.png){kind=icon}
.png){kind=icon}
Return
Improved Distribution Difference Driven Diffusion Generative Method for AMOSR
.png){kind=icon}
PRE
AI
DOI：10.1109/TCCN.2025.3613534.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Automatic modulation open set recognition (AMOSR) is crucial for enhancing reliable recognition of dynamic wireless communication systems. It enables accurate identification of known modulation types (KM) while effectively rejecting unknown modulation types (UKM). This capability has drawn increasing attention in military and civilian domains. However, existing research remains limited and struggles to strike a balance between the empirical and open space risks, thereby hindering AMOSR performance. To address this challenge, we propose an improved distribution difference driven diffusion generative method (I4D), designed to effectively reject UKM while maintaining high recognition accuracy for KM. Specifically, we propose a knowledge conditioned diffusion model (KCDM) that learns the structural and intrinsic information of signals. KCDM incrementally aligns the distributions between the original and generated signals for KM, and progressively amplifies their differences for UKM. Additionally, we construct an improved distribution space that balances class-common and class-specific information by employing a progressive feature calibration strategy to hierarchically constrain the features. I4D minimizes the empirical risk and establishes clear decision boundaries for UKM. Extensive experiments conducted on the public datasets RML2016.10A, RML2016.10B, RML2018.01A and real-world signal demonstrate that our proposed I4D outperforms existing OSR and AMOSR approaches, establishing its superiority. Our code is available at: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/thy-wq/AMOSR-I4D-main</uri>
Keywords:
Automatic modulation open set recognition
diffusion model
improved distribution space
progressive feature calibration
Journal
I
IF:
7
Papers:
1.5K
Citations:
5.5K