Search

Hybrid Filter Bank (HFB) A/D converters are a good candidate for realizing the future versatile, intelligent and wide-band communication systems. However, the HFB structures exhibit a large sensitivity to the analog non-idealities of analysis part so that the classical HFB ADCs are not practically useful unless these errors are corrected. The efforts have been made in this thesis to more profoundly study this problem and to propose a group of possible solutions. Firstly, the design phase of related HFBs is described in the matrix form. Considering the simply realizable first- and second-order analog circuits as analysis filter bank and FIR digital synthesis filters, the HFB structures are designed for A/D conversion in this thesis. Simulating some exemplary HFB ADCs, it is shown that the sensitivity of HFB to analog errors is so large because the related analysis matrix is ill-conditioned, particularly in the case of oversampling process. Using Second-Order and Higher-Order Statistics, it is shown that the analog imperfections of analog circuits may digitally be estimated through the output samples. However, these techniques appear not to be applicable to the conventional HFB structure because of undersampling process included at each branch of HFB-based ADC. Thus, for exploiting the digital techniques to estimate and then correct the analog imperfections of analysis filter bank, new Multiple-Input Multiple-Output (MIMO) HFB structures are proposed so that there exist no under-sampling operation anymore between the related input-output signals. The simulations show that the MIMO (TDM and subband) HFB architectures provide not only a better output resolution but a less sensitivity to the realization errors of analysis filter bank than the classical HFB. Finally, using the TDM and subband MIMO HFBs, it is proposed to use the blind methods such as blind deconvoluton or Automatic Noise Cancelation (ANC) for compensating and then reducing the sensitivity to the analog non-idealities., Les Convertisseurs Analogique-Numérique (CAN) à Bancs de Filtres Hybrides (BFH) sont de bons candidats pour répondre aux exigences des futurs systèmes de communication devant être versatile, intelligent et à large-bande. Cependant, les BFH montrent une grande sensibilité aux non-idéalités analogiques du banc d'analyse, de sorte que les CAN à BFH classiques ne seraient pas pratiquement utilisables à moins que ces erreurs ne soient corrigées. Les efforts, dans cette thèse, ont porté sur l'étude de ce problème afin de proposer des pistes de solutions. A cet égard, la conception des BFH est, d'abord, décrite sous la forme de matrice. Puis, en utilisant des circuits analogiques simplement réalisables ainsi que des Filtres numériques à Réponse Impulsionnelle Finie (RIF), les BFH sont conçus pour la conversion A/N. Selon la simulation des CAN à BFH, nous montrons que la sensibilité de ceux-ci aux erreurs analogiques est très élevée puisque la matrice d'analyse associée est mal-conditionnée, surtout dans le cas oµu le suréchantillonnage est utilisé. Pour estimer numériquement les imperfections des circuits analogiques, nous proposons l'utilisation de méthodes d'estimation aveugle, basées sur des statistiques de seconde-ordre ou d'ordre supérieur. Cependant, ces techniques semblent ne pas être applicables aux BFH classiques en raison du sous- échantillonnage inclus µa chaque branche du CAN à BFH. Ainsi, pour exploiter les techniques numériques pour la correction des imperfections analogiques des Filtres d'analyse, nous proposons de nouvelles structures µa Entrée-sortie Multiple (ESM). Dans ces structures, il n'existe plus aucune opération de sous-échantillonnage entre les entrée-sortie associées. Les simulations prouvent que les BFH µa ESM (à sous-bande et à multiplexage temporel) mènent non seulement à une meilleure résolution mais aussi µa une sensibilité moins élevée par rapport aux BFH classique. En conclusion, en utilisant les BFH à ESM, les méthodes aveugles telles que la déconvolution ou l'annulation du bruit peuvent être employées afin de réduire encore la sensibilité aux non-idéalités analogiques.

International audience; The analog electronic circuits are always subject to some imperfections. Analog imperfections cause deviations from nominal values of electronic elements. In the case of Linear Time-Invariant (LTI) circuits, the coefficients of the transfer function include some deviations from related typical values leading to the differences between the typical (i.e. design) and the actual transfer functions. In this paper, the analog imperfections are digitally estimated using only the output samples, without any access to the input signal nor to the analog system (blind method). Super Exponential Algorithm (SEA) is used as the blind equalization technique, since it provides rapid convergence. The only assumption is that the input is a non-Gaussian independent and identically distributed (i.i.d.) signal. Using this algorithm, the effects of analog imperfections in the analog circuits can be digitally estimated and possibly compensated without any dependance on the types and the sources of the analog imperfections. It provides the possibility to have an online compensation of the imperfections (realization errors, drifts, etc.). The analog imperfections have been estimated with a precision of §0:2% and §1:3% for the exemplary RC and RLC circuits respectively.

International audience; The analog parts of mixed analog-digital systems are always subject to some imperfections. Considering Linear Time-Invariant (LTI) analog circuits, the real transfer function is then practically different from the desired nominal one and includes some deviations. The goal of this paper is to offer a model which digitally estimates the deviations from typical values supposing that the only available information is the sampled analog output (blind estimation). The model is independent from the type of the imperfections sources and is applicable when the input is a white noise (either Gaussian or non-Gaussian). The model has been applied to several RC and RLC circuits and the performance of the estimation is studied. The simulations show that the model can estimate the analog imperfections of §20% with a precision §4% in first- and second-order circuits which may be useful in correction purposes such as compensation.