This thesis studies the harsh quantization of radar signals. More specifically, what can be achieved in terms of localization of targets using FMCW radars from 1-bit dithered measurements and processing. The first part of this thesis leverages the framework of Quantized Compressive Sensing to achieve high quality localizations using coarse 1-bit measurements from an FMCW radar. The gain provided by the added dither is highlighted through simulations and actual radar measurements and are compared with the developed reconstruction bounds. Range and angle estimations are achieved using the PBP and QIHT algorithms. The second part highlights some difficulties inherent to adding a random dither to radar signals and in response, studies an alternative way of dithering the measurements by altering instead their phases. This method, compared to the additive case, is shown to be a viable alternative in the search for an implementation of 1-bit quantization of radar signal that has theoretical guarantees and is cost-effective to implement. This new way of dithering radar signals is compared using Monte-Carlo simulations against its additive counter-part and using actual radar data. This alternative way of dithering is linked to the Phase-Only acquisition, that only measures the phase of complex signals, and its reconstruction performances are studied through the lens of the guarantees provided to PBP using the (l1,l2 )-Restricted Isometry Property. This property is proved for complex Gaussian random matrices. The thesis does not finish by the study of yet another way of acquiring a quantized version of a signal but by studying the quantization of the processing itself. Indeed, using low resolution processing could enable more power-efficient implementations. To that end, we study the reconstruction guarantees of the Projected Back Projection algorithm in the setting where the back-projection used is a 1-bit quantized version with additive dithering of the one used in high resolution processing. We show a uniform bound on the l2-reconstruction that behaves as O(m^-2 ). This study is then extended to the case of back-projection operators that have a factorized representation. These factorized representations, among which the FFT is the most well-known, can often be computed efficiently thanks to their sparse and factorized structures. This thesis shows that in cases where either the power or the amount of data that one can use for the estimation is limited, lowering the individual resolution of the measurements and possibly of the processing, can allow for better results than sub-sampling those high-resolution measurements to fit within the limitations. This was shown throughout the thesis using both theory and simulations often accompanied by real radar measurements. (FSA - Sciences de l'ingénieur) -- UCL, 2021