• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 1
  • 1
  • Tagged with
  • 2
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

Multi-precision Function Interpolator for Multimedia Applications

Cheng, Chien-Kang 25 July 2012 (has links)
A multi-precision function interpolator, which is fitted in with the IEEE-754 single precision floating point standard, is proposed in this paper. It provides logarithms, exponentials, reciprocal and square root reciprocal operations. Each operation is able to dynamically select four different precision modes in demand. The hardware architecture is designed with fully pipeline in order to comply with hardware architectures of general digital signal processors (DSPs) and graphics processors (GPUs). When considering the usefulness of each precision mode, it is designed to minimize the error among various modes as far as possible in the beginning. According to the precision from high to low, function interpolator can provide 23, 18, 13 and 8-bit accuracy respectively in spite of the rounding effect. This function interpolator is designed based on the look-up table method. It can get the approximation value of target function through the calculation of quadratic polynomial. The coefficient of quadratic polynomial is obtained by piecewise minimax approximation. Before implementing the hardware, we use the Maple algebra software to generate the quadratic polynomial coefficients of aforementioned four operations, and estimate whether these coefficients can meet IEEE-754 single precision floating point standard. In addition, we take the exhaustive search to check the results generated by our implementation to make sure that it can meet the requirements for various operations and precision modes. When performing one of the above four operations, only the tables of the operation are used to obtain the quadratic polynomial coefficient. Therefore, we can take the advantage of the tri-state buffer as a switch to reduce dynamic power consumption of tables for the other three operations. In addition, when performing lower precision modes, we can turn off a part of hardwares, which are used to calculate the quadratic polynomial, to save the power consumption more effectively. By providing multi-precision hardware, we hope users or developers, those who use the battery device, can choose a lower precision mode within the permissible error range to extend the battery life.
2

Robust tools for weighted Chebyshev approximation and applications to digital filter design / Outils robustes pour l’approximation de Chebyshev pondérée et applications à la synthèse de filtres numériques

Filip, Silviu-Ioan 07 December 2016 (has links)
De nombreuses méthodes de traitement du signal reposent sur des résultats puissants d'approximation numérique. Un exemple significatif en est l'utilisation de l'approximation de type Chebyshev pour l'élaboration de filtres numériques.En pratique, le caractère fini des formats numériques utilisés en machine entraîne des difficultés supplémentaires pour la conception de filtres numériques (le traitement audio et le traitement d'images sont deux domaines qui utilisent beaucoup le filtrage). La majorité des outils actuels de conception de filtres ne sont pas optimisés et ne certifient pas non plus la correction de leurs résultats. Notre travail se veut un premier pas vers un changement de cette situation.La première partie de la thèse traite de l'étude et du développement de méthodes relevant de la famille Remez/Parks-McClellan pour la résolution de problèmes d'approximation polynomiale de type Chebyshev, en utilisant l'arithmétique virgule-flottante.Ces approches sont très robustes, tant du point de vue du passage à l'échelle que de la qualité numérique, pour l'élaboration de filtres à réponse impulsionnelle finie (RIF).Cela dit, dans le cas des systèmes embarqués par exemple, le format des coefficients du filtre qu'on utilise en pratique est beaucoup plus petit que les formats virgule flottante standard et d'autres approches deviennent nécessaires.Nous proposons une méthode (quasi-)optimale pour traîter ce cas. Elle s'appuie sur l'algorithme LLL et permet de traiter des problèmes de taille bien supérieure à ceux que peuvent traiter les approches exactes. Le résultat est ensuite utilisé dans une couche logicielle qui permet la synthèse de filtres RIF pour des circuits de type FPGA.Les résultats que nous obtenons en sortie sont efficaces en termes de consommation d'énergie et précis. Nous terminons en présentant une étude en cours sur les algorithmes de type Remez pour l'approximation rationnelle. Ce type d'approches peut être utilisé pour construire des filtres à réponse impulsionnelle infinie (RII) par exemple. Nous examinons les difficultés qui limitent leur utilisation en pratique. / The field of signal processing methods and applications frequentlyrelies on powerful results from numerical approximation. One suchexample, at the core of this thesis, is the use of Chebyshev approximationmethods for designing digital filters.In practice, the finite nature of numerical representations adds an extralayer of difficulty to the design problems we wish to address using digitalfilters (audio and image processing being two domains which rely heavilyon filtering operations). Most of the current mainstream tools for thisjob are neither optimized, nor do they provide certificates of correctness.We wish to change this, with some of the groundwork being laid by thepresent work.The first part of the thesis deals with the study and development ofRemez/Parks-McClellan-type methods for solving weighted polynomialapproximation problems in floating-point arithmetic. They are veryscalable and numerically accurate in addressing finite impulse response(FIR) design problems. However, in embedded and power hungry settings,the format of the filter coefficients uses a small number of bits andother methods are needed. We propose a (quasi-)optimal approach basedon the LLL algorithm which is more tractable than exact approaches.We then proceed to integrate these aforementioned tools in a softwarestack for FIR filter synthesis on FPGA targets. The results obtainedare both resource consumption efficient and possess guaranteed accuracyproperties. In the end, we present an ongoing study on Remez-type algorithmsfor rational approximation problems (which can be used for infinite impulseresponse (IIR) filter design) and the difficulties hindering their robustness.

Page generated in 0.1347 seconds