AMCRN: Few-Shot Learning for Automatic Modulation Classification

Modulation signal recognition based on feed‐forward attention mechanism

“…There are two main reasons for this: on the one hand, some of the modulated signals are extremely similar in themselves and become more difficult to identify under the influence of noise, and on the other hand, when interference from noise is received between the signals, the difference between their characteristics decreases, leading to mutual misjudgment. As shown in Figure 4, we analyse the recognition of convolution kernels as (1,5), (1,6), (1,7), and (1, 8) under the RLADNN model. As the convolutional kernel gets smaller, its recognition rate as a whole shows a gradually decreasing trend, and the smaller the convolutional kernel, the smaller its perceptual field, the less relevant it is to the feature as a whole, and when a wash feature is encountered, it may not be able to represent its features when using a relatively small convolutional kernel.…”

“…Zhou et al. [7] proposed an automatic modulation classification relation network (AMCRN) structure to improve the performance with few samples, which uses a one‐dimensional convolution to effectively reduce the dimensionality, reduce redundancy, and remove some noise. The global average pooling (GAP) method used by Hao et al.…”

“…For example, Liu et al [6] proposed a network model based on CNN and LSTM (DCN-BiLSTM), using deep complex network (DCN) to extract the phase and amplitude features of the signal and bi-directional long-and short-term memory (BiLSTM) to extract the sequence features present on the signal. Zhou et al [7] proposed an automatic modulation classification relation network (AMCRN) structure to improve the performance with few samples, which uses a onedimensional convolution to effectively reduce the dimensionality, reduce redundancy, and remove some noise. The global average pooling (GAP) method used by Hao et al [8] to reduce the dimensionality of the highdimensional feature maps makes the model more simplified, requires a smaller number of parameters, and has a higher training speed.…”

Deep learning based automatic modulation recognition: Models, datasets, and challenges

“…For example, a DL-AMR model can be trained on offline datasets such as RML2016.10a, and fine-tuned using online data obtained in real-world by adopting the transfer learning strategy. If no training data is available only a few training samples are available for some signal classes, then zero-shot [11] and few-shot [94,37,91] learning are possible solutions for signal recognition. To improve the robustness of a model, methods that can withstand adversarial attacks need to be further investigated [58].…”

Automatic Modulation Classification using Relation Network with Denoising Autoencoder

“…The authors in [7], proposed a new FSL framework, Attention Relations Network (ARN), to utilize channel and spatial attention, thereby effectively extracting features from a support set. In addition, a novel network architecture called AMCRN, an AMC algorithm using a relational network, is proposed in [8]. The architecture reached a maximum classification accuracy of 93% and showed a performance improvement of 10 to 50% compared to the existing baseline schemes.…”