Print Researchers from the National Time Service Center (NTSC) of the Chinese Academy of Sciences (CAS) proposed a demodulation algorithm based on the multiclass support vector machine (MSVM). The results were published in the journal Remote Sensing under the title "Research on eLoran Demodulation Algorithm Based on Multiclass Support Vector Machine"
As a vital supplement and backup to GNSS, eLoran systems play a pivotal role in ensuring the reliability of high-precision timing services. Traditional demodulation methods are based on the phase comparison of time-domain signal sampling points, and their anti-interference and noise resistance performance is limited.
By utilizing the principle of support vector machines, Researcher transformed the multi-classification problem into multiple binary classification problems, and classification is carried out according to the sample confidence scores, significantly improving the demodulation accuracy and stability (See Figure 1). When constructing the algorithm, an in-depth study of the kernel function was conducted, and it was found that the linear kernel function has an excellent effect on processing eLoran signal demodulation. It can efficiently separate samples in high-dimensional space and suppress noise. The construction of feature vectors focuses on a specific range of signal pulse groups to extract key features, thus balancing accuracy and computational cost.
Figure 1. Flowchart of the MSVM algorithm. It includes several core steps such as feature vector construction, kernel function processing, support vector machine training, MSVM model building, and demodulation by confidence score classification.
According to Dr. LIU Shiyao, the first author of this research, the experimental results show that the MSVM algorithm demonstrates excellent performance under different noise and interference environments. Compared with the traditional EPD algorithm, it has a significant performance advantage. Especially in high-noise and strong-interference scenarios (such as SNR = -10 dB), the improvement in its demodulation accuracy rate is as high as more than 12.5%. Compared with other various machine learning algorithms designed by the research team, such as random forest (RF) and K-nearest neighbors (KNN), the MSVM algorithm also has obvious advantages in terms of classification accuracy and computational efficiency (see Figure 2 and Figure 3).
Figure 2. DAR comparison curve graph of different algorithms without continuous wave interference (CWI).
Figure 3. DAR comparison curve graph of different algorithms with different numbers of CWI. Subgraphs (a)~(d) respectively represent 1~4 CWI interferences.
This research innovatively introduces the idea of AI and brings innovative ideas for eLoran demodulation technology. The study not only opens up a brand-new way of thinking for the design of eLoran system receivers but also provides strong technical support for enhancing the high-precision timing service ability of the system.