A channel map assisted multi-cell user positioning method based on federated learning

Abstract

The application discloses a kind of channel map auxiliary multi-cell user positioning methods based on federal learning, it is related to wireless communication technical field, its core step is: 1) obtain the data set for positioning, extract beam domain channel covariance matrix to data set, and calculate cosine difference degree of measurement matrix;2), central server initializes global model parameter, and broadcast this parameter to each cell;3) each cell is based on local covariance matrix and difference degree of measurement matrix, using the channel map auxiliary positioning algorithm based on federal learning is trained, generates local model;4) each cell uploads local model parameter to central server, server is updated global model parameter by aggregating each cell parameter and rebroadcast to each cell;5) repeat step 3-4 until loss converges or reaches preset communication round T , finally output the position estimation result of the sample to be positioned.The application improves the positioning performance by more than 35%, providing conditions for achieving 6G high-precision and low-latency positioning.

Images