GNSS observation simulator with denoising diffusion probabilistic model based generating network in urban occlusion environment

Global Navigation Satellite System (GNSS) positioning can be heavily influenced in urban occlusion environment. Enhancement algorithms such as anti-multipath and GNSS based integrated navigation play an important role in improving positioning precision. Those algorithms’ development relies on sensor data with high quality in complex occlusion environment. However, expensive truth system and time-consuming outdoor work make data collection difficultly accessible. In this paper, an observation level GNSS simulator with denoising diffusion probabilistic model (DDPM)-based generating network is proposed to achieve low cost and high quality simulation of raw GNSS observations in urban occlusion environment. To evaluate the performances of simulation, a series of indexes, both in time-series shape as well as in probabilistic distributions, are applied. Our model achieved 88.26% in the fidelity of standard point position (SPP) errors simulation. Compared with the max value of environmental errors is usually over 10 m, our model achieved a mean RMSE of 0.630 m in pseudo range environmental errors simulation. Meanwhile, the non-environmental error information has been preserved as the mean RMSEs between simulated data and open-sky observations are 0.778 total electron content unit (TECU), 0.020 m, 0.129 m for slant total electron content (STEC), zenith tropospheric wet delay (ZWD) and receiver clock bias. As for artificial intelligent (AI) application, our simulated observations achieved an accuracy of 80.60% in pre-trained positioning scene recognition network. Our simulator is now available at https://github.com/alxanderjiang/GNSS_diffusion.