数字调制信号的分类识别在信号传输中应用广泛，不仅可用于多址多用户的信道估 计和多用户检测，实现更有效的信号分离和接收处理；还可以用于频谱感知和动态频谱 分配，实现更灵活的频谱资源管理和优化。通信系统中的信道环境日益复杂，导致信号 受到噪声干扰的影响加重，使得特征提取变得困难。此外，调制信号数量过大也会增加 特征提取的复杂度，从而影响信号的识别率。针对噪声信号特征参数提取困难，计算复 杂度高，且低信噪比环境下识别率低的问题，本文提出一种基于特征提取的识别算法， 通过高阶累积量和瞬时信息两种特征参数结合识别算法对数字调制信号的识别展开研究。 本文主要研究内容如下： （1）基于高阶累积量和变分模态分解（Variational Mode Decomposition，VMD）的 数字调制信号识别算法。针对 VMD 依靠传统阈值分解得到的 IMF 分量不能有效重构信 号特征分量的问题，提出了基于VMD的调制信号识别算法，对MASK、MPSK、MFSK、 16QAM、64QAM这 11类单载波调制信号进行识别。首先，通过 SSA-VMD算法，将功 率谱熵与峭度的比构造为适应度函数，来对 VMD 中的两个参数进行优化，实现 VMD 自适应分解。其次，将含噪信号经过优化后的 VMD 分解，剔除含噪信号并重构有效信 号。最后构造能够区分 11 类调制信号的高阶累积量参数，在提取有效信号的高阶累积 量后，根据特征参数的阈值来对这 11 类调制信号进行识别。仿真结果表明，本文提出 的优化算法相比于传统算法，降低了特征参数的计算复杂度。同时，通过该预处理手段 提升了信号信噪比，且在同一信噪比的通信环境下，能够达到识别率更高的效果。 （2）基于瞬时信息和反向传播神经网络（Back Propagation Neural Network，BPNN） 的数字调制信号识别算法。针对高斯环境下数字调制信号特征参数提取繁琐，识别算法 复杂度较高，低信噪比环境下不能有效识别信号的问题，提出一种基于瞬时信息和 BP 神经网络的调制信号识别算法，来对 2ASK、4ASK、2PSK、4PSK、2FSK、4FSK 这 6 类调制信号进行识别。首先，搭建决策树模型，构造基于瞬时信息的特征参数。其次， 针对决策树识别存在低信噪比下信号识别率低的问题，引入自适应学习速率梯度下降法 训练 BP 神经网络并设计网络结构。最后，通过实验仿真确定隐含层神经元个数和训练 样本数，搭建 BP 神经网络来对六类调制信号进行识别。实验结果表明，本文算法降低 了识别算法的复杂度，同时低信噪比下的信号识别率也得到有效提升。

Classification and identification of digital modulated signals are widely used in signal transmission, not only for channel estimation and multi-user detection in multi-access multi-user channels, to achieve more effective signal separation and reception processing, but also for spectrum sensing and dynamic spectrum allocation, to achieve more flexible spectrum resource management and optimization. The channel environment in communication systems is becoming increasingly complex, causing signals to be affected by noise interference, making feature extraction difficult. In addition, the large number of modulated signals will also increase the complexity of feature extraction, thereby affecting the recognition rate of signals. To solve the problem of difficulty in extracting feature parameters of noise signals, high computational complexity, and low recognition rate in low signal-to-noise ratio environments, this paper proposes a feature-based recognition algorithm by combining the high-order cumulants and instantaneous information feature parameters to identify digital modulated signals. The main research content of this paper is as follows: (1)An algorithm for identifying digital modulated signals based on high-order cumulants and VMD. To solve the problem that the IMF components obtained by traditional threshold decomposition in VMD cannot effectively reconstruct the signal feature components, a modulation signal identification algorithm based on VMD was proposed, which can identify the 11 single-carrier modulation signals including MASK, MPSK, MFSK, 16QAM, and 64QAM. Firstly, the power spectral entropy and roughness ratio were constructed as the fitness function by the SSA-VMD algorithm to optimize the two parameters in VMD, achieving the adaptive decomposition of VMD. Then, the noisy signal was decomposed by the optimized VMD, the noisy signal was removed, and the effective signal was reconstructed. Subsequently, the highorder cumulant parameters that can distinguish the 11 modulation signals were constructed. After extracting the high-order cumulant parameters of the effective signal, the 11 modulation signals were identified based on the threshold value of the feature parameters. Simulation results show that the proposed optimization algorithm is more efficient than traditional algorithms in terms of feature parameter calculation complexity. At the same time, the signal-to-noise ratio is improved by the preprocessing method, and the same signal-to-noise ratio communication environment can achieve higher recognition rate. (2)An algorithm for identifying digital modulation signals based on instantaneous information and BP neural network. To solve the problem of complicated feature parameter extraction and high complexity of the identification algorithm in Gaussian environment, as well as the low signal recognition rate in low signal-to-noise ratio environment, a modulation signal recognition algorithm based on instantaneous information and BP neural network is proposed to identify the 6 types of modulation signals, including 2ASK, 4ASK, 2PSK, 4PSK, 2FSK, and 4FSK. Firstly, a decision tree model is built to construct feature parameters based on instantaneous information. Secondly, to solve the problem that the decision tree recognition has a low signal recognition rate in low SNR environment, an adaptive learning rate gradient descent method is introduced to train the BP neural network and the BP neural network structure is designed. Finally, the number of hidden layer neurons and the number of training samples are determined through experimental simulation to build the BP neural network for signal recognition. The experimental results show that the proposed algorithm reduces the complexity of the recognition algorithm and effectively improves the signal recognition rate in low SNR environment.