In the case of parallel chip amps (e.g. the PA100 LM3886), which rails of which ICs need decoupling capacitors close to the pins?
My gut says "all of them!" (Option 1 in the image attached), but I notice many schematics floating around of the PA100 that show both Option 2 and Option 3, where there are only two sets of caps - one IC gets the caps on its + rail and the other IC gets the caps on its - rail, or only one IC gets both + and - caps.
In addition, I notice many parallel LM3886 boards feature only two sets of decoupling caps, not four (one for each rail) as Option 1 might imply.
Would I be correct in assuming that the rails can therefore share the same set of caps - so, both the + rail on the 1st IC and the + rail on the 2nd IC can share a single set of (say) 100uF and 100nF capacitors? And that this therefore implies needing to space them on the PCB that there's a compromise between the ideal distance of the caps to the pins (i.e. as close as possible) and keeping them equidistant for the sake of sharing them?
Unless I totally misunderstand
Thanks!
My gut says "all of them!" (Option 1 in the image attached), but I notice many schematics floating around of the PA100 that show both Option 2 and Option 3, where there are only two sets of caps - one IC gets the caps on its + rail and the other IC gets the caps on its - rail, or only one IC gets both + and - caps.
In addition, I notice many parallel LM3886 boards feature only two sets of decoupling caps, not four (one for each rail) as Option 1 might imply.
Would I be correct in assuming that the rails can therefore share the same set of caps - so, both the + rail on the 1st IC and the + rail on the 2nd IC can share a single set of (say) 100uF and 100nF capacitors? And that this therefore implies needing to space them on the PCB that there's a compromise between the ideal distance of the caps to the pins (i.e. as close as possible) and keeping them equidistant for the sake of sharing them?
Unless I totally misunderstand
Thanks!
Attachments
@adason - Ha ha, this is less about feelings and more about understanding what I can get away with and why!
I’d really like to know why conventional wisdom (put all decoupling caps on all rails, as close as possible) doesn’t seem to be upheld in most parallel amplifier cases. Anyone got some insight?
I’d really like to know why conventional wisdom (put all decoupling caps on all rails, as close as possible) doesn’t seem to be upheld in most parallel amplifier cases. Anyone got some insight?
You evidently put low-impedance resistors in series with the outputs?
The local decoupling capacitors serve to avoid the voltage at the connection point to drop due to rapid changes in the current consumption.
If the local decoupling capacitors are missing, it may worst case leave the amplifier to oscillate.
How many are needed? If you put them in at both chips for a start and you have no problems, it did not cost you much to be on the safe side. If you put the local decoupling only at one amplifier and the other makes problems it will be more difficult to add the remaining decoupling later.
It also depends on your PCB layout. I would put at both.
The local decoupling capacitors serve to avoid the voltage at the connection point to drop due to rapid changes in the current consumption.
If the local decoupling capacitors are missing, it may worst case leave the amplifier to oscillate.
How many are needed? If you put them in at both chips for a start and you have no problems, it did not cost you much to be on the safe side. If you put the local decoupling only at one amplifier and the other makes problems it will be more difficult to add the remaining decoupling later.
It also depends on your PCB layout. I would put at both.
My gut feeling is to use a simple set of big (+/-1000uF) electrolytic per pcb and smaller polymer+x7r at each pin. At least if the pcb layout has low impedance tracks between the big caps and the various chips.
If you build a bridged amp on the same pcb, then you only need decoupling caps between the rails, not from rail to gnd.
The LM3886 data sheet says: 470 uF bulk cap on the board, 10 uF electrolytic or tantalum + 100 nF ceramic right at the chip.
I recommend 470 uF (or more) on the board, 22 uF electrolytic + 1 uF X7R right on the chip. You can see my justification here: LM3886 chip amp supply decoupling.
If you reduce the 10 uF, you'll get oscillation when the output swing exceeds 30 V peak. Note that the voltage coefficient of many ceramics is pretty horrid. Thus, a ceramic cap is not a good choice for the 10 uF decoupling cap.
You'll need decoupling on each rail of each chip.
Tom
I recommend 470 uF (or more) on the board, 22 uF electrolytic + 1 uF X7R right on the chip. You can see my justification here: LM3886 chip amp supply decoupling.
If you reduce the 10 uF, you'll get oscillation when the output swing exceeds 30 V peak. Note that the voltage coefficient of many ceramics is pretty horrid. Thus, a ceramic cap is not a good choice for the 10 uF decoupling cap.
You'll need decoupling on each rail of each chip.
Tom
Tom, great comment, as always, you know your stuff.
But how often do you listen at 30 volt output.
Me, never. I rarely use more than few hundred millivolts under normal listening conditions. My speakers are bi amplified, so i use two stereo chip amps.
Those chips barely know they are being used...
But how often do you listen at 30 volt output.
Me, never. I rarely use more than few hundred millivolts under normal listening conditions. My speakers are bi amplified, so i use two stereo chip amps.
Those chips barely know they are being used...
Haha, speakers too, best not to ask too much, they're less likely to lie to you.
Never. However, can you guarantee that the output of the chip will never exceed 30 V even during a turn-on/off transient pop? Can you guarantee that it won't oscillate at lower output voltages once it heats up or if you get an LM3886 that's closer to the production test limits than the one I measured?
I don't see the point of risking my speakers to save a few cents on bypass caps. Then again, I tend to take the engineering approach and design my circuits to perform well when mass-produced. The occasional hacker who doesn't mind swapping tweeters out when they die from amplifier oscillations may have different design criteria than those I use.
Tom
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.