Power Supply filter capacitors

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Member
Joined 2011
Paid Member
The Akitika kit is a stereo power amplifier using one LM3886 per channel. Unusually, it includes a (discrete component) voltage regulator on its own PCB with its own heatsink. This provides unusually stable supply voltage with unusually low output impedance (compared to unregulated supplies typically used with chipamps).

The construction manual, including schematics and theory-of-operation discussion, is online and freely downloadable. Might be worth 3 minutes to peruse it and appropriate the ideas you deem good.
 
Physics tell us that bigger caps (within reason) give less ripple; less ripple is a good thing for sound reproduction.

That is true, but you're making the assumption that the caps have a reasonable impedance curve when you make this statement. In reality, caps vary a lot in this regard and in general larger caps have higher impedance at higher frequencies.

Power supply caps almost always have good characteristics at ripple frequency (120 Hz) but not always at higher frequencies. One nice thing in this regard is that with the ubiquity of switching power supplies, many improved caps are now available that do have lower inductance and esr. I've used them in power supplies for line level circuitry and they work perfectly. I don't see why bigger ones couldn't be paralleled for a power amplifier supply. I will investigate this as it relates to 1875 and 3886 chips once I complete about 20 other projects :D. In fact I did just this to a piece of vintage (90s) equipment (see above) and the results were gratifying to say the least.

So with awareness of this situation, it is possible to design around this drawback. It is imperative to at the very least place smallish value electrolytic caps (say 220-470 uF) as close as possible to the chip's supply pins. It even points this out in the datasheets.
 
....................
So with awareness of this situation, it is possible to design around this drawback. It is imperative to at the very least place smallish value electrolytic caps (say 220-470 uF) as close as possible to the chip's supply pins. It even points this out in the datasheets.
It's more important that the HF decoupling is located as close as possible to the chip's supply pins and that the current route around the loops is kept very short and very compact.
The MF decoupling you describe, then fits around the HF decoupling and necessarily must have longer, higher impedance connections.
you seem to have misread the datasheet/s
 
It's more important that the HF decoupling is located as close as possible to the chip's supply pins and that the current route around the loops is kept very short and very compact.

I know that. I solder them directly to the chip leads on the underside of the board. Your chip is sure to be unstable without this. I have some similar to this ECW-F2224JAQ Panasonic Electronic Components | Capacitors | DigiKey that I use.

The MF decoupling you describe, then fits around the HF decoupling and necessarily must have longer, higher impedance connections.
you seem to have misread the datasheet/s

Of course. I wasn't clear, but the whole discussion was about what size of electrolytic caps to use.

Do you think the inductance is necessary, or just a consequence of the necessary layout? I've only built a few prototypes and always put the MF decoupling as close as possible to the chips, less than an inch. Always worked with no funny stuff.
 
The inductance, leading to increased impedance, is a consequence of the size of the components. One can make the routes short, but ultimately the component size prevents making zero loop area.

I have even fitted the smd HF decoupling on the top side direct to the pins of a chipamp. Can't get shorter/smaller loop areas than that.
Then fitted the electro cans as close as possible. probably 10mm away leaving >20mm loop length.
 
@ ggidzinski : the lm3886 has nothing that makes it "allergic" to a high capacitance supply and only some kind of anthropomorphic thinking can lead to conclude that higher capacitance lead to a slower sound (whatever that means). Trust your "old school" intuition on this.

Btw, if we take layout examples, this is pretty much ideal and to the point.

Decoupling caps are c14-15, c3-13, c4-12. It all goes to a groundplane under the board for low impedance.

See LM3886 Done Right. I'm not pushing Tom's products (I think these particular boards are a bit overpriced to be true) but the engineering is solid.
 

Attachments

  • LM3886DR_PCB.jpg
    LM3886DR_PCB.jpg
    225.4 KB · Views: 330
Fast Eddie D said:
That is true, but you're making the assumption that the caps have a reasonable impedance curve when you make this statement. In reality, caps vary a lot in this regard and in general larger caps have higher impedance at higher frequencies.
By "ripple" I meant mainly the low frequency stuff, up to a few kHz. That is what a reservoir cap is meant to deal with. Higher frequency is dealt with by smaller filter or decoupling caps.
 
By "ripple" I meant mainly the low frequency stuff, up to a few kHz. That is what a reservoir cap is meant to deal with. Higher frequency is dealt with by smaller filter or decoupling caps.

What's important is that there's electrolytic caps on board with the chip. Many (probably most) consumer grade designs, from cheap "hi-fi" systems to receivers, do not have an intermediate set of decoupling caps. They have one large pair of reservoir caps located close to the output devices, be they integrated or discrete. These caps have to cover a wide frequency range of decoupling because they are usually complemented with only a small value (0.1 to 0.22 uF) capacitor. I know this because I have repaired, hacked, and disassembled for salvage many recycle bin units.

DIY types usually put the power supply circuitry on another board. I know I do. This necessitates the use of smaller (220-470 uF) decoupling caps on board with the chip.

Is there an advantage to the DIY way? Idon't really know. But I for one mix, match, and reuse stuff I build since 90% of them are prototypes anyway, so it's an advantage for me.
 
The fast current changes are supplied by the on board local supply rail decoupling capacitors.
The very fastest by the HF decoupling and the slower changes by the MF decoupling.

The PSU is separated from the amplifier by long traces/cables. The smoothing capacitance can only meet LF and DC current demand due to the inductive reactance of those longer connections.

You have to interpret the current demands of the amplifier you are building and include HF and MF decoupling that can meet those demands.
 
Which design do you advocate, the one with the snubberized/35,000uF design or the last one with no filter caps at all and only the bridge output directly to the chip amp pins with the chip amp only decoupling?

George

With the right layout, a 1000uF capacitor will only have about 0.1Ω to 0.2Ω at 1MHz. The bigger problem is the impedance at low frequencies. Look at what happens below 500Hz.

But then again, shouldn't we be asking, at what frequencies do the capacitors work at and why.
 

Attachments

  • Kondensator-Impedanzverläufe-Wiki-1.jpg
    Kondensator-Impedanzverläufe-Wiki-1.jpg
    106.5 KB · Views: 291
Which design do you advocate, the one with the snubberized/35,000uF design or the last one with no filter caps at all and only the bridge output directly to the chip amp pins with the chip amp only decoupling?

George
The last scheme can be a recipe for disaster. While the filter effect (in terms of ripple voltage) might be sufficient with only close-chip caps, you have all the huge charge current spikes running at your PCB locally. With the slightest error in layout these spike currents will couple directly into your signal path and ruin all the performance. Actually, this a circuit design & layout challenge of the tougher variety. It can be done reaching "OK levels" of error if you are a seasoned expert in the field. Not recommended, that is.

Like other have suggested, locate the main reservoir electrolytics off-board, no bypasses (so you don't ruin their damping ESR). Then place smaller caps (value about 1/10th of the main caps) at the chip. These also don't need snubbers/bypasses when correctly chosen (eg paralleled multiple smaller caps) and using a good layout.

Don't be tempted to use only small film caps of way too low value, because this will end up in an supply impedance (what the chip is actally seeing) like in below screen-shot. It shows the impedance of a 150nF film cap in parallel with a 1000uF electro which is located 30cm away and connected with a set of twisted wires. The inductance from this wiring rings with the 150nF and results in a supply impedance peak of about 6ohms(!) at 1MHz, not a good idea (and with more capacitance it gets even worse as the peak wanders down in frequency)
Note that @1Mhz the impedance is actually lower when the 150nF cap is removed, not what one would expect...
 

Attachments

  • 33cm_twisted_pair+1000u_63V_PanaFC+150n_imp.jpg
    33cm_twisted_pair+1000u_63V_PanaFC+150n_imp.jpg
    70 KB · Views: 286
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.