| Title |
Performance Enhancement of Automatic Modulation Recognition through a Weighted-sum-based CFO Preprocessing Method |
| Authors |
박나윤(Nayun Park) ; 전민욱(Min-Wook Jeon) ; 정진우(Jinwoo Jeong) ; 심이삭(Isaac Sim) ; 윤상범(Sangbom Yun) ; 서정현(Junghyun Seo) ; 김형남(Hyoung-Nam Kim) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.3.99 |
| Keywords |
Automatic modulation recognition; Carrier frequency offset estimation |
| Abstract |
Automatic modulation recognition is a technology that automatically identifies the modulation method of a signal and is used in various wireless communication fields. Especially in the defense field, it is used as a key step for detecting, identifying, and responding to unknown enemy signals. However, in a real environment, the carrier frequency offset (CFO) caused by the transmission/reception clock mismatch and Doppler rotates the phase of the signal, deteriorating the modulation recognition performance. To solve this problem, a preprocessing process of estimating a CFO is required before performing the modulation recognition process. However, since prior information for CFO estimation is often limited, a non-data-aided (NDA) method should be used in this case. Recently, attempts have been made to mitigate the impact of CFO by using deep learning such as fully feedforward network (FNN) and ResNet, but real-time application is limited due to large-scale learning and high computational complexity. This paper introduces a weighted sum-based CFO preprocessing algorithm that does not require a learning process. Instead of exploring the entire frequency domain, the weighted-sum-based technique calculates the size of the complex index correlation for a ?? of candidate frequencies set in advance, and applies an index weighting with a sensitivity parameter ?? to the result to calculate the final CFO as a weighted sum. The received signal is corrected with the estimated CFO and used as an input for automatic modulation recognition. As a result of the simulation, when compared with RMSE, the weighted-sum technique showed superior performance in CFO estimation accuracy compared to the existing maximum likelihood estimation technique, and when the corrected signal is input into the deep learning model, the modulation recognition accuracy is significantly improved. Through this, it was confirmed that this method can stably improve the automatic modulation recognition performance in the CFO environment without prior information or additional learning. |