Real-time noise identification in DSL systems using computational intelligence algorithms

Despite the advances and improvements in the Digital Subscriber Line (DSL) technology, noise is still the main impairment. In special, far-end crosstalk, Radio Frequency Interference (RFI) and Impulsive Noise (IN) are of greatest concern and study. In DSL world, there are many noise mitigation techniques, but to know the impairment as a priori knowledge is a step necessary to apply the appropriate technique. In this paper we propose a new methodology for noise identification on real-time. Computational Intelligence (CI) algorithms are used in order to classify in real time the absence of noise or the predominance of IN, crosstalk or RFI. The algorithms are applied to a database composed by management information base (MIB) metrics. In order to ensure the database diversity, several DSL topologies using real cables were created and evaluated. In order to choose the best CI algorithm, a benchmarking was performed comparing the results achieved by naive Bayes, Bayesian belief networks and artificial neural networks based on backpropagation and on Radial Basis Function (RBF). The results demonstrate the potential use of CI for noise identification in DSL networks through MIB metrics and the most difficult noise to be identified is pointed. Tests indicate the RBF algorithm achieving the best result with 99.6% of accuracy.