Hi !
Do you think it's ok this ground wiring of my home made NAS ? The yellow rectangles are separated modules and the connections are wires, not pcb traces. I'm not sure if I should connect those ground wires between Raspberry and Nas Latch, and Raspberry and Fan. I need to control the Nas Latch (a mosfet switch) and the Fan (PWM) with the Raspberry, so I need these devices have common ground to work. But they have the GND connected to the power supply too. Isn't that a ground loop ?

 

But they have the GND connected to the power supply too. Isn't that a ground loop ?

Yes.
Any ground wires carrying the fan current should have just one connection to the star power supply ground.
You don't want any of that ground current going through the circuit ground back to the supply.

Similarly any supply wires carrying the fan current should have a separate connection to the power supply.

 

So, if I cut the GND between the Raspberry and the Fan, and I let the pwm signal gnd to go throug the power supply gnd, is better ?

 

So, if I cut the GND between the Raspberry and the Fan, and I let the pwm signal gnd to go throug the power supply gnd, is better ?

From a ground-loop perspective, yes.
The PWM signal is digital, so shouldn't be much affected by any ground noise from the fan current.
Worst-case, you could use an opto-isolator for the PWM signal.

 

What about the ground loop made between the Power Supply - Raspberry - USB Hub, through the USB cable ? I Cannot cut the USB cable ground... I already tested with this ground loop of the Hub (but not the others 2) and it is working fine.

 

What about the ground loop made between the Power Supply - Raspberry - USB Hub, through the USB cable ?

Sometimes, as here, it is difficult to eliminate all loops.
If it works with the loop, then you are okay.
Using larger wires to carry the high currents will minimize the problem.
Also use proper 100nF decoupling caps across all the module's power to ground pins.

 

I worked primarily in the industrial CNC field where there was 5v processors , servo motors, contactors and other "noisy" devices, all interconnected.
The practice was to take all earth grounded points to a central Star Point GND, which included the Service GND.
There is an interesting publication by Siemens on the subject, included the recommendation now to GND both ends of a shielded cable.

 

I think now it should be fine...
I drew with yellow the control signal path from Raspberry to NAS Latch, and with green the pwm signal from Raspberry to the Fan. Now it will use the power ground as return. I hope it is ok...

 

The actual "ground loop" consists of the current from more than one device flowing in the same conductor. Of course at some point that is going to happen because it ends at the power source. So the "star point" that Max mentioned is the common work-around. A similar concept applies also tp the other side of the power connections as well. So there is the explanation, it goes right along with what Max described.
and it really does work.

 

As the Siemens article shows, equi-potential bonding of a machine connecting multiple ground points effectively eliminates ground loops which refer to a different potential level between in more than one GND point.
Prior to this custom, it was recommended that any shielded cable be earth grounded at one end of the shield only, due to the different possible potential at each end of a shield.
The only down side was, the shield could act as a 'radiator' of EMF or 'noise'.
Equipotential bonding eliminates this and allows the earthing of both ends of a shield.