Binarized Neural Networks and Trainable ISTA based Signature Code with Channel Estimation for Multiple Access Rayleigh Fading Channel

摘要

User Identification (UI) and Channel Estimation (CE) schemes are essential issues in wireless networks with massive users. Due to the high spectral efficiency, the signature code-based UI and CE schemes are widely concerned. The traditional signature code uses the discrete sensing matrix as a dictionary to generate codewords. Then, the sparse vector recovery algorithm is used to recover the user state information and channel state information in the received signal to complete the UI and CE. We proposed an end-to-end machine learning aided signature code scheme under multiple access Rayleigh fading channel called Machine Learning-Signature Code (ML-SC). The ML-SC consists of a binarized neural networks-based trainable encoder and a trainable iterative soft threshold algorithm-based trainable decoder. The dictionary is optimized by minimizing the mean squared error between original and recovered information to improve the accuracy. Our proposed scheme achieved better performance and more efficiency than the conventional schemes in the simulation. Moreover, it confirmed that the dictionary generated by ML-SC is suitable for various conventional decoders.