Telecommunication systems today have increasing complexity and require higher energy consumption. Therefore, energy-efficiency becomes one of the most important goals in a design of embedded systems. This thesis proposes a dynamic scaling precision mechanism to reduce energy consumption in an embedded system, especially in a wireless mobile terminal. Energy--efficient implementation of digital signal processing algorithms in embedded systems requires using fixed-point arithmetic in order to satisfy the cost, power and area constraints. In traditional approaches, the data bit-widths are calculated based on the worst case so that they are satisfied in all cases. We propose another approach to dynamically change the specification according to the environment (e.g. the transmission channel quality) with the goal of reducing energy consumption under certain conditions. Firstly, the relation between the quantization noise and the bit error rate at a receiver noise level is studied for a QPSK transmission system. The result is then applied in Direct Sequence Code Division Multiple Access (DS-CDMA) systems. Among several telecommunications systems using the DS-CDMA technique, we demonstrate how to dynamically adapt the fixed-point algorithm accuracy of a WCDMA 3G receiver. The fixed-point conversion uses a combinatorial optimization algorithm for the determination of each bit-width under an accuracy constraint. The second part of this thesis concentrates on optimization algorithms. We propose new algorithms for problems having a single constraint or a series of constraints corresponding to different accuracy levels of a self-adaptive system. The result of Multi-Objective Genetic Algorithms (MOGA), a Pareto front, allows determining the bit-widths for each quantization noise level. An improved version of MOGA combined with elitism and tabu search is proposed. In addition, we suggest using GRASP, a stochastic local search algorithm, to find a result in a relatively short time in comparison with MOGA.
