Research On Modulation Recognition Of Software-defined Radio Signals Based On Machine Learning

The modulation identification of a communication signal is to determine the modulation type of an unknown signal.The modulation and recognition of signals mainly play an important role in the military and civilian fields.At the same time,modulation and recognition technology is also one of the important technologies in cognitive radio.After many years of technology accumulation and innovation,modulation and recognition technology has emerged for modulation recognition.Algorithms and methods,such as those based on maximum likelihood ratios,which use manual extraction of features and classification through statistical machine learning algorithms to obtain signal modulation types,and also use branch deep learning techniques in the field of machine learning to perform modulation identification,each modulation The complexity of the recognition method algorithm is different,and the recognition efficiency and accuracy are also very different.However,the emergence of each modulation recognition technology is to make the signal modulation recognition more automated and simpler.In this paper,we mainly use some algorithms in machine learning,such as using deep learning techniques in supervised learning to build,use the dimensionality reduction algorithm in unsupervised learning algorithms to study modulation recognition technology,and also build a high recognition accuracy rate.Modulation identifier,so as to achieve automatic identification and detection of different types of modulation signals in the communication environment.In the context of the rapid development of modern communication technology,more and more different modulation technologies continue to appear,making the communication environment have many different modulations.Modulation signals of different types also have higher and higher requirements for modulation recognition technology.With the improvement of computer computing and big data processing capabilities,it cannot meet the modern communication environment based on early statistical machine learning,so it is more popular to use Deep learning technology and dimensionality reduction algorithms are used to study modulation recognition technology.In addition,considering the increase of modulation types,more and more types need to be identified,considering the different modulation complexity of different modulation types,and the actual environment of different modulation tasks.Different,so in Paper constructed based recognizer modulator at two different data types,which are low order modulation and the analog modulation type mixing modulator and modulator identifier identifying the high-order modulation,and the low order modulation on analog mixing conditions.Compared with the traditional statistical learning algorithm-based modulation recognizer,the modulation recognizer constructed in this paper has the characteristics of low algorithm complexity,high recognition efficiency,and high recognition accuracy.This paper mainly uses a large amount of signal data of known modulation types to update the neural network.A large number of parameters in the model,thus using the constructed neural network model to identify unknown signal sample types.In the process of constructing a complete modulation recognizer,the effects of neural network models with different structures on modulation recognition results were studied.At the same time,the effects of constructing modulation recognizers under a variety of different parameter conditions and different data conditions were studied.In-depth research and introduced the technical difficulties of different models,etc.At the same time,with the iterative update of the technology,the use of some new machine learning techniques for modulation recognition construction in the future is also the development direction and trend of social modulation recognition technology in the future.The modulation identification of a communication signal is to determine the modulation type of an unknown signal.The modulation and identification of signals has a very wide range of applications,mainly playing an important role in the military and civilian fields.At the same time,modulation and identification technology is also one of the important technologies in cognitive radio.With the continuous development of modulation and identification technology,many emerge For modulation recognition algorithms and methods,for example,there are those based on maximum likelihood ratios,which use manual extraction of features and then use statistical machine learning algorithms to classify the signal modulation type,and also use branch deep learning techniques in the field of machine learning to perform modulation recognition.The complexity of each modulation recognition method algorithm is different,and the recognition efficiency and accuracy are also very different.However,the emergence of each modulation recognition technology is to make signal modulation recognition more automated and simpler.Machine learning technology can be classified into supervised learning algorithms,semi-supervised learning algorithms,and unsupervised learning algorithms according to whether it is supervised learning.The supervised learning algorithm mainly used in this paper performs modulation recognition on signals.Supervised learning algorithms mainly use models to learn known data samples to achieve classification of unknown samples.Supervised learning has been in existence for many years from its appearance to development.Supervised learning is not a new concept.In the early days,people tried to use some methods to simulate the learning process of the human brain.At first,it developed from some simple models,that is,early statistics-based machine learning algorithms appeared.With the improvement of computer computing and big data processing capabilities,some simple artificial neural networks have appeared on the basis of early statistical machine learning,and then further development of complex neural network structure models has emerged,that is,current deep learning technologies,The deep learning technology has strong characteristics of fitting highdimensional data.It can fit complex data relationships through some non-linear processing,so as to be as close as possible to the mode of human thinking and solve some complex problems at this stage.This paper uses deep learning techniques in machine learning algorithms such as neural networks and some unsupervised machine learning algorithms such as dimensionality reduction algorithms to implement and study modulation recognition technology.Compared with traditional statistical learning algorithms,it has low algorithm complexity,high recognition efficiency,and recognition.With high accuracy,this article mainly uses a large amount of signal data of known modulation types to update a large number of parameters in the neural network model,so the constructed neural network model is used to identify unknown signal sample types,which is similar to the process of modeling.This article mainly uses the Keras deep learning framework with Tensor Flow deep learning as the framework to study the differences between the recognition results in the high-order samples and the low-order samples in the modulated signal,and the neural network models of various different structures to the modulation recognition results.The impact of different convolutional layers on the recognition results is also studied.In addition,a comparative analysis is performed on the data,such as data size and sample length.With the continuous development of machine learning technology,some new machine learning technologies will be used in the future.Optimizing modulation recognition is also the direction of future signal modulation development.