Compound Interference Identification Based on Multi-scale Feature Pyramid using CNN

With the commercial development of 5G, the construction of a large number of base stations will cause 5G base stations to suffer from a lot of intra-system and out-of-system interference. Operators need to perform a lot of routine maintenance work and constantly analyze the types of interference. An...

Full description

Saved in:
Bibliographic Details
Published in2022 IEEE/CIC International Conference on Communications in China (ICCC) pp. 320 - 325
Main Authors Li, Boran, Yue, Lei, Zhao, Zhimin, Zhang, Lei, Dai, Jingwei, Ma, Jian
Format Conference Proceeding
LanguageEnglish
Published IEEE 11.08.2022
Subjects
5G interference identification
5G mobile communication
Base stations
Classification algorithms
CNN
Interference
Maintenance engineering
Measurement
multi-label tasks
Training
Online AccessGet full text
DOI10.1109/ICCC55456.2022.9880681

Cover

More Information
Summary:With the commercial development of 5G, the construction of a large number of base stations will cause 5G base stations to suffer from a lot of intra-system and out-of-system interference. Operators need to perform a lot of routine maintenance work and constantly analyze the types of interference. And it is found that there are a large number of compound interferences in the maintenance work of the current network. To improve operation and maintenance efficiency and save operating costs, artificial intelligence algorithms need to be used to accurately identify compound interference. Therefore, this paper uses a multi-scale convolution kernel to construct a feature pyramid, and performs multi-label classification through the binary cross-entropy loss function. We use the collected and calibrated current network datasets for training, and make a test set from multiple current network datasets in different regions. The algorithm uses the metric mAP to measure the classification performance, and the multi-classification accuracy rate reaches 86%. It shows that the algorithm has good classification performance and generalization ability. And the algorithm has been successfully deployed in many regions and has been practically applied in the current network. The algorithm has fewer learning parameters and lower time complexity, and obtains higher recognition performance, which provides the possibility for the algorithm to be embedded in intelligent instruments.
DOI:10.1109/ICCC55456.2022.9880681