Iterative sparse recovery from high-dimensional measurements
Date
2017
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Compressive-sensing techniques have gathered strength over the last decade since they overcome the requirements of the Nyquist-Shannon sampling theorem, asserting that a signal can be recovered from far fewer measurements than traditional methods use. In this field, the theory of graphical models has been applied to simplify the solving of large-scale regression problems. Such iterative algorithms as approximate message passing have been developed to solve problems with a high inner complexity in a simple computational manner. ☐ One of the topics where this techniques are really suitable is the nonlinear suppression in communications. Current trends in mobile communications systems seem to agree that matching a moderate increase of data rate with high spectral and power efficiencies are key challenges. New standards of wireless digital communications are pushing the design of power amplifiers (PAs) towards challenging requirements in terms of linearity and efficiency. 4G systems have also incorporated the advances accomplished during the last decade in the processing of spread spectrum and multicarrier signals, together with Multiple Input Multiple Output (MIMO) techniques. While these new systems demand active devices like PAs to work near saturation seeking power efficiency, the created nonlinearity may cause the overall system not to meet the requirements of out-of-band emissions and in-band distortion as well. The need of digital compensation techniques and evolution in the designing of new digital signal processing architectures make digital predistortion (DPD) a convenient linearization approach. Digital predistorters usually rely on behavioral models such as memory polynomials (MPs), generalized memory polynomials (GMPs), pruned Volterra series, etc. Volterra system identification suffer the "curse of dimensionality", since they tend to grow exponentially in number of coefficients as the memory and nonlinear orders become larger. Practical use of DPDs require pruning of the models, where only the most important coefficients of it are retained, selecting them by some criteria.
Description
Keywords
Applied sciences, Digital predistortion (DPD), Power amplifier, Sparse signal processing, Volterra series