A recent article in EETimes about using Transmitting Video over Cat5 Cable reminds an engineer of a problem he encountered when his employer was upgrading the data network.
We had a 4-wire leased line from our main site about 4 Km outside town, to our Customer Service Centre (CSC) in the town center. It had been running at 384 Kbps which had, until then, been adequate. New software which was, as always, more resource-hungry, necessitated an increase in the bandwidth to the CSC office.
These were all-copper leased circuits, and we used baseband modems made by Pair Gain. They produced a later model which could provide up to 2 Mbps over the same lines. The technology was much the same as that used for the ADSL broadband services that so many of us use today.
The previous modems at 384 K had been super-reliable. They had been in service since before I joined the company and I had never had much to do with them. But they seemed simple enough to use and so one evening after the CSC had closed for the day I replaced the modem at the head office end and went into town to change the one there.
The modems had a front-panel display that provided various parameters, notably the signal-to-noise margin at each end. With the 384 K units we had always had a comfortable 15 - 20 dB or better margin. The new units, however, had a good margin at the head office end but a terrible margin at the CSC end Ė around 2 - 4 dB Ė and we were only running them at 1 Mbps.
Our Data Network Administrator who was monitoring the status of the circuit confirmed that we were getting a fairly high error rate into the CSC. Because of the length of the circuit, we had not expected to get the full 2 Mbps capability of the new units, but the manuals indicated that we could expect at least 1.5 Mbps with the length of circuit that we had. And in any case, why the bad results at one end only? I checked all my connections, but since the new units were just a plug-in replacement, I did not find anything wrong there.
We reduced the speed a bit and at around 750 Kbps we were down to a usable, if not good, error rate. Since this was still better than we had previously, we left it at that and retired to consider our next line of attack.
Over the next few days I read the manuals and tried a few things after hours when I could take the units offline. I changed the units at both ends and got the same result. I checked the mains voltage in case that was low. It was fine. I even took the units out of circuit and with the aid of one of my co-workers, checked the resistance of the lines. They were well within spec and according to the manual, we should probably have been able to get the full 2 Mbps bandwidth out of them. While I had the units off the lines I put on a level meter and checked for noise. The lines were quiet.
In desperation I emailed the agents who sold us the equipment, describing the symptoms and my findings and asking if they had any suggestions. They didnít, but they helpfully sent me an article on the theory of the HDSL transmission used by these units. In it was a mention of the twists in the pairs of the cables used for telephone transmission.
I was familiar with network cable theory, and how the twists in the pairs are important to prevent crosstalk between pairs at high data rates. And some time previously, an inexperienced building cabler had connected some 10 Mbps network cables using only the Blue, Orange, Green and Brown wires of the Cat 5 Cable to pins 1,2,3 and 6 of the connectors, ignoring the white wires of each pair, and of course they had not worked at all and Iíd had to reterminate them correctly myself.
The penny dropped, and a little light went on above my head. The tail cable between the modems and the line patch panel at the CSC end was about 2 metres of flat telephone cable, a relic of the previous installation. But could just 2 metres of bad cable affect the signal so much at just 1 Mbps? Telephone cable does not have nearly as much twist in it as Cat5 network cable, and we were working at much lower frequencies than even 10 Mbps networking. I was dubious, but made up a cat-5 cable to suit and that evening I went to try it.
The results were immediate and dramatic. Our signal-to-noise margins were back up to around 20 dB and the error rate dropped to almost zero. We upped the rate to 1.5 and then to 2 Mbps and the line behaved impeccably.
All the pieces of the puzzle fell into place. Because of the flat cable and its poor crosstalk performance, the strong transmit signal at the CSC end was able to couple across the pairs and degrade the relatively weak received signal. And the crosstalk from the received signal was too weak to interfere with the transmit signal, which got to the head office in one piece. The units came with a suitable twisted pair connection cable, but it turned out our data network guys had unpacked everything and tested the units on the bench and not given me the supplied cable, knowing that cables were already in place at both sites.
It was a salutary lesson in cable theory. When I first worked with ADSL I wondered briefly why it worked OK with flat cables, but it only uses one pair, which canít interfere much with itself. Nonetheless I still always use twisted pair patch cables for ADSL circuits. Our CSC was this year connected via fibre-optic cable at 1 Gbps. And thatís a cable you DONíT want to twist!
About Author David Ashton: "Iím not sure what I amÖ.. I was born in London, UK, raised and trained and worked in Rhodesia, then Zimbabwe, and I now live in Australia. (So Iím a Pom-Rhodie-Zimbo-Aussie??) Work-wise it's much the same. I have run electronics labs and managed telecoms centres, run my own comms business, and I am now working as a telecoms specialist keeping a large comms network going. Iím a jack of all trades, and yes, admit Iím master of none, but I kinda like it that way. It makes it difficult to get bored."