Equalization and echo cancellation in DMT-based systems

Digital subscriber line (DSL) is a technology to provide broadband communications over the existing twisted pair telephone network. The signals received by a DSL modem are typically corrupted by channel induced noise, background noise, radio frequeny interference (RFI) and undesired echo. In this thesis we focus on the design of digital signal processing algorithms to improve the bit rate and/or the loop reach of current and future DSL systems. Furthermore, in the proposed algorithms we aim at keeping the hardware cost as low as possible. The transmission format of many DSL systems is based on discrete multitone modulation (DMT). To combat channel induced noise, DMT-based receivers perform an equalization step by means of a time domain equalizer (TEQ) and a one-tap frequency domain equalizer (FEQ) per used tone. Despite the variety of TEQ design methods presented in the literature, we show that almost all designs can be formulated as the maximization of a product of generalized Rayleigh quotients. This unified framework allows us to highlight the subtle differences between the different design methods. Special attention is paid to minimum mean square error (MMSE) TEQ design with three distinct unit-energy constraints. We illustrate that the resulting TEQs have equal performance and can be obtained from a single iterative initialization method. The modem structure can also be extended with receiver windowing to reduce the sensitivity to narrowband RFI. We show how to design a TEQ in combination with a time domain window in order to maximize the achievable bit rate. Simulations indicate that equalizer taps can be exchanged for additional windowing coefficients, resulting in substantial complexity savings. The design procedure also allows to design an equalizer and a window for small groups of tones or for each tone separately (per tone equalization). Besides equalization and receiver windowing, we also investigate structures that effectively cancel undesired echo. In a first step, we enhance existing mixedtime/frequency domain echo cancelers by adding double talk cancellation and by minimizing the complexity, independent of the misalignment between transmitted echo symbols and received far end symbols. In a second step, an alternative structure is proposed, where less transmit power is required to update the echo canceler coefficients. Finally, we propose an efficient method for fast initialization of a receiver structure based on an equalizer and an echo canceler for each tone separately (per tone equalization and echo cancellation). We show that the computational complexity and memory requirements can be substantially reduced compared to existing methods.