Connector impedance matching

Question

The importance of termination matching the cable is emphasized all the time. But before the wires reach the termination resistor they usually pass the connector. I've only seen the impedance specified for some F-type connectors, while CAN, RS485 and other differential lines are routinely connected with DB9, RJ45 and even simple screw terminals. If the impedance is a property of geometry and dielectric, I fail to see how those can be even close to the cable itself. Would not they create reflection points in the wiring?

From my own experience, long time ago I've tried splicing Ethernet and USB cables by soldering. I was keeping untwisted length to the minimum (under 10mm) and used very little solder. Neither did work anyway. This is even more puzzling since the cables soldered to ubiquitous USB connector breakout boards work just fine.

Similar question here has a bunch of contradictory answers but no definite or accepted one.

Answer 1

Yes, connector impedance matching is crucial when the high speed signals need it to pass through without reflections.

Now, CAN and RS-485 are far from being called high speed.

High speed signals require fast slew rate on signal edges and transitions to achieve fast data rates.

Generally you only want as fast edges as you need, and for e.g. CAN and especially RS-485, you can select a chip or configure it for certain data rate which means it will have the required slew rate to still work but not faster than you need to create EMI/EMC problems and impedance mismatches or discontinuities causing reflections.

So CAN and RS-485 are generally not high enough speed for connector impedance discontinuity to matter a lot. Of course it may matter a bit. For example RS-485 is used on DMX-512 theatre/venue lighting buses and use XLR connectors - not especially high speed connector either.

These buses anyway allow for small stubs, such as 30cm to 100cm on a 100m bus, and that is because slew rate and data rate are slow enough to allow it.

Faster buses may develop problems simply by having a too large resistor or capacitor pad on the PCB track, or a few millimeters of stub, or incorrectly designed via between PCB layers. That's why faster buses require suitable connectors, such as USB Type-C, DisplayPort, HDMI, or coaxial BNC - these are designed to pass the signals with roughly the expected impedance, and you may need to do special tricks with the PCB design to compensate for the effect of the connector, such as leaving out ground plane on the second layer under the first layer connector pads.

In short, for RS-485, the signal transition time is so long that even bad connectors are electrically too short have much effect on the signal.

Answer 2

I fail to see how those can be even close to the cable itself. Would not they create reflection points in the wiring?

It's all about the "mismatch" producing near-zero effect at lower frequencies but, progressively getting worse as frequency rises towards 1 GHz and beyond. The general rule of thumb is that if a "connection anomaly" has an electrical length that is within one tenth of the wavelength of the useful frequencies being transmitted then, you might get into difficulty.

300 MHz (for example) has a wavelength of 1 metre but, in electrical systems (due to the slower speed of cables), the wavelength is more like 67 cm. And, for example, if your data rate is 6 Mb/s (six million bits per second) then, that is equivalent to 3 MHz (as a frequency) and, 3 MHz has an electrical wavelength of of 67 metres.

But, your data isn't a sinewave and, you might consider that up-to the tenth harmonic (30 MHz) is of significance. That brings the electrical wavelength down to 6.7 metres.

So, the rule-of-thumb suggests that you don't want a "connection anomaly" to have an electrical length greater that 67 cm and, generally speaking, a cable interface connector will be a tiny fraction of this length.

But, the whole cable run might be many tens of metres hence, you do need terminating resistors at both ends on RS485 for example.

Up at ethernet speeds, it's a whole different ball-game and, a few millimetres of mismatched line or mismatched connection-interface is enough to throw a complete spanner in the works.

Associated posts: -

• Does the impedance of a PCB track matter if the length of the track is far smaller than the wavelength of the signal?
• Are termination resistors needed for UART, I2C and SPI?
• Why don't I need to consider trace width for a high speed oscillator crystals?
• Why do we not care about matching the input impedance of non-RF amplifiers?
• Why is power reflected when there is an impedance mismatch?

Answer 3

Eric Bogatin's book: "Signal and Power Integrity - Simplified" (published by Prentice Hall) spends a whole chapter on this and is absolutely worth buying if you want a clearly written, intuitive understanding of the physics, but here is a simple figure from that book demonstrating why you don't need to match the impedance at low bit rates:

Essentially, if the rise time of a signal is long relative to the time delay of the connector (length divided by signal velocity), then its impedance can be very different (in the plot a connector with half the impedance is shown) without introducing reflections. So when you look at a low frequency connector like RS485, the connector length/delay is very short relative to the rise time, so its impedance is irrelevant. However, if you look at a higher frequency connector like SMA, you can buy them in different impedances (e.g. 50 ohm or 75 ohm) because they may transmit signals with rise times comparable to their delay and so matching matters.

As for USB, it is very forgiving of poor signal and reflections, so you may get away with poorly made wires. However, I suspect if you try soldering a USB 10 or 20gbit cable together like that you'll find that it works poorly or not at all. Lower speeds will be more tolerant for the above reasons.

Answer 4

The frequency of interest has a lot to do with all the variations you've come across. RF connections are more sensitive to discontinuities than are kHz serial signals.
If you ever look inside an IT closet, the Ethernet cables often times are wired similar to telephone systems, using punch down panels. This is changing as we go above 1Gig bit.
I've also seen RS422 routed horribly across a PCB, with over 1 inch between the traces (with components and other traces in between), for a length of about 6" long. I had little hope of it working, but to my surprise it did.

Working in the world below 2.4Ghz, I've come to conclude that as long as the majority of the signal path is correct, a small discontinuity is just that, a small distortion of the signal that can usually be tolerated.

Answer 5

Allowable mismatch depends primarily on bandwidth and noise margin.

Noise margin is generally high, because binary signaling is used, at low to modest voltages (say 100s mV and up). Usually the source is extra strong compared to threshold (e.g. 3V vs ~100mV for RS-485), which also allows for long distances (attenuation due to cable losses).

A mismatch causes under/overshoot or ringing of the switching edge, and as long as the transient doesn't cross the receiver threshold within the receiver's response time, and the transient settles out within a reasonable margin before the next edge, then you're good to go.

CAN and RS-485 have extremely slow pulses (at most, 10s of Mbps) and modest edge rates (10s ns?), making them very tolerant. They are not generally considered "high speed" interfaces, and indeed many applications run extremely slowly (~kbps). The corresponding mismatch length is some meters, for say a 2x mismatch. Which for typical cable dimensions, connector geometry, and wire spacing/routing, would be a huge impedance ratio (and may well have more important consequences, like CM-DM mode conversion, or loops enclosing interference sources bringing DM interference directly).

Ethernet is generally "high speed", but with 100BASE-T and gigabit still the most common, it's only a modest step up from RS-485 rates -- 125MHz bandwidth plus low harmonics. Likewise the mismatch length (or stub length, say for routing between magnetics and PHY IC) is some 10s of cm.

USB is a mixed bag, because it ranges from RS-485-like rates (12Mbps or below), to low voltages at 480Mbps or more. It's common that a newbie/rookie is confused about which rates apply -- it depends on both devices on the link, what they are able to negotiate, even what cable is used (since USB-C brings more pairs to support the higher data rate, that USB-A simply doesn't have). It's a vast and diverse standard; easy to get overwhelmed with at first, and frankly a bit surprising it Just Works(TM) as often as it does.

Mind, actually implementing a device with such impairments, might well fail standard tests, or be much more marginal than is desirable -- but it's uncommon that equipment is tested at all to the standards, or at maximum length with worst-case cabling, or that the interface ICs don't handle the poor conditions despite it all (Ethernet in particular is quite robust to poor media conditions).

(Not that [not bothering to test] is any excuse, more to say it's something that happens all-too-often. Mediocrity is the rule, not the exception, more or less by definition. If you've ever had some device that seems mostly normal, but some things, some parts about it, it just... yeah. Well, this is how it happens.)

It's entirely possible your attempted splices failed for other reasons, like incorrectly matched wires, bad insulation, missing or improperly joined shield, etc. USB in particular requires shielding.