A centralized multi-level water-filling algorithm for dynamic spectrum management

  • Authors:

  • Hao Zou;Aakanksha Chowdhery;John M. Cioffi

  • Affiliations:

  • Stanford University, Department of Electrical Engineering, Stanford, CA;Stanford University, Department of Electrical Engineering, Stanford, CA;Stanford University, Department of Electrical Engineering, Stanford, CA

  • Venue:
  • Asilomar'09 Proceedings of the 43rd Asilomar conference on Signals, systems and computers
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

In digital-subscriber-line (DSL) networks, the interference, or crosstalk, between multiple DSL lines can severely limit the data-rate of the lines. Level-2 Dynamic Spectrum Management (DSM) optimizes the transmit spectra of multiple DSL lines to mitigate such mutual interference. This paper proposes a centralized multi-level water-filling (MLWF) Level-2 DSM algorithm for optimizing the transmit spectra for multiple DSL lines at a Spectrum Management Center (SMC). The algorithm is a practical approximation to the more complex Optimal Spectrum Balancing (OSB), which is known to present computational issues in practice. The intuition behind the algorithm is that a DSL user should first try to use frequency bands where less interference is emitted unless its target data rate cannot be met. The frequency bands are characterized by one or more cut-off frequencies that separate the bands. MLWF uses a centralized gradient-descent algorithm to search for such cut-off frequencies. After obtaining the cut-off frequencies, MLWF first uses a water-filling algorithm to compute the power spectrum density (PSD) of each user, and then moves bits from the user's high-interference band to its low-interference bands to minimize its crosstalk to the adjacent DSL lines. The PSDs obtained by MLWF are subsequently passed to each DSL user as PSDMASKs for transmit spectra shaping. Simulations showed that MLWF achieves close to optimal performance with significantly less computational complexity and much faster convergence.