A Deep-Learning Approach to a Volumetric Radio Environment Map Construction for UAV-Assisted Networks

Table 1

3D-GAN model parameters for volumetric REM construction.

Model parameters	Values
	Generator		Discriminator
	Encoder	Decoder
Basic building layers (BL)	3D-Conv.	Transposed 3D-Conv.	CNN
	3D-Conv.	Transposed 3D-Conv.
	3D max-pooling (2)	3D-stride (2)
Layers	4 BL-enc	Transposed 3D-Conv. + 4 BL-dec	6
Input resolution (pixel size of the image)	(128, 128, 128, 1) + (128, 128, 128, 1)		(128, 128, 128, 1) + (128, 128, 128, 1)
Filters	(10, 10)	(512) (512, 512)	64 128
	(64, 64)	(512, 512) (256, 256)	256 512
	(128, 128)	(128, 128) (64, 64)	512 1
	(256, 256)	(32, 32) 1
	(512, 512)
Kernel size	4	4	4
Activation	LeakyReLU (alpha = 0.2)	LeakyReLU (alpha = 0.2)	LeakyReLU (alpha = 0.2), sigmoid (output later)