Support Vector Clustering Analysis Of Radar Emitter Signals

Radar reconnaissance and countermeasure include interception, deinterleaving and recognition of the radar emitter signal. Recognition of the radar emitter signal is the key of signal processing in radar countermeasure and an importance part of electronic intelligence system and electronic support system. The recognition accuracy of the radar emitter signal can decide whether radar countermeasure will be successful or not. Because the military must keep secret, we can hardly get the prior probability and conditional probability of the radar emitter signal. The feature's parameters of the radar emitter signal which we get though some kinds of channels are also incomplete and uncertainty. So, traditional methods encounter great difficulties.Unsupervised learning is a good method to solve the problem of recognition of unknown radar emitter signal. Support vector clustering (SVC) is a kind of Unsupervised clustering method which based on support vector machine. SVC can obtain cluster boundaries of arbitrary shape and soft margin constant allows the data samples to lie outside the feature-space sphere. These samples are called outliers. So SVC can also deal with overlapping clusters. But, SVC not only has a high time complexity but also gets low recognition rate when distribution of the data set is complex and uneven. We do some research on how to lower the time complexity, at the meanwhile improve the recognition rate when distribution of the data set is complex and uneven. So, we can get fast and accurate recognition of radar emitter signal. The main work and research results of this paper as follows:1. Learn the basic theory of SVC and its status of research. Then we can find out its advantages and disadvantages.2. SVC is applied in radar emitter signal recognition. The experimental results show that SVC not only has a high time complexity but also gets low recognition rate when distribution of the data set is complex and uneven.3. Based on edited nearest neighbor, we combine the advantages of K-nearest neighbor and SVC. We made a strategic decision of two-step to divide the data set. First, SVC is employed to cluster all of the unknown samples. The wrong clusters are edited by editing rules. Then, K-Nearest Neighbor will deal with the samples of these clusters based on the distribution of the correct clusters. This method is tested by using radar emitter signals. The experimental results show that this method can greatly improve the recognition rate.4. Combining the advantages of SVC and K-means, a novel SVC based on K-means (K-SVC) is presented. First, we use k-means to divide the data set into several subsets. Then, SVC is employed to cluster each subset. K-SVC can decompose a large quadratic-programming problem into smaller ones. So, it can reduce the computational cost. When the data set is divided into several subsets, more suitable parameters can be easily found for each subset. The introduced method is tested by using radar emitter signals, and experimental results show that K-SVC can obtain better clustering performances in terms of computational efforts and correct recognition rates than other methods.