基于深度学习的干扰信道功率控制算法

Deep Learning Algorithm for Interference Channel Power Control

多用户干扰信道功率控制是无线通信领域的基本问题之一,不仅要求较高的频谱效率,而且还需要较低的计算复杂度。笔者基于注意力机制和卷积神经网络提出AT-UNet深度学习模型,通过在真实信道中测试,发现其性能超越了现有的深度学习方法,媲美经典数学优化算法。同时,笔者进一步将迁移学习运用在模型训练过程中,一定程度上减少了训练集和训练时间,降低了深度学习模型的训练成本。

Multi-user interference channel power control is one of the fundamental problems in the field of wireless communication, which not only requires high spectral efficiency, but also requires low computational complexity. The author proposes the AT-UNet deep learning model based on the attention mechanism and convolutional neural network. Tested in the real channel, its performance surpasses the existing deep learning methods and is comparable to the classical mathematical optimization algorithm. At the same time, the author further Applies transfer learning in the model training process, which reduces the training set and training time to a certain extent, and reduces the training cost of the deep learning model.