Abstract:Automatic classification of modulation type in detected signals is an intermediate step between channel equalization and signal demodulation. This subsystem is an essential task for an intelligent receiver in various civil and military applications. Most of automatic modulation classification algorithms have been evaluated in SNR aware case which is obviously an unrealistic case. In this paper, we propose a semi-supervised online passive-aggressive classifier that uses a self-training approach for additive white Gaussian noise channels with unknown or variable signal to noise ratios to classify the modulated signals. An appropriate set of selected feature helps the general system work for a set of initial samples of each class. In the adaptation phase, unlabeled high confidence samples are used to adapt to system. An appropriate confidence measure is proposed to collect confident samples. Simulation results shows that adding unlabeled input samples to the training set improve the generalization capacity of the presented classifier in the target SNR. In addition to online characteristics of the algorithm that make the approach suitable for cognitive radio systems, this algorithm converges with a few number of signal samples. Simulation results in SNR unaware conditions show that employing this learning method results in high classification rate close to SNR aware conditions.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the lack of accuracy and robustness of automatic modulation classification (AMC) in an unknown or changing signal - to - noise ratio (SNR) environment. Specifically, most traditional AMC algorithms are evaluated under known SNR conditions, which is inconsistent with the situation in practical applications. Therefore, this paper proposes a self - training method based on the semi - supervised online passive - aggressive algorithm to improve the performance of AMC in an unknown or changing SNR environment. ### Specific description of the problem 1. **Requirements of real - application scenarios**: In many real - application scenarios, the receiver needs to detect and demodulate the received signal without any prior information. This means that the receiver must be able to work without knowing the modulation type, carrier frequency, bandwidth and other parameters of the transmitted signal. 2. **Limitations of traditional methods**: - Most existing AMC algorithms are evaluated under known SNR conditions, which is inconsistent with the non - stationary and changing SNR environment in practical applications. - These algorithms usually rely on a large amount of labeled data for training, and it is very difficult to obtain a large amount of labeled data in practical applications. 3. **Goals of the new method**: - Propose an AMC algorithm that can adapt to an unknown or changing SNR environment. - Use a small amount of labeled data and a large amount of unlabeled data for training to improve the generalization ability and robustness of the algorithm. ### Overview of the solution The solution proposed in this paper includes the following aspects: 1. **Semi - supervised learning framework**: Use the self - training method to expand the training set through the initial small amount of labeled data and gradually adding unlabeled data with high confidence. 2. **Online passive - aggressive algorithm (Online PA Algorithm)**: Apply the PA algorithm to the online learning scenario so as to continuously update the model during the real - time signal interception and processing. 3. **Feature selection**: Select an appropriate feature set, including high - order moments and high - order cumulants, etc., to improve the discriminative ability of the classifier. 4. **Confidence measurement**: Propose a confidence measurement method for selecting unlabeled samples with high confidence to be added to the training set, thereby avoiding the influence of wrong labels on the model. ### Experimental results The experimental results show that the proposed algorithm performs well in an unknown or changing SNR environment. In particular, after introducing unlabeled samples, the performance of the classifier is significantly improved. Specifically: - In a constant SNR environment, the classification accuracy is close to or even exceeds the performance under known SNR conditions. - In a changing SNR environment, the classifier also shows good robustness and adaptability. In short, by introducing the semi - supervised online passive - aggressive algorithm, this paper solves the problem of insufficient accuracy and robustness of AMC in an unknown or changing SNR environment, and provides an effective solution for practical applications.

A Self Training Approach to Automatic Modulation Classification Based on Semi-supervised Online Passive Aggressive Algorithm

An Improved Modulation Recognition Algorithm Based on Fine-Tuning and Feature Re-Extraction

Automatic Modulation Classification for Low SNR Digital Signal in Frequency-Selective Fading Environments

Blind Modulation Classification via Combined Machine Learning and Signal Feature Extraction

Automatic Modulation Classification of Signals Using Wavelet Transform and Pattern Recognition

An efficient automatic modulation recognition using time–frequency information based on hybrid deep learning and bagging approach

Automatic modulation classification of digital modulation signals with stacked autoencoders

Generative Adversarial Networks-Based Semi-Supervised Automatic Modulation Recognition for Cognitive Radio Networks

Data-and-Knowledge Dual-Driven Automatic Modulation Recognition for Wireless Communication Networks

SOAMC: A Semi-Supervised Open-Set Recognition Algorithm for Automatic Modulation Classification

A Novel Automatic Modulation Classification Scheme Based on Multi-Scale Networks

AutoSMC: An Automated Machine Learning Framework for Signal Modulation Classification

Machine learning empowered Modulation detection for OFDM-based signals

Efficient Cumulant-Based Automatic Modulation Classification Using Machine Learning

An Efficient and Lightweight Model for Automatic Modulation Classification: A Hybrid Feature Extraction Network Combined with Attention Mechanism

Automatic Modulation Classification using a Waveform Signature

Automatic Modulation Classification with Deep Neural Networks

Data-Transform Multi-Channel Hybrid Deep Learning for Automatic Modulation Recognition

A Joint Automatic Modulation Classification Scheme in Spatial Cognitive Communication

Radio Modulation Classification Optimization Using Combinatorial Deep Learning Technique

Machine Learning Based Automatic Modulation Recognition for Wireless Communications: A Comprehensive Survey