# Effect of splices on 120 Ω twisted pair cable impedance a.k.a correct way to attach stubs to the bus?

I am currently in the process of designing a wiring loom for a race car control system which must also connect a number of devices via a CAN bus.

This is my first encounter with CAN bus (J1939-15) systems which call for a 120 Ω UTP bus cable terminated by 120 Ω resistors on each end, and my question relates to the correct method for connecting stubs (device cables) to the bus.

Asking around within my industry, the solutions proposed by colleagues are the following:

1. Strip insulation only and solder stub connections onto the bus without cutting the cable.
2. Cutting the cable and using Y or T splitter connectors. (Amphenol and Deutsch seem to make J1939 specific Y-splitter connectors. No mention on either product datasheets about impedance matching however.)
3. Cutting and using 2-1 environmental splices instead of Y-splitter connectors (e.g MIL-S-81824/1 or Raychem MiniSeal splices).

Is there a better or correct way to attach stubs to the main bus?

If not, of the three above options, would the use of the splices linked in the third option adversely affect the network?

Note: In this instance the job specification does not allow me to use solder unless the joint is also encapsulated in a specified potting compound, and I would also prefer avoid the Deutsch Y-splitter connectors due to bulk.

The baud rate is 1 Mbit/s, and bus length is 20 ft (6 m). Stubs in most cases will be less than 1 ft (0.3 m) length.**

• If you are designing the loom, why do you need stubs? To keep mass down? – Martin James Mar 16 '17 at 1:53
• What's the bit rate and edge slew rate? – alex.forencich Mar 16 '17 at 1:58
• Don't use stubs just loop in and loop out of each module. – Andy aka Mar 16 '17 at 8:20
• In general, the correct way to add stubs to a controlled impedance bus is, don't. I don't know CANBUS but I'm not surprised to see the same recommendation here. – Brian Drummond Mar 16 '17 at 9:23
• There is no 'correct' way to add stubs to a bus, it's always better to loop-in, loop-out. However, for a given edge and data rates, stubs not exceeding a certain length can be tolerated. At CAN data rates, the length of the stub is important, the type of splice made to the main bus is almost totally irrelevant, unless it physically gets to a size where it is vying with the length of the stub. – Neil_UK Mar 16 '17 at 9:37

"Don't use stubs just loop in and loop out of each module" as Andy Aka said. UTP rigid cable may not be the best solution. Use a shielded twisted pair cable with characteristics impedance 120 ohm. At each device/stub wrap the wires or crimp them, don't solder them. The type of splitters you have mentioned are in my opinion too expensive and they just don't add any benefit to your system, except if you are doing hi-tech military stuff. I have installed many CAN devices in industrial environment, kind of monitoring system. As a base rule, I used a good shielded cable, correct impedance (UNITRONICS BUS CAN) and no stubs. I'd suggest you not to use UTP, as the rigid wire can break when twisting it, also UTP has no shield.

EDIT:

@Conway @Lundin:I agree. I have misunderstood the question about vehicle CAN aka J1939. However, I would like to underline that it would be good if you crimp wires and use some standard vehicle connector where you will join also power wires. As for characteristics impedance it matters the conductor diameter and overall diameter (as well the $\varepsilon_r$ of insulating material). I have found some online calculator: https://www.allaboutcircuits.com/tools/twisted-pair-impedance-calculator/

The $\varepsilon_r$ of the insulator has a standard value regarding the material. If you need perfect impedance match, you have just to choose a suitable wire and wrap them to make a twisted pair. Alternatively you can also inspect CAN wires from a car at junkyard.

AWG20 = 0.032 in

distance between conductors == overall diameter = 0.085 in

dielectric constant PVC = 3

• I don't think there exists a general case. Looping in and out of each module might not be an option: suppose the CAN nodes are for example scattered all over a vehicle? You'd have to use double amounts of wires and possibly drill twice as many holes. I think stubs are perfectly fine in such a case. In my experience, CAN will work just fine with stubs as long as there's no great distances and you terminate the bus where it makes the most sense. Something like this could be an option. – Lundin Mar 16 '17 at 10:43
• @Lundin The OP mentioned the industrial ue, not vehicle. I don't know the speed of vehicle CAN, but I don't think it is 1M. My car has a CAN gateway for splitting a network in more subnetworks, not all devices are connected in a single CAN network. – Marko Buršič Mar 16 '17 at 11:07
• @Lundin is correct in that looping in and out of each module is not an option in this case. Basically I'm trying to find out if my only option is using T connectors like the one suggested (and the ones in the OP).None of them mention anything about impedance in their datasheets so if I can avoid their use, and considerable added bulk to the harness that would be great. – Conway Mar 16 '17 at 14:32
• @MarkoBuršič Most often cars use different buses (with different baud rates) depending on the nature of the data - keeping safety-critical data separate from non-critical data. But also because they want to decentralize functionality, so that there's no "tight coupling" between every non-related part of the car. Modern, commercial cars is kind of a special case because of the sheer complexity of their electronics. On less complex vehicles, where J1939 is more likely to appear, there is usually just one or two buses, but probably no great bus lengths and baudrates typically around 250k. – Lundin Mar 16 '17 at 14:45