H
HAYK
The problem with CFP outputs is the double cut off of the two transistors separately. As the driver needs to bias the output, it needs extra current, hence switches off later linearizing the crossover current. To get a perfect crossover the currents must increase with square law. To get back square law, I simply added a biasing current to the output so that the drivers don't need the extra linearizing current.
Bellow shows the difference of P3a after Sakis @East electronic and my version.
The third curve is the output voltage of the amplifier while a triangular current generator pouring +/-2A.
Bellow shows the difference of P3a after Sakis @East electronic and my version.
The third curve is the output voltage of the amplifier while a triangular current generator pouring +/-2A.
Attachments
Interesting, having built a P3A i feel the urge to experiment, I also notice you reduced the emitter resistors to 0,05ohm from the original 0,33ohm is there a specific reason for that?
When looking at it I also wonder how necessary it is to use a constant current generator, the VBE variations of the output transistors would be very small compared to the total 70V supply voltage, could you just use a 12K resistor instead?
When looking at it I also wonder how necessary it is to use a constant current generator, the VBE variations of the output transistors would be very small compared to the total 70V supply voltage, could you just use a 12K resistor instead?
H
HAYK
The emitter resistor has two purposes, thermal and linearity. Normally with single transistors, 0.22 ohms is enough to provide negative feedback to bring the thermal positive feedback bellow 1 and avoid avalanche. This value depends on the gm, Ic/Vbe of the transitor. With CFP, the gm of the driver is multiplied by the Hfe of the output which makes the emitter resistor to be effective 60 times. 0.05 ohms gives as much negative current feedback as 3 ohms emitter resistance on condition that the drivers are thermally compensated and cooled separately from the outputs. For linearity, Indeed if you look the Sakis curve, it is more linear than mine crossover put aside, but the other stages are not yet adapted.
The rails if supplied by a regulated PSU, indeed a resistor is enough maybe with PTC even better, it gives 0.5 volt bias to the outputs which can be better to decrease with temperature.
The rails if supplied by a regulated PSU, indeed a resistor is enough maybe with PTC even better, it gives 0.5 volt bias to the outputs which can be better to decrease with temperature.
The Threshold "STASIS" used similar CFP outputs, but inserted small-valued series resistors in the emitters of Q6, Q7, Q8, and Q9 in the post#1 schematic. This idea trades off gm against linearity & stability.
You should have a look at this thread (etc) about the Monticelli output, which uses a ~cascode cross-coupling to linearize "rail-rail" outputs. See the thread https://www.diyaudio.com/community/threads/monticelli-ultra-efficient-ultra-low-distortion.401389/ .
This is used in some of the newer ultra-low distortion R-R audio op-amps. Using a small capacitor to cross-couple CE outputs is very simple and cheap and improves distortion but there is the issue of supply noise rejection, PSRR. The Monticelli circuit may not be as suited to discrete amplifiers as it is to ICs where device matching is a given.
This is used in some of the newer ultra-low distortion R-R audio op-amps. Using a small capacitor to cross-couple CE outputs is very simple and cheap and improves distortion but there is the issue of supply noise rejection, PSRR. The Monticelli circuit may not be as suited to discrete amplifiers as it is to ICs where device matching is a given.
H
HAYK
Degenerating the emitters is to linearize the square law function. What concerns linearity, is a secondary problem to deal with after the crossover is perfected.
This is a pre_bias current source using standard 10k NTC. Bolted upon one of the outputs, it decreases 20uA/°C, R3 adjusts the cold current, the zener is protection. I used BF862 as it can also be used in the LTP IPS .
This is a pre_bias current source using standard 10k NTC. Bolted upon one of the outputs, it decreases 20uA/°C, R3 adjusts the cold current, the zener is protection. I used BF862 as it can also be used in the LTP IPS .
Attachments
H
HAYK
I am looking for linear outputs to be better than C5200 A1943, in vain. What puzzles me is the njw1302g. The DS shows very linear Hfe curves for 20 Vce and 5 Vce to be about 100. The simulator says something different, bellow is the comparison draft1 is A1943. Any explanation?
I will try out IRFP240/9240 MOSFETS.
I will try out IRFP240/9240 MOSFETS.
Attachments
H
HAYK
The output stage can be called complimentary feedback pair only if it is driven with voltage, or else the feedback doesn't occur and the two transistors are just Hfe cascaded. This is why the bootstrapping as it presents lower impedance. I replaced the input transistors with dn2540 depletion MOSFETs which results 3 times less odd harmonics. In open loop mode, 1mv peak input results 23v output with 0.1% THD. In closed loop I get 0.0009%, as it is mainly second harmonic generated by the VAS, I left it as it is. Remark that the CCS is replaced by bootstraped sourced resistor, it replaces without any degradation.
I will change the compensation to get at least 60db NFB at 20khz.
The pre_bias CCS can also be made with the DN2540, as I have it with the 10k NTC, I will dimension it in real.
I will change the compensation to get at least 60db NFB at 20khz.
The pre_bias CCS can also be made with the DN2540, as I have it with the 10k NTC, I will dimension it in real.
Attachments
H
HAYK
It is possible to reduce the THD to 0.0005% by adjustments with odd harmonics dominance. I was right when I said the CFP can function only with a low impedance source. As I decreased the bootstrap feedback and decreased the VAS impedance, the THD fell to 0.00062%.
Attachments
Hi Hayk,
I ran your circuit using Post 1 attachment with Cordell models re-trim the idle current with R10 & I2 and mods for temp sensitivity of the driver transistors:
The capacitors are bypassed and V1 added to the bootstrap so they do not affect LF Gm plots. I was interested to see clip recovery and check slewing at 20kHz and 50kHz. I found cross-conduction at 20kHz when clip leaves the positive rail (see below):
"xC" is cross conduction causing a spike in Q9 (NPN) dissipation. But strangely, there is no xC in Q7 - so it some asymmetry. Increasing C6 to 680pF seems to fix this, and is still OK at 50kHz - although the crossover mid region dissipation is about 50% higher at 50kHz compared to 20kHz. (It may overheat the amp if left running at 50kHz for a long time but otherwise IMO not a serious problem).
Now the temp sensitivity. The driver transistors are very sensitive to temperature change - eg only 10 deg C increase caused the idle current to go from 240mA (as per your original) to 550mA. So you thermal linkage of the driver transistors to the bias needs to be quite precise and hopefully fast - and decoupling the driver heatsink from the main heatsink as you suggest seems the best. The main heatsink does not seem to need any thermal linkage to the bias generator but the current source does need to reduce with the main heatsink temperature to keep the power transistors very slightly ON with a few milliamps.
So I see your circuit is a great way to reduce crossover distortion.
But success in implementation is all to do with the thermal aspect - more than for a conventional output stage. Bench testing is essential.
The main reason for the high thermal sensitivity is the very low value emitter resistors R14, R16.
But low value emitter resistors requires more idle current for best linearity which is helpful because it gives a wider Class-A region (about a watt in your circuit) and better linearity or lower THD.
Keep up the great work!
IanH
I ran your circuit using Post 1 attachment with Cordell models re-trim the idle current with R10 & I2 and mods for temp sensitivity of the driver transistors:
The capacitors are bypassed and V1 added to the bootstrap so they do not affect LF Gm plots. I was interested to see clip recovery and check slewing at 20kHz and 50kHz. I found cross-conduction at 20kHz when clip leaves the positive rail (see below):
"xC" is cross conduction causing a spike in Q9 (NPN) dissipation. But strangely, there is no xC in Q7 - so it some asymmetry. Increasing C6 to 680pF seems to fix this, and is still OK at 50kHz - although the crossover mid region dissipation is about 50% higher at 50kHz compared to 20kHz. (It may overheat the amp if left running at 50kHz for a long time but otherwise IMO not a serious problem).
Now the temp sensitivity. The driver transistors are very sensitive to temperature change - eg only 10 deg C increase caused the idle current to go from 240mA (as per your original) to 550mA. So you thermal linkage of the driver transistors to the bias needs to be quite precise and hopefully fast - and decoupling the driver heatsink from the main heatsink as you suggest seems the best. The main heatsink does not seem to need any thermal linkage to the bias generator but the current source does need to reduce with the main heatsink temperature to keep the power transistors very slightly ON with a few milliamps.
So I see your circuit is a great way to reduce crossover distortion.
But success in implementation is all to do with the thermal aspect - more than for a conventional output stage. Bench testing is essential.
The main reason for the high thermal sensitivity is the very low value emitter resistors R14, R16.
But low value emitter resistors requires more idle current for best linearity which is helpful because it gives a wider Class-A region (about a watt in your circuit) and better linearity or lower THD.
Keep up the great work!
IanH
Attachments
H
HAYK
Hello Jan.
Thank you giving interest for this subject. The cross conduction is due to the Vbe multiplier, once bypassed by 100nF, the problem get solved. Bellow is 50khz saturared
Thank you giving interest for this subject. The cross conduction is due to the Vbe multiplier, once bypassed by 100nF, the problem get solved. Bellow is 50khz saturared
Hi Hayk,
Yes, adding 100nF across the Vbe multiplier fixed it. So is C6 needed? What is it for?
Are you seeing increased dissipation in Q7, Q8 at 50kHz with large swing or clip recovery?
If so, can this excess dissipation be reduced back to worst case dissipation somehow? (Worst case is at half full power - around 25W in your amp). I have had a BJT amp oscillate at 50kHz (due to stray coupling from the speaker lead to the input stages) and not realised it was oscillating until it overheated and blew (the speaker had no tweeter, just fizzer cone, so it survived, thanks to the output cap!).
Cheers, IanH
Yes, adding 100nF across the Vbe multiplier fixed it. So is C6 needed? What is it for?
Are you seeing increased dissipation in Q7, Q8 at 50kHz with large swing or clip recovery?
If so, can this excess dissipation be reduced back to worst case dissipation somehow? (Worst case is at half full power - around 25W in your amp). I have had a BJT amp oscillate at 50kHz (due to stray coupling from the speaker lead to the input stages) and not realised it was oscillating until it overheated and blew (the speaker had no tweeter, just fizzer cone, so it survived, thanks to the output cap!).
Cheers, IanH
H
HAYK
Concerning the temperature compensation. People consider that the Vbe multiplier drifts 2x 2mv/°C, in reality with high Hfe transistors as BC550C the drift is only 3mv/°C, with BD139 it is only 2mv where 4mv is required. This subject seams not much discussed. The NTC is a better choice as it allows to design the drift that you want even over compensate for heatsink-junction temperature difference. Bellow is a low impedance <1 ohm Vbe multiplier using the same NTC10k B3.977k to drift 4mv/°C.
ef2/ef3 , why do ya chase the cfp "ghost" ... it don't sound better. It's not 'faster" EF3 + CFA = 300V/us.
"mental masturbation" ??
"mental masturbation" ??
H
HAYK
Normal CFP is not better then EF as D. Self concluded in his book designing the Blamless. Here I am introducing a new approach that makes the class AB output get as linear as a class A can. You can't get with EF.
Sakis @east electronics who repairs over a 1000 amplifiers a year, preaches the best sounding amp he has heard to be P3a of Rode Elliot.
Myself in 86, I designed one 35W with CFP output gain driven by TL82 to be bedside FM/CD it had very agreeable sound.
I remembered that the driver and the power should be nearest possible to each other to avoid oscillations, as the loop frequency is very high. Then, the two drivers being far from each other, can't be mounted on the same heatsink. I designed bellow with BD139 to be mounted upon one driver and the NTC upon the other as each drift 2mv/°C.
Sakis @east electronics who repairs over a 1000 amplifiers a year, preaches the best sounding amp he has heard to be P3a of Rode Elliot.
Myself in 86, I designed one 35W with CFP output gain driven by TL82 to be bedside FM/CD it had very agreeable sound.
I remembered that the driver and the power should be nearest possible to each other to avoid oscillations, as the loop frequency is very high. Then, the two drivers being far from each other, can't be mounted on the same heatsink. I designed bellow with BD139 to be mounted upon one driver and the NTC upon the other as each drift 2mv/°C.
I have been drawing and making prototypes like these for many years after being introduced to this technique with the Blomley circuit the idea is that the output devices never turn off Then the faster drivers can then do their thing
H
HAYK
The speed is not a problem, the driver pours in the needed while the output wakes up. If too fast, the opposite still conducting it will shoot through, by the way, the input filter capacitor must be higher to avoid high slope signals.
That is technically impossible. Class AB relies on non-linearity for its operation.Here I am introducing a new approach that makes the class AB output get as linear as a class A can.
Your driver transistors cut-off.
Ed
- Home
- Amplifiers
- Solid State
- Linearizing the CFP crossover - P3a of Rod Elliot as example