Differential or Not?Differential signaling is an important way of maintaining signal integrity in the presence of adverse electrical noise, and of controlling radiated emissions and immunity. The advantages of differential signaling is the ability to ignore common mode voltages, as are typically encountered in longer transmission distances. This is taking on increased importance at all levels not only for long cable runs but for differential signals at the board level. the circuit board to the chip itself high speeds are demanding impedance control even for short runs. Two situations need closer attention when dealing with differential signaling. Let’s look at the characteristics of the transmitter at the chip and the signal path. Chip Transmitter Figure 1. Common mode currents generated by timing skew First to note, most differential drivers are not really differential there are two complementary outputs, each referenced to circuit ground. In principle, these two outputs could be routed to separate destinations, rather than as a differential pair. If the signals were truly differential, or even truly complementary, the sum of currents in the two legs would be equal to zero. But any difference between the two legs, whether amplitude, signal timing, impedance or path length, will result in a net common mode current, which will return on circuit ground, perhaps by a circuitous route. An example of timing skew is shown in Figure 1. For short runs, say, chip to chip, the skew is generally ignorable (but then, why use differential at all). Most differential applications involve going off board, perhaps for extended runs left unchecked, this common mode component will go out the cable and radiate from there. This is not an emission problem if you are using well-terminated shielded cables, but without shielding or with mediocre shield termination, you need to block the common mode currents. The simplest choice is simply a common mode choke, selected to have significant impedance at the fundamental data rate, and extending up to the fifth harmonics. Impedance Control at the Circuit Board Figure 2. Differential current paths with and without a ground plane. In parallel to the assymetries mentioned above, we need to consider impedance control in the signal path, especially where impedance control is needed. As we have cited in past articles, the return path is equally important for impedance control. We often hear that the ground return path is not important for differential signals, which serve as their own return path. This would be true if the signals were truly differential or if the path was exclusively between the signal pairs, but as we said above, the signal origins are not differential, only complementary. Figure 2 shows the actual signal and return current paths. Over the ground plane, some of the return path flows in the ground path as indicated (how much depends on the signal line tracing). The signal currents continue when you cross the ground plane boundary to, say a connector, but the return path is blocked at the end of the ground plane, creating an impedance discontinuity. Now let’s look at the characteristic impedance in the signal path. Figure 3 shows the impedances where the signal paths are over the ground plane and with the ground plane removed. Each signal line, considered separately, has its impedance to the ground plane under the trace, Z1. In addition, there is an impedance between the two signal paths, Z2. At the circuit board level, Z1 will be the dominant factor, with Z2 being secondary. So the effective line to line impedance is approximately 2*Z1.  Figure 3. Characteristic impedance with and without a ground plane But what happens when the signal pair leaves the board to a wire pair? Z1 drops out of the picture, leaving only Z2 as the impedance of the signal path. There must be a transition between the two signal to circuit ground impedances and the wire pair. Unless you handle this one carefully, there will be an impedance discontinuity at the boundary, resulting in signal degradation and possibly emission issues. We suppose with careful modeling, you might come up with a way to transition without a significant discontinuity, but even if you can, you will need to custom-build the connector not a palatable alternative. So What Can We Do?First, we place the transmitter or receiver close to the connector boundary, the intent to allow the associated impedances to be treated as lumped rather than distributed. We may have to empirically juggle impedances at the board level to get best results, but this would be much easier than trying to match up distributed transmission line mismatches by modeling metallic members at the boundary. How close do we need to be? Reflections become significant when the roundtrip propagation delay time is equal or greater than the signal risetime. For a typical circuit board dielectric, the propagation delay is about 15cm/nanosecond of trace length. At the higher frequencies, the risetime is approximately that which you would get in a sine wave, so as a quick approximation: L = 2/f, where f is frequency in GHz,
Thus, for a 1GHz signal, we find the maximum run length is 2cm. In the simplest case, this is the distance from the transmitter or receiver to the connector: all termination activity must be within this 2cm. Once you leave the circuit board, you may have a shield or not actually, twisted pair wiring is very effective for reducing differential mode currents. Ethernet uses a common mode choke at this boundary to block common mode currents this works for emissions of the signal cable and for RF immunity, as well. Of course, these magnetic elements do take up some room. USB eliminated the magnetics, putting a greater requirement on the chip transmitter/receiver. SummaryDifferential mode signaling, especially at high frequencies, is increasingly common in high speed data streams. You need to pay attention to the impedance paths don’t let the textbook lore on differential signaling prevent you from considering all the impedance factors involved, especially at the transition from the circuit board to the cable. You can avoid a lot of problems if you consider these factors and take care of them immediately at the boundary. | advertisement |
