Over-the-Air Modulation Classification using Deep Learning in Fading Channels for Cognitive Radio

Abstract

Background/Objectives: The ability to recognize the type of modulation is a critical function of Cognitive Radio. The objective of this study is to increase the modulation classification efficiency in Over-The-Air (OTA) signals by utilizing channel characteristics that are strong. Methods: In this work, we demonstrate how to classify Over-The-Air modulation using a deep learning technique under various fading channels simulating real-time data. The network recognizes eight different digital modulation schemes and three different analogue modulation methods. Each modulation scheme will consist of 10,000 frames with 1024 samples per frame and a sampling rate of 200 kHz. Each sample will pass through fading channels prior to training, with 80% of samples are for training, 10% for validation, and 10% for testing. Six convolutional layers and one fully linked layer comprise our network. The final convolution layer is followed by a batch normalization layer, an activation layer utilizing rectified linear units (ReLUs), and a maximum pooling layer. As a result, the final convolution layer contains soft-max activation instead of the maximum pooling layer. Findings: Modulation categorization OTA is done with two separate ADALM-PLUTO SDRs working in various channel configurations. Network-I has a forecast accuracy of 91.4 percent using 12 Mini-Batch Size and 256 Epochs, whereas Network-II has a prediction accuracy of 95.3 percent using 24 Mini-Batch Size and 128 Epochs. There are a number of ways in which SDR technology can aid to make computer-generated data more realistic, such as adopting alternative channel models. Novelty: Using Software Defined Radio hardware; the same network was used to analyze various fading situations, such as Rayleigh, Rician or Lognormal distributions, and to optimize the network topology by adjusting hyper-parameters to increase accuracy.

Automatic Modulation Classification Using a Support Vector Machine-Based Pattern Recognition Algorithm

“…From the detailed literature, it is also found that some of the other approaches have also been developed for AMC. They are statistical approaches [17][18][19][20][21][22], where different statistical features of the signal, such as correlations, moments, and cumulants in the complex envelope of the signal, are extracted and then a multilevel classification algorithm is applied for classifying the signals. The accuracy of the Back Propagation Neural Network (BPNN) is higher than that of the Kolmogorov Smirnov (KS) and higher-order statistics (HoS) approaches [23].…”