Recently, deep neural network (DNN) has been widely adopted in the design of intelligent communication systems thanks to its strong learning ability and low testing complexity. However, most current offline DNN-based methods still suffer from unsatisfactory performance, limited generalization ability, and poor interpretability. In this article, we propose an online DNN-based approach to solve general optimization problems in wireless communications, where a dedicated DNN is trained for each data sample. By treating the optimization variables and the objective function as network parameters and loss function, respectively, the optimization problem can be solved equivalently through network training. Thanks to the online optimization nature and meaningful network parameters, the proposed approach owns strong generalization ability and interpretability, while its superior performance is demonstrated through a practical example of joint beamforming in intelligent reflecting surface (IRS)-aided multi-user multiple-input multiple-output (MIMO) systems. Simulation results show that the proposed online DNN outperforms conventional offline DNN and state-of-the-art iterative optimization algorithm, but with low complexity.
翻译:最近,由于学习能力强和测试复杂性低,在智能通信系统的设计中广泛采用了深神经网络(DNN),但目前大多数基于DNN的离线方法仍然表现不佳,普遍化能力有限,解释能力差。在本篇文章中,我们建议采用基于DNN的在线方法,解决无线通信中的一般优化问题,为每个数据样本专门培训一个DNN。通过分别将优化变量和目标功能作为网络参数和损失函数处理,优化问题可以通过网络培训等同地得到解决。由于在线优化性质和有意义的网络参数,拟议方法拥有很强的通用能力和可解释性,而其优异性则通过在智能反映表面(IRS)的多用户辅助多投入多输出(MIMO)系统中联合组合的实用实例来证明。模拟结果显示,拟议的DNNN将常规离线 DNN和最先进的迭相优化算法(但复杂性低)。