DKDNet: Dual Knowledge and Data-Driven Network for Cross-Domain Automatic Modulation Classification

2026-07-11 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Wireless Communications & Signal Processing · Depth: Expert, extended

Summary

DKDNet, a dual knowledge and data-driven network, is proposed for cross-domain Automatic Modulation Classification (AMC) to address performance degradation from communication environment distribution shifts. The network integrates prior-guided signal representations with data-driven feature learning, selecting in-phase/quadrature (IQ), amplitude–phase (AP), and autocorrelation function (ACF) as compact, complementary inputs. DKDNet employs a Multi-Representation Feature Encoder (MRFE) for unified feature extraction and a Dynamic Lightweight Fusion Unit (DLFU) for adaptive feature fusion. Experiments on the RML2025 Series and public RML22 datasets demonstrate its superiority, achieving 47.58% adapted accuracy and a 12.83% UDA gain on the AWGN→Ri1 setting, and improving all-SNR accuracy by 8.03–15.04 percentage points across various tasks. It also shows enhanced robustness, flexibility, and sample efficiency.

Key takeaway

For Machine Learning Engineers developing Automatic Modulation Classification (AMC) systems facing performance degradation due to communication environment distribution shifts, you should integrate signal prior knowledge into your models. Consider DKDNet's approach of using IQ, AP, and ACF representations with adaptive fusion (MRFE and DLFU) to learn features that are both discriminative and transferable. This strategy significantly improves cross-domain generalization, robustness, and sample efficiency, particularly at higher SNR levels.

Key insights

Integrating signal prior knowledge like IQ, AP, and ACF with data-driven learning enhances cross-domain automatic modulation classification generalization.

Principles

• Signal priors vary in modulation discriminability and domain stability.
• Complementary signal representations improve cross-domain generalization.
• Adaptive fusion of heterogeneous features is crucial for integration.

Method

DKDNet uses MRFE for representation-specific encoding of IQ, AP, ACF, then DLFU adaptively fuses features, optimized by modulation classification and adversarial domain alignment objectives.

In practice

• Utilize IQ, AP, and ACF as prior-guided inputs for AMC.
• Employ the RML2025 Series for controlled cross-domain evaluation.
• DKDNet is compatible with various AMC backbones and UDA methods.

Topics

• Automatic Modulation Classification
• Cross-Domain Adaptation
• Signal Priors
• Deep Learning
• Unsupervised Domain Adaptation
• Feature Fusion
• RML2025 Series

Code references

• FireTracer/DKDNet-AMC
• radioML/dataset
• FireTracer/RML2025-Series

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• DKDNet: Dual Knowledge and Data-Driven Network for Cross-Domain Automatic Modulation Classification — Integrating modulation-specific signal prior knowledge with data-driven deep learning significantly enhances cross-domain automatic modulation classification.
• UF-AMA: A unified framework for cross-domain emotion recognition via adaptive multimodal alignment — UF-AMA unifies multimodal physiological signal fusion with adaptive alignment and multi-level domain adaptation for robust cross-domain emotion recognition.
• Cross-Domain Generalization in Optical Networks via Joint Contrastive and Classification Learning — Joint contrastive and classification learning enhances ML model generalization across diverse optical network domains.
• DSFNet: Learning Dual-Domain Spectral Operators for Multi-Modality Spatio-Temporal Forecasting in Urban Transportation Systems — DSFNet uses dual-domain spectral filtering to model complex spatio-temporal and cross-modality couplings in urban traffic.
• Domain-Incremental Remote Sensing Change Detection via Difference-Guided Adaptation and Frequency-Decoupled Distillation — DG-FDD mitigates catastrophic forgetting in remote sensing change detection by integrating difference-guided adaptation and frequency-decoupled distillation.
• Beyond KV-Cache: Test-Time Training & Invariant Latent Topologies for ICL — New methods enable LLMs to learn during inference and generalize across diverse domains by modifying internal weight structures or identifying invariant neural subspaces.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.