When I built my first tube amp - my 6LU8 Spud - I was inspired by Pete Millett's use of a capacitance multiplier as a ripple filter. He uses it in his Engineer's Amplifier. The cap multiplier offers rock solid ripple rejection - at least up to a few kHz - and is much less expensive than an inductor. Hence, I used a similar circuit in my Spud.
One thing that's always bugged me about my Spud amp was that hum would be emitted from the speakers every time a heavy load turned on. So if I was using a space heater or had my laser printer going, I'd get hum every time the heater or fuser turned on.
I finally got around to figuring out the root cause of this issue. You guessed it! It's the ripple filter.
I tested this by measuring the output of my Spud amp with an oscilloscope (triggered to the AC line frequency) while plugging in a heat gun (1400 W load - enough inrush current to make the lights dim when turned on). I got a rather obvious 400 mVpp hum on the output of the amp. By tweaking the resistors in the cap multiplier, I was able to reduce this to slightly less than 100 mVpp of hum on the output.
I ran a simulation to show what's happening. The issue is that when the heavy (1400 W) load is turned on, the input voltage to the supply filter drops. This causes C9 to discharge through R8, D1, and the body diode of the NMOS. Due to the time constant of R6, C9, it takes a while for the voltage on C9 to stabilize. While this voltage stabilizes, the cap multiplier is just passing the supply ripple without any filtering effect.
It's worth noting that any low-frequency disturbance on the input voltage will make it through the filter, regardless of the filter type. However, the CLC and CRC filters maintain their ripple rejection even when the input voltage droops. Hence, the 120 Hz ripple is still attenuated by the filter. The capacitance multiplier loses its filtering effect when the input voltage droops and doesn't recover its filtering effect until the voltage on the gate of the MOSFET has stabilized. Due to the time constants involved, its recovery time is much longer than those for CLC, CRC filters.
It is possible to make the cap multiplier immune to supply disturbances, but basically, this means that the output voltage of the circuit needs to be lowered such that it is always lower than the biggest droop expected on the input voltage. In that case, one would be better off using a actual supply regulator.
I suspect these findings is one of the reasons Pete's reverted back to the CRC, CLC style of filtering in his more recent amps. Perhaps he can comment on this.
The good news is that if you already have an amp that uses a cap multiplier and you're experiencing these hum issues, you can always take the cap multiplier out and replace it with an inductor.
Note that in a single-ended amp without global negative feedback, all this supply ripple is only attenuated by the turns ratio of the output transformer before it hits the speaker.
Just thought I'd share...
~Tom
PS: In the sim, I used 2 x 0.5 ohm as a rough model for the impedance of the house wiring (transposed to the secondary of the power transformer). The 40 ohm resistors (R1~R4, R9) represent the DCR of the transformer secondary. Two transient sims were run. One where the load is on for 100 ms, another where it's on for 500 ms.
One thing that's always bugged me about my Spud amp was that hum would be emitted from the speakers every time a heavy load turned on. So if I was using a space heater or had my laser printer going, I'd get hum every time the heater or fuser turned on.
I finally got around to figuring out the root cause of this issue. You guessed it! It's the ripple filter.
I tested this by measuring the output of my Spud amp with an oscilloscope (triggered to the AC line frequency) while plugging in a heat gun (1400 W load - enough inrush current to make the lights dim when turned on). I got a rather obvious 400 mVpp hum on the output of the amp. By tweaking the resistors in the cap multiplier, I was able to reduce this to slightly less than 100 mVpp of hum on the output.
I ran a simulation to show what's happening. The issue is that when the heavy (1400 W) load is turned on, the input voltage to the supply filter drops. This causes C9 to discharge through R8, D1, and the body diode of the NMOS. Due to the time constant of R6, C9, it takes a while for the voltage on C9 to stabilize. While this voltage stabilizes, the cap multiplier is just passing the supply ripple without any filtering effect.
It's worth noting that any low-frequency disturbance on the input voltage will make it through the filter, regardless of the filter type. However, the CLC and CRC filters maintain their ripple rejection even when the input voltage droops. Hence, the 120 Hz ripple is still attenuated by the filter. The capacitance multiplier loses its filtering effect when the input voltage droops and doesn't recover its filtering effect until the voltage on the gate of the MOSFET has stabilized. Due to the time constants involved, its recovery time is much longer than those for CLC, CRC filters.
It is possible to make the cap multiplier immune to supply disturbances, but basically, this means that the output voltage of the circuit needs to be lowered such that it is always lower than the biggest droop expected on the input voltage. In that case, one would be better off using a actual supply regulator.
I suspect these findings is one of the reasons Pete's reverted back to the CRC, CLC style of filtering in his more recent amps. Perhaps he can comment on this.
The good news is that if you already have an amp that uses a cap multiplier and you're experiencing these hum issues, you can always take the cap multiplier out and replace it with an inductor.
Note that in a single-ended amp without global negative feedback, all this supply ripple is only attenuated by the turns ratio of the output transformer before it hits the speaker.
Just thought I'd share...
~Tom
PS: In the sim, I used 2 x 0.5 ohm as a rough model for the impedance of the house wiring (transposed to the secondary of the power transformer). The 40 ohm resistors (R1~R4, R9) represent the DCR of the transformer secondary. Two transient sims were run. One where the load is on for 100 ms, another where it's on for 500 ms.
Attachments
Last edited:
Tom -
Looks like a good analysis.
I guess my line voltage is stable enough that I never noticed this.
Most of the problem can be alleviated by adding a diode in parallel with the resistor that charges the cap on the MOSFET gate. This rapidly discharges this cap when the input voltage droops. See attached.
In any case... nothing is free, it seems. A good LC filter works very well, but it's big, and expensive. With solid state rectifiers (where you can stand large peak currents) you can go with bigger caps and smaller chokes, like I did in the "807" amp design. But then you have to deal with high peak charging currents (low power factor), and the resultant noise issues that can result.
Pete
Looks like a good analysis.
I guess my line voltage is stable enough that I never noticed this.
Most of the problem can be alleviated by adding a diode in parallel with the resistor that charges the cap on the MOSFET gate. This rapidly discharges this cap when the input voltage droops. See attached.
In any case... nothing is free, it seems. A good LC filter works very well, but it's big, and expensive. With solid state rectifiers (where you can stand large peak currents) you can go with bigger caps and smaller chokes, like I did in the "807" amp design. But then you have to deal with high peak charging currents (low power factor), and the resultant noise issues that can result.
Pete
Attachments
Tom,
What mains voltage disturbance magnitude do you think is being simulated in your examples?
A sim that just looks at external load-on step and then another sim with just external load-off step would be more applicable to your own load examples (ie. space heater). And I guess you'd hear the click of the thermostat in the space heater if it was in the room you were in, and similarly the whine of the laser printer as it was printing.
Ciao, Tim
What mains voltage disturbance magnitude do you think is being simulated in your examples?
A sim that just looks at external load-on step and then another sim with just external load-off step would be more applicable to your own load examples (ie. space heater). And I guess you'd hear the click of the thermostat in the space heater if it was in the room you were in, and similarly the whine of the laser printer as it was printing.
Ciao, Tim
Pete: Thank you for replying and for the little diode trick. Now, of course, the voltage takes its own sweet time to recover after a transient. Tradeoffs, tradeoffs. I can think of two reasons I get hum while you don't:
1) In my Spud I push the limits pretty hard on the cap multiplier. I use 33k/1M5 for the voltage divider, hence, only shave off about 2 % of the B+ voltage. Your design used 47k/1M0, hence, shave off about 5 % of the B+ voltage. I changed mine to 68k/1M5 and that made a significant improvement. The hum went from 400 mVpp to maybe 80 mVpp.
2) I'm using a single-ended amp. You're using push-pull. I suspect the single-ended amp is more susceptible to supply ripple. But that's an educated guess.
Tim: In the lab, I measured about 6 % drop in mains voltage when I turned on my heat gun load. I just found a bug in my simulation. I've fixed it and readjusted the resistors modeling the AC line DCR such that the resulting voltage drop is about what I measure in the lab. There was also some coupling between the different sections in the simulation, so I put some diodes in to separate them. It's a bit of a hack, but it works well enough.
~Tom
1) In my Spud I push the limits pretty hard on the cap multiplier. I use 33k/1M5 for the voltage divider, hence, only shave off about 2 % of the B+ voltage. Your design used 47k/1M0, hence, shave off about 5 % of the B+ voltage. I changed mine to 68k/1M5 and that made a significant improvement. The hum went from 400 mVpp to maybe 80 mVpp.
2) I'm using a single-ended amp. You're using push-pull. I suspect the single-ended amp is more susceptible to supply ripple. But that's an educated guess.
Tim: In the lab, I measured about 6 % drop in mains voltage when I turned on my heat gun load. I just found a bug in my simulation. I've fixed it and readjusted the resistors modeling the AC line DCR such that the resulting voltage drop is about what I measure in the lab. There was also some coupling between the different sections in the simulation, so I put some diodes in to separate them. It's a bit of a hack, but it works well enough.
~Tom
Attachments
Last edited:
I found that the MOS cap multiplier took care of the background hum very nicely. Except for those "random" bursts of hum. It took a while to figure out that they were caused by loads turning on/off. Once I figured that out, the root cause was easily found. 
I never tried with a bigger inductor. I didn't want to pay for one... Perhaps I should. I like the performance I'm simulating with the 7 H Edcor. I can also take advantage of the bigger caps available now.
~Tom
I never tried with a bigger inductor. I didn't want to pay for one... Perhaps I should. I like the performance I'm simulating with the 7 H Edcor. I can also take advantage of the bigger caps available now.
~Tom
Actually that's a very good question. Hopefully this doesn't launch a religious debate.
In this particular amp, I really didn't want a true regulator for two reasons: One, I didn't want to dissipate all the power needed for a B+ regulator - a cap multiplier only has to drop a bit over the peak-to-peak input ripple, whereas a regulator would have to drop enough voltage so that it still regulates at minimum mains voltage, so dissipates a lot of power at nominal and especially high mains voltages.
The other reason relates to the AC response of regulators.
The problem with true closed-loop regulators is that they becomes an active participant in the AC circuit, so can have some effect on the AC performance (read: sound) of the amplifier circuit.
The cap multiplier has no feedback. It is kind of an open-loop regulator that just tracks slightly below the average input voltage.
You can put zeners in parallel with the cap on the FET gate, and you have a simple regulator that should be pretty benign in the AC domain.
But add an error amp and feedback, and now you have introduced much complexity into the AC response of your circuit.
One way I've avoided this becoming a problem is to rely on big capacitors or even an LC filter AFTER the regulator to carry most of the AC signal current. At least keep signal currents above, say, 400Hz out of the regulator. Unfortunately, to be useful in reducing ripple the regulator has to have good gain up to 120Hz, so the regulator will always have some impact.
I'm not saying that it always causes problems. Just that it's not as simple a situation as it first appears.
One thing I always do is to apply an AC current to the output of a regulator to see how it behaves. You can do this easily in simulation, or you can AC-couple the output of a power amp directly on to the regulator output (loaded with a resistor). A square-wave load shows you quite a bit about the regulator's response (like ringing and overshoot), as does an AC sweep. You may find some surprising things - I've seen regulators that will actually INCREASE the amplitude of the applied signal at some frequency. You can imagine that this has a pretty big impact on an amplifier if it happens in the audio band.
Pete
Last edited:
159S Hammond Manufacturing | Mouser
chokes are expensive ?
what is semiconductor cost in you supply? add also cooling solutions
chokes are expensive ?
what is semiconductor cost in you supply? add also cooling solutions
That's basically the point I arrived at.
And that's the advantage of the cap multiplier right there. It only drops a certain fraction of the supply voltage, hence, doesn't dissipate much power. And as long as any glitches on the supply are smaller than the voltage drop across the cap multiplier, it actually works quite well.
A 4 H inductor and a 220 uF cap will set you back about $30. This filter will provide 60 dB of ripple attenuation at 120 Hz.
The component cost of my 21st Century Maida Regulator runs about $18~20 as I recall. It provides 120 dB of ripple attenuation and offers extremely low output impedance (0.05 ohm across much of the audio band, further improvements under investigation). Yes. You do have to provide a heat sink. I happen to have a handful of heat sinks that I fished out of a tub skid at a scrap yard about 15 years ago. They're "NOS" and I paid a few bucks per pound for them. If you aren't as fortunate as I, a 3" section of 5.375" profile from Heatsink USA shouldn't wreck the budget.
So yeah... You get what you pay for.
A few bucks will buy you an RC filter providing less than 30 dB of attenuation (120 Hz).
For $10 you can have a cap multiplier that actually works pretty well - except for its handling of transients on the input voltage. It provides 80 dB of attenuation at 120 Hz, but less at higher frequencies, which concerns me a bit.
$30 buys you an LC filter that can provide 60 dB of attenuation at 120 Hz. Spending more on a bigger inductor will result in closer to 70 dB of attenuation for about $60.
$50~60 buys you one of my 21st Century Maida Regulator boards, the components to stuff it, and a heat sink. This solution provides over 120 dB of ripple attenuation across a wide frequency range. It also provides a nice low output impedance.
But, of course, I'm biased... My goal here is just to do a sanity check to see what lower cost solutions are out there and what their relative performance would be. That's all.
~Tom
Actually, that won't solve the problem either. When the input voltage drops, the output voltage will drop as well. In this scenario, the voltage on C9 has to drop, otherwise, you blow the gate oxide on M1 (Vgs violation). In fact, that's what the zener diode is there for. To prevent the Vgs excursions from blowing the MOS device. Inserting another device as a darlington-type helper won't solve this. It just moves the problem to the helper device...
Funny how such a small circuit can cause so much trouble...
~Tom
Funny how such a small circuit can cause so much trouble...
~Tom
Last edited:
Of course with regulator you will have rock star ripple rejection and output impedance as you said , but since in your schematic you already have voltage divider , R6 and R7 , you can change these values to get lower output voltage ( that is what I have in my mind when I said " small voltage reducer " ) and the circuit will remain a capacitance multiplier , because we don't have any zener or voltage reference or any error amplifier e.t.c , In conclusion the most important is to have a circuit that will SUCCESSFULΥ do your job no matter what that circuit is called .
Last edited:
If you optimize the layout keeping the current flows in mind, there shouldn't be a need for a cap multiplier. This assumes that the output impedance of the regulator and supply connections are low. Should this not be the case, the cap multiplier will give you a little more rejection. I would only use the cap multiplier on the input stage of the amp. Feed the output stage from the regulator.
~Tom
An externally hosted image should be here but it was not working when we last tested it.
/fix
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.