You could try calculation.ggidzinski said:
A reservoir capacitor C fed from mains frequency f via full wave rectification (e.g. a bridge) with current draw I will drop by approximately
I/(2 f C) volts
between charging pulses.
The worst case is having to reproduce a low frequency square wave. Then peak current is V/R, where V is supply rail voltage and R is speaker resistance. You don't want the supply rail to droop very much, so say you can cope with a fractional drop of k. So
I/(2 f C) < k V, but I=V/R so
V/(2 f C R) < k V so
1/(2 f C R) < k
For f=50Hz (I am in Europe) and R=8 ohms this means that
1/(800 C) < k
It is up to you to pick a value for k. In the past people might have been happy with k=20% but nowadays k=10% might be better. It depends on the amplifier PSRR, and how much hum you can tolerate. k=10% gives C > 12.5mF. (10mF in 60Hz countries)
LM3886 decoupling
Some speak about what capacitance to use after the rectifier, other speak about magic. I like magic and LM3886 can supply a magic sound from a symmetrical power supply. But, let us face facts.
If the power supply is a traditional unregulated type, the 100Hz ripple is determined by the size of the total decoupling capacitance (when more capacitors are used in parallel). I believe that a 10000uF capacitor looses 1V in 10ms (the double rectification rate) with a discharge current of 1A.
Thus, on each rail you should use some 10000uf-22000uF. Such good-size decoupling capacitors can be bought in low ESR versions (for instance from ELNA) and if the PCB copper-tracks to the LM3886 power pins can be kept short, nothing more is needed.
But, another reasonable option exists: Use ordinary electrolytic capacitors (of the same capacity) with two low ESR capacitors (470uF-1000uF) connected in parallel and eventually also two 1uF ceramic/foil capacitors connected in parallel. Then you only need to arrange the two smaller capacitors close to the LM3886 power terminals and the big ones further away. The parallel coupling of the three capacitors cost less than one high quality ELNA and still performs well because the two smaller capacitors reduce the ESR. This trick is frequently used for decoupling of class D amplifiers.
If the power supply is of the regulated type, the big buffer capacitors may be less in capacity. The capacitance across the LM3886 power terminals should always be enough to ensure that the LM3886 remains stable (and the copper-tracks should be short). Some 1000uf should be sufficient. For the rest, the regulator must be quick in response and have good surge current margins. These two parameters are a matter of the regulator design
I often see amplifier datasheets where the power decoupling capacitors are suggested to be surprisingly small (100uF-1000uF). I assume that these are values that ensure the amplifiers remain stable and do not oscillate. Such small values are clearly not sufficient with just a 50Hz/60Hz rectifier bridge connected upstream.
Some speak about what capacitance to use after the rectifier, other speak about magic. I like magic and LM3886 can supply a magic sound from a symmetrical power supply. But, let us face facts.
If the power supply is a traditional unregulated type, the 100Hz ripple is determined by the size of the total decoupling capacitance (when more capacitors are used in parallel). I believe that a 10000uF capacitor looses 1V in 10ms (the double rectification rate) with a discharge current of 1A.
Thus, on each rail you should use some 10000uf-22000uF. Such good-size decoupling capacitors can be bought in low ESR versions (for instance from ELNA) and if the PCB copper-tracks to the LM3886 power pins can be kept short, nothing more is needed.
But, another reasonable option exists: Use ordinary electrolytic capacitors (of the same capacity) with two low ESR capacitors (470uF-1000uF) connected in parallel and eventually also two 1uF ceramic/foil capacitors connected in parallel. Then you only need to arrange the two smaller capacitors close to the LM3886 power terminals and the big ones further away. The parallel coupling of the three capacitors cost less than one high quality ELNA and still performs well because the two smaller capacitors reduce the ESR. This trick is frequently used for decoupling of class D amplifiers.
If the power supply is of the regulated type, the big buffer capacitors may be less in capacity. The capacitance across the LM3886 power terminals should always be enough to ensure that the LM3886 remains stable (and the copper-tracks should be short). Some 1000uf should be sufficient. For the rest, the regulator must be quick in response and have good surge current margins. These two parameters are a matter of the regulator design
I often see amplifier datasheets where the power decoupling capacitors are suggested to be surprisingly small (100uF-1000uF). I assume that these are values that ensure the amplifiers remain stable and do not oscillate. Such small values are clearly not sufficient with just a 50Hz/60Hz rectifier bridge connected upstream.
After trying many different ways to power the LM3886, i found out, that less is more. Mick Feuerbachers 3 resistor copper amp sounded best to me. It uses only 2200uF caps per rail, nothing else. This amp delivers very much detail and very fast response. Mids and highs are beautiful.
Forget 10000uF and more smoothing caps if you dont snubberize them. High cap values make the sound very muddy with the lm3886.
I was not sure if a single lm3886 is enough for me and use 2 225va torroids, but this is to have the option to go parallel with 2 lm3886s. My torroids are way too strong, but dont go below 100va per lm3886 if you want a punchy nice low frequency.
Forget 10000uF and more smoothing caps if you dont snubberize them. High cap values make the sound very muddy with the lm3886.
I was not sure if a single lm3886 is enough for me and use 2 225va torroids, but this is to have the option to go parallel with 2 lm3886s. My torroids are way too strong, but dont go below 100va per lm3886 if you want a punchy nice low frequency.
What works for me
I put snubbers on the large filter caps (10K ufd.). The snubber is a 0.1 ohm resistor and a 47 nfd stacked film cap in series. Then 470 ufd caps a few inches away from the chip. Then 22 ufd caps soldered right to the power pins of the chip. And a 0.22 ufd film connected rail to rail close to the chip.
The 22 ufd caps soldered to the legs of the chip make a big difference.
I put snubbers on the large filter caps (10K ufd.). The snubber is a 0.1 ohm resistor and a 47 nfd stacked film cap in series. Then 470 ufd caps a few inches away from the chip. Then 22 ufd caps soldered right to the power pins of the chip. And a 0.22 ufd film connected rail to rail close to the chip.
The 22 ufd caps soldered to the legs of the chip make a big difference.
Seemed like a mystery here so I thought I would bring it up...
The 1000uf parts add a nice sound when close to the outputs, but the 100uf on many amps are usually closer to the input stages from what I’ve noticed, am sure they help the outputs as well though.
Last edited:
What are examples of other LM3886 implementations that you felt were inferior to the MF 3 resistor design?
No, my oscilloscope tells something different. Distortion is rather low, dont know the exact numbers any more. If i find some time i will measure once again.
Last edited:
I tried the datasheet implementation before and the carlos version with snubberized high capacity. Micks version sounds more „honest“.
Yes, 2200uF could be small enough to ensure some ripple intermodulation in loud passages. This could make the sound 'richer'.
It is generally a good idea not to 'bypass' low ESR electrolytics, as you might increase the HF impedance. If you like bypassing then better to use normal electrolytics with higher ESR, so the ESR can damp the parallel resonance you have created.
It is generally a good idea not to 'bypass' low ESR electrolytics, as you might increase the HF impedance. If you like bypassing then better to use normal electrolytics with higher ESR, so the ESR can damp the parallel resonance you have created.
Audiosector (Peter) uses low smoothing capacitance. All his Followers do likewise.
This is not a trick. It is the standard way that the manufacturer tells us all to wire up the local supply rail decoupling SEPARATE from the main smoothing capacitance back at the transformer/rectifier.
It is better to implement the manufacturer's standard local supply rail decoupling.
Audiosector uses this and he claims improved audio performance. He does not claim reduced audio performance.
Once again, Tomchr did an analysis on decoupling for the lm3886 here: LM3886 chip amp supply decoupling.
Others came to similar conclusions about using a triplet made of :
- a standard high value electrolytic (470u-1000u),
- a 20u or so low esr cap (polymer or the like),
- a highish value x7r cap (2.2u-10u).
But all this assume a decent pcb layout and a proper power supply.
Others came to similar conclusions about using a triplet made of :
- a standard high value electrolytic (470u-1000u),
- a 20u or so low esr cap (polymer or the like),
- a highish value x7r cap (2.2u-10u).
But all this assume a decent pcb layout and a proper power supply.
What is it about a chip amp that makes it "allergic" to high smoothing capacitance values back at the rectifier bridge, causing it to sound muddy or "lose some of its magic". To an old school guy like me that is certainly counterintuitive where you designed power supplies with enough smoothing to bring up the bottom of the charging ramp to make your minimum rail voltage requirement and too much smoothing capacitance/low ripple was not a thing. I understand most of the discussion on decoupling and stability but not why big power supply capacitance and the resulting low ripple and tighter supply rails can be a negative with regards to sound quality. Heck, in the last "three resistor" design example the power supply/rectifier capacitance is completely eliminated and it has bridge straight to local decoupling. I do understand that some smoothing is occurring in this example of local only capacitance that can have total values in the low thousands of uF but there are some designs with 30,000 uF back at the bridge before decoupling vs this example of zero. It's fascinating and certainly thought provoking to see those minimalist implementations.
As a side note, my favorite guitar amp pedals are implemented with op amps, even for distortion and some of the best sounding power amps use tube rectifiers which "suffer" from voltage rail sag. I realize that in this case a certain type of distortion and compression is a good thing and the opposite of most of the goals in audiophile design but that has been suggested as a possible explanation for desirable sound attributes in a post above (IM distortion sounds rich).
I have had the privilege of hearing some of the worlds best audio in massively expensive listening rooms in my work life but I don't prefer the sound of some of these technically perfect implementations. This is all very interesting to me because apparently my ears/brain prefer a less than perfect sound chain or certain coloration held in high regard. Mark Levinson mono blocs and B&W 801's that weigh hundreds of pounds with one inch thick cables does look cool. I guess I will have to put cement in the bottom of my chip amp design and work on my logo. That's some audio humor (my favorite amps are pretty heavy). I did sell my 801's though.
Thanks,
George
As a side note, my favorite guitar amp pedals are implemented with op amps, even for distortion and some of the best sounding power amps use tube rectifiers which "suffer" from voltage rail sag. I realize that in this case a certain type of distortion and compression is a good thing and the opposite of most of the goals in audiophile design but that has been suggested as a possible explanation for desirable sound attributes in a post above (IM distortion sounds rich).
I have had the privilege of hearing some of the worlds best audio in massively expensive listening rooms in my work life but I don't prefer the sound of some of these technically perfect implementations. This is all very interesting to me because apparently my ears/brain prefer a less than perfect sound chain or certain coloration held in high regard. Mark Levinson mono blocs and B&W 801's that weigh hundreds of pounds with one inch thick cables does look cool. I guess I will have to put cement in the bottom of my chip amp design and work on my logo. That's some audio humor (my favorite amps are pretty heavy). I did sell my 801's though.
Thanks,
George
Last edited:
I replaced the old dried up caps (7800uF/70V) in my old Pioneer receiver with an array of caps. I made a couple of PC boards that stand up where the old caps were and put 5x2200 uF/70V + 22 uF/70V paralleled on each board. The 22 uF caps were closest to the "motherboard" and less than 2" from the output devices.
The bass, the bass, the bass! I did this because the receiver worked fine but sounded "thin" compared to the Luxman I had been using. It came out of a dumpster so I figured it wasn't working properly. This hack did everything good and nothing bad for the sound of this receiver. I'm now using it for the power amp only (another hack
) and while I do intend to design and build my own power amp, the performance of this setup does not motivate me to hurry up.
The 3886 has special bugaboos I'm sure. I haven't gone all the way down that rabbit hole yet but it's on my short list. But I have yet to be convinced that a well thought out power supply filter (which would include a low inductance, low esr electrolytic capacitor filter very close to the chip that is separate from the smoothing capacitors) would muddy the sound. And I don't see how using smallish value smoothing capacitors would be able to prevent the power supply from collapsing under a heavy, relatively constant load - which is exactly what the power supply sees when playing bass heavy music loud. (You can hear this happening on a cheapo amplifier - a sustained note causes the amplifier to "fuzz out" and I have measured a 35% droop under such conditions.) I suspect it could be fine if the 3886 were driving a midrange in an active crossover system - but I could be wrong.
The bass, the bass, the bass! I did this because the receiver worked fine but sounded "thin" compared to the Luxman I had been using. It came out of a dumpster so I figured it wasn't working properly. This hack did everything good and nothing bad for the sound of this receiver. I'm now using it for the power amp only (another hack
) and while I do intend to design and build my own power amp, the performance of this setup does not motivate me to hurry up.The 3886 has special bugaboos I'm sure. I haven't gone all the way down that rabbit hole yet but it's on my short list. But I have yet to be convinced that a well thought out power supply filter (which would include a low inductance, low esr electrolytic capacitor filter very close to the chip that is separate from the smoothing capacitors) would muddy the sound. And I don't see how using smallish value smoothing capacitors would be able to prevent the power supply from collapsing under a heavy, relatively constant load - which is exactly what the power supply sees when playing bass heavy music loud. (You can hear this happening on a cheapo amplifier - a sustained note causes the amplifier to "fuzz out" and I have measured a 35% droop under such conditions.) I suspect it could be fine if the 3886 were driving a midrange in an active crossover system - but I could be wrong.
Last edited:
"Minimalistic" only on the mechanical/assembly side of things
But on the inside ..... OMFG !!!!!!
There is a big assumption there: that big reservoir caps do actually degrade the sound of a chip amp. What is the evidence for that? Physics tell us that bigger caps (within reason) give less ripple; less ripple is a good thing for sound reproduction. Psychoacoustics tells us that some people prefer distorted sound, provided that it is not too obviously distorted. Small PSU caps might, under some circumstances, cause gain pumping to take place; some might misinterpret this is an increase in 'dynamics'.ggidzinski said:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.