Turbo-AI, Part II: Multi-Dimensional Iterative ML-Based Channel Estimation for B5G

Targeting to potential evolution of future wireless systems, precise channel estimation is regarded as one of the fundamental prerequisites. In this paper, we focus on Machine Learning (ML) based channel estimation with Turbo-AI, which is an iterative training approach, and can monotonically reduce the post-processing noise variance of Gaussian inputs after each iteration, by updating the Neural Network (NN) models with re-training. After discussing about the initial results of Turbo-AI as Part I in our introductory paper, we will now deploy the same principle in a multicarrier system from a more practical view point. Multi-dimensional Turbo-AI will estimate the channel in an iterative manner through frequency, time and spatial domain cooperatively. Since the complexity of traditional channel estimation method will be extremely high due to the multi-dimensional data structure, Turbo-AI can be regarded as a complementary solution to balance the performance and complexity. Throughout this paper, we exploit 5G compliant link level simulator to show that genie-aided upper bound can be approached by Turbo-AI. Challenges of realizing Turbo-AI based channel estimation in practical systems for future wireless communication towards Beyond 5G (B5G) are focused on.