Velocity measurement improvement of landing radar considering irradiated surface using neural networks

In this study, a signal-processing method for the landing radar of a lunar lander is proposed using deep learning. To perform a precise landing, measurement of the relative velocity with respect to the surface is essential. To measure the velocity, the landing radar irradiates the surface with a pulse wave and observes the Doppler shift. High-precision measurement on complex terrains, such as a crater or slope, has always been the problem of landing radar because the irradiated terrain causes a deformation of the reflected pulse wave and strongly affects the measurement accuracy. A system is proposed in this study that performs measurements with high accuracy on complex terrains using convolutional neural networks. In the proposed method, spectrograms are used as input data to consider the effect of irradiated terrain on the measurement data. Experiments show that our method not only improves the measurement accuracy compared with the existing method but also can be implemented from the viewpoint of execution time. Moreover, this paper attempted to deepen the network architecture and input irradiated terrain data simultaneously. It was confirmed that the measurement accuracy was further improved by this enhancement.