Part 2 - Mitigating common mode noise

In my previous blog post, I outlined the fact that differential signaling can still be negatively impacted by common mode noise. In this post, I will outline some methods that you can use to reduce the impact of common mode noise on differential signal paths. I’ll cover three main categories of devices: passive circuits, high-impedance active circuits and low-impedance active circuits.

Let’s start with the simplest example, passive circuits. System designers often use an LC filter to eliminate noise. As commonly implemented, a differential filter will be floating with respect to ground. This is fine for the differential response, as shown in Figure 1. However, as shown in Figure 2, there is no attenuation of common mode signals. By implementing a simple change in the filter topology, as shown in Figure 3, we get the response shown in Figure 4.  

Figure 1: Filter differential response

Figure 2: Filter common mode response with 2.5k Ohm of common mode “termination”

Figure 3: Simple filter change to shunt common mode energy with one extra capacitor

Figure 4: Filter common mode response with change of Figure 3

Like filters, some active differential devices do not have any ground reference (or termination) for common mode signals. One example is the LMH6521 and other amplifiers with inductors connected to the output pins. The LMH6521 amplifier is a very high-speed, digitally controlled variable gain amplifier (DVGA). It has a fully differential signal path, with a 2.5k-Ohm common mode resistance, which is not sufficient for it to be a major attenuator of common mode energy, as shown in Figure 2. 

You can add more common mode energy dissipation to the LMH6521 by adding some resistors to the circuit, as shown in Figure 5. Through the beauty of differential signaling, the two 500-Ohm resistors add only 1000 Ohms of load to the differential circuit, but they provide 250 Ohms of common mode termination. The amplifier output load changes from 180 Ohms before the resistors are added to 152 Ohms after. This change in load condition will have very little impact on the amplifier’s performance, yet the common mode termination is much improved. The same technique could also be used at the amplifier input.  

Figure 5: Common mode termination of a differential amplifier (additional resistors shown in green)

Not all differential devices are high impedance with respect to the common mode. Some amplifiers, such as the LMH3401, have common mode control circuits that set the amplifier output common mode with a low-noise, low-impedance circuit. The LMH3401 has a common mode circuit that is basically a unity gain amplifier. It takes the voltage at CM and buffers it at the amplifier outputs, such that the common mode of the amplifier output pins is fixed. 

In Figure 6 below, I show the impact of this circuit as dotted capacitors. With respect to the common mode, the amplifier outputs are a virtual short to ground for any AC noise in the system. 

Figure 7 shows the common mode output impedance of the LMH3401. By selecting an amplifier with these features, you achieve built-in common mode rejection as an extra benefit, and you don’t have to do anything extra with your design.

Figure 6: Differential amplifier with low impedance common mode control

Figure 7: LMH3401 common mode output impedance (including on chip resistors)

As you can see, there are several ways you can help mitigate common mode noise pickup on differential signal paths. Some of them require a bit of planning and some extra components. An easier approach is to select an amplifier that’ll do the work for you.

Bonus:

Do you see the error in Figure 3? Leave a comment if you find it… 

Related resources:

Anonymous
  • Richard;  The 10M resistors are also an error of sorts.  With "perfect" inductors and capacitors TINA-TI cannot find a DC operating point for differential circuits.  The solution is to use imaginary ground ties at different points of the circuit.  Another solution is to edit the properties of the capacitors and use 1G Ohm instead of infinite for the parallel resistance.  Likewise the inductors should have an equivalent series resistance entered into their properties.  Most RF inductors are around 50 to 400 milli Ohms of resistance.  Once these properties have been changed the resistors to ground can be removed.  

  • Thanks for the comments!

    Eric has the correct response.  Two capacitors in series have 1/2 the capacitance of a single capacitor.  In order to be equivalent they each must be double the original value.

    The single ended drive is correct, this is how you must drive the circuit to model the common mode response.

  • I agree with Eric but not the Watcher as we are testing common mode response...

    The newly added 10M resistors though are not symmetrical!  Surely one should be added to each arm of the differential signal.

  • The error is that the source is single-ended when it should be differential. R2 should connect to the (-) terminal of VG1 instead of R1 and the ground connection should be removed.

  • The error in figure 3 is C1 and C3 should be 17pF in the changed circuit to be equivalent to the original circuit.