A Fast Retraining Scheme for G.Lite ADSL Receviers

• Main Authors: Jia-Feng Jiang, 江嘉峰
• Other Authors: Chin-Liang Wang
• Format: Others
• Language: en_US
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/55050703852663299879

碩士 === 國立清華大學 === 電機工程學系 === 89 === In order to satisfy the demand for reliable high-speed data transmission through billions of telephone lines around the world, the American National Standards Institute (ANSI) and the European Telecommunication Standards Institute (ETIC) have chosen the discrete multi-tone (DMT) modulation as the standard line code for ADSL to combat the severely distorted channel since 1993. In a DMT-based splitter-less ADSL receiver, the coefficients of equalizers may be reinitialized whenever a gross channel changes abruptly. Fortunately, some of these situations may be determined from the predetermined states and can be stored as a profile used to compare with the results from fast retraining and was chosen as the next steady state. In this thesis, the invoking mechanism of fast retraining was first proposed to determine when we should enter the fast retraining stage. Subsequently, the retraining method of equalizers will be discussed during fast retraining. Finally, the profile-updated method will be proposed to improve the efficiency of updating profiles, i.e., the power-saving issue of updating profiles when the states of profiles are not infinite and some of them appear frequently. Per Tone Equalization (PTE) for DMT-based systems, an alternative receiver structure with a complex multi-taps frequency-domain equalizer (FEQ) per tone, was proposed to maximize the signal-to-noise ratio (SNR) for each individual tone instead of the usual structure consisting of a real time-domain equalizer in combination with a complex 1-tap FEQ per tone. Moreover, PTE has a reduced sensitivity to the synchronization delay. However, the channel model length is truncated to N maximum samples where N is the IFFT size in the DMT-based system, and use the indirect method of equalization to calculate taps in FEQ and SNR per tone. Whenever the channel response is longer than N, then the size of the compound matrix for calculating modified cost function grows with the channel length. Moreover, this method needs the exact channel estimation. In this thesis, the simple training method (the direct method) will be introduced for multi-tap FEQ, and will add several feedback taps per tone to slightly raise the SNR during the fast retraining interval.