LAUSR

With the development of 5G network construction and the influx of users, the anomaly detection of base stations become more onerous. The types of uplink interference from base stations have grown significantly, and they will be affected by more than one type. It will be difficult to identify the uplink compound interference for the network operator. Meanwhile, the same interference type has flexible frequency location and variable frequency bandwidth. For multiple interference source cases, the classification accuracy and generalization ability of traditional algorithms cannot meet the needs of the actual production environment. Therefore, this paper proposes the FETTrans network to transform the compound interference identification task into a multi-label classification in natural language processing. Inspired by the attention mechanism in Transformer and combined with the pattern recognition task, the correlation between the constructed output vector and input features is obtained through the attention mechanism. On this basis, the network transforms the unidirectional static projection method into a bidirectional dynamic adaptation method, which effectively improves the identification accuracy of the variable compound interference frequency band. After verification by different test sets collected from the current network, the classification accuracy has steadily increased by more than 15%, and the mAP has reached 92%. The FETTrans network, as an end-to-end algorithm, can be adapted to another fault diagnosis.