| In the past years, a number of researchers have investigated the use of digital echo cancellation systems for transmission technologies based on discrete multi-tone (DMT) modulation, such as that used in asymmetrical digital subscriber lines (ADSL) and the likes. Almost all the existing papers report on equal transmission rate systems, while little attention has been given to the multi-rate case. In fact, the first multi-rate echo canceller structure was developed by Ho et al., and has always been referred since.; In this thesis, two sets of simplified techniques are developed. The first one reports on a fast initialization technique of the echo canceller. The method consists of replacing the first least mean square (LMS) iteration of the echo channel estimate by the solution of a system cost function in which circular convolution is assumed. The simplification of the method resides in the fact that no known sequence is used, and thus, no protocol is required.; The second simplified technique developed aims to reduce the computational complexity of the multi-rate echo canceller structure at the lower transmit rate side of the link, e.g. the line terminal side in ADSL. In this case, the echo canceller located in the terminal, processes the signals in the entire frequency bandwidth (higher rate). During the processing, the echo channel that is estimated corresponds in fact to the imperfect hybrid channel combined with the lowpass filtering action of the digital-to-analog converter (DAC) and the time-domain equalizer(TEQ). In such a situation, the frequency taps representing the estimated echo channel decay rapidly above the cutoff frequency of the DAC filter. In this work, we exploit this specific structure of the echo channel to reduce the computational complexity of the echo cancellation processing by setting a portion of the high frequency echo tap coefficients to zero, and by exploiting certain properties of the inverse Fast Fourier Transform (IFFT) algorithm.; Experimental simulations for an ADSL-like configuration involving 32 tones on the upstream and 256 tones on the downstream, showed that by employing the new proposed approach, the computational complexity can be reduced by a factor of as much as 53% while at the same time improving the achievable bitrate by 35 kbps. Furthermore, in regards of the initialization technique, the echo channel estimate convergence can be reached after around 6 iterations as compared to 15 for usual LMS algorithm. |
