Early-Exiting DNN MIMO Detector Design with Fast Signal Estimation Process

Abstract

Multi-input multi-output (MIMO) receiver design has been continuously investigated in order to satisfy the massive requirements of six-generation (6G) wireless networks regarding the accuracy-centric signal estimation under the dynamic channel state information. In addtion, the architectures MIMO detector is a pioneering work for applying artificial intelligence techniques. Eventhough the DL-based MIMO detector can be able to provide accurate signal estimation, huge execution time and energy consumption of immensely large deep neural network (DNN) models in size has not been studied. To this end, we introduce a concept of early-exiting DNN MIMO detector and fast signal estimation, which outputting the estimated transmitted signal without a full forward and back propagation computation of the conventional DNN model. The early-exiting model and the fast signal estimation concept is believed to be suitable for scenarios of practical time sensitive and resouce limited applications. © 2023 IEEE.