Keantoken's CFP cap multiplier

IMHO, negative version will perform much better, since base of Q4 is "feeded" through Q3. In positive version, base of Q2 is feeded through 100R resistor.

According to Walt Jung in Figure 1b, 40dB rejection is possible. No matter if anyone stack 2, 3,...10 transistors, if pass transistor's base is feeded with noise, you have "garbage in, garbage out" situation.

Swap R1 and Q1 positions.
Or feed base of Q1 through resistor, but then you need heavy R/C filtering, which will add complexity.
Or add JFET.
Or.....do anything.
But 100R resistor there is wery bad choice. No way to get better than -40dB (in theory).

Another important thing is capacitor. You want low leakage type there. Film caps are the best, low ESR solid-polymers (OSCON,....) are the worst.

Just my personal opinion.
 
Last edited:
Storm, the article doesn't say anything about my configuration, only stuff about opamps and zener diodes.

There is no difference between the positive and negative versions; the bases of the outputs are "fed" by the drivers, and the 100R resistors give ~6mA bias for the drivers, to reduce output impedance. There may be a difference if Early effect is different for PNP's and NPN's, but if so even Andy_C's models haven't revealed this.

Let me draw the schematic a different way.

- keantoken
 

Attachments

  • KmultEuro_AlterSchem.png
    KmultEuro_AlterSchem.png
    108.7 KB · Views: 889
Storm has a good point about low-leakage caps. Unfortunately the rubycon ZL datasheet is down, so I can't check.

Also, some models DO suggest that input rejection for the positive reg would be less because of the PNP transistor, by ~10db. SPICE has trouble modeling Early effect, so we can't know how drastic the changes are in real life.

- keantoken
 
Storm, let's make sure we've got our facts straight before you write my name on anything of yours (awkward moment, lol). We only need low enough current to not turn on D7. Across 1k, this means less than 300uA. 300uA at 16V mean 53.3kohms. As long as the cap's leakage is say 80kohms, we're okay. What say you? I don't think ANY reasonable lytic will have such high leakage current.

- keantoken
 
I drove to RS Components today and bought the BC550C/560C. Repaired the damaged negative rail on board 1 (a diode was destroyed) and put on the BC550C/560C.

Measurements of the positive rail are basically identical to the negative rail after two decimal digit round up, so only the numbers on the positive rail are given. Measurements were done with a bench transformer and 390R resistor load, not in the Marantz real circuit.

Load current: 43mA
Input voltage after rectification and filtering: 18.2V
Input ripples: 85mV peak to peak (ripple frequency: rectified mains)
Output voltage: 16.8V
Output ripples (with 4,400uF at output): roughly 1.5mV peak to peak
Voltages at top of C1 and after the two diodes: 17.4V

So under the test conditions, no obvious difference between the positive and negative rails has been found.

Note that my scope does not have high resolution. The maximum resolution is 10mV per grid so anything under 2mV is only indicative. Also I am not using high quality probes so on long ground tracks I am skeptical about readings under a few mVs.

I am now progressing to modify board 2. More test data is coming soon.
 
Last edited:
Question for Kean: I notice that in the initial simulation of your published circuit, you used 200mA. Is the ripple rejection dependant of load current or not? I suspect yes, as the transistors behave differently with different current loads, may be not much with the transistors chosen based on the test current. Nevertheless, perhaps you could simulate with 300mA current and see if the simulation matches the test result. I guess that we can accept difference of some dBs for sure, but if the difference were as high as 30dB then some underlying conditions must be found.
 
Last edited:
O.K., Here is some fun:

Continued from the replacements of drivers with BC550C/560C on board 1, with the test result above, I put this board 1 into the Marantz player.

Due to difficulty in accessing the negative rail, I did not risk burning some components again so I only measured the output voltages and ripples. Voltages are as expected just above 14VDC, but:

Ripples at output of positive rail: 10mV
Ripples at output of negative rail: 3mV

Comparing to using 2N5088/2N5401 with ripples of 4mV and 0.5mV respectively, the BC550C/BC560C has higher output ripples at 10mV and 3mV. These transistors were bought at RS Components, so fake parts are nearly impossible.

I have not replaced any other parts other than a broken diode so all other parts are identical to yesterday's test.

Now it is over to you guys who are more advanced than me.
 
Last edited:
The fun continues...

I gave the new player a listen. Now this is subjective and the result is strongly influenced by human psychie.

I have to say that the result is not reliable, as I earlier also replaced the output cap from the Marantz reg, changing from Rubycon ZL 25V 100uF to Rubycon ZL 25V 47uF. The original Marantz used a Elna Silmik (or Cerefine, I can't recall exactly) 25V 470uF.

But I have to say that I like the sound much better. Don't know if it is the cap change or the BC550C/BC560C change, or both.

I guess that the Marantz reg has a theoretical -60dB rejection, combining with the K-Multiplier and the capacitors it has a total of -100dB rejection, so it really does not care if there are a few mV ripple difference in mains frequency. However, if the high frequency filtering is better in the K-Multiplier, then there should be sonic improvements. If the improvement is made only by the cap change, then it would be possible that the 100uF ZL has too low impedance that the Marantz reg does not like it. If this is the case, it can only prove that the Elna Silmik (or Cerefine) cap of 470uF (as famous as Blackgate) must have a higher impedance / ESR, that is higher than the Rubycon ZL 100uF.
 
Load current: 43mA
Input voltage after rectification and filtering: 18.2V
Input ripples: 85mV peak to peak (ripple frequency: rectified mains)
Output voltage: 16.8V
Output ripples (with 4,400uF at output): roughly 1.5mV peak to peak................................ I put this board 1 into the Marantz player.

Due to difficulty in accessing the negative rail, I did not risk burning some components again so I only measured the output voltages and ripples. Voltages are as expected just above 14VDC, but:

Ripples at output of positive rail: 10mV
Ripples at output of negative rail: 3mV
how does 1.5mVpp become 10mV and 3mV when fitted to the Marantz?

What is changing?
 
While you guys may try to perfect the K-Multiplier, I am very happy with it. It solves the problems I had.

I am now starting thinking about adopting it to the power amps. Kean has mentioned that the power transistors need to be 15A ones. I have little knowledge and experience with transistors. Any recommendations here?

My thought is this. Most power amps have high ripple rejection at low frequencies but low ripple rejection at high frequencies. There may not be much gain in smoothing the rectified mains ripples. But a capacitor multiplier can use relatively large film cap bypass which, after multiplication, would appear to be a very large film cap with very low impedance at high frequencies. Film caps sound a lot better than electrolytic caps. Would this give a large sonic improvement? I hope so. At leat it is worth experimenting.

But we also need to consider other things. The K-Multiplier expects low ripples at the input. I guess that I could increase the diode strings to reduce that requirement, but am not sure that if the impedance can still remain low in that case. A K-Multiplier for a power amp does not need to be the same as the one for the Marantz, which requires low voltage drop. All requirements can be relaxed except low output impedance. A 0.04mR output impedance must be kept. My experience made me believe that even a 0.1R series resistance in the power amp power supply degrades the sound.

For class AB amps, we have to use the worse case scenarios and estimate 10A current at least. If the circuit draws 10A, I would expect the input to the K-Multiplier has much higher ripples.

Would it work?
 
Have you compared the JLH "ripple eater"?
Maybe as a pre-reg.

I built and played with the JLH "ripple eater" prevously and found it improve the LM317 regs marginally. But at the time, I did not have sufficient knowledge and experience to implement the JLH properly.

However, in this particular case, no other circuits I know of are suitable for the Marantz CD player upgrade, as it requires (1) low output impedance; (2) very low voltage drop; and (3) relatively good isolation. Only the K-Multiplier delivers all these.
 
HiFi, I have added varying amounts of ESR and ESL to each capacitor. As it turns out, C1 has a significant affect on the pulse/HF behavior of the multiplier. If I understand correctly, capacitance is multiplied, while resistance and inductance are divided by the current gain of the multiplier. However, Ib grows with frequency, meaning HF load behavior will influence the caps' resonances more. So in this way it does seem Cob of the driver is important.

Your results still mystify me. The only way I can explain it is if the simulator isn't modeling Vcesat effect right (did I confuse this with Early effect?). Perhaps it can model Vcesat effect, but even Andy_C hasn't accomplished this: look at the textbook graphs produced with his models, and compare to the datasheet.

I just simulated with load currents from 100-400mA. As I predicted, input isolation didn't degrade much, about 4db. Still, I don't know where the faulty Vcesat modeling put us.

HiFi, I think we could this if you would add another diode to the string, and see if specs improve. However I cannot be sure.

Also, if the total capacitance for C1 is less than 1000u, the rolloff point will be increased from 100Hz where it is, and ripple isolation will decrease (120Hz).

If there are jagged or toothy traces anywhere, it means a transistor is saturating or a diode is turning on. The output voltage should show a smooth trace, perhaps with some ripple but no sharp corners.

Another possibility is that it's oscillating. But I think that if it was, you would have identified it. However it's still possible that it's faster than your scope can trace. To test this, try inserting a 100ohm resistor in series with the base of the driver and see if anything changes.

- keantoken
 

Attachments

  • Hfe-Vce_SPICE.png
    Hfe-Vce_SPICE.png
    68 KB · Views: 761