Measuring FEXT

Written on 6:47 AM by ooe

FEXT is a well-understood problem for telephony cables at low frequencies. It is, however, not well characterised for higher frequencies in a real network. In a real network, FEXT is not just a function of the crosstalk in the cables, but also a function of joints, gauge changes, and other physical properties of the installed transmission plant. In order to understand the implications for VDSL, equipment has been developed to measure FEXT in a live network (one carrying existing narrowband telephony or ISDN).

Measuring FEXT is both time consuming even in a controlled laboratory environment; in the access network it is even more problematic. In a live network not only is it more difficult to make any measurement, but also the disruption to the network must be kept to an absolute minimum.

In the laboratory, though cumbersome, it is straight forward to measure the FEXT in a cable. Typically the cable is either new or well-maintained with few faults; also the relative arrangement of the pairs in the cable is known. FEXT between two pairs drops off markedly with their separation in a cable, and the couplings between two adjacent pairs will be similar from pair to pair. Therefore the FEXT signal-to-noise ratios (SNRs) do not have to be measured for all pair combinations to have sufficient confidence in the characterisation of the cable. This, though, is not the case in the access network, and to adequately characterise a cable here, it is necessary to measure as large a set of the pair-to-pair combinations as possible.

A test signal generator is constructed that generates ten separate, distinct and broadband output signals. These signals are connected to up to ten of the VDSL connections at the DP. A scanning measurement then samples the signals on each of the up to 100 VDSL connections at the cabinet sequentially. As it measures each pair it is able to separate contributions to that pair from each of the up to ten connected distinct transmitters, so measuring up to ten FEXT couplings at up to 100 frequency points.

This is repeated at all the DPs present, thereby measuring all the required FEXT couplings to adequately characterise that cable.

The data is then analysed to extract the complete coupling matrix for the cable being measured of size n by m by l where n is the number of pairs scanned, m is the number of those pairs that had one of the transmitters connected at the DP at some stage during the testing, and l is the number of different frequency points measured. The cable attenuations are accounted for, and the matrix can then be used to make predictions for FEXT.

The outcome of the analysis is that the Werner model is correct except for one point for the BT access network. The value of the constant is not −55.8 (at a frequency of 100 kHz, a length of 1 km and 49 like disturbers) but better characterised by a constant of −50.0. The difference is accounted for by noting that the previously quoted constant was taken from measurements upon cables in a laboratory with a shorter lay length of the pairs than is common in the BT access network. References [12,13] for example state that FEXT is reduced by having a shorter lay on the pairs in a cable and that the variance of the lay is important in the worst case. Intuitively this is correct since a long lay means that the wires lay nearly flat as opposed to keeping their shape in a tight spiral. This increases the likelihood of deviation from an ideal helix and so increases FEXT.