Automatic Modulation Classification with Gaussian Distributed Frequency Offset

Automatic Modulation Classification (AMC) is an important technology in various communication systems. However, AMC is vulnerable to the frequency offset. Previous works treat the frequency offset as a constant while the frequency offset is a stochastic variable in some communication systems. Thus in this paper, we propose an unsupervised clustering based method, termed as Clustering based Dynamic Identification(CDI), which can blindly identify signals with stochastic frequency offset. First, we locate one of the cluster centers in constellation through hill-climbing method. Then the modulation order is derived via calculating the number of signals in the certain section. We adopt the clustering method to identify the modulation type. Different from traditional clustering methods which use the Euclidean metric, our specially designed metric is adopted in CDI to diminish the influence of stochastic frequency offset. Finally, experimental results based on hardware measurement verify that our method outperforms than previous methods. It is shown that the Bit Error Rate (BER) for classification decreases by 0.98% for 16QAM and 1.16% for 8PSK, compared with the k-means method.