Thanx for the reply but I don't understand it
Then I suggest working on gaining understanding. Play with the different grounding scenarios in a circuit simulator. Or build it several ways and measure it. The former is the less expensive and faster option.
The whole GND star is just 4mm's away from the Load GND.
The use of a ground star is problem #1. Here's why: You want as little voltage drop across the ground as you can possibly get. To accomplish this, you need the lowest impedance you can possibly get. Recall, impedance is both real (resistive) and imaginary (reactive (inductive/capacitive)). At audio frequencies, the resistance and inductance are the ones that matter the most and you need to drive both of them as close to zero as you can get as this will result in the lowest error voltage. Any error voltage will degrade the THD of the amp, resulting in crappy "chipamp" sound (rising THD versus frequency). The lowest possible impedance you can get is a ground plane, so use a ground plane.
The argument against a ground plane often goes, "but the currents will mix". That's true. They will. However, by designing the ground plane correctly, for example by keeping all the high currents flowing primarily in one end of the plane with the other end being relatively quiet, you can get good separation between currents. That way you can keep the supply/load currents and the feedback currents separate, for example. Both supply/load currents (A) and the feedback currents (mA) are nasty if they get into your reference ground.
To keep the reference/input ground separate, you can either pour a separate plane for it or you can make a moat in your current plane using anti-copper to make an island in the plane. The small input/reference ground plane island should join with the main ground plane at the output ground connection.
Pouring a ground plane in high-end audio is not as simple as just plopping down a rectangle in the CAD tool and calling it done. You have to put some thought into where the currents flow and make sure you keep the large currents away from the sensitive spots, either by careful component placement or by moating. Most likely you'll be using both methods. Do beware that whenever you slice the ground plane, you'll increase its impedance, so you need to think about where to put the input section to avoid excessive hacking of your main ground plane.
The only substantial length is the GND for the in+ resistor but that comes out by the load gnd. How should a load current result in a measuring fault?
The load current creates a voltage drop across your ground connection. With a ground star, you maximize this voltage drop. You then proceed to put it in series with your input signal ground through that 4 mm connection to the load ground. That's the problem; or rather part of it.
Well then you didn't fully read or understand what he wrote, Tom goes to great length to explain what's important and why.
Thank you.
we all start out ignorant and work our way to enlightenment.
Yep. It takes work. The results are well worth it, though.
Since this is germane to this thread and it popped into my head when I was reading through your "Taming the LM3886" grounding page I thought I would post it here.
I have always wondered WHY the speaker return "ground" is tied into the amplifier PCB at all.
As I write in the first paragraph of the An Improved Grounding Scheme section (second time I quote it on this thread, by the way):
From a feedback theory point-of-view, it seems apparent that if the voltage across the load is what we intend to control, the feedback voltage should be measured as close to the load as possible. This means that GND_SIG should connect to GND_LOAD.
Maybe I should add a qualifier: This means that GND_SIG should connect to GND_LOAD with a low-impedance connection.
I would also invite you to think about where GND_LOAD should connect. At frequencies above a few hundred Hz the vast majority of the current flows from the local bypass caps, through the IC, through the load, and back to the local bypass caps. At least it does if you follow my suggestions for bypassing on my "Taming the LM3886" pages. This means GND_LOAD should connect to the PCB, preferably as close to the local bypass caps as you can get and tied to them via a ground plane. This means that GND_LOAD and GND_SIG will meet on the PCB as well.
Thinking of loop area or return path lengths, I suppose that returning the current to the ground pins of the PS caps on the amplifier board makes sense.
Bingo!
On the other hand, the currents that are driving the amplifier output devices are coming from the main power supply caps and not just the caps on the amp PCB itself (unless they are the same because of onboard PS).
Correct. That current flow is from the main reservoir cap, through a wire to the board, through the local bypass caps, back through a wire to the main reservoir cap. Keep those wires relatively beefy (low impedance) and tightly coupled either with a very tight twist or with zip ties (low loop area).
This would seem to eliminate problems with high return currents from the loudspeaker in the PCB, which can corrupt the ground by inducing voltage, etc.
No, it just moves the problem to the supply wiring and increases the loop area. There is NO problem having the load currents on the PCB if you pay attention to where the current flows, minimize the impedance the currents flows through, and minimize the loop area.
It is where you connect the feedback loop to the load that is important.
That is 100 % true
as long as you keep in mind that the feedback loop has two connections: signal and ground. You are trying to control the voltage across the output of the amp. This means, the input to the feedback loop needs to measure the output of the amp. Connecting the signal part of the feedback loop is easy. Just tap off at the pin of the output connector (or the "inside" pin of the Thiele network, if used). The feedback ground connection is usually the one that causes trouble.
There are two reasons for the trouble. One issue is that with a crappy layout, part of the load or supply current ends up flowing in the feedback ground, adding to the feedback signal. I addressed how to keep the currents separate earlier in this post. The other issue is that the feedback current is pretty nasty. It basically contains the error-correction signal imposed by the loop. You definitely want to keep this away from the input signal.
One way to keep the feedback current separate is to run a long and skinny trace to the ground reference (where the GND_LOAD and GND_SIG join). That will keep the current separate, but it will also drastically increase the inductance of the feedback ground connection. This will result in a degradation in THD+N at frequencies above a few hundred Hz due to the reduction in gain caused by the increasing magnitude of the feedback impedance. Recall the magnitude of the impedance of an inductor rises with frequency. This is why we use a ... wait for it ... ground plane to connect the feedback ground.
The final issue is the input ground. In a single-ended amp, adding a voltage on the input ground adds a voltage directly in series with the input signal. It should go without saying that the input ground should be kept quiet for this very reason. That's why I suggest using a small island in the main ground plane if you can do so without mucking up the main ground plane as I described above.
Tom
