An open source layout for LM3886?

Like the last version, still not convinced about the missing GND plane under the feedback network though. I would prefer less of a voltage drop at such low frequencies.

It would be great if you made PSU companion boards as well!
With decent LM3886 board alone, biggest improvement would come from better PSU!

I was looking into PSU solutions for class AB, and the situations is not all that great:
-CRC does not work well with class AB current pulses due to the V drop on R
-CLC, good luck finding a suitable inductor, even if you do find one, you have the R component of inductor and you are back to CRC problem
-capacitance multipliers, not sure if they would work that great with class AB, they have their own problems even with constant current load
-high quality premade SMPS(connex for example, maybe the best and most economic solution in this situation)
-brute force with paralleling caps.
 
About the PS.

I wasn't planning something very fancy. Here is a first look. It's on a 100*60mm pcb, holes on the 10mm grid. Connections either by faston or screw terminals.

The rectifier on board is a vishay GSIB1540, a 15A/400V bridge. It is best heatsinked. Btw, heatsinks for bridges aren't that common and the mouser search system doesn't help. I finally found the aavid 530101B15G but it's pricey at 5.36€. :mad: There is provision for c+rc snubbers before the bridge.

There are 8 caps on boards, 2200uF/50V, either nichicon UPW or pana FC, for a total of 8800uF per rail. Prices and performances are similar. That is more than enough for a pair of lm3886 in my view. Someone willing to go all out could use two boards...

Expect 30€ vat included to fill the board from mouser.

Now for the fancy things... it's about the same price to order a 60*100mm board or a 100*100. So I've got 40*100mm I could fill. I was considering two options. One would be DC offset protection, with mosfets as ssr. The other would be an auxilliary +/-15Vdc regulated supply.

Some ideas ?
 

Attachments

  • lm3886-ps.PNG
    lm3886-ps.PNG
    84 KB · Views: 1,461
I do the last one with chip amps...CCC
Easiest, most economic and probably best solution.

Could you elaborate on this? I can extend a ground plane under those resistors. But which one? And what for exactly? Shielding?

From what I have understood, Mark is worried that PWR ground carries high current under the feedback network, I don't see that in this layout as the currents at such low frequencies act as DC - shortest path.
Bigger groundplane=less inductance, less of a voltage drop, shielding, etc...

I finally found the aavid 530101B15G but it's pricey at 5.36€
I have used this one , two times cheaper, you only need some thermal grease and d=3.6 mm m3 screw.

Try these guys-

PCB Prototype & PCB Fabrication Manufacturer - JLCPCB

Might be a good option for cheap PCBs. 100mm square for $2/10!
That is CHEAP!! $15 for a four layer board. Did you ever use their services, if so are they any good? I used PCBWay and I have been pretty happy.
 
That is CHEAP!! $15 for a four layer board. Did you ever use their services, if so are they any good? I used PCBWay and I have been pretty happy.

This guy seems very happy with them-

Custom PCB Board from JCBPCB

I'm going to try them out soon for some tube stuff. I might be a willing guinea pig for their services once you get a layout and gerbers finalised, as have a dozen or so LM3886TF I otherwise wouldn't use.
 
@bozoc: if the aim is to lower the power groundplane impedance, then it gets back under the feedback resistors. In order to lower it even further, I made use of the empty space in the middle to put another groundplane on the bottom, stitched with the top one.

Thx for the heatsink. I had seen similar models from other brands but they were out of stock.

I attach the eagle files in the zip, if anyone wants to play with them.

@Lingwendil: that's very cheap, so cheap I'm a bit wary of it... I'm pretty happy with the quality and speed of smartprototyping as it is. Still, I might try it once.
 

Attachments

  • lm3886th.png
    lm3886th.png
    72.2 KB · Views: 1,431
  • lm3886th-bot.png
    lm3886th-bot.png
    39.6 KB · Views: 1,407
  • lm3886th-top.png
    lm3886th-top.png
    36.9 KB · Views: 1,391
  • lm3886th.zip
    71.7 KB · Views: 206
  • Like
Reactions: 1 user
@bozoc: sadly that heatsink didn't fit. Not wide enough for the diode bridge :(
@Lingwendil: yes, that's worth a try. On 2 layers boards, their cost of shipping is actually a bit higher than smartproto, so the final price is about the same.

Anyway, here are two auxiliary pcb. One is the PS (60*100mm), one is a simple DC protect (60*40mm). Eagle files in the zip. There's a small error to correct on the ps, the top molex connector should of course be labeled v- and not v+.
 

Attachments

  • lm3886-aux.zip
    180 KB · Views: 227
  • protect.png
    protect.png
    45.3 KB · Views: 2,479
  • protect-sch.png
    protect-sch.png
    12.4 KB · Views: 2,462
  • lm3886-ps.PNG
    lm3886-ps.PNG
    38.9 KB · Views: 2,461
  • lm3886-ps-sch.PNG
    lm3886-ps-sch.PNG
    9.7 KB · Views: 2,523
  • Like
Reactions: 1 user
The DC protect board is mostly coming from the discussions here and the detector is a simplified version from ESP protection project. The added led indicates when the protection triggers.

It works well in sims and I've used that detector before so I know it works well. This implementation is pretty bare with no mute at turn-on, turn-off but the lm3886 is well behaved under that respect. It's made to be powered from the positive power rail.
 
Here is a link to updatemydynaco.com, it shows why an HBR works and how to implement it. As pointed out before, just place a wire if you don't think you need it.

A layout is always a compromise. I connect the feedback resistor close to pin3, therefore it is difficult to extend the AGND under the resistors. What is the effect of this?

If we look at the loop area of pin 9 and 10 to PGND (feedback+input). Placing the same plane under these circuits will effectively reduce loop area.
 
Sorry for the long answer but I'm trying to understand how this goes...

Here is a link to updatemydynaco.com, it shows why an HBR works and how to implement it.
In plain terms, you'd rather have the feedback cap ground pin return to the input signal groundplane ? I did reproduce the sim from akitika + Tom's sims on current returns and it's indeed the best solution.

A layout is always a compromise. I connect the feedback resistor close to pin3, therefore it is difficult to extend the AGND under the resistors. What is the effect of this?
I'm not quite clear on what you call AGND. Is it the input gnd plane ?

Ok, here's my understanding of the problem.

Let's consider the loops formed by the feedback resistors. Basically, the feedback network is a divider at the output of the amp. There's a first current loop formed by the feedback resistor and pins 3-9. Gnd aren't included into that loop. So it's only a matter of keeping the loop small, to avoid to pick up radiated crap. It's also a matter of capacitive coupling to the gndplane under it. It's best not to run dirty high current tracks under it. And it's finally a matter of parasitic capacitance at pin 9. For that loop, the compromise is whether to go searching for an optimal sense point, at the risk of picking more radiated noise.

My take on this is that as long as the sense track (from the feedback resistor to the amp output) stays close to the output pour (which is low impedance) to minimize loop area and is not too contaminated by capacitive coupling from the power ground plane (hence a thin track), we're good. That's why I followed Tom's suggestion on this. Btw, that's apparently how it is done on the lm3886dr.

Then we have a second loop, with more current. It's the loop going from the output of the amp, through the two feedback resistors, the feedback cap, the ground loop breaker resistor, power gnd, the power supply, the supply pins of the lm3886. If we want to minimize the loop area, we actually have to run the feedback resistors over the power ground plane, not the input one. While the input gnd is tied to the loop, no current from the loop has to run through it.

The compromise here is in between loop area, which would require the power gndplane to run under the feedback network and capacitive coupling of noise. With through holes resistors and short tracks, probably no big deal. Another aspect of compromise is how we're using the output connector as sense gnd so the layout has to conform to that.

If we look at the loop area of pin 9 and 10 to PGND (feedback+input). Placing the same plane under these circuits will effectively reduce loop area.
By pgnd, you mean the signal/input groundplane or the power gndlplane ?

BTW, could we agree on a common terminology ? Let's say there a signal groundplane (sgnd) and a power groundplane (pgnd) on this pcb. AGND sounds too much like analog ground.
 
Well, there are two issues here. One is how things are joined and the impact of common tracks' impedance. No objection on that, the feedback cap has to go to the input gndplane. The other thing, less clear, is about loop area and radiations pickup. Groundplanes are relevant to both issues as provide a low impedance connection but also allow to reduce some loops.

So, here's an attempt to satisfy all the compromises. Yeah, I wish... As before, the zip is for the eagle files.

The feedback cap now returns to the input gnd. The feedback resistor going to gnd is over the split between the signal and the power gndplanes. If you consider the loop mentioned in the post above, it's quite tight.

The two gndplanes are joined near the output connectors by a resistor. It can be jumpered in case of a mono amp.

Better now ?
 

Attachments

  • lm3886th.png
    lm3886th.png
    62.8 KB · Views: 3,604
  • lm3886th-bot.png
    lm3886th-bot.png
    32.1 KB · Views: 874
  • lm3886th-top.png
    lm3886th-top.png
    35.3 KB · Views: 640
  • lm3886.zip
    71.8 KB · Views: 474
For my TDa7294 pcb I went for a copper pour for the low level audio ground.
I starred the power supply ground and Zobel ground connection to the zero volt pcb input pin.
This way I don't have any real current going through the ground plane and modulating it.
The smoothing capacitors are important too. The input from diodes are from one side only of the smoothing caps. And the output from smoothing goes to TDA7294.

The results were pretty much zero noise and hum.