Using the IBM Cabling System for Fast Ethernet

Many large institutions installed the IBM Cabling System (ICS) years ago for Token Ring Networks. IBM dropped active support for Token Ring a couple of years ago in the face of their customers' overwhelming support for Ethernet. The low cost of Fast Ethernet (100 megabits per second) has made it the network of choice for general office connections. Companies, like Cisco, who make switches and routers are faced with either telling their customers to rewire their buildings with Cat5e or buy converters that will allow using the existing ICS.

Rewiring costs several hundred dollars, while converters only tens of dollars per user. Furthermore, Cat6 is still not ready for general installation, due to continuing problems of cross company compatibility. So, forward looking companies would be wise to delay rewiring for a couple of years. The converters are a cheap way to delay rewiring.

So the question is whether the ICS with converters is up to the job. As we will show later on, the ICS was well designed, its parameters generally exceeded the equivalent ones for Cat5e. It turns out, however, that not all the Type 1 cable installed is up to the requirements. The earlier Type 1 only had to meet 20 MHz capabilities. The later cabling, labeled Type 1A, was specified up to 300 MHz. Some Type 1, e.g., from Belden, was built such that when Type 1A was defined, they merely had to re-label their existing stock. Not all the wire companies were so far sighted.

The good news is that you don't have to test all the cable, just a small sample to assure that you don't have some of the Type1 that is bad. So, assuming your cable is not some of the bad stuff, the issue becomes how well the converters match up the Cat5 source and the Cat5 receivers to the ICS that runs between them. The equipment suppliers need to be able to tell their customers what works. Since there are several companies that supply converters with very different specifications, which converters are best and who supplies them is the subject of this paper.

There are a couple of issues that the converter must address. These are return loss (reflections due to an impedance mismatch) and crosstalk. Return loss is created when the wire and connections between the source and the receiver are of differing impedance. The ICS system is a 150-ohm system while all the categories of wiring (e.g. Cat5e) are 100 ohms. So there is a mismatch. How bad is it? It turns out not to be that bad.

The general belief of engineers is that impedance matching is important. And it is true that there will be less noise in the system if the connectors are impedance matched. That said, IBM ran some tests and found that for longer distances impedance matching, due to the added losses in the impedance converting transformers, somewhat reduced the maximum distance achievable for a given error rate. Unfortunately, IBM assumed, without testing that very short links were also OK (no increase in error rates) because of signal strength. A potential problem with short links is resonant built-up (reinforcing reflections). So we do not currently know if there are critical short lengths. The following analysis says there should not be a problem.

Between 150 and 100 ohms, the reflection is 20 % of the signal. For a reflection to do damage, it must interact with the desired signal. That only happens when there are two reflections. Two reflections (worst case) is the square of the single reflection or 4% of the original signal. And, with a long run, the squared reflection has to go three times the signal path, increasing noise attenuation. However, with a short link virtually all the doubly reflected signal gets to the receiver. Note, the doubly reflected noise will do another double reflection (and another, etc.) reinforcing the noise with a fraction of a percent (0.16% in this case). So with a good installation, reflections should not be an issue.

There is one qualification. The source impedance must absorb any reflections from the first mismatch. The first mismatch is where the converter transfers the signal from the 100 connector (RJ45) to the 150 ohm Universal Data Connector (UDC). Thus, the first reflection that counts is the reflection created when the signal exits the ICS. The second reflection occurs when that exit reflection propagates back to the entrance and is reflected to combine with the wanted signal as noise.

We have established that with a clean installation (two reflecting points, one going into the ICS, the other exiting), reflected signal noise should not be a problem. What about the other issue, crosstalk? Since Fast Ethernet was designed to run over Cat5 (this was before Cat5e) we need to determine how much crosstalk is being added by the converters.

Going to the latest ANSI/EIA/TIA 568B Standard we find at 100 MHz:

NEXT
(near end crosstalk) Cable Connectors
Cat5 -32.3 dB - 40 dB
Cat5e -35.3 dB - 43 dB
ICS -38.5 dB - 46.5 dB (Type 1A)

If we combine these numbers into a simple link (two connectors and the cable) we find the following:

NEXT Link
Cat5 -27.1 dB
Cat5e -30.4 dB
ICS -33.4 dB

We need some additional data. The UDC must attach to an RJ45 (an inherently poorer connector than the UDC). Using the AMP UDC (they have a patent on it and historically most people used it in their converters) and an ordinary RJ45, the crosstalk varies from -32 dB to -38 dB. Calculating the link crosstalk we find a range of values from -24.2 to -28.3 dB. Unfortunately, the average value is not as good as a cat5 link even though the ICS components are superior. The uncompensated 45 is the problem. I had one vendor tell me that crosstalk cancellation was unnecessary because the cable was so good. A particular physical cable may be very good, but the Standard must be used in qualification and certification

My company, ETS, has developed a converter with crosstalk cancellation and guarantees at least -44 dB crosstalk. When we calculate link NEXT with two of these converters and the cable, the crosstalk drops to 32.2 dB. It's not only better than Cat5, but also better than Cat5e.

Conclusion:

Anyone who wishes to use converters, and wants assurance of proper operation the first time, would do well to consider the low NEXT converters without impedance matching. However, for the lowest system noise, low NEXT with impedance matching should be considered. For lowest cost, a simple converter consisting on an UDC attached to an RJ45 can be considered. The cost of using only impedance matching, without low NEXT, makes it an unwise decision.