DISTURBER IDENTIFICATION FOR SINGLE-ENDED LINE TESTING (SELT) Tom Bostoen Alcatel Francis Wellesplein 1, 2018 Antwerpen Belgium Margherita La Fauci and Marco Luise Universit di Pisa Via G. Caruso, 56122 Pisa Italy Patrick Boets Vrije Universiteit Brussel Pleinlaan 2, 1050 Brussel Belgium Abstract One of the most challenging problems related to single- ended line testing (SELT) is the estimation of the noise power spectral density (PSD) at the customer-premises (CP) side of the line under test, because only measurements at the central-office side (CO) are possible. A solution for the noise caused by crosstalk could consist of first identifying the disturbers present in the cable binder of the line under test based on the measured noise PSD at the CO side and secondly estimating the noise PSD at the CP side by means of crosstalk models and the transmit PSD’s of the identified disturbers. In this paper we focus on the first part of the solution, the disturber identification. For the identification non-parametric and parametric modeling approaches are explored. The estimator is based on a maximum-likelihood (ML) cost function. The proposed algorithms have been experimentally verified by means of measurements on a France Telecom cable. Key Words Digital subscriber line (DSL), near-end crosstalk (NEXT), far-end crosstalk (FEXT), single-ended line testing (SELT), disturber. 1. Introduction In order to sustain the mass deployment of a myriad of different DSL (Digital Subscriber Line) services, telecom operators need more intelligent broadband subscriber line testing tools than are available on the market today. Recently testhead manufacturers have upgraded their POTS (Plain Old Telephony Service) testing systems towards DSL. These broadband testheads may be external to the DSLAM (DSL Access Multiplexer) connecting to a particular line via the metallic test bus of the DSLAM. However in order to reduce cost there is a trend towards test boards integrated into the DSLAM and in the longer run even integrated test boards could become obsolete, when DSL chipsets will include line-testing capabilities. It is expected that the DSL modem will only perform the measurements. The interpretation of the measurement results is done by an expert system that would become part of the OSS (Operation Support System). The task of this interpretation unit is to deduce as much useful information about the line under test as possible starting from the measurement data acquired by the DSL modem as well as a priori knowledge stored in the OSS database. For single-ended line testing (SELT) there are still challenging problems with respect to the interpretation of the limited measurement data. Indeed one mainly has to rely on reflectometry and noise power spectral density (PSD) measurements. However a complete loop characterization includes the loop topology, the cable types and lengths of all line sections, the end-to-end loop transfer function, the noise PSD at the central-office (CO) side and customer-premises (CP) side of the loop, the bit rates achievable on the loop for different DSL technologies, and the disturbing DSL services present in the same cable binder. During the last years a lot of progress has been made with respect to loop identification based on reflectometric measurements. [1] [2] However one major challenge remains, namely the estimation of the noise PSD at the CP side given the SELT is performed at the CO side. The noise PSD at the CP side of the loop is required for downstream capacity estimation. In this paper a partial solution is presented with respect to the problem of the estimation of the noise PSD at the CP side based on a noise PSD measurement at the CO side. The main assumption is that the noise on the subscriber lines is primarily caused by crosstalk, hence we are not taking into account RFI (radio frequency ingress). Hence the noise measured on a particular line at the CO side is dominated by near-end crosstalk (NEXT) from the DSL services on the other lines in the same cable binder. This paper presents a solution for identifying the dominant disturbers in the cable binder of the line under test improving and extending the algorithms developed in [3] and [4]. In a second step the knowledge of the dominant disturbers could be used in a crosstalk modeling approach to estimate the noise PSD at the CP side, but this is out of the scope of this paper. In general identification consists of matching a model to a measurement minimizing a cost function, which quantifies the difference between model and measurement. In section II we will discuss the models, in section III the estimator (cost function and minimizer), 408-109