SPATE-NetA
Open Access DepositedA Predictive Model for Automatic Modulation Classification in Non-Cooperative Wireless Communication Systems
Downloadable Content
• Multiscale convolutional neural network (CNN).• Squeeze-and-excitation channel-wise attention. • Bidirectional gated recurrent units (BiGRU). • Multi-head cross attention. • Novel feature set of in-phase/quadrature (I/Q), amplitude/phase (A/P), fractional Fourier transform (FrFT), and stationary wavelet transform (SWT). The SigMod-45 dataset was introduced, which contains signals based on the characteristics found in non-cooperative environments. SigMod-45 includes 45 modulations types across a signal-to-noise range of +30 dB to -8 dB. In addition to SigMod-45, two supplementary datasets were employed in the testing phase to substantiate the findings
RadioML2016.10B and HisarMod2019.1. Ablation testing with single-stream variants of SPATE-NetA revealed superior accuracy performance of the quad feature set across all datasets by a minimum and maximum macro-average of 2.92% to 22.17%, respectively. Moreover, SPATE-NetA surpassed 12 benchmark DL models in the same evaluation by 5.63% to 29.36%. The performance improvement margins varied between datasets, revealing a less expansive distance when dataset complexity decreased. This study suggests that the use of manual features can lead to increase AMC performance in non-cooperative wireless communications. The postulated advantages are more prevalent with an escalation in classification complexity.
Automatic modulation classification (AMC) plays an instrumental role in non-cooperative wireless communication systems, wherein a receiver autonomously detects signal modulation types without prior information of its source. Historically utilized for military functions like signal jamming and interception, AMC's applicability has broadened to civilian domains, including spectrum management, dynamic spectrum access, and cognitive radios. Traditionally, likelihood-based and distribution test methods defined AMC research and implementations. Interest in deep learning (DL) methods has surged due to their ability to identify nuanced patterns in signal data without manual feature extraction. Non-cooperative wireless networks experience significant challenges in regards to channel impairments, a class of phenomena that obscure signal characteristics. Under these situations, the AMC model must overcome undesirable radio frequency (RF) spectrum conditions. This study examined the potential performance improvement of a DL model through manual feature extraction within the context of a challenging non-cooperative scenario. Hence, the implementation of SPATE-NetA (SPAtioTemporal Enhanced Network with Attention). SPATE-NetA is a multi-stream DL architecture that several defining SOTA concepts
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.