The future applications of cognitive radio require digitization systems being capable to perform a flexible conversion in terms of bandwidth and Resolution. The digitization systems based on Hybrid Filter Bancs (HFB) provide an attractive solution for achieving this purpose. The HFBs consist of a bank of analog filters, a bank of analog/digital converters and a bank of digital filters. However, they are so sensitive that the presence of analog errors renders them impossible to carry out. Therefore, the goal of the thesis was to propose and study a calibration method for the analog errors to be corrected in the digital part. Furthermore, the proposed method had to be implementable in an embedded system. Based on Multichannel Adaptive Equalization (MCAE), we proposed a new calibration method. The digital filter coefficients are adjusted according to the real analog filters. To perform this calibration process, a known test signal is injected into the HFB which output is compared to a linked desired signal, their difference is used to adjust the digital part iteratively until the goal is achieved. For different reconstruction goals (wideband, subband or a particular narrow band), we proposed two ways to generate the test and desired signals. With the filters achieved by using method Wiener-Hopf, these signals have been validated by the evaluation of the reconstruction performances. In order to approach the optimal coefficients with a minimal computational complexity, we have implemented an algorithm of stochastic gradient. The robustness of the MCAE method has been studied both in presence of the thermal noise in the analog part and in presence of quantization errors in the digital part. A more robust test signal against the analog noise has been proposed. According to our analytical expressions, for the reconstruction goal (i.e. resolution of 14 bits), the numbers of bits needed for coding the different data of the digital part can be indicated. This thesis is a step forward for realizing future digitization systems based on HFBs.
