幅度与相位分步识别的QAM调制模式识别算法

正交幅度调制(Quadrature Amplitude Modulation,QAM)信号的调制模式识别一直以来是人们研究的热点,通过星座图来进行调制模式识别也是一种常见的方法。然而,大多数调制模式识别算法会受到频偏和相偏的干扰,因此提出了一种幅度相位分步识别的QAM识别算法来识别调制模式。先利用卷积神经网络(Convolutional Neural Networks,CNN)识别出未消除频偏相偏的QAM星座图的幅度层数,对信号进行第一次分类;再检测每个信号点的瞬时相位进行差分,得到每个点之间的相位跳变幅度;经过减法聚类确定相位跳变次数,由此对信号在相位上进行二次分类,最后识别出QAM信号的调制模式。该方法虽然步骤比传统方法繁琐,但是不依赖于信号的频偏消除和相偏消除,能够起到很好的抗频偏作用。此外,因为没有频偏消除和相偏消除的步骤,所以使得信号不至于在频偏消除和相偏消除等预处理过程中损失信息量。经过试验,这种方法在识别率上比传统的神经网络识别方法在低信噪比下有更好的识别率。

QAM Modulation Pattern Recognition Algorithm by Step-by-step Identificationof Amplitude and Phase

The modulation mode recognition of QAM signals has always been a research hotspot,and modulation mode recognition implemented through constellation diagrams is also a common method.However,most modulation pattern recognition algorithms are subject to interference from frequency offsets and phase offsets.Therefore,a QAM identification algorithm by step-by-step identification of amplitude and phase is proposed to identify the modulation mode.Firstly,the convolutional neural networkCNN(Convolutional Neural Networks)is used to identify the amplitude layers of the QAM constellation map without eliminating the frequency offset and skew,and the signals are classified for the first time.Then the instantaneous phase of each signal point is detected,and a difference is made to obtain the phase jump amplitude between each point.The number of phase transitions is determined through subtraction clustering,so that the signals are classified twice in phase.Finally,the modulation mode of the QAM signal is identified.Although the steps of this method are more complicated than the traditional method,it does not depend on the frequency offset cancellation and phase offset cancellation of the signal,and can play a good role in resisting the frequency offset.In addition,because there is no step of frequency offset elimination and phase offset elimination,the signal is prevented from losing the amount of information during the pretreatment processes of frequency offset elimination and phase deviation elimination.Experiments indicate that this method has a better recognition rate than traditional neural network recognition methods at low SNR.