A method of semi-supervised modulation classification model based on improved scdae

Abstract

The invention belongs to the technical field of information retrieval and its database structure, and discloses a method for classifying models of semi-supervised modulation modes based on improved SCDAE, normalizing wireless signal data sets; initial training sets and test sets; unlabeled training sets The samples are input into the improved SCDAE, and the network weight gradient of each layer is calculated; the labeled samples in the training set are input into the supervised classification model, and the network weight gradient of each layer is calculated; with the goal of reducing the value of the loss function of the semi-supervised model, the gradient is used to The descent method adjusts the network weights of the semi-supervised model; the test set data samples are input into the supervised classification model to obtain the classification accuracy of the modulation method. The invention can directly extract features that are beneficial to the classification of modulation modes from the original signal, saving a lot of labor costs; it can use a large amount of easily obtained unmarked wireless signal data to enhance the generalization performance of the model and improve the accuracy of modulation mode classification.

Description

technical field [0001] The invention belongs to the technical field of information retrieval and its database structure, and in particular relates to a method for classifying a semi-supervised modulation mode based on an improved stacked convolution noise reduction autoencoder SCDAE. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: The classification of wireless signal modulation needs to extract useful information from the received signal, and there is a lot of noise interference in the actual channel environment, even if the prior information of the modulation is known, this task is still challenging. When modulation a priori information is not available, traditional modulation classification methods will not be feasible, since these methods rely on expert knowledge of the modulation. Automatic modulation classification (AMC) techniques fall into two categories: likelihood-based (LB) methods and feature-base...

AI Technical Summary

Problems solved by technology

[0004] (1) The traditional modulation classification method needs to manually extract wireless signal features, which requires a lot of labor costs, and the quality stability of the extracted features is often not high, which affects the classification accuracy

[0005] (2) The traditional modulation classification method cannot use a large amount of unlabeled signal data to improve the classification accuracy, which is a great waste of data resources

[0009] Unlabeled wireless signal data is easy to obtain, but unlabeled data is difficult to use due to the lack of label information

Images