 ohm
1A 0.14%; 4A 0.18%; 6.5A 0.12%; 11.5A 0.17%; 16A 0.4%
100 hz - 1 ohm
1A 0.14%; 7A 0.09%; 11.2A 0.13%; 16A 0.39%
1khz - 1 ohm
1A 0.17%; 7.5 A 0.45%; 12A 0.67%; 13.5 A 0.7%; 16A 3%
15 khZ - 1ohm
10A 0.5%; 12A 2%;13A 3.5%
Last edited:
...thanks for the hint, and with this it looks much better.
The last two screen shots show the key signals, but we really need a timeRequested wave forms(i hope
scale of 50ns/grid (and high sampling rate) to see the details.
With the given time scale, we can see that the gate drive is slow and
that there is ringing. But no chance to say OK or not.
P.S.
Altogether it seems that your results are already on a level, were many DIYers would not really notice that there are still issues.
The gate switching shows the effects of Miller charges. Slow gate cahrge doesn't really means good or bad. Shorter driving loops, or BJTs next to the fets would decrease recharging, or redischarging effects of Miller charge, but also would increase dV/dt which would increase Miller charge effects.
Better zoom on gate drive signal would point out details, on how to's.
The measured distortions seems reasonable for a DIY Class-D amp. Exspecially the 3,5 Ohm measurements are pretty good. I would be really satisfied with them. Less than 0,1% THD at low load (1A whole spectra), n*0,1% THD at about 100W is good. Distortions from other component (like the loudspeaker) would e greater than that.
It would be better to state the supply votlages, and state the distortions at not current, but output votlages (or power at the nominal load), or the best percent of output voltage per supply voltage.
Driving an 1Ohm load impedance is really difficult. To that you must scale all impedances to 1/4th from the 4 Ohm (original design). Look at for example some the Coilcraft inductors which are mead of low-profile wire. YOu should also use an output capactiro with lower ESR. (so very good MKP capacitor is needed), because you have to scale it to output load. And of course properly scaled outptu fitler is needed, as I sead earlier.
Why do you need 1 Ohm loading? For an 1 Ohm load at high power,loudspeaker cable widths could be great, like a thumb.
p.s.: Sorry Chocoholic, this answer is for CPX of course, but I clicked on quite for your post. I just realize it now.
Ok..new wave forms :
rising positive half wave
falling positive half wave
rising negative half wave
falling negative half wave
Something i said wrong...at 15khz real load impedance is arround 1.8 ohm not 1 ohm because my resistors and cables are inductive..
All the results and measurements in the last 2 posts are with irf540z.when i will switch to irfb4321 again i will have other values..also this amp has to work in bridge mode(in 2 ohm speakers since the 1 ohm requirement )..and then distorsions will get doubled,and eficiency has to to go up to at least 86% so there is still enough work to be done..
Thanks for your patience and support!
rising positive half wave
An externally hosted image should be here but it was not working when we last tested it.
falling positive half wave
An externally hosted image should be here but it was not working when we last tested it.
rising negative half wave
An externally hosted image should be here but it was not working when we last tested it.
falling negative half wave
An externally hosted image should be here but it was not working when we last tested it.
Something i said wrong...at 15khz real load impedance is arround 1.8 ohm not 1 ohm because my resistors and cables are inductive..
All the results and measurements in the last 2 posts are with irf540z.when i will switch to irfb4321 again i will have other values..also this amp has to work in bridge mode(in 2 ohm speakers since the 1 ohm requirement )..and then distorsions will get doubled,and eficiency has to to go up to at least 86% so there is still enough work to be done..
Thanks for your patience and support!
Ok..new wave forms :
rising positive half wave
An externally hosted image should be here but it was not working when we last tested it.
falling positive half wave
An externally hosted image should be here but it was not working when we last tested it.
rising negative half wave
An externally hosted image should be here but it was not working when we last tested it.
falling negative half wave
An externally hosted image should be here but it was not working when we last tested it.
Something i said wrong...at 15khz real load impedance is arround 1.8 ohm not 1 ohm because my resistors and cables are inductive..
All the results and measurements in the last 2 posts are with irf540z.when i will switch to irfb4321 again i will have other values..also this amp has to work in bridge mode(in 2 ohm speakers since the 1 ohm requirement )..and then distorsions will get doubled,and eficiency has to to go up to at least 86% so there is still enough work to be done..
Thanks for your patience and support!
CPX: If you want so high power, then you must sacrifice some of the quality (here THD), or you need a much better design, but that would fall out of DIY category.
As I see you have a supply about +-50V, so you want to get around some kWs. (from +-50V you can get about 1kW to 1 Ohm, 2kW to 2 Ohm bridged)
This is a special need and this circuit without major modifications is not capable for that.
You would need a much better comparator. Then a major rework to full bridge-mode, including the elimination of level-shift. Then IR2110 is still good, but IR2010 or better gate drive would be needed, with more parraleled FETs with their own totem-poles. Of course this from almost full SMD, very-well designed double-board. Then you could have the kW poer with low THDs.
At least it does not automatically mean, that a catastrophy will happen.
And I agree that the faster we go, the more miller effect and parasitic inductances will strike back.
On the other hand slow gate drive does mean that times for current commutation are long. Naturally commutation times vary with the applied load. If we bring together both then we will see that this is an distortion mechanism that hurts more and more with increasing load currents, means at low load impedances.
By chance things are fitting together with a posting that I send a few days ago in another thread. The posted wave forms are from a preliminary breadboard, that I had set up for the evaluation of the power stage - before stepping to the full amp on a nice PCB.
In fact there I am using two bipolars at each MosFet. The Fets are similar to IRFB4321.
....posting#23 and onwards....
http://www.diyaudio.com/forums/class-d/190100-new-mosfets-block-3.html#post2809376
Don't bother, I think information exchange is important for all readers here.
Lorylaci i only want 450-500W power... and i have +-44 volts at 7 A max from smps(~ 600W so higher efficiency for the amp is needed),but for that in 2 ohm i need 16A rms.The amplifier will have to be able to work in stereo and bridge also as needed so i can not elimnate the level shifter..I am thinking in changing comparator to lt1016,it is very fast, it has smaller offset and i hope smaller noise also(like lm319 versus lm311)
I tested some 470nF MKT capacitors and output residual was bigger than with the poliester capacitors i am using now..so they seem to have higher esr..so i will probably use 3 poliester capacitrors in parralel mode(i will post a new board version soon)
I tested some 470nF MKT capacitors and output residual was bigger than with the poliester capacitors i am using now..so they seem to have higher esr
..so i will probably use 3 poliester capacitrors in parralel mode(i will post a new board version soon)
I understand this, and read the post you linked. I understand a bit, but I have to think about is. (I am a chemist, so I need little more time with english elctronic texts.)
So what do you suggest CPX should change?
Last edited:
Lorylaci i only want 450-500W power... and i have +-44 volts at 7 A max from smps(~ 600W so higher efficiency for the amp is needed),but for that in 2 ohm i need 16A rms.The amplifier will have to be able to work in stereo and bridge also as needed so i can not elimnate the level shifter..I am thinking in changing comparator to lt1016,it is very fast, it has smaller offset and i hope smaller noise also(like lm319 versus lm311)
I tested some 470nF MKT capacitors and output residual was bigger than with the poliester capacitors i am using now..so they seem to have higher esr
MKT = metalised polyester capacitor
I suggested MKP = meatlized polypropylen, which has less dielectric lossa at n*100kHz, and usually less ESR than MKT capacitors.
How did you measured efficiency? Do your FETs run so hot?
By the way, you mentioned that you think that your switching losses are high. The difference in using IRF540Z and IRFB4321, that IRFB4321 will have higher switching losses (bigger gate charge, worse diode - almost 8x worse, bigger output capacitance).
Last edited:
Hi CPX,
most wave forms are looking acceptable for me.
Except: 'falling positive half wave'
And I am wondering that this shows massive ringing while the
'rising negative half wave' is looking OK.
This might indicate a reason outside the half bridge.
Which type of diode are you using for the bootstrap supply of the high side driver?
most wave forms are looking acceptable for me.
Except: 'falling positive half wave'
And I am wondering that this shows massive ringing while the
'rising negative half wave' is looking OK.
This might indicate a reason outside the half bridge.
Which type of diode are you using for the bootstrap supply of the high side driver?
I fully agree, if you can handle the current and the voltage with the IRF540Z, then better not change to IRFB4321
Lorylaci :
Sorry ..they were wima MKP10 0.47uF..my mistake..
I measured efficiency by measuring output current,output voltage,input current and input voltage and making needed calculations(Pout/pin).
Fets run hot...at 16 amp rms (50W measured losses)they reach 90 degrees in 2 minutes with small computer smps heat sink(the one posted in past images).
I chose Irfb4321 because it can easily withstand 18A rms at 100 degrees celsius and has acceptable gate charge for direct ir2110 driving and it is very close related to irfb4227 and irfb4228 wich are class D recomended.
irfp540z will fail in theese conditions or will be an the edge...what better mosfet do you suggest that will make efficiency go higher?
ChocoHolic:
I am using byv28-200 in series with a 2.2 ohm resistor.
Sorry ..they were wima MKP10 0.47uF..my mistake..
I measured efficiency by measuring output current,output voltage,input current and input voltage and making needed calculations(Pout/pin).
Fets run hot...at 16 amp rms (50W measured losses)they reach 90 degrees in 2 minutes with small computer smps heat sink(the one posted in past images).
I chose Irfb4321 because it can easily withstand 18A rms at 100 degrees celsius and has acceptable gate charge for direct ir2110 driving and it is very close related to irfb4227 and irfb4228 wich are class D recomended.
irfp540z will fail in theese conditions or will be an the edge...what better mosfet do you suggest that will make efficiency go higher?
ChocoHolic:
I am using byv28-200 in series with a 2.2 ohm resistor.
Last edited:
But you have admitted that switching losses dominate. Part of the swithcing losses always there, so it is usually to use FETs with a bit bigger rdson, but less Qrr, less Coss, less Qgate etc... since music is compromised of rare but high peaks, and low average power content.
The worst part of IRFB4321 is its diode, it has 300nC typical Qrr. At high current, with hard switch ringing, this have an important role.
I could reccmoned FDP3652, it has 16mOhm Rdson, and about the same Qgate and Qrr as IRF540Z. But of course it mostly depends on what you can get. The best part that Rdson is even stated for 175°C conditions. (I love good datasheets, where are more infomration tahan you can imagine). And the best of all, it has even lower Miller-charge than IRF540Z!
The best would be to make an Excel spreadsheet and calculate all losses. I could reccomend AN-1070 from IRF, or attached you will find an excel spreadsheet of mine. I made it a long time ago, based on AN-1070, so there can be errors in it. YOu can put the data of the FETs you can get into it, check the Excel function, according to AN-1070.