Signal Classification Recovery Across Domains Using Unsupervised Domain Adaptation

TL;DR

This work tackles the problem of deep-learning-based signal modulation classification failing under domain shifts between simulated and over-the-air channels. It evaluates five unsupervised domain adaptation methods—DANN, MCD, Deep-CORAL, STAR, and JAN—within a unified ResNet-18 IQ framework across three scenarios: cross-SNR, SNR-matched, and stepwise adaptation. Results show that STAR and JAN provide consistent, substantial gains, with DANN being less stable in some settings, and that transfer is best when the source SNR is in a mid-range. The study demonstrates practical viability for deploying DL-based wireless signal classifiers and offers guidance on method choice and source-domain selection for real-world deployments.

Abstract

Signal classification models based on deep neural networks are typically trained on datasets collected under controlled conditions, either simulated or over-the-air (OTA), which are constrained to specific channel environments with limited variability, such as fixed signal-to-noise ratio (SNR) levels. As a result, these models often fail to generalize when deployed in real-world scenarios where the feature distribution significantly differs from the training domain. This paper explores unsupervised domain adaptation techniques to bridge the generalization gap between mismatched domains. Specifically, we investigate adaptation methods based on adversarial learning, statistical distance alignment, and stochastic modeling to align representations between simulated and OTA signal domains. To emulate OTA characteristics, we deliberately generate modulated signals subjected to realistic channel impairments without demodulation. We evaluate classification performance under three scenarios, i.e., cross-SNR, SNR-matched cross-domain, and stepwise adaptation involving both SNR and domain shifts. Experimental results show that unsupervised domain adaptation methods, particularly stochastic classifier (STAR) and joint adaptive networks (JAN), enable consistent and substantial performance gains over baseline models, which highlight their promise for real-world deployment in wireless systems.

Figures (8)

• Figure 1: Overview of proposed framework for deep learning based signal classification using unsupervised domain adaptation. Domain shifts under consideration are shifts from simulated to OTA signals and SNR offsets.
• Figure 2: Overview of the studied domain shift methods. We evaluate on three domain shift possibilities, isolating the SNR offsets within a data domain, matching SNR offset and adapting cross-domain from simulated to over-the-air and finally stepwise SNR adaptation from simulated to over-the-air with no isolated domains.
• Figure 3: Base ResNet architecture, layers of residual and Blocks. Classifiers of DANN, CORAL, MCD, STAR, JAN use a dense block for end signal classification.
• Figure 4: Cross-SNR classification probability for simulated signals only with source as 22dB.
• Figure 5: Cross-SNR classification probability for OTA signals only with source as 22dB.