Automated deep learning-based wide-band receiver
Automated deep learning-based wide-band receiver
复制标题
DOI：
10.1016/j.comnet.2022.109367
复制
发表时间：
2022-10-07
影响因子：
5.6
通讯作者：
Erdogmus, Deniz
中科院分区：
文献类型：
--
作者：
Azari, Bahar;Cheng, Hai;Erdogmus, Deniz
我们提出了一个模块化和成熟的物理层接收机设计的正交频分复用（OFDM）无线系统，利用深度神经网络（DNN）的进步。我们采用了详细的模块化设计，其中包括每个元素中适当和最大的领域知识，并使用通过模拟以及空中和模拟无线传输收集的数据对其进行训练。然后，我们将所有模块统一到一个端到端的自动化基于深度学习的宽带接收器中，并对其进行微调，以进一步提高其准确性。我们的组合流水线分析表现出上级的性能，通过显示误码率值高达8倍，如果相比，传统的无线通信方法。
We propose a modular and full-fledged physical layer receiver design for Orthogonal Frequency Division Multiplexing (OFDM) wireless systems leveraging the advances of deep neural networks (DNN). We adopt a detailed modular design that includes proper and utmost domain knowledge in each element and train it using data collected both via simulations as well as over-the-air and emulated wireless transmissions. We then unify all the modules into an end-to-end automated deep learning-based wide-band receiver and fine-tune it to further improve its accuracy. Our combined pipeline analysis exhibits superior performance by showing bit error rate values up to 8 times lower if compared to the traditional approaches for wireless communications.