I have gained full understanding of the cross-conduction problem and I have solved it too.
Cross-conduction was arising because of two pitfalls:
1 - The discrete splitter and level-shifter that I use to carry the output of the LM311 to the inputs of the IR2010 does not always take the same time to turn on than to turn off. When the output of the LM311 toggles too often, like twice in a 500ns period, some overlap happens at the inputs of the IR2010. This is because the circuit uses only 3mA collector currents and 5.6k load resistors. Overlap arises because input capacitances of the IR2010 and capacitances of the level shifting transistors are charged quite fast by those 3mA on the rising edges, but they discharge a bit slower and exponentially through the 5.6k on the falling edges. A minimum of 500ns clearance between LM311 toggling is required.
2- Cross-conduction only happened when coming out of clipping. I investigated this further and I found that the poor little LM311 was suffering brief indecision problems 😀😀😀 during the first oscillation period after clipping events. This happened because in this first switching period the amount of carrier residual being sensed at the output of the amplifier is so small that anything else, like a few milivolts of EMI and noise, can fool the comparator and toggle it. Indeed, primary-side turn-off EMI spikes radiated by the PSU toroid were making the problem worse.
Pitfall no.1 would be solved by replacing the 5.6k load resistors of the level shifter by 1mA current sinks (with anti-clipping diodes), but this adds a lot of complexity to the circuit and I don't feel like throwing away the existing channel boards. Also, the level shifter would work fine if the LM311 wasn't producing too short pulses.
So I tried to solve pitfall no.2 by adding a little bit of RF hysteresis to the LM311 by means of a 2.4pF capacitor from collector output to IN+. This seems to be working fine as I can no longer see any traces of abnormal LM311 toggling or the resulting cross-conduction pulses.
Next time I will do a better level shifter with as little "memory effect" on short pulses as possible. Uniform delays everywhere are a must for low distortion anyway.
Level shifter schematic:
On the other hand, I have rewound the PSU toroid transformer in a more tight way and I have soldered it back to the PCB with no extra magnet wire lenght. The push-pull primaries are now truly bi-filar. As a result, the resonance of leakage inductance togerher with primary and secondary side capacitances has risen from 15Mhz to 22Mhz. This means that I have achieved a 2:1 reduction in leakage inductance 😀😀😀 I also increased secondary turns from 13+13 to 14+14 to get some more power, now rails are +/-36V and they only sag to +/-31V. Peak current in the primary side is now approx. 54A yet the Vds turn-off spikes of the IRFZ48V hardly reach 30V.
Some pictures of the improved transformer:
Cross-conduction was arising because of two pitfalls:
1 - The discrete splitter and level-shifter that I use to carry the output of the LM311 to the inputs of the IR2010 does not always take the same time to turn on than to turn off. When the output of the LM311 toggles too often, like twice in a 500ns period, some overlap happens at the inputs of the IR2010. This is because the circuit uses only 3mA collector currents and 5.6k load resistors. Overlap arises because input capacitances of the IR2010 and capacitances of the level shifting transistors are charged quite fast by those 3mA on the rising edges, but they discharge a bit slower and exponentially through the 5.6k on the falling edges. A minimum of 500ns clearance between LM311 toggling is required.
2- Cross-conduction only happened when coming out of clipping. I investigated this further and I found that the poor little LM311 was suffering brief indecision problems 😀😀😀 during the first oscillation period after clipping events. This happened because in this first switching period the amount of carrier residual being sensed at the output of the amplifier is so small that anything else, like a few milivolts of EMI and noise, can fool the comparator and toggle it. Indeed, primary-side turn-off EMI spikes radiated by the PSU toroid were making the problem worse.
Pitfall no.1 would be solved by replacing the 5.6k load resistors of the level shifter by 1mA current sinks (with anti-clipping diodes), but this adds a lot of complexity to the circuit and I don't feel like throwing away the existing channel boards. Also, the level shifter would work fine if the LM311 wasn't producing too short pulses.
So I tried to solve pitfall no.2 by adding a little bit of RF hysteresis to the LM311 by means of a 2.4pF capacitor from collector output to IN+. This seems to be working fine as I can no longer see any traces of abnormal LM311 toggling or the resulting cross-conduction pulses.
Next time I will do a better level shifter with as little "memory effect" on short pulses as possible. Uniform delays everywhere are a must for low distortion anyway.
Level shifter schematic:
On the other hand, I have rewound the PSU toroid transformer in a more tight way and I have soldered it back to the PCB with no extra magnet wire lenght. The push-pull primaries are now truly bi-filar. As a result, the resonance of leakage inductance togerher with primary and secondary side capacitances has risen from 15Mhz to 22Mhz. This means that I have achieved a 2:1 reduction in leakage inductance 😀😀😀 I also increased secondary turns from 13+13 to 14+14 to get some more power, now rails are +/-36V and they only sag to +/-31V. Peak current in the primary side is now approx. 54A yet the Vds turn-off spikes of the IRFZ48V hardly reach 30V.
Some pictures of the improved transformer:
TOINO said:
In my opinion Anti-clip or clip limiting are just "BAND-AID SOLUTIONS"........what I prefer is 0-100% dutycycle approach.......Thats something which accounts for reliability[though some fools would like to say that for audio amp full dutycycle approach has no relevance, that simply describe their inability to get the things done under control]
Eva said:
What is your switching frequeny and what is the slew-rate involved ?
I once wanted to use such an approach but refrained from doing so due to the high coupling capacity of such converters (a few tens of pF) and the possible current spikes through this capacity caused by the high slew rates involved.
Regards
Charles
I have been doing THD measurements with a noisy built-in laptop soundcard.
In this first attempt the output of the amplifier was attenuated and fed to the line-in of the laptop, without any attempt to remove carrier residuals. The THD (plus noise) floor of the laptop seems to be between 0.025% and 0.035%.
Dummy load was 3 ohms (resistor bank). A single channel was driven. Supply rails were +/-36V sagging to +/-31V at full power. Power supply was a heavy duty 12V 17V Ah battery.
I measured at 100Hz and 1Khz and I got similar results.
Until +/-10V output (16W), the THD of the laptop seems to dominate and I get a reading between 0.025% and 0.035%
Between +/-10V and -3dB full power (66W), THD increases slowly up to 0.1%.
Between -3dB and -2dB THD increases to 0.25%.
Between -2dB and -1dB THD increases to 0.5%.
Between -1dB and the threshold of clipping as seen on oscilloscope THD increases to 1%.
The self oscillating frequency compensation is not optimum. With UcD style compensation, the frequency was dropping to 55Khz before clipping. Now I'm using my own style of compensation, but the frequency still drops from 320Khz to 100Khz before clipping. In simulation the optimum ratio is around 2.5*:1, but that compensation would produce over -3dB at 20Khz, thus requiring additional EQ.
I will try to use an active filter to avoid feeding the laptop with stuff above 20Khz. The carrier residuals are probably been aliased by the ADC of the laptop in the same way that the amplifier aliases all the RF junk that comes out through the line-out of the laptop.
EDIT: I forgot to mention that I went back to a slower gate drive with approx 10ns dead time because I wanted to reduce quiescent current due to hard switching (0.83A before, 0.67A after). Remember that this is going to be used in a motorbike!!
In this first attempt the output of the amplifier was attenuated and fed to the line-in of the laptop, without any attempt to remove carrier residuals. The THD (plus noise) floor of the laptop seems to be between 0.025% and 0.035%.
Dummy load was 3 ohms (resistor bank). A single channel was driven. Supply rails were +/-36V sagging to +/-31V at full power. Power supply was a heavy duty 12V 17V Ah battery.
I measured at 100Hz and 1Khz and I got similar results.
Until +/-10V output (16W), the THD of the laptop seems to dominate and I get a reading between 0.025% and 0.035%
Between +/-10V and -3dB full power (66W), THD increases slowly up to 0.1%.
Between -3dB and -2dB THD increases to 0.25%.
Between -2dB and -1dB THD increases to 0.5%.
Between -1dB and the threshold of clipping as seen on oscilloscope THD increases to 1%.
The self oscillating frequency compensation is not optimum. With UcD style compensation, the frequency was dropping to 55Khz before clipping. Now I'm using my own style of compensation, but the frequency still drops from 320Khz to 100Khz before clipping. In simulation the optimum ratio is around 2.5*:1, but that compensation would produce over -3dB at 20Khz, thus requiring additional EQ.
I will try to use an active filter to avoid feeding the laptop with stuff above 20Khz. The carrier residuals are probably been aliased by the ADC of the laptop in the same way that the amplifier aliases all the RF junk that comes out through the line-out of the laptop.
EDIT: I forgot to mention that I went back to a slower gate drive with approx 10ns dead time because I wanted to reduce quiescent current due to hard switching (0.83A before, 0.67A after). Remember that this is going to be used in a motorbike!!
Sometimes it even pays to use a passive filter first in order to avoid slew-induced distortion in the active filter !
Regards
Charles
Eva said:
How could you expect everybody to remain calm with what you have done is simply igniting the fires........
😀 😀 😀 😉
phase_accurate said:
did 2000V in 140nS (I posted scope shots made with an HP Infinium oscilloscope and a 2500V 100:1 100MHz probe in a previous thread). WRT the coupling capacitance current spikes, they seem to measure less than the "60pF isolation capacitance" spec listed on the datasheet would indicate, and I'm just living with it.
I've tried really hard to blow the converters up with fast switching, but they just keep chugging away without complaint.
Cheers,
Glen
G.Kleinschmidt said:
So your experience with encapsulated DC-DC Converter has met with considerable success
G.Kleinschmidt said:
Thank you for your further abus...uhm...testing. Do you mind sharing which low cost device this DC converter is?
Workhorse said:
Hmmm… I didn’t say that anti clip and 0-100% duty-cycle is mutually exclusive Mr. Smart…[though some fools would like to say that heavy clipping amplifier sound is healthy for loudspeaker and ears]
This discussion is off-topic here, but if an input signal limiter is a band aid, then, aren't band aids too things like soft-start, current limiting, overheating shutdown, over-voltage shutdown, under-voltage lockout, etc.. ??? And what about temperature-controlled fans? 😀😀😀
BTW: Regulated power supplies? 😀
BTW: Regulated power supplies? 😀
Ouch!!
List price for one of those converters, a 4W 12V to 12V version, model: HB04U12S12Q , is US $54 !!!
Link: http://www.mouser.com/catalog/625/1511.pdf
And they are not even regulated !!! No load output voltage is 15.2V and it falls to 11.7V at full load. Note that there are not dV/dt ratings and that isolation is rated at 60Hz.
Datasheet: http://www.cd4power.com/data/power/ncl/tdc_hb04u.pdf
No, thanks, I'm not of that kind of people that has to pay US $54 to get 4 watts of "floating" unregulated DC power... See picture...
BTW: The first big SMPS prototype that I did (2KW, long time ago) already employed 5 fully-isolated active-buffered gate-drive cells that received both drive signals and power through the same drive transformer... I'm tired of winding transformers (the configuration shown produces high leakage inductance but it's OK when you want to pass a few mA of power to a floating circuit with absolute minimum capacitance and no dV/dt problems). I ended up not liking transformers because they take a lot of time to specify and build and custom models are expensive...
List price for one of those converters, a 4W 12V to 12V version, model: HB04U12S12Q , is US $54 !!!
Link: http://www.mouser.com/catalog/625/1511.pdf
And they are not even regulated !!! No load output voltage is 15.2V and it falls to 11.7V at full load. Note that there are not dV/dt ratings and that isolation is rated at 60Hz.
Datasheet: http://www.cd4power.com/data/power/ncl/tdc_hb04u.pdf
No, thanks, I'm not of that kind of people that has to pay US $54 to get 4 watts of "floating" unregulated DC power... See picture...
BTW: The first big SMPS prototype that I did (2KW, long time ago) already employed 5 fully-isolated active-buffered gate-drive cells that received both drive signals and power through the same drive transformer... I'm tired of winding transformers (the configuration shown produces high leakage inductance but it's OK when you want to pass a few mA of power to a floating circuit with absolute minimum capacitance and no dV/dt problems). I ended up not liking transformers because they take a lot of time to specify and build and custom models are expensive...
Attachments
....hm, my selfmade flyback shows a capacitve coupling around 15pF.... And with a changed transformer I should be able to go below 10pF... This is looking less catastrophic than expected. So in first step I will simply try to use my existing flyback proto and see how it performs in floating.
- Status
- Not open for further replies.