The op amp ground plane and the power amplifier star are connected together this constitutes the power ground and this is connected to the audio ground through the 4.7 ohm resistor, this is my understanding of the schematic.
Rotating the cap doesn't really shorten the loop by much and it adds to the board length. Putting the cap on the right of the input connector defeats the purpose because the input pin is on the other side resulting in even longer loop.
An externally hosted image should be here but it was not working when we last tested it.
Here is another version...
An externally hosted image should be here but it was not working when we last tested it.
Hi all,
Another version of the layout with some changes...
* Added a regulated supply for the op amp instead of the zener regulator (7815/7915).
* LM3886 supply decoupling is now 2200uF || 1uF (film) || 100nF (film) per rail.
* Shuffled the input filter components around a bit to minimize the input loop, still think it's long but the dimensions of the 1uF cap are the limiting factor here.
Question: The mute cap ground isn't a signal return though i see many boards with this cap connected between the chip and the signal ground, why is that?
Any comments are appreciated.
Another version of the layout with some changes...
* Added a regulated supply for the op amp instead of the zener regulator (7815/7915).
* LM3886 supply decoupling is now 2200uF || 1uF (film) || 100nF (film) per rail.
* Shuffled the input filter components around a bit to minimize the input loop, still think it's long but the dimensions of the 1uF cap are the limiting factor here.
An externally hosted image should be here but it was not working when we last tested it.
Question: The mute cap ground isn't a signal return though i see many boards with this cap connected between the chip and the signal ground, why is that?
Any comments are appreciated.
The input looks a lot more compact.
Add extra pads for all film capacitors to accommodate different Pin Pitch packages.
Change the 100nF film decoupling to X7R ceramic decoupling.
Are you able to solder these X7R directly to the chipamp power pins on the top side?
You may need to add 1r0 into the lead of the 1uF film decoupling, to help attenuate supply rail ringing.
Add extra pads for all film capacitors to accommodate different Pin Pitch packages.
Change the 100nF film decoupling to X7R ceramic decoupling.
Are you able to solder these X7R directly to the chipamp power pins on the top side?
You may need to add 1r0 into the lead of the 1uF film decoupling, to help attenuate supply rail ringing.
Here is a more compact one...
An externally hosted image should be here but it was not working when we last tested it.
Already got all the components (and some spare) so there'll be no need for that.
Why is that? reference to a paper or a book if possible.
I prefer the one with mounting holes in the corners. It may be a little bit worse electrically (probably not a measurable difference) but will be a bit easier to deal with mechanically. Either is fine. It's up to you how many holes you'll want to poke in the chassis before the board fits... 🙂
I suggest 4.7 uF X7R ceramic + 22 uF OSCON + 1000 uF for These Reasons.
~Tom
Hi Tom,
I also prefer the board with the mounting holes in the corners so to keep both advantages i am going to extend the board a few millimeters in the y direction.
Very informative topic but the problem is the largest ceramic i can get is 1uF so if you would suggest another low impedance combination that would be great.
Also the mute cap ground isn't a signal return though i see many boards with this cap connected between the chip and the signal ground, why is that?
Regarding the Zobel network, i am using a 22nF capacitor + 15 Ohm/2W resistor, the values used in your PCB version. Are these values OK or should i use the more popular 100nF+10Ohm combination?
Last edited:
As I state in Post #9 of the bypassing thread, a good bypassing strategy for the LM3886 is to get as much energy storage as close to the IC pin as possible. If 1 uF is all you can get, use that. Just make sure it's an X5R or X7R ceramic as these have the lowest voltage coefficient of the high-k dielectric.
Following the Equivalent Schematic on page 7 of the LM3886 data sheet, it can be deduced that the GND pin serves as a reference for the mute circuit. When the mute function is activated, the IC is servoed to the voltage present on the GND pin (pin 7). If you care immensely about the sound quality when the amp is muted, connect pin 7 to small signal ground. But if you aren't concerned about the sound quality when the amp is muted, connect pin 7 to power ground.
I haven't investigated this fully. If in doubt, stick with the 2R7, 100n used by NSC. I wouldn't go higher than 10 Ω and I would try to keep R*C the same 270 ns as the NSC circuit.
~Tom
Last edited:
Here is the 1uF i have...
http://www.epcos.com/inf/20/10/db/cc_07/X7R_Leaded.pdf
It's an X7R, it's small and i've plenty of those so could you suggest the optimum decoupling combination that provides min. supply impedance and also avoids ringing.
* 2200uF || 1uF (film) || 100nF (film).
* 2200uF || 1uF (X7R) || 100nF (film).
* 2200uF || 1uF (X7R) || 100nF (X7R).
* 2200uF || 1uF (X7R) || 1uF (X7R) || 100nF (film).
or any other combination of the above. 🙂
I already connected pin 7 to the power ground, i was asking about the mute cap +ve terminal connection, signal ground or power ground?
But you used a 15 Ohm resistor in your PCB?
Something that really confuses me is the Zobel resistor power rating, you posted before this equation: P = F*C*(V^2) which yields a rating of about 2~3 Watts depending on the amp supply and the de-rating but i see many boards for chip amps and solid state amps ignore this value and uses a tiny 0.5 or 1 Watt resistor, what gives?
7.5cm*6.5cm.
Last edited:
Scrap the 100 nF. Once you have the 1 uF in there, you don't need it as the lead inductance of the LM3886 dominates the supply impedance at HF.
You do need something to tame the impedance in the 100 kHz to 1 MHz range. A small electrolytic (10~47 uF) would be handy here.
The best way to figure out what would work the best in your board is to run a simulation of the supply impedance as seen by the LM3886. You can see how I did it in the LM3886 Bypassing thread. I linked to it yesterday as well.
Connect Cmute ground to the same ground as you connect pin 7.
I used 15 Ω because that's what I had available. I never said it was the final value or that it was the optimal value. In fact, I kept restating that the PCB was not the final version and that it was not optimized.
Many "designers" don't know how to do the math. They just poke around and if it doesn't blow up, it must be good...
One thing, though: You need to use the RMS voltage in the equation above. Say on a ±28 V rail:
Drop-out across LM3886: 1.6 V --> output swing = 28-1.6 = 26.4 Vpeak.
26.4 Vpeak = 26.4/sqrt(2) = 18.7 V RMS.
So at 20 kHz, full swing: 20e3 * 100e-9 * 18.7^2 = 700 mW.
I tend to derate resistors by 3~4x, so I'd probably use at least a 2 W type if not a 3 W type. If you dissipate the full rated power into a resistor, it'll get screaming hot. As in 250 ºC for 25 ºC ambient. Just look at a data sheet for a power resistor...
Now, one could argue that few people run their amps a full bore at 20 kHz for any length of time. And even modern music has some crest factor, so for a given peak power, the RMS power of the music is significantly less than that of a sine wave. So maybe a 1 W type is OK for use with an LM3886 on ±28 V rails. Just be careful when testing with a function generator... 🙂
Not bad.
~Tom
So now we're back to the following bypassing scheme...
2200uF || 22uF Tantalum (close to the chip) || 1uF X7R (as close as possible).
Which software did you use to simulate the power supply distributed parameters?
The distributed parameters were all estimated from back-of-envelope calculations. 1 mm of PCB trace or wire has a self-inductance of about 1 nH. The closest I can place a leaded ceramic cap to the LM3886 is about 2.5~3 mm from the LM3886, so I figure about 3 nH of trace inductance.
The 22 uF electrolytic is about 6 mm in diameter as I recall, so the trace between the ceramic and the electrolytic caps is probably on the order of 7 mm (= 7 nH).
Etc...
We're not dealing with GHz frequencies here, so there's no reason to turn it into rocket science, but if you wanted a better inductance estimate, you could type your layout parameters into one of the many PCB inductance calculators (such as this one). Standard "1 oz PCB" has a copper thickness of 35 µm (which is why it's called 35 µm board in the rest of the world). The dielectric is FR-4 fiber glass which has a dielectric constant of about 4.2~4.4 and a thickness of 1.6 mm.
Once you have the lumped parameter models (i.e. the trace inductance), you can use any circuit simulator. I used TINA-TI (free download from TI.com).
~Tom
The 22 uF electrolytic is about 6 mm in diameter as I recall, so the trace between the ceramic and the electrolytic caps is probably on the order of 7 mm (= 7 nH).
Etc...
We're not dealing with GHz frequencies here, so there's no reason to turn it into rocket science, but if you wanted a better inductance estimate, you could type your layout parameters into one of the many PCB inductance calculators (such as this one). Standard "1 oz PCB" has a copper thickness of 35 µm (which is why it's called 35 µm board in the rest of the world). The dielectric is FR-4 fiber glass which has a dielectric constant of about 4.2~4.4 and a thickness of 1.6 mm.
Once you have the lumped parameter models (i.e. the trace inductance), you can use any circuit simulator. I used TINA-TI (free download from TI.com).
~Tom
Last edited:
No harm at all.
It would be nice with access to a good EM field solver, but, honestly, any frequency relevant to the LM3886 is basically DC as far as those are concerned. Once the working frequencies get up in the 1~10 GHz range, life becomes much different - and in many ways more interesting.
~Tom
It would be nice with access to a good EM field solver, but, honestly, any frequency relevant to the LM3886 is basically DC as far as those are concerned. Once the working frequencies get up in the 1~10 GHz range, life becomes much different - and in many ways more interesting.
~Tom
- Status
- Not open for further replies.