计算机科学
正交频分复用
无线
探测理论
电子工程
稳健性(进化)
信号处理
通信系统
计算机网络
实时计算
变压器
无线传感器网络
信噪比(成像)
信号(编程语言)
嵌入式系统
作者
Fengyang Jiang,Huahua Wang
标识
DOI:10.1109/jiot.2026.3719141
摘要
Reliable signal detection is essential in OFDM-based wireless communication systems, especially in practical deployment scenarios where receivers must operate under noisy channels and limited computational resources. However, conventional detectors often suffer from performance degradation under noise variation, channel mismatch, and complex multipath propagation, while many existing deep learning-based receivers remain computationally heavy. To address these issues, this paper proposes NDL-Former, a noise-aware lightweight Transformer for robust OFDM signal detection in wireless communication systems. The proposed model integrates three key components: noise-aware feature modulation for adaptive robustness, normalized linear attention for reduced computational complexity, and a dynamic early-exit mechanism for efficient inference. Experiments on DeepMIMO and the 3GPP TDL-E channel model show that NDL-Former consistently outperforms classical detectors and recent neural receivers. Additional sensitivity analyses under SNR-estimation errors and imperfect CSI further examine deployment-related robustness. In particular, it achieves up to 18% BER reduction under noisy conditions while reducing model parameters by about 35% compared with existing Transformer-based detectors. These results demonstrate that NDL-Former provides an effective and efficient solution for intelligent wireless receivers.
科研通智能强力驱动
Strongly Powered by AbleSci AI