An important case of multiuser channel estimation is considered here, the problem of identifying the crosstalk that disturbs a DSL signal. Crosstalk originates from signals transmitted on nearby pairs in a telephone cable, and couples over unknown pair-to-pair crosstalk coupling channels into the pair carrying the signal. While crosstalk is generally the dominant impairment for current DSL systems, only recently have papers appeared addressing the problem of multiuser crosstalk channel estimation. In Galli et al. (2001), it was proposed to identify crosstalk sources by finding the maximum correlation with a "basis set" (dictionary) of representative measured coupling functions. It is shown here that this can be considered equivalent to finding an optimal sparse representation of a vector from an overcomplete set of vectors. A well-known algorithm that solves this problem is the matching pursuit (MP) algorithm (Mallat et al. (1993)), a greedy algorithm for choosing a subset of vectors from an overcomplete dictionary and finding a linear combination of that subset which approximates a given signal vector. A method based on singular value decomposition (SVD) for reducing the size of the dictionary is also discussed.

Measurements of near-end crosstalk (NEXT) and far-end crosstalk (FEXT) in 26-AWG and 24-AWG polyethylene insulated cable (PIC) twisted pair cables are presented. This data is used to develop crosstalk models that indicate existing T1.417 Standard NEXT and FEXT models are probably conservative by about 5 dB for 25-pair binder groups. However, many cables start from wire centers as pulp cables of 100-pair binder groups. Even if most of these older pulp cables are 26-AWG, prudence dictates that a more conservative model be used for these cases. On a going forward basis for new and rehab construction, it should be recognized that PIC (air core and filled) cables with not larger than 25-pair binder groups are the predominant cable media supplied by manufacturers for feeder and distribution loops. Results for NEXT in 12-pair binders are also analyzed and show that 26-AWG cable has about 2 dB more NEXT loss than 24-AWG cable and 12-pair binders have about 2 dB more NEXT loss than 25-pair binders. In addition, the NEXT loss separation between adjacent binders of 12 and 13 pairs is found to be not more than 10 dB as compared with the more optimistic 15.5 dB assumed within the industry.

Crosstalk between multiple services transmitting through the same telephone cable is the primary limitation to digital subscriber line services. From a spectrum management point of view, it is important to have an accurate map of all the services that generate crosstalk into a given pair. If crosstalk is measured via modem-based methods, i.e., while a digital subscriber line (DSL) system is running, what is measured is the crosstalk in the bandwidth of the considered DSL system. For this reason, DSL services running on adjacent pairs may not be detected if their bandwidth is not significantly overlapping with the bandwidth of the disturbed system. This is a major drawback of modem-based system identification techniques since, from a spectrum management point of view, it is important to be able to identify all crosstalkers. We address the important problem of crosstalk identification when the pair under test does not bear DSL services, i.e., via a non-modem-based approach. Crosstalk sources are identified in the frequency domain by finding the maximum correlation with a "basis set" of representative measured crosstalk couplings. The effectiveness of the proposed technique is also verified on the basis of real crosstalk measurements performed on actual cables. Finally, new techniques based on multiple regression and best basis selection are also discussed.