The Amazing Twisted Pair

By William D. Kimmel, P.E.
and Daryl D. Gerke, P.E.
Kimmel Gerke Associates, Ltd.

We continue to be amazed by the effectiveness of the lowly twisted pair wire in reducing field coupling to and from the wire pair. Let’s take a closer look at the twisted pair and see how it works and where it doesn’t work.

The Basic Principle

The arguments apply for coupling to and from the wire pair, but it is easier to visualize for fields emanating from the wire. Start with an untwisted wire pair (Figure 1a). For a long wire, fringing effects can be ignored, and the magnetic field surrounding each wire is simply H = I/2R. In the ideal case, the signal current and return current are equal, in which case the field from the return path nearly cancels that of the signal wire. The net field in the worst direction (magnitude measured in the plane of the two wires) will be H = I*s/πR2, where s is spacing between the two wires.

Thus, we see that the closly spaced wire pair is a poor antenna – the H field falls off at the square of the distance – and the closer the spacing, the poorer the antenna. Now, let’s make it even poorer by twisting the wire pair (Figure 1b). Now we see that the H field reverses direction each half twist – thus, at even a small distance from the wire pair, the magnetic field cancellation is nearly complete. Effectively, the loop area from one half twist is cancelled by a negative loop area from an adjacent twist.

So we have the serendipidous situation that the already weak magnetic field from the closely spaced wire pair is made much weaker by twisting.

It’s hard to appreciate how effective the cancellation is, until you’ve tried it. Even at very close spacing (as would occur between adjacent twisted wire pairs), cancellation is very good, providing some simple conditions are met (as described below).

What Are the Necessary Conditions?

First, cancellation only occurs to the extent of balance between signal current and return current. Any difference currents will not be canceled – common mode currents fall in this category.

Second, the lengths of the twists must be short enough that distributed effects between adjacent twists can be ignored. Our rule for satisfying this condition is that the twist must be less than 1/20 wavelength of the highest frequency component expected. For digital signals we expect to see significant frequency components up to about 1/tr, where tr is risetime – a one ns risetime will produce frequency components up to about 300MHz. It is not mandatory, nor is it even desirable that the twists be uniform for cancellation to be effective.

Third, twisting of adjacent twisted pairs must not be uniform. If adjacent twists are exactly the same, cancellation is pretty much negated – effectively, the second twist untwists the first twist. Thus, twisted wire bundles are twisted randomly to prevent this situation from occurring.

Where Does the Twisted Pair Fail?

Well, the twisting fails where the above conditions are not met. Let’s take a look at the effects and how they can occur.

First, the twisting is not tight enough. As indicated above, this will start to occur when the wavelength is shorter than 1/20 of the twist. Taking 300MHz signal as an example, wavelength is one meter, so the twist much be less than 5cm, or about two inches. As is seen, this is not a significant restriction – two twists per inch puts you up past one GHz, and it’s unusual to see twisting that loose.

In practice, the principal deficiency is not in the cable bundle occurring but in the connector. Clearly, the connector pins aren’t going to be twisted, and you will actually have to run the wires straight for a little while on each end of the connector as well. That’s one reason you will find high fields occurring at the connector, even for unshielded cables.

Second, the currents aren’t equal. This can occur in two ways, either because the signal is not symmetrical (one leg is a little slower than the other, producing a net glitch each transition) or because there are common mode currents originating on the driver circuit board.

How Do We Fix It?

Ethernet is an outstanding example of the effectiveness of twisted pair wiring. Figure 2 shows a typical circuit design. First, we see a common mode choke is used to equalize the signal currents. One of the common problem with differential drivers is that they aren’t quite exactly differential – if one leg lags the other by a small amount, the result is a net current difference at transition time. The common mode choke suppresses the difference.

The second is the transformer, which provides isolation as well, being effective for common mode currents, whether generated by the ethernet driver or anywhere else on the circuit board.

Note that the key is to block common mode currents – they will not be reduced by twisting. If you can’t adequately block the common mode currents, you will need shielding – fortunately, shielded twisted pair (STP), commonly used in high speed ethernet, is readily available.

Interestingly, we need not restrict ourselves to a wire pair. If we have a wire bundle, and our concern is magnetic field coupling to and from the bundle, we can twist the entire bundle and still get good cancellation – not as good as with individually twisted wires, but still good. A typical case would be the power harness. Note that this approach does not minimize crosstalk within wires in a bundle – if you need that, stick with the individual twisted pairs.

Summary

The twisted pair is an unsung hero of EMI control. It is inexpensive to implement, tolerant of variation, and very effective. The downside is that it only works for differential mode currents – any deviation from complete balance and the cancellation is lost.

You should make it a categoric practice to twist your wires and harnesses – you’ll eliminate lots of problems before they start. And while you are at it, route the wires along a ground plane to keep common mode loop areas to a minimum.