3D Radio Map Reconstruction Based on Generative Adversarial Networks Under Constrained Aircraft Trajectories

Three-dimensional (3D) radio map, which characterizes the spatial distribution of the received signal strength (RSS) across a 3D space, can be a significant tool for wireless resource management and spectrum surveillance. However, incomplete 3D radio map is usually obtained in practice due to the infinite number of positions where aircraft perform RSS measurement suffering from constrained trajectories. In order to tackle this problem, we propose a 3D radio map reconstruction scheme based on generative adversarial networks (GANs), where a novel GAN variant with unsupervised learning is proposed. Specifically, the cost function of the variant integrates a reconstruction loss and an adversarial loss, while the network structure exploits ResNet and dilated convolution under the idea of patchGAN. Numerical results reveal that the network structure and the cost function can benefit the reconstruction scheme, yielding low-volatility and lower average mean squared error (MSE). Moreover, it is shown that the proposed scheme can outperform baselines in terms of the average MSE, even if the number of measured samples for inference significantly decreases.