Compressed sensing in mobile systems

Résumé

We are interested in investigating the compressed sensing paradigm in the context of mobile systems to deal with the logistical and computational challenges in future communications. Due to the increasing number of connected objects (Internet of Things), the data exchanged in turn grows exponentially: this is the era of big data. This adds a level of intelligence to devices enabling them to communicate real-time data without a human being involved, effectively merging the digital and physical worlds. Hence the need to propose efficient techniques for compressing data from a variety of sources. Firstly, we propose simpler speech codecs based on quantization of compressed sensing measurements. The results show that the proposed codecs can be promising alternatives to current speech codecs. Secondly, we focus on the communication system. The behaviour of compressed sensing-source coding within the transmission chain is studied when undergoing real mobile communication-conditions. More specifically, we design new end-to-end mobile communication schemes based on compressed sensing. The proposed designs incorporate the compressed sensing-based speech codec instead of sampling signals at Nyquist rate then using a complex speech codec. Additionally, efficient techniques are chosen for channel compensation. The proposed systems show a simplified design, and allow reducing bit rate and processing load compared to actual communication systems based on adaptive multi-rate wideband (AMR-WB) speech codecs. The recovered speech has good quality and fair intelligibility scores when dramatic communication conditions are experienced (Rayleigh environment). In addition to reducing the computational burden for all the transmission steps, compressed sensing allows secure communications without additional costs. Thirdly, we consider the background noise coming from the environment. We propose a new speech enhancement method based on compressed sensing. In this approach we perform noise subtraction in the measurement domain before sparse recovery. Significant results are obtained showing that the proposed method is a good alternative to classical as well as prior compressed sensing-based speech enhancement methods, especially at low signal-to-noise ratios.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 4

Téléchargements : 0