If it's used before the regulator then there's no point in it having low Zout. It better have very good filtering performance. Just my 2c.
Ikoflexer,
I have been taking every single word you said as gold except this one.

Regards,
Bill
kean, no need for complexity. Ue the same number of parts, just rearange them.
Make Q2 NPN. Make Q1 PNP. Swap positions of R1 and Q1. R1 to the right, Q1 to the left.
Make Q2 NPN. Make Q1 PNP. Swap positions of R1 and Q1. R1 to the right, Q1 to the left.
I think Salas' idea to use the BC5x6 transistors is a good one.
Do you mean it is beneficial replacing 2N5551 with BC550C and 2N5401 with BC560C?
I support any refinements, but maybe someone should start a new thread about it.
I will be very keen to follow that.
Ikoflexer,
I have been taking every single word you said as gold except this one.Can you do me a favour? Add a 10R resistor between your raw supply and the input of the Iko reg and tell us how it sounds when playing some good violin music.🙂
Regards,
Bill
You shouldn't take any of my words at face value.
Anyway, I don't see what you mean. A shunt reg, if it works well, will provide the circuit after it with a high degree of isolation from whatever components before it. The shunt reg is solely responsible for the impedance the circuit it powers sees. Think about this. You can replace the entire CCS part of the shunt reg with a resistor. In fact some people do that, preferring a resistor instead of the CCS. A good CCS will provide a very high input impedance to the shunt part of the regulator. You will have a hard time to convince me that a 10R resistor before the CCS can make a difference. 🙂
I support any refinements, but maybe someone should start a new thread about it.
- keantoken
I think this would make sense, as it seems to have become a subject in itself.
You shouldn't take any of my words at face value.
Anyway, I don't see what you mean. A shunt reg, if it works well, will provide the circuit after it with a high degree of isolation from whatever components before it. The shunt reg is solely responsible for the impedance the circuit it powers sees. Think about this. You can replace the entire CCS part of the shunt reg with a resistor. In fact some people do that, preferring a resistor instead of the CCS. A good CCS will provide a very high input impedance to the shunt part of the regulator. You will have a hard time to convince me that a 10R resistor before the CCS can make a difference. 🙂
Ikoflexer,
Yes I have learnt that theory and used that. I did find in practice that I preferred C to CLRC in the raw supply for subjectively better sound, and if the last C, especially the high frequency caps are far from the input of the CCS, the sound is subjectively degraded. I did those experiments again and again. Of course, I can still be wrong, as such experiments may still be subject to many other conditions.
I guess that in simulations we use ideal capacitors and ideal components. But most capacitors don't work well in high frequencies therefore in high frequencies the result is not as good as simulated. The noise beyond the audible frequency range can impact on the audible range. I know that this does not really give an explanation, but I don't have explanation on why different capacitors in the raw supply can make audible differences, knowing that those differences must be so tiny and should not have an effect on the regulator.
I also applied the same to my power amps and moved the capacitors close to the CCS of the driver stage and added film bypass with quite substantial improvements on sound.
It would not take long to add a 10W R between the raw supply and the regulator so I encourage some of you to do this experiment and report back here. I can learn a great deal if I am proven to be wrong.
Regards,
Bill
kean, no need for complexity. Ue the same number of parts, just rearange them.
Make Q2 NPN. Make Q1 PNP. Swap positions of R1 and Q1. R1 to the right, Q1 to the left.
I thought about this and it seems reasonable.
Would Keantoken please simulate this?
Regards,
Bill
I think this would make sense, as it seems to have become a subject in itself.
It sounds good. Unfortunately I am not a "tech" person so I can't start the new thread on behalf of Keantoken.
I am happy to be a tester again just to join the fun. I am already very happy with the existing K-Multiplier as it was designed for the exact purpose and it works very well.
Good job Salas, thanks for the effort you put into this.
Bill, I'll try it first time I get a chance.
Bill, I'll try it first time I get a chance.
kean, no need for complexity. Ue the same number of parts, just rearange them.
Make Q2 NPN. Make Q1 PNP. Swap positions of R1 and Q1. R1 to the right, Q1 to the left.
I don't know what you mean by less complex, if it has the same number of parts?
There are many things you can do with 2 transistors and a resistors, but few of them will show less output impedance than a diode. It looks like this proposal will make a buffer more suited to driving the base, which will make it fast, but the output impedance will not be any better than the usual capacitance multiplier.
One argument is that Q2's base is driven current-style by Q1. If Q2 demands more Ib, it first has to get it from R1 until NFB catches up and Q1 can respond. If the load happens to have a low impedance, then it will actually see Q2's Cob through Q1's emitter. I don't think one can remedy this without decreasing input rejection.
If we want more input rejection, we need some form of cascode. The main culprit at DC is Early effect. Vbe changes with Vce because Vce changes Ib, blah blah...
In short, to do better I doubt we can maintain a low voltage drop. I don't think we can get past the input rejection "floor" without a fancy cascode.
Thanks Salas, I didn't realize I had been redirected to the new thread until I tried looking for the new thread...
- keantoken
HiFi, Salas' changes are for the better (mainly for lower noise), although you will will need the upgraded output transistors for output current over 200mA, to avoid the dreaded Beta Droop.
- keantoken
- keantoken
Thanks, Keantoken.
A few more days listening has confirmed that the K-Multiplier works very well. There is a large perceivable sound improvement.
Technically, I have not been able to get it working 100% as simulated. The first one (with the better 2SA1930 and 2SC5171) put in the player had a near flat negative rail but about 6mV ripples (peak to peak) on the positve rail (of the frequency corresponding to the rectified mains frequency). I failed to find any problems with any parts so I built a second one with completely new parts. The result was the same. Ripples at input is about 400mV, current draw is about 300mA, ripples on the negatively rail output is definitely less than 0.5mV, while on the positive rail 6mV. All other voltage measurements show identical numbers for both rails.
Before I plugged it in, I tested it on resistor dummy load of 390R (about 40mA constant current), and did not find any problems. Both the positive and the negative had a nearly flat output. Measured figures on the positive rail are as follows: input rectified mains ripples: 100mV. Output current draw: 40mA. Input voltage: 18V. Voltage on R2(1k) after two diodes: 17.2V. Voltage at the base of pass transistor: 17.5V. Voltage at the bass of driver transistor: 16.2V. Voltage at output: 16.5V. Basically the same absolute numbers measured on the negative rail.
Any ideas?
A few more days listening has confirmed that the K-Multiplier works very well. There is a large perceivable sound improvement.
Technically, I have not been able to get it working 100% as simulated. The first one (with the better 2SA1930 and 2SC5171) put in the player had a near flat negative rail but about 6mV ripples (peak to peak) on the positve rail (of the frequency corresponding to the rectified mains frequency). I failed to find any problems with any parts so I built a second one with completely new parts. The result was the same. Ripples at input is about 400mV, current draw is about 300mA, ripples on the negatively rail output is definitely less than 0.5mV, while on the positive rail 6mV. All other voltage measurements show identical numbers for both rails.
Before I plugged it in, I tested it on resistor dummy load of 390R (about 40mA constant current), and did not find any problems. Both the positive and the negative had a nearly flat output. Measured figures on the positive rail are as follows: input rectified mains ripples: 100mV. Output current draw: 40mA. Input voltage: 18V. Voltage on R2(1k) after two diodes: 17.2V. Voltage at the base of pass transistor: 17.5V. Voltage at the bass of driver transistor: 16.2V. Voltage at output: 16.5V. Basically the same absolute numbers measured on the negative rail.
Any ideas?
Last edited:
Maybe you need even heavier filtering? Like cascading two in a row.
P.S. Can the NPN-PNP have differences in Rbb' that are not accounted for in the simulator, Kean?
P.S. Can the NPN-PNP have differences in Rbb' that are not accounted for in the simulator, Kean?
keantoken said:Storm, the 100R resistors could be replaced with Jfets, with some reward but I wanted the circuit to start simple BEFORE getting complex.
Kean, I was thinking about your sentence above and answered with ".... no need for complexity. Ue the same number of parts, just rearange them"
You are right Storm that the Jfet wouldn't add any extra parts, but it may be more involving if you have to tweak IDSS (though I don't think it's necessary..?).
Salas, it is possible that the driver transistor has less Hfe than I accounted for (Rbb shouldn't affect performance unless grotesque, AFAIK). 300mV across the 1k resistor means 300uA, which is too large. This current may turn on D7 which will drastically reduce filtering. You can try using another small-signal transistor. However it is equally possible that the output transistor has low beta too, or has worse specs than the datasheet.
1: Remove R1 and see if performance improves.
2: If not, try replacing Q1 with a different TO92 with higher Hfe.
Also, I received an Email requesting for a version equivalent to the JLH:
The Class-A Amplifier Site - The Capacitance Multiplier
The design is easily scaled, but the issue becomes stability. This has potential to be a great output stage regulator, but the increased parasitic capacitances mean more opportunities for mischief. We need to test a version with a variety of currents and voltages.
- keantoken
Salas, it is possible that the driver transistor has less Hfe than I accounted for (Rbb shouldn't affect performance unless grotesque, AFAIK). 300mV across the 1k resistor means 300uA, which is too large. This current may turn on D7 which will drastically reduce filtering. You can try using another small-signal transistor. However it is equally possible that the output transistor has low beta too, or has worse specs than the datasheet.
1: Remove R1 and see if performance improves.
2: If not, try replacing Q1 with a different TO92 with higher Hfe.
Also, I received an Email requesting for a version equivalent to the JLH:
The Class-A Amplifier Site - The Capacitance Multiplier
The design is easily scaled, but the issue becomes stability. This has potential to be a great output stage regulator, but the increased parasitic capacitances mean more opportunities for mischief. We need to test a version with a variety of currents and voltages.
- keantoken
Last edited:
Salas, it is possible that the driver transistor has less Hfe than I accounted for (Rbb shouldn't affect performance unless grotesque, AFAIK)
- keantoken
Was referring to Q2,Q4. Can their Rbb' play a role? Q1,Q3 must be high hFE & low noise, no doubt.
Wait, voltage at R1 is 17.2 and at the base of Q1 is 16.2? This means 1V across D7, which means something is shorted or broken.
- keantoken
- keantoken
- Home
- Amplifiers
- Power Supplies
- Keantoken's CFP cap multiplier