A Transfer Learning Approach for Compressed Sensing in 6G-IoT

The data in the Internet of Things (IoT) and the sixth generation (6G) wireless networks increases dramatically with higher dimensions compared to the traditional wireless networks. Compressed sensing (CS) has been adopted to effectively reduce the amount of transmitting signal with sparsity and recover accurately at the receiver. It has been proved that better recovery performance can be achieved via deep learning based CS approaches. However, these methods require a mass of relevant data to train neural networks (NNs), not adapted for the case of small sample data. In this paper, a convolution-based transfer compressed sensing (CTCS) model is proposed to reconstruct the compressed signal based on transfer learning. Ultra-wide band (UWB) radar echo signal and Mnist hand-written dataset are selected to evaluate the performance of CTCS. It is verified the proposed model outperforms other traditional reconstruction algorithms in 6G-IoT under different noise levels, measurement numbers, and signal sparsities.