Receiver-based Protection against Time-varying Noise in xDSL Systems
Digital Subscriber Lines (DSLs) have gained importance over the last years as services such as high-definition television (HDTV), three-dimensional television (3DTV), Voice over Internet Protocol (VoIP), video streaming and interactive video applications, which pose high requirements on data rates, latency and line stability, have become part of our every day life. The constantly increasing number of DSL users and trying to match the growing bandwidth demand by using high frequencies have made far-end crosstalk (FEXT) between copper wires the dominant impairment in current DSL systems. The noise perceived on a DSL line is comprised of noises from several sources, but the largest and most dominant noise on DSL lines is the summation of FEXT from all other active lines in the bundle. The number of active lines is strongly daytime dependent and consequently, the distribution of noise on DSL lines is not stationary. Furthermore, FEXT fluctuates when DSL users change their operational state. This work investigates low-complex and practical approaches for the protection against time-varying noise in DSL systems. In order to be in compliance with current DSL systems, in this work, the DSL modems operate independently without any coordination with other lines. Furthermore, the same transmit spectra are assumed for all the modems, and only receiver-based approaches are considered.
One approach to protect against time-varying non-stationary noise is to set the target bit rate such that it can still be achieved at times with peak noise levels. Traditionally, a fixed signal-to-noise-ratio (SNR) margin is subtracted from the measured SNR during initialization when calculating the target data rate. In practical DSL systems, the value of the SNR margin is set ad-hoc to 6 dB and is independent of the SNR at initialization and of the noise distribution, and therefore, the state of the art SNR margin is not optimal in terms of data rate and outage probability. To improve the protection against time-varying noise and increase link stability, the concept of Virtual Noise (VN) has been introduced. In this work, we jointly optimize the VN mask and the SNR margin in order to maximize the target data rate while satisfying a target outage probability defined over 24 hours. Another problem with the state of the art SNR margin is that it is tone-independent. In practical DSL systems, the received noise might increase only at some tones such that the per-tone SNR margins of the affected tones become negative. To prevent re-initializations in such cases, modems have the ability to equalize the SNR margin across tones by adapting their bit-loading to slowly changing SNR by using bit-swapping. However, bit-swapping might be too slow to move bits from the affected tones such that non-negative per tone margins are restored and the connection is interrupted. In this work we solve this problem by adjusting the VN mask and SNR margin such that in addition to achieving the maximum target data rate while satisfying a target outage probability, the probability of the line going to a state with negative per-tone SNR margins is reduced. The VN mask and the SNR margin are expressed as functions of the target outage probability and the distributions of the maximum noise power measured over 24 hours at each tone. Comparison with state of the art approaches reveals that the proposed approach improves the data rate and outage probability performance.
Another approach to protect against time-varying non-stationary noise is to adapt the target data rate to the varying noise. The concept of Seamless Rate Adaptation (SRA) was introduced to seamlessly adapt the data rate of a connection to the changing SNR while in operation without any service interruption. Interleavers used in transmission for the protection against burst errors impose a constraint on how much the data rate that can be changed by an SRA procedure. Moreover, to keep an equal error probability over all tones, it is very likely that the bit-loadings of all tones have to be modified in each SRA procedure. Hence, the state of the art SRA is too slow and provides means of adaptation only to slowly changing SNRs and will not prevent reinitializations under certain conditions such as large rapid changes in FEXT. In this work, two adaptation approaches are proposed. In the case where multiple adaptation procedures are required to modify the bit-loadings of the tones to the target bit-loadings that correspond to the new SNR, both approaches satisfy the data rate constraint imposed by interleavers. Furthermore, both approaches equalize the error probability over all tones only after the last procedure when the target bit-loadings is achieved at all tones. The first approach modifies the tones to their target bit-loadings in the order that leads to the fastest decrease in the average bit error rate (BER). The second one modifies the bit-loadings of the tones in groups and finds the per group constant bit reductions that decrease the average BER as rapidly as possible. Moreover, low complex versions of the proposed approaches are presented such that they can be used in practice by DSL modems with limited computational power. Performance results reveal that all the proposed approaches shorten the adaptation time tremendously and decrease the number of errors that occur during the adaptation when compared to the state of the art SRA.
DSL users often use services with different requirements on quality of service in terms of outage probability. In such cases, we propose in this work a hybrid VN-SRA approach that iteratively allocates tones to be used by the proposed approach where the VN mask and the SNR margin are jointly optimized for services that require a BER that is smaller than or equal to the target BER with an outage probability that is smaller than or equal to the target outage probability. Moreover, the hybrid VN-SRA approach allocates tones to be used by the proposed adaptive data rate approaches for services that can tolerate temporary increases in the BER above the the target BER during the adaptation to the varying noise. Performance results reveal data rate gains when compared to the case where the data rate achieved over all tones is guaranteed at a BER that is smaller than or equal to the target BER with an outage probability that is smaller than or equal to the target outage probability.
The approaches presented in this work can improve the performance of today’s DSL systems. Furthermore, the implementation of those approaches requires only minimal changes of the current DSL standard.