Analysis of the impact of impulse noise in digital subscriber line systems
Nedev, Nedko H
In recent years, Digital subscriber line (DSL) technology has been gaining popularity as a high speed network access technology, capable of the delivery of multimedia services. A major impairment for DSL is impulse noise in the telephone line. However, evaluating the data errors caused by this noise is not trivial due to its complex statistical nature, which until recently had not been well understood, and the complicated error mitigation and framing techniques used in DSL systems. This thesis presents a novel analysis of the impact of impulse noise and the DSL framing parameters on transmission errors, building on a recently proposed impulse noise model. It focuses on errors at higher protocol layers, such as asynchronous transfer mode (ATM), in the most widely used DSL version, namely Asymmetric DSL (ADSL). The impulse noise is characterised statistically through its amplitudes, duration, inter-arrival times, and frequency spectrum, using the British Telecom / University of Edinburgh / Deutsche Telekom (BT/UE/DT) model. This model is broadband, considers both the time and the frequency domains, and accounts for the impulse clustering. It is based on recent measurements in two different telephone networks (the UK and Germany) and therefore is the most complete model available to date and suited for DSL analysis. A new statistical analysis of impulse noise spectra from DT measurements shows that impulse spectra can be modelled with three spectral components with similar bandwidth statistical distributions. Also, a novel distribution of the impulse powers is derived from the impulse amplitude statistics. The performance of a generic ADSL modem is investigated in an impulse noise and crosstalk environment for different bit rates and framing parameters. ATM cell and ADSL frame error rates, and subjective MPEG2 video quality are used as performance metrics. A new modification of a bit loading algorithm is developed to enable stable convergence of the algorithm with trellis coding and restricted subtone constellation size. It is shown that while interleaving brings improvement if set at its maximum depth, at intermediate depths it actually worsens the performance of all considered metrics in comparison with no interleaving. No such performance degradation is caused by combining several symbols in a forward error correction (FEC) codeword, but this burst error mitigation technique is only viable at low bit rates. Performance improvement can also be achieved by increasing the strength of FEC, especially if combined with interleaving. In contrast, trellis coding is ineffective against the long impulse noise error bursts. Alien as opposed to kindred crosstalk degrades the error rates and this is an important issue in an unbundled network environment. It is also argued that error free data units is a better performance measure from a user perspective than the commonly used error free seconds. The impact of impulse noise on the errors in DSL systems has also been considered analytically. A new Bernoulli-Weibull impulse noise model at symbol level is proposed and it is shown that other models which assume Gaussian distributed impulse amplitudes or Rayleigh distributed impulse powers give overly optimistic error estimates in DSL systems. A novel bivariate extension of the Weibull impulse amplitudes is introduced to enable the analysis of orthogonal signals. Since an exact closed-form expression for the symbol error probability of multi-carrierQAM assuming Bernoulli-Weibull noise model does not exist, this problem has been solved numerically. Multi-carrier QAM is shown to perform better at high signal-to-noise ratio (SNR), but worse at low SNR than single carrier QAM, in both cases because of the spreading of noise power between subcarriers. Analytical expressions for errors up to frame level in the specific case of ADSL are then derived from the impulse noise model, with good agreement with simulation results. The Bernoulli-Weibull model is applied to study the errors in single-pair highspeed DSL (SHDSL). The performance of ADSL is found to be better when the burst error mitigation techniques are used, but SHDSL has advantages if low bit error rate and low latency are required.