ALWNN Empowered Automatic Modulation Classification: Conquering Complexity and Scarce Sample Conditions

TL;DR

This work tackles AMC under stringent resource and data constraints by introducing ALWNN, a lightweight encoder that fuses depthwise separable convolutions with a learnable adaptive wavelet transform, achieving substantial reductions in FLOPS and MACC while maintaining competitive accuracy. Building on ALWNN, MALWNN extends to few-shot modulation classification via a prototype-network framework, enabling efficient learning with limited labeled samples and unseen modulations. Comprehensive evaluations on RadioML benchmarks and real-edge deployments (USRP and Raspberry Pi) show ALWNN’s strong efficiency-accuracy balance and MALWNN’s robust few-shot performance, highlighting practical viability for edge-enabled spectrum monitoring and cognitive radio systems. The proposed approach advances AMC by delivering near-state-of-the-art accuracy with orders-of-magnitude reductions in computational complexity, enabling real-time, low-power operation in constrained environments.

Abstract

In Automatic Modulation Classification (AMC), deep learning methods have shown remarkable performance, offering significant advantages over traditional approaches and demonstrating their vast potential. Nevertheless, notable drawbacks, particularly in their high demands for storage, computational resources, and large-scale labeled data, which limit their practical application in real-world scenarios. To tackle this issue, this paper innovatively proposes an automatic modulation classification model based on the Adaptive Lightweight Wavelet Neural Network (ALWNN) and the few-shot framework (MALWNN). The ALWNN model, by integrating the adaptive wavelet neural network and depth separable convolution, reduces the number of model parameters and computational complexity. The MALWNN framework, using ALWNN as an encoder and incorporating prototype network technology, decreases the model's dependence on the quantity of samples. Simulation results indicate that this model performs remarkably well on mainstream datasets. Moreover, in terms of Floating Point Operations Per Second (FLOPS) and Normalized Multiply - Accumulate Complexity (NMACC), ALWNN significantly reduces computational complexity compared to existing methods. This is further validated by real-world system tests on USRP and Raspberry Pi platforms. Experiments with MALWNN show its superior performance in few-shot learning scenarios compared to other algorithms.

Figures (9)

• Figure 1: Architecture of our proposed ALWNN.
• Figure 2: Architecture of our proposed few shot framework MALWNN.
• Figure 3: Classification accuracy on different SNR of ALWNN and other models on RML2016.10a.
• Figure 4: Classification accuracy on different SNR of ALWNN and other models on RML2016.10b.
• Figure 5: Classification accuracy on different SNR of ALWNN and other models on RML2018.01a.