The usual resonances between MOSFET capacitances, magnetic snubber primary winding capacitances, MOSFET lead inductances and supply capacitor inductances are making me work hard as usual to tame the resulting resonant peaks.
According to my measurements and math, the 1500uF 250V supply electrolytics have 16nH inductance that combines with the rest of the parasitistics to resonate at 27Mhz. This resonance is excited by fast switching transients and causes trouble because the "center tap" of the supply capacitors "moves" too resulting in plenty of common-mode EMI. The first goal is to damp the ringing because a spike is not that bad when it does not have an oscillation tail. The second goal is to keep the "center tap" at ground plane potential at RF.
Switching MOSFETs are protected against resonance by means of turn-on ferrite beads connected in series with the gates. Turn-off is brute force through a schottky diode, which seems to produce the best results. I like the beads because they result in a slower gate charging to the threshold voltage (thus less RF excitation is produced) but a faster charge to the final voltage later (thus Rds-on is reached faster).
Most people would just suggest sticking a small film cap in parallel with the electrolytics, and this is what I tried... I took a pair of 100nF 250V MKT stacked-film caps from Philips with ultra short leads and I soldered them directly on the leads of the electrolytics. As a result I shifted the main supply resonance up to 38Mhz and got another ugly resonance at 3.2Mhz that causes a lot of trouble when duty cycles become short and it gets excited periodically enough for resonance "buildup"... Paralleling caps in class D circuits is particularly funny.
Again, according to math and measurements, my 100nF 250V stacked-film MKTs exhibit 8nH so they are of little help and don't represent a particular improvement over the big electrolytics, whose impedance is already low enough up to 10Mhz.
As I don't want to give up too much switching speed or end up wasting a few watts with dissipative RC snubbers across the MOSFETs (the usual solutions), I'm going to try SMD ceramic chip capacitors. According to datahsheets, 10nF 500V X7R chips seem to resonate at 40Mhz and five of these connected in parallel will result in a non-inductive supply impedance below 0.15 ohms in the 20 to 100Mhz range. The 100nF 250V caps will be also required, but with 0.33 ohm in series. Their new purpose is just to damp the huge 5.6Mhz resonance that arises from paralleling the electrolytics and the SMD ceramics (this is according to simulation since I haven't got the SMD chips yet).
Finally, I also tried paralleling one of those fancy 100nF 250V MKT stacked-films with the 2.2uF 250V wound-film output capacitor, which I believe to be inductive. I soldered it for minimum layout inductance. This certainly reduced the magnitude of the "blips" of high frequency leaking on the output carrier-residual sine wave, but it also resulted in a 1Mhz-or-so resonance that became triggered everytime the duty cycle became low (or high), thus screwing up the self oscillating modulator completely. I had to put a 3.3 ohm resistor in series with the 100nF film to avoid the resonance completely.
BTW: I was also curious about how much capacitive coupling were the MPP cores providing between the beggining and the end of the windings (2mm clearance, wire tie method) so I arranged one core with 42 turns evenly spread, then I cut the winding in two 21-turn halves and measured only 27pF between each half. I think that this is too low to be a problem.
According to my measurements and math, the 1500uF 250V supply electrolytics have 16nH inductance that combines with the rest of the parasitistics to resonate at 27Mhz. This resonance is excited by fast switching transients and causes trouble because the "center tap" of the supply capacitors "moves" too resulting in plenty of common-mode EMI. The first goal is to damp the ringing because a spike is not that bad when it does not have an oscillation tail. The second goal is to keep the "center tap" at ground plane potential at RF.
Switching MOSFETs are protected against resonance by means of turn-on ferrite beads connected in series with the gates. Turn-off is brute force through a schottky diode, which seems to produce the best results. I like the beads because they result in a slower gate charging to the threshold voltage (thus less RF excitation is produced) but a faster charge to the final voltage later (thus Rds-on is reached faster).
Most people would just suggest sticking a small film cap in parallel with the electrolytics, and this is what I tried... I took a pair of 100nF 250V MKT stacked-film caps from Philips with ultra short leads and I soldered them directly on the leads of the electrolytics. As a result I shifted the main supply resonance up to 38Mhz and got another ugly resonance at 3.2Mhz that causes a lot of trouble when duty cycles become short and it gets excited periodically enough for resonance "buildup"... Paralleling caps in class D circuits is particularly funny.
Again, according to math and measurements, my 100nF 250V stacked-film MKTs exhibit 8nH so they are of little help and don't represent a particular improvement over the big electrolytics, whose impedance is already low enough up to 10Mhz.
As I don't want to give up too much switching speed or end up wasting a few watts with dissipative RC snubbers across the MOSFETs (the usual solutions), I'm going to try SMD ceramic chip capacitors. According to datahsheets, 10nF 500V X7R chips seem to resonate at 40Mhz and five of these connected in parallel will result in a non-inductive supply impedance below 0.15 ohms in the 20 to 100Mhz range. The 100nF 250V caps will be also required, but with 0.33 ohm in series. Their new purpose is just to damp the huge 5.6Mhz resonance that arises from paralleling the electrolytics and the SMD ceramics (this is according to simulation since I haven't got the SMD chips yet).
Finally, I also tried paralleling one of those fancy 100nF 250V MKT stacked-films with the 2.2uF 250V wound-film output capacitor, which I believe to be inductive. I soldered it for minimum layout inductance. This certainly reduced the magnitude of the "blips" of high frequency leaking on the output carrier-residual sine wave, but it also resulted in a 1Mhz-or-so resonance that became triggered everytime the duty cycle became low (or high), thus screwing up the self oscillating modulator completely. I had to put a 3.3 ohm resistor in series with the 100nF film to avoid the resonance completely.
BTW: I was also curious about how much capacitive coupling were the MPP cores providing between the beggining and the end of the windings (2mm clearance, wire tie method) so I arranged one core with 42 turns evenly spread, then I cut the winding in two 21-turn halves and measured only 27pF between each half. I think that this is too low to be a problem.
Hi Eva,
your cap resonances are fitting perfectly to my unpleasant findings.... in the meantime I am becoming a fan of the SMD chip caps with X7R ceramic. Geometry results in low L and the lossy X7R results in nice damping of HF resonances. If you have a good reason not to go this path ( except costs & handling/cracking) then please let me know. (Don't need to rediscover all pit falls on my own...
)
your cap resonances are fitting perfectly to my unpleasant findings.... in the meantime I am becoming a fan of the SMD chip caps with X7R ceramic. Geometry results in low L and the lossy X7R results in nice damping of HF resonances. If you have a good reason not to go this path ( except costs & handling/cracking) then please let me know. (Don't need to rediscover all pit falls on my own...
)...for the output filter I would also not like the X7R, but for all the rest.... and ooohps, I did not read your last past carefully, now I see that you are also using the X7R. Nice to see.
...hm, would there be an advantage in paralleling multiple smaller X7R, or would the PCB inductance cause to much issues?
...hm, would there be an advantage in paralleling multiple smaller X7R, or would the PCB inductance cause to much issues?
Hi Choco. Nice to hear from you. I still don't have 500V X7R chips but after reviewing the datasheets I have decided to order some 10nF parts (resonance close to 40Mhz) to parallel them. Paralleling a few small (1206) identical parts seems like the best alternative to lower RF supply impedance. Bigger 100nF 500V X7R are more inductive (they come in a larger package) and seem to exhibit higher ESR (they are probably more expensive too). I promise to explain my findings once I try them
Concerning PCB parasitistics, my usual almost-continuous power-ground plane seems to create a very low inductance path and reduces track inductance substantially as expected. The RF voltage drops that I have measured across the ground plane are at least 20 times lower than the RF voltage drops that I'm seeing now across supply capacitors.
BTW: I only have 50V X7R parts but I haven't tried these either (I recently bought a 32-value 2000-part kit on ebay from some... Chinese!! seller for quite cheap )
Concerning PCB parasitistics, my usual almost-continuous power-ground plane seems to create a very low inductance path and reduces track inductance substantially as expected. The RF voltage drops that I have measured across the ground plane are at least 20 times lower than the RF voltage drops that I'm seeing now across supply capacitors.
BTW: I only have 50V X7R parts but I haven't tried these either (I recently bought a 32-value 2000-part kit on ebay from some... Chinese!!
seller for quite cheap )... not everything from China is low quality. You can also get good components for low price here. Hope you had good luck with your choice. In any case you have to take care regarding cracks in the X7R ceramic....
Currently I also do not have high voltage X7R on hand. For my first trials I built a 100nF/150V cap from 9 pieces 100nF/50V (each 0805 size) and already with this far from optimum setup, I got quite improved results vs wired MKTs... (see my Gen2 thread at DIYHIFI).
Looking forward to your findings
Currently I also do not have high voltage X7R on hand. For my first trials I built a 100nF/150V cap from 9 pieces 100nF/50V (each 0805 size) and already with this far from optimum setup, I got quite improved results vs wired MKTs... (see my Gen2 thread at DIYHIFI).
Looking forward to your findings
ChocoHolic said:... not everything from China is low quality. You can also get good components for low price here. Hope you had good luck with your choice. In any case you have to take care regarding cracks in the X7R ceramic....
Currently I also do not have high voltage X7R on hand. For my first trials I built a 100nF/150V cap from 9 pieces 100nF/50V (each 0805 size) and already with this far from optimum setup, I got quite improved results vs wired MKTs... (see my Gen2 thread at DIYHIFI).
Looking forward to your findings
choco
back to guangzhou again?
i have 100v 1uf(2024 perhaps) and 0.1uf(1206) smd caps , good for my mhz class-d (500w), but not from china.
if u need , call me
eva , the queen
i think you should change ur project to smd , expensive, but better than dip.
best rg
fumac
Now I'm doing my tests with 150V (equivalent to +/-75V), music and an old 8 ohm 3-way speaker as a load (10 inch woofer), and all that I get is painfully loud output with a dead cool heatsink (3ºC over ambient), despite the 0.2ohm rds-on of the SPW20N60CFD (I also have SPW35N60CFD but these are too expensive to risk destroying them during experimentation). I suppose that the cross-conduction free operation provided by the magnetic snubber, the optimized gate drive, the 80Khz operating freq and the crazy fast switching have to do with this.
My gate drive scheme is explained in Chocoholic's thread in the other forum, it uses only a ferrite bead and a schottky diode.
Note that this amplifier was only intended for 50Hz-5Khz operation, but it produces at least moderate levels of 15Khz trebble with no visible waveform degradation (loud trebble results in slew-rate limiting, though). The filter is resonating at 10Khz or so but this does not alter at all the closed-loop frequency response. The oscillation frequency is still 48Khz when output is +/-60V (15V away from the rails) and drops to 24Khz just before clipping. A frequency sweep watched on the oscilloscope in dual trace mode reveals ruler flat (better than +/-1dB) response from 20Hz to 24Khz with no phase shift at high frequencies and no peaking. At 20Khz and 30Vp-p the output is still sine shaped.
My gate drive scheme is explained in Chocoholic's thread in the other forum, it uses only a ferrite bead and a schottky diode.
Note that this amplifier was only intended for 50Hz-5Khz operation, but it produces at least moderate levels of 15Khz trebble with no visible waveform degradation (loud trebble results in slew-rate limiting, though). The filter is resonating at 10Khz or so but this does not alter at all the closed-loop frequency response. The oscillation frequency is still 48Khz when output is +/-60V (15V away from the rails) and drops to 24Khz just before clipping. A frequency sweep watched on the oscilloscope in dual trace mode reveals ruler flat (better than +/-1dB) response from 20Hz to 24Khz with no phase shift at high frequencies and no peaking. At 20Khz and 30Vp-p the output is still sine shaped.
Ouroboros said:
Who? I swear I didn't
UPDATE:
I have been working in other similar but higher priority projects for the past two months.
However, now I have got plenty of SMD chip capacitors to cut parasitistic inductance and solve the ringing problem, including luxury stuff like 220nF 200V, 100nF 500V, and 220nF 630V (dual chip TDK which I manually split to get two 100nF). I have also developed a new magnetic snubber with much lower winding losses due to eddy currents, much lower shunt capacitance and no longer involving flyback peaks over 2KV. Furthermore, I have developed an active clamp circuit that recovers the energy from magnetic snubber leakage inductance too.
I will apply these new resources to this project as soon as I finish the higher priority stuff. I think that the future is quite promising, for example, with the new magnetic snubber and active clamp the switching frequency may be increased to 200Khz without too high losses.
Check the datasheet, TDK P/N is CKG45NX7R2J224M :
http://www.tdk.co.jp/tefe02/e415_ckg.pdf
I got them from RS-Amidata Spain. They were expensive but I needed something rated at 630V.
http://www.tdk.co.jp/tefe02/e415_ckg.pdf
I got them from RS-Amidata Spain. They were expensive but I needed something rated at 630V.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.