| 摘要: |
| 针对大部分基于深度学习（Deep Learning，DL）的信道估计算法估计高维信号时出现的训练开销过大、泛化能力差等问题，提出了一种不需要训练的基于深度学习的高维信号信道估计算法，即UTCENet（Untrained Channel Estimation Network）。在UTCENet中，信道信息上的复杂分布转换为模型参数上的简单分布，即通过神经网络参数化来获得隐式先验知识并将其应用于信道估计。虽然该算法不需要任何训练，但保证了估计的性能，其原因在于专门设计的网络模型可以有效利用时频网格中元素的相关性。仿真结果表明，与传统方法以及现有的深度学习方法相比，所提出的算法在归一化均方误差和误码率方面性能提升明显。 |
| 关键词: 多输入多输出正交频分复用（MIMO OFDM） 信道估计 高维信号 深度学习 |
| DOI： |
|
| 基金项目:国家自然科学基金青年科学基金项目(61701433)；云南省科技厅面上项目(2018FB099) |
|
| A channel estimation algorithm for high-dimensional signals based on deep learning |
| QIAN Rongrong,LYU Xiaoming,REN Wenping |
| (School of Information Science and Engineering,Yunnan University) |
| Abstract: |
| For the problems of excessive training overhead and poor generalization ability of most channel estimation algorithms based on deep learning(DL) when estimating high dimensional signals,this paper proposes a high dimensional signal channel estimation algorithm based on deep learning that does not require training,called Untrained Channel Estimation Network(UTCENet).In UTCENet,the complex distribution on the channel information is converted into the simple distribution on the model parameters,that is,implicit prior knowledge is obtained by neural network parameterization and applied to channel estimation.Although the algorithm does not require any training,it guarantees the estimated performance.The reason is that the specially designed network model can effectively utilize the correlation of the elements in the time frequency grid.Simulation results show that compared with traditional methods and existing deep learning methods,the proposed algorithm has significantly improved the performance in terms of normalized mean square error(NMSE) and bit error rate(BER). |
| Key words: multiple input multiple output orthogonal frequency division multiplexing（MIMO OFDM） channel estimation high dimensional signal deep learning |
