Cross Ambiguity Function Resolution Enhancement Using Neural Networks for Direct Position Estimation Computational Cost Reduction

Conventional methods for the Global Navigation Satellite System (GNSS) positioning involves a two-step (2SP) process, where intermediate measurements such as Doppler shift and time delay of received GNSS signals are computed and then used to for the receiver's position estimation. Alternatively, Direct Position Estimation (DPE) was proposed to estimate the position directly from the sampled signal without intermediate variables, providing high sensitivity and reliable operation in challenging scenarios. Similar to the conventional 2SP method, search methods are designed in the DPE approach to estimate position by searching indices corresponding to the peak in the joint Cross Ambiguity Function (CAF). Currently, the popular search methods for DPE includes the grid-based method and the accelerated random search (ARS) method, but both methods need a large set of candidate points to achieve an accurate position estimation and therefore they involve high computational costs. This paper proposes a neural network based search method, aiming at reducing the high computational cost in the DPE approach. To validate the proposed algorithm, a static receiver positioning scenario is simulated, and the positioning accuracy as well as the running time of each search method is compared.