31 March 2023 in IEEE Internet of Things Journal
Traffic classification is indispensable for the Internet of Things (IoT) in intrusion detection and resource management. Deep-learning (DL)-based strategies are the key tools for traffic classification due to high accuracy but still have some challenges: 1) it is hard to deploy complex DL models on resource-constraint IoT devices and 2) performance is limited because of the ignorance of the similarity between IoT traffic. To address these issues, we propose lightweight but accurate models for traffic classification. First, we adopt a network-in-network basic model to reduce model size. Second, the basic model is trained with self-distilled response, feature map, and similarity among traffic types to enable its identification accuracy. Next, redundant filters are removed from the basic model to achieve compressed architectures. Then, a teacher model updating scheme with knowledge distillation is proposed to train compressed models without compromising performance. Experimental results demonstrate that compared to the state-of-the-art deep packet model, the compressed model can achieve the highest accuracy, deal with imbalanced traffic, and reduce nearly 99% of computation overhead in two encrypted traffic classification scenarios, thus, emphasizing its efficiency.
10 February 2023 in IEEE Communications Letters
Channel autoencoders adopt neural networks to represent and optimize previous block-driven communication systems from an end-to-end perspective. The existing deep fully connected autoencoder needs retraining when the input length of bit sequences changes since it can only handle fixed-length data. Convolutional neural network (CNN)-based autoencoder can accept arbitrary lengths and is widely adopted. But it has limitations in 1) error floor in high signal-to-noise ratio (SNR) regions, and 2) poor performance against interference. Therefore, we first adopt a residual attention module to enhance the representation ability of the autoencoder, where channel and spatial attention focus on finding fine-grained clues between signals and noise. An adaptive data-flow merging scheme for demapper is further proposed to satisfy dynamic environments. Simulation results show that the proposed method can achieve great gains in coded and uncoded transmission scenarios. Our method is robust and generalizable against error floor and burst interference, as compared to a conventional system.
1 January 2026 in IEEE Wireless Communications Letters