D. Self has started designing his Blamless amp by using CFP output in 1993.
http://www.douglas-self.com/ampins/dipa/dipa.htm
Rode Elliot transformed the CCS into bootstrap to call it P3 and in 2007 the VAS got upside down with with PNP to be called P3a.
The P3a is highly appreciated for its sound the Sakis @east electronics who repairs over 1000 amps a year has never heard any amp as good.
Recently I could considerably enhanced the CFP output stage to be as good as a class A can be, so I will take back the aborted Blamless and design it for best sounding.
The P3a will be the starting point to start enhancing, as it is a tuned amplifier.
https://sound-au.com/project03.htm
It has an NFB of 58db up to 1khz, the harmonc distortion is even order dominant.
This is the circuit of the actual amp that will be regularly updated
The CFP output stage is elaborating on, https://www.diyaudio.com/community/...rossover-p3a-of-rod-elliot-as-example.402148/
http://www.douglas-self.com/ampins/dipa/dipa.htm
Rode Elliot transformed the CCS into bootstrap to call it P3 and in 2007 the VAS got upside down with with PNP to be called P3a.
The P3a is highly appreciated for its sound the Sakis @east electronics who repairs over 1000 amps a year has never heard any amp as good.
Recently I could considerably enhanced the CFP output stage to be as good as a class A can be, so I will take back the aborted Blamless and design it for best sounding.
The P3a will be the starting point to start enhancing, as it is a tuned amplifier.
https://sound-au.com/project03.htm
It has an NFB of 58db up to 1khz, the harmonc distortion is even order dominant.
This is the circuit of the actual amp that will be regularly updated
The CFP output stage is elaborating on, https://www.diyaudio.com/community/...rossover-p3a-of-rod-elliot-as-example.402148/
The first thing to consider is the PSRR. It is "horrorifying" particularly the +rail that beside the supply ripples it provokes bad feedback from the output current. Bellow is the circuit with PSRR original right up, with diode 100 resistor and two 1000uF, right bellow and capacitor multiplier on the left.
The -rail needs also to be cleaned. The bad item is the CCS. I replaced it with much simpler and higher impedance. If the bootstrap is to be also replaced by a CCS as D. Self wanted it, the diode and the RC maybe canceled.
The up right is the original, bellow it is with diode, 500 ohm and 1000uF, on left is thar of the shown circuit.
The -rail needs also to be cleaned. The bad item is the CCS. I replaced it with much simpler and higher impedance. If the bootstrap is to be also replaced by a CCS as D. Self wanted it, the diode and the RC maybe canceled.
The up right is the original, bellow it is with diode, 500 ohm and 1000uF, on left is thar of the shown circuit.
The input stage.
What I don't like is the DC stopper electrolytic capacitor. If I look at a P3a board, there is a huge input 4.7uF polypropylene capacitor. As if someone exige crocodile leather shoe on one foot and accepts a slipper on the other.
The solution for this was proposed by Dudek the circuit bellow.
It works only if the input has low bias current.
Another problem presented by Sakis, is mismatch of the pair or offset that un balances the inputs, the sound isn't at its best. This is due to no degenerated emitters.
To give a remedy for these problems, FET transistors are much better suited. The consequence of using FET is great loss of gm. The image bellow gives a comparative gm for 1.3ma and 2.6ma bias currents. The result is loss of 10db in the loop gain.
Solution, is either CPF the inputs, this needs two extra transistors and increase the odd harmonics or render the VAS darlington, which will increase the even ones. I will chose the darlington as it gives 10db more gain than the BJT version. The input transistors I think of 2N7000 or K170, the voltage of BF862 is too low it will need bypassed zener.
What I don't like is the DC stopper electrolytic capacitor. If I look at a P3a board, there is a huge input 4.7uF polypropylene capacitor. As if someone exige crocodile leather shoe on one foot and accepts a slipper on the other.
The solution for this was proposed by Dudek the circuit bellow.
It works only if the input has low bias current.
Another problem presented by Sakis, is mismatch of the pair or offset that un balances the inputs, the sound isn't at its best. This is due to no degenerated emitters.
To give a remedy for these problems, FET transistors are much better suited. The consequence of using FET is great loss of gm. The image bellow gives a comparative gm for 1.3ma and 2.6ma bias currents. The result is loss of 10db in the loop gain.
Solution, is either CPF the inputs, this needs two extra transistors and increase the odd harmonics or render the VAS darlington, which will increase the even ones. I will chose the darlington as it gives 10db more gain than the BJT version. The input transistors I think of 2N7000 or K170, the voltage of BF862 is too low it will need bypassed zener.
The VAS, I will do it D. Self way with CCS as I found very simple way to do. It has -100db PSRR without any rail filter and this will be 35db lower with overall feedback. The Amp will be nested feedback the iner one is shown with BD140, 2SB647a and KSA1381 responces highest is KSC, lowest is the BD. Only the later needs some pF Miller.
The distortion is mainly even order, to be highest with BD 4 times lower with 2SB and 10 times lower with KSA. The distortion can be increased by adding a load resistor, here 80kohm load represents reflected 8ohms.
The distortion is mainly even order, to be highest with BD 4 times lower with 2SB and 10 times lower with KSA. The distortion can be increased by adding a load resistor, here 80kohm load represents reflected 8ohms.
This is the closed loop response of the amplifier.
To remind from precedent post the open loop response, you can see how they are identical except the feedback decreased the gain without touching the frequency character of the amplifier as no any compensation has been applied.
How is this possible?
Attachments
These are the transients without and with 100nF upon 8 ohms load. The scale is 500ns. Note that this amp doesn't need Zobel or Thiel. I biased the VAS to insane 30ma as the 2sa/2sb can go down to 1V with 100ma, it sources the OPS with 4 ohms . The result is high slew rate saturation and very low OPS distortion.
Attachments
Last edited:
You may find strange to use a virtual ground at the input to provoke ground circuit breaker.
Designers consider the supply ground as zero impedance, and it is not at all specially with linear PSU. At first view the reservoir capacitors make the impedance, but every one knows when listening at low level when these capacitors have far more charge than needed, it gets better bass with higher capacitance,why? Positive feedback. When the positive input is grounded, it increases the impedance of the supply ground through the load, this why the reservoir capacitors act as if they are higher impedance and obliges to increase the value. This phenomenon doesn't exist with single supply bridged amps as the ground is always Vcc/2. Here, the security capacitor 100nF and 2x3.3k determine the frequencies to follow the virtual ground or the supply ground, by this the offset problem is avoided.
Attachments
I found it needs a Zobel when a cable+speaker model is used (attached). Open J3 to disable the Zobel and it oscillates without the Zobel and with a speaker+ cable model.
Closing J2 enables a Zobel at the speaker end, but this cannot stop the oscillation. I have found some other amps that have Zobel plus the output inductor will oscillate with this speaker+ cable model when J2 is open (no Zobel at the speaker end) but be stable with a Zobel at the speaker end.
Previous discussion on cable plus speaker modelling is here https://www.diyaudio.com/community/...peaker-cable-interactions.346475/post-6003221 and Cyril Batemans unpublished articles links are given here https://www.diyaudio.com/forums/sol...r-amplifier-book-post5990267.html#post5990267.
In the attached I use a 20m cable model of Cyril Bateman's home brew #55 cable (the #55 refers to the Raychem tubing he used). This is deliberately a long cable to give the "worst case" likely for a home setup - so if you don't see an oscillation with this cable (and no Zobel at the speaker end) then you can be fairly confident that it is a rock-solid stable amp for anyone to use with any cable type.
But if you know your cable length is much shorter and more like "tame" zip cord then you probably don't need to use a transmission line cable model. A transmission line has the peculiar characteristic of "inverting" the driving impedance to the cable when the terminating speaker end has a high impedance at ultrasonic frequencies (which is typical) - giving low (near shorting) impedances at very high frequencies much like a capacitor but with a time delay, which can have different different stability issues than a capacitor across the load at the amp itself. Bob Cordell has said (and its in his book) that these strange cable induced oscillations make some amplifier's sound different, and these oscillations are often hard to find, but usually can be stopped or prevented by adding the extra Zobel at the speaker end.
I eliminated the darlington VAS. By increasing the currents I could get same result, even better in distortion spectrum which is 0.025%THD. It is similar to Aksa's Nirvana. The PSRR is -85db for 100Hz.
The output needs just 50nH inductance to be perfect with 100nF load that can be obtained just by wiring from PCB to the chasis.
It looks more P3a with CCS.
Remark that there is no any compensation, just input RF filter.
The outputs will be MJL ones identical but plastic casing.
The output needs just 50nH inductance to be perfect with 100nF load that can be obtained just by wiring from PCB to the chasis.
It looks more P3a with CCS.
Remark that there is no any compensation, just input RF filter.
The outputs will be MJL ones identical but plastic casing.
Last edited:
I created this semi floating PSU to get powerful bass and better PSRR. Without capacitor multiplier, it has -90db PSRR @100Hz. The 10,000uF are rectifying while the 2200uF are low esr audio capacitors.
The generator and its replicat represent the secondary of the transformer and the L3 1mH 1ohm represents the central line linked to ground by a fuse and 4 ohms. If the amp is defect, the DC current passes through the resistor and the fuse to protect the loudspeaker.
I will add current limiters to protect the amp from shorted output.
The generator and its replicat represent the secondary of the transformer and the L3 1mH 1ohm represents the central line linked to ground by a fuse and 4 ohms. If the amp is defect, the DC current passes through the resistor and the fuse to protect the loudspeaker.
I will add current limiters to protect the amp from shorted output.
Attachments
I wanted to see the residual ripple on the output. To help the simulater start up, I added 30v supplies meanwhile the AC PSU charge the capacitors. Unexpectedly, the negative rail was the bad guy poluting the output with ripples with less than -56db rejection. For now, I don't understand why high impedance current sources can pass ripples when by applying a generator it showed very high rejection. The positive rail that showed very sensitive with generator, on the contrary is inert to ripples only a detail.
Bellow is the ripples of the negative rail and output with and without base to base capacitor.
Although the peak is high, the rms value is low.
Bellow is the ripples of the negative rail and output with and without base to base capacitor.
Although the peak is high, the rms value is low.
Attachments
Last edited:
The negative ripple was due to DN2540 impedance, I add a second one to make it run at constant voltage and the problem resolved. The base to base capacitor was acting unbalanced because one base resistor was higher as the Vbe was different. Once equaled most of the ripple decreased to less than 5uV pp. Any filtering on the positive rail makes the ripple increase.
Attachments
I added Becker clamps on both VAS and CCS, more important, I added a current limiter for the VAS of 70ma. Without current limiter, if the output is shorted, the VAS will be pouring over 700ma to destroy the npn driver base the output current protector and itself.
How it works. The other branch of the differential drives a low Ron MOSFET SSM3J340 and shorts the current limiter resistor 15 ohms. When the differential pours all its current to the VAS, the MOSFET gets disabled and let's the resistor limit the current.
This smd MOSFET is available from major suppliers for less then $0.5.
How it works. The other branch of the differential drives a low Ron MOSFET SSM3J340 and shorts the current limiter resistor 15 ohms. When the differential pours all its current to the VAS, the MOSFET gets disabled and let's the resistor limit the current.
This smd MOSFET is available from major suppliers for less then $0.5.
Attachments
I added an optional SOA current protection, mainly for output short protection.
The Shottky diodes SR360 have 0.2V@0.1A and cost10 pieces, $0.5 with shipment from Ali.
The advantage of this protector, it doesn't limit the current distorting before limit current reached, it triggers a shutdown to be reseted by off-on.
The Shottky diodes SR360 have 0.2V@0.1A and cost10 pieces, $0.5 with shipment from Ali.
The advantage of this protector, it doesn't limit the current distorting before limit current reached, it triggers a shutdown to be reseted by off-on.