< Back to previous page
Project
Resource Allocation in Modulation and Equalization Procedures in DSL Modems
Digital subscriber line (DSL) technology is a very popular broadband access technology. It uses the existing telephone infrastructure to provide broadband access. In order to cope with the increased bandwidth demandto support broadband services, such as, Video on Demand (VoD), real time multimedia streaming, it is important to further improve the DSL.
The main performance degradation of the DSL system is caused by channel impairments, such as, crosstalk and inter-symbol interference (ISI). Furthermore, the discrete Fourier transform (DFT) based discrete multitone (DMT) system has very poor spectral properties, which prohibit the use of tones at the band edges in order to meet the power spectral density (PSD) constraints of the system, thus reducing the achievable bit rate.
In order to mitigate the channel impairments as well as to combat the poor spectral properties of the DFT based DMT, sophisticated signal processing techniques are employedboth in single-user DSL and multi-user DSL scenarios. These signal processing techniques increase the overall run-time complexity of the DSL system, which makes them hard to realize in practice. However, the DSL channel is a very slowly time-varying channel, which allows for offline optimization of the signal processing techniques such that their run-time complexity is reduced. This necessitates efficient algorithms to reduce the overall run-time complexity without affecting the system performance significantly.
In this dissertation, various resource allocation algorithms are investigated in order to reduce the run-time complexity of the signal processing techniques without sacrificingthe performance of the system.
In the first part, resource allocation algorithms for the single-user DSL scenario are investigated. In particular, the per-tone pulse shaping filter based DMTtransmitter and per-tone equalization (PTEQ) based DMT receiver will beconsidered. Owing to various properties of the DSL channel, the use of a constant number of non-zero filter taps in both the pulse shaping filter and the PTEQ is seen to unnecessarily increase the run-time complexity and hence an allocation procedure for the non-zero filter taps is called for. It can be seen that the filter tap allocation problem is a combinatorial problem. In this dissertation, we propose two approaches to reduce this combinatorial complexity: The first approach is based on a contiguous filter tap selection and the second approach is based on sparse approximation based filter design. It is shown that with both of these methods the number of required non-zero filter taps and hence the run-time complexity is reduced significantly. Furthermore, it is shown that the filter tap allocation performs even better when employed in combination with transmit power allocation.
In the second part, resource allocation algorithms for the multi-user DSL scenario are investigated, where crosstalk cancellation is a crucial component. In the presence of only additive white Gaussian noise (AWGN), it has been shown that instead of performing a full so called linear zero-forcing (ZF) crosstalk cancellation only few major crosstalkers may be canceled without significant performance loss. Therefore, an optimum canceler allocation algorithm, known as partial crosstalk cancellation, only cancels the major crosstalkers per line and per tone, resulting in a significant reduction in run-time complexity. In this dissertation, we extend this partial crosstalk cancellation concept to the scenario where spatially correlated background noise is present. It is shown that in this scenario, minimum mean square error (MMSE) based cancelers outperform linear ZF cancelers. However, the optimal canceler allocation problem then has a prohibitively high complexity. In this dissertation, we propose two approaches to reduce the complexity: the first approach is based on a selection metric, which combines the information of in-domain crosstalk with information of spatially correlated background noise. Based on this selection metric the crosstalkers are ordered and selected for inclusion in the linear MMSE cancellation. This approach is then extended to include non-linear MMSE cancelers. The second approach, based on sparse approximation, directly chooses the cancelers to be taken into account by first designing a sparse linear MMSE canceler. Both of these approaches reducethe run-time complexity while performing near optimally under the resource constraint. Furthermore, it is shown that the spectrum balancing, i.e., transmit power optimization, combined with partial crosstalk cancellation further reduces the number of canceler taps required without sacrificing performance.
The main performance degradation of the DSL system is caused by channel impairments, such as, crosstalk and inter-symbol interference (ISI). Furthermore, the discrete Fourier transform (DFT) based discrete multitone (DMT) system has very poor spectral properties, which prohibit the use of tones at the band edges in order to meet the power spectral density (PSD) constraints of the system, thus reducing the achievable bit rate.
In order to mitigate the channel impairments as well as to combat the poor spectral properties of the DFT based DMT, sophisticated signal processing techniques are employedboth in single-user DSL and multi-user DSL scenarios. These signal processing techniques increase the overall run-time complexity of the DSL system, which makes them hard to realize in practice. However, the DSL channel is a very slowly time-varying channel, which allows for offline optimization of the signal processing techniques such that their run-time complexity is reduced. This necessitates efficient algorithms to reduce the overall run-time complexity without affecting the system performance significantly.
In this dissertation, various resource allocation algorithms are investigated in order to reduce the run-time complexity of the signal processing techniques without sacrificingthe performance of the system.
In the first part, resource allocation algorithms for the single-user DSL scenario are investigated. In particular, the per-tone pulse shaping filter based DMTtransmitter and per-tone equalization (PTEQ) based DMT receiver will beconsidered. Owing to various properties of the DSL channel, the use of a constant number of non-zero filter taps in both the pulse shaping filter and the PTEQ is seen to unnecessarily increase the run-time complexity and hence an allocation procedure for the non-zero filter taps is called for. It can be seen that the filter tap allocation problem is a combinatorial problem. In this dissertation, we propose two approaches to reduce this combinatorial complexity: The first approach is based on a contiguous filter tap selection and the second approach is based on sparse approximation based filter design. It is shown that with both of these methods the number of required non-zero filter taps and hence the run-time complexity is reduced significantly. Furthermore, it is shown that the filter tap allocation performs even better when employed in combination with transmit power allocation.
In the second part, resource allocation algorithms for the multi-user DSL scenario are investigated, where crosstalk cancellation is a crucial component. In the presence of only additive white Gaussian noise (AWGN), it has been shown that instead of performing a full so called linear zero-forcing (ZF) crosstalk cancellation only few major crosstalkers may be canceled without significant performance loss. Therefore, an optimum canceler allocation algorithm, known as partial crosstalk cancellation, only cancels the major crosstalkers per line and per tone, resulting in a significant reduction in run-time complexity. In this dissertation, we extend this partial crosstalk cancellation concept to the scenario where spatially correlated background noise is present. It is shown that in this scenario, minimum mean square error (MMSE) based cancelers outperform linear ZF cancelers. However, the optimal canceler allocation problem then has a prohibitively high complexity. In this dissertation, we propose two approaches to reduce the complexity: the first approach is based on a selection metric, which combines the information of in-domain crosstalk with information of spatially correlated background noise. Based on this selection metric the crosstalkers are ordered and selected for inclusion in the linear MMSE cancellation. This approach is then extended to include non-linear MMSE cancelers. The second approach, based on sparse approximation, directly chooses the cancelers to be taken into account by first designing a sparse linear MMSE canceler. Both of these approaches reducethe run-time complexity while performing near optimally under the resource constraint. Furthermore, it is shown that the spectrum balancing, i.e., transmit power optimization, combined with partial crosstalk cancellation further reduces the number of canceler taps required without sacrificing performance.
Date:2 Jun 2008 → 26 Oct 2011
Keywords:Telephony network, ADSL
Disciplines:Other engineering and technology
Project type:PhD project