Proof that diff pair comms cables need 4 wires minimum.

[cupoftea]

Hi,
This concerns differential signalling comms, on a cable, between two PCBs.
The attached is proof that a differential signal pair needs two ground returns to be sent with the diff pair….ie 4 wires in total. Do you agree with this?

The currents in the diff pair certainly do not return via each other...they return via the respective ground. Therefore, in order to reduce the size of current loops, the grounds need to be taken with the diff pair. Surely you agree?

LTspice simulation, and PDF schem, attached.

 

[KlausST]

Hi,

I don´t understand.
In best case a differential pair just needs two wires, as in your PDF.

One has to guarantee to be within the receivers common_mode_input_voltage_range.(Mind GND bounce between transmitter and receiver)
In some cases it´s a benefit to add a GND wire (shield) for this purpose.

But the GND is not needed as a return path. The "pair" can be seen as it´s own "drive and return system".

BTW:
* your driver circuit is not symmetric regarding HIGH_impedance and LOW_impedance.
* your reciver circuit is not symmetric in N_impedance and P_impedance.

A good differential pair communication system has all three equal
* source impedance
* characteristic cable impedance
* load impedance

I miss all this with your design.

Maybe it´s a good idea to read some application notes on how all this works together and what happens on mismatch.

Klaus

 

[betwixt]

Think of the receiving end as being like a DVM connected across the wires. All it needs to know whether a '1' or '0' is being sent is whether the voltage reading is positive or negative. As Klaus clearly stated, you can't have a high voltage difference between the sending and receiving ends simply because it could be outside the common mode range of the receiving device. A ground connection isn't mandatory if you can be certain the common mode range will never be exceeded but adding one will ensure that is always the case.

Brian.

 

[dick_freebird]

A balanced differential pair returns only the imbalance residue along the ground.

 

[cupoftea]

Maybe it´s a good idea to read some application notes on how all this works together and what happens on mismatch.

Thanks, unfortunately no app notes show the transceiver circuitry .
The LTspice sim of a diff signalling system of the top post is the very simplest it could be. It seems a waste of time to add further components to it in order to make the impedances equal.
I appreciate the requirements of the theory, but its like in offline SMPS, all the App notes say that the impedance in line and neutral must be equal otherwise you will get common mode noise problems.....but few people bother with the expense of doing that, and they still pass EMC.

The Sim of the top post shows the diff pair going into a differential opamp, which means only the difference is amplified, and the noise is cancelled out.....so as far as i can see, the LTspice sim of the top post, is fine as it is, surely?

If a differential signalling transeiver is more complex than that, then you would have to ask if its worth while?

 

[FvM]

As already pointed out, the simulation is completely missing the typical differential signalling setup by using an intentional asymmetrical driver, supplemented by asymmetrical load impedances.

I understand that you arrive at "four" wires by treating the lines of a differential pair as separate signals which is already misunderstanding the nature of differential signals that have ideally no common mode component in the signal band and at its harmonics.

An appropriate analysis of differential and common mode components in data interfaces will model the cable as transmission line with diffrential and common mode impedance rather than "four wires". Also use realistic driver and receiver models.