GAN-Based Joint Activity Detection and Channel Estimation for Grant-Free Random Access in IoT Networks
Joint activity detection and channel estimation (JADCE) for grant-free random access relates to the fundamental problems to adequately address in supporting massive connectivity in the IoT networks. In a model-free paradigm, existing work can either achieve activity detection or channel estimation, but not both. In this paper, a new approach to the JADCE problem is introduced using the generative adversarial network (GAN) based model-free learning method. Instead of using the traditional GAN generator, we build the generator’s U-net architecture that takes as-in sink a pre-estimated value containing activity information. The generator is enhanced using an affine projection and a skip connection to guarantee that the generator remains measurement-consistent by making use of pseudoinverse properties. Additionally, a two-layer fully connected neural network is constructed for designing a pilot matrix aimed at minimizing the effects of noise due to the receiver. The simulations depict that the current approach is more effective than the previous approaches in high SNR regimes due to improvement in the learning Capability from both data consistency projection and the pilot matrix optimization.
