An Online Automatic Modulation Classification Scheme Based on Isolation Distributional Kernel

TL;DR

This paper tackles the challenge of online Automatic Modulation Classification under time-varying channel conditions by introducing a distribution-based signal representation using Isolation Distributional Kernel (IDK). It couples IDK with Online Gradient Descent (OGD) to form IDK-OGD, an online, multi-class AMC classifier that updates incrementally only when ground-truth labels are available. The method avoids information loss from signal-to-image/sequence conversions and achieves linear-time classification, outperforming state-of-the-art DL baselines under mismatched channel scenarios. Empirical results on RadioML2018.01A and synthetic channels demonstrate robust online adaptability and superior performance in realistic streaming conditions, highlighting practical potential for real-time cognitive radio and non-cooperative networks.

Abstract

Automatic Modulation Classification (AMC), as a crucial technique in modern non-cooperative communication networks, plays a key role in various civil and military applications. However, existing AMC methods usually are complicated and can work in batch mode only due to their high computational complexity. This paper introduces a new online AMC scheme based on Isolation Distributional Kernel. Our method stands out in two aspects. Firstly, it is the first proposal to represent baseband signals using a distributional kernel. Secondly, it introduces a pioneering AMC technique that works well in online settings under realistic time-varying channel conditions. Through extensive experiments in online settings, we demonstrate the effectiveness of the proposed classifier. Our results indicate that the proposed approach outperforms existing baseline models, including two state-of-the-art deep learning classifiers. Moreover, it distinguishes itself as the first online classifier for AMC with linear time complexity, which marks a significant efficiency boost for real-time applications.

Figures (16)

• Figure 1: Functional diagrams of existing DL classifiers and the proposed IDK-OGD classifier. The key differences are: (i) none of the existing signal representations used in DL classifiers include distribution representation; and (ii) DL classifiers are in batch mode and have difficulties in designing a practical online AMC scheme. The distribution representation, called Isolation Distribution Kernel (IDK), enables a kernel-based OGD classifier to be used readily for AMC
• Figure 2: Sketch of distribution-based signal representation (different letters, i.e., A, B & C, stand for different modulation formats). Each point-pair of the I/Q baseband signal (shown in I) is translated into a point in a two-dimensional I/Q signal space of a modulation format (shown in II). All the points in the I/Q space (as a sample set from an unknown distribution) can then be transformed via a distributional representation into a point in a feature space (shown in III)
• Figure 3: Baseband signals of 8PSK & 16APSK at SNR = 15 & 20 dB in distribution representation
• Figure 4: The similarity matrix of the baseband signals, computed using a distributional kernel ting2021-IDK when signals are in distribution representation
• Figure 5: Baseband signals in image representation