On the Design of Channel Coding Autoencoders With Arbitrary Rates for ISI Channels
On the Design of Channel Coding Autoencoders With Arbitrary Rates for ISI Channels
复制标题
ISI通道任意速率通道编码自编码器的设计
DOI：
--
复制
发表时间：
2022
影响因子：
6.3
通讯作者：
M. Coates
中科院分区：
文献类型：
--
作者：
Yitian Zhang;Huihui Wu;M. Coates
这封信提出了一种基于自动编码器的信道编码方案，在存在符号间干扰（ISI）和加性白色高斯噪声（AWGN）的情况下，支持任意编码速率。所提出的自动编码器的发射器和接收器都采用双向门控递归单元（Bi-GRU）层。在发射机的末端和接收机的开始处应用额外的额外密集层，分别用作可学习的穿孔和解穿孔模块。通过调整额外密集层的输出维度可以实现不同的码率。实验结果表明，建议的自动编码器显着优于传统的卷积码在ISI信道，多个码率。建议的自编码器也优于LDPC码在低信噪比（SNR）制度。神经编码仍然需要改进以在高SNR区域中具有竞争力。
This letter presents an autoencoder-based channel coding scheme in the presence of inter-symbol interference (ISI) and additive white Gaussian noise (AWGN), supporting arbitrary coding rates. Both the transmitter and receiver of the proposed autoencoder employ bi-directional gated recurrent unit (Bi-GRU) layers. Additional extra dense layers are applied at the end of the transmitter and at the beginning of the receiver, serving as learnable puncture and depuncture modules, respectively. Different code rates can be achieved by adjusting the output dimension of the extra dense layers. Experimental results demonstrate that the proposed autoencoder significantly outperforms conventional convolutional codes over ISI channels, for multiple code rates. The proposed autoencoder also outperforms LDPC codes in the low signal-to-noise ratio (SNR) regime. The neural codes still require improvement to be competitive in the high SNR regime.