.png){kind=icon}
Return
Micro-Dose Augmented Wiener-Robust Automatic Modulation Classification
.png){kind=icon}
PRE
AI
DOI：10.1109/lwc.2026.3747850.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Automatic modulation classification (AMC) in millimeter-wave and terahertz bands is limited by non-stationary Wiener phase noise, which existing deep-learning solutions counter by estimating and removing a deterministic phase/frequency offset, or by generic augmentation, rather than at the feature level. We exploit an asymmetry: amplitude is rotation invariant and absolute differential phase has an innovation-variance distortion bound, whereas absolute phase and I/Q depend on the accumulated walk state and progressively lose coherence. PNV-RPNet couples amplitude and differential representations with online I/Q-level phase-noise injection. On RadioML2016.10b, with 284.30K deployed parameters, it attains 79.28% high-SNR accuracy, averaged over SNR > 0 dB, under σ = 15° phase noise, 9.53 pp above the strongest baseline under matched augmentation.
Keywords:
Automatic modulation classification
data augmentation
differential features
feature fusion
recurrence plot
Wiener phase noise
Journal
I
IF:
5.5
Papers:
1.3K
Citations:
0
Organization
Cited Papers
No cited papers available