Approximating Techniques for Low-Power GNSS Receivers

Abstract

Global Navigation Satellite Systems (GNSS) is a key technology used by billions of devices today. Its free global access and accurate positioning capacity makes it an extremely attractive technology to embed in various devices that requires precise outdoor navigation. The general improvements in modern electronics have led to a continuous miniaturization of the GNSS receivers, giving rise to the concept of “low-power GNSS” and their integration into new generations of energy-constrained embedded devices. However, GNSS receivers remain high-energy consuming sensors. Thus, it complicates their integration into the new Internet of Things (IoT) platforms with limited size and strict energy consumption constraints (e.g., asset or wildlife tracking). Novel processing methods are needed to meet these new energy constraints. A recent paradigm called Approximate Computing (AxC) has been introduced, and is aiming at trading off computational accuracy while still fulfilling the application requirements in exchange for energy consumption reduction. As AxC is still relatively new, its application beyond the field of computer science (e.g., Machine Learning, data compression), for instance wireless communication, have been limited.

This thesis aims to find new methods to reduce the energy consumption inside GNSS receivers, through the novel field of AxC. The goal is to assess the application of AxC techniques to GNSS processing, and to review its effect on the processing quality and energy savings. Consequently, the work is divided in three major axes.

Firstly, a review of modern GNSS is proposed, with a focus on the legacy and modernized signals structure and their impact on processing algorithms’ complexity. Legacy signal are seen to be less complex to process than modernized signals. Yet, a large part of the research over the last decade focused on improving the modernized signals, as the offer numerous advantages, including more precise and robust positioning. Consequently. the current state-of-the-art algorithms targeting lowix complexity processing in both legacy and modernized signals is assessed. Difficulties in comparing and assessing the algorithms complexity were highlighted, leading to the need for specialized framework for algorithms comparison.

Secondly, a novel benchmarking framework named “SyDR” is developed, providing a complete end-to-end processing chain. It enables a relative algorithmic complexity comparison and it highlights areas with potential optimization opportunities for AxC. Using SyDR, we showed that acquisition and tracking operations are the most computationally complex operations of the GNSS receiver processing. The correlation operations in both acquisition and tracking stages were seen as the complexity bottleneck, where algorithmic optimization is not always possible and novel computing techniques, such as AxC, could provide a new outlook.

Thirdly, the application of AxC techniques in a GNSS receiver processing is reviewed. Focus was given on inexact arithmetic, and its application to the correlation operations. Inexact arithmetic can result in substantial energy savings with limited impact on computation accuracy. Thus, its implementation inside the GNSS processing shows limited degradation in SNR and the correlation operations. Moreover, successful acquisition and tracking of GNSS signals was then performed in SyDR on real-world RF recording, using an inexact multiplier unit with a high error probability and magnitude. This multiplier can save up to 87% power consumption of the multiplication operations.

In summary, our results validate the hypothesis that AxC can be used in GNSS processing, and lead to consequent energy reductions for GNSS receivers with a proper selection of the technique.