Automatic Modulation Classification Of MQAM Signals Based On Manifold Learning

Automatic Modulation Recognition (AMR) of communication signals is an intermediate step between signal detection and demodulation. The modulation types of the received signals are determined by analyzing and processing the received signal in complex environment where is full of noise. AMR has been widely used in military or civilian fields in recent years. Feature based(FB) method is a classic approach to solve AMR problem which is widely used in signal recogniton and cluster analysis. Starting with Manifold Learning, this paper focuses on the AMR of M-ary quadrature amplitude modulation (MQAM) signal recognition based on local gometry has been studied deeply. Details are as follows:The Local Preserve Projection (LPP) can preserve the local geometry between the data, But LPP do not use the discrimination information of data in training. To conquer this problem, a supervised LPP is proposed so the data with the same class labels in local adjacency graph get more closely after projection. Five modulation types are considered here:8QAM,16QAM,32QAM,64QAM and 128QAM. High order cumulants have been used as input features. The result shows that the proposed method performs well at various channel SNR levels.In terms of the description of the geometry of data with LPP, only in-class geometry of data is considered without between-class geometry. Therefore, the performance in modulation recognition is not satisfactory. Marginal Fisher Analysis (MFA) takes this into consideration by building a localized between-class adjacency graph. This method can not only maintain local spatial geometric properties, but also maintain distinguish geometric properties of data. It is proposed an AMR approach based on MFA.6 high order cumulants have been used as features. Nearest neighbor classifier is used in this paper. With this method, a better effect on modulation recognition can be obtained. The experiment show that compared with other algorithms, the recognition rate of MQAM signal can be improved significantly based on MFA algorithm. This method not only possesses anti-noise performance, but also has the good robustness.