Recent advancements in machine learning for communications show that channel autoencoders could revolutionize conventional communication systems through end-to-end optimization. For channel autoencoders to reliably transmit over the air, a scheme to enable adaptive use of resources is needed. Thus, we propose ConvAE-Advanced, an improved channel autoencoder structure that can adaptively transmit across multiple timeslots. ConvAE-Advanced utilizes an unexploited input dimension in ConvAE by the use of the resource-aware residual block and whole resource power normalization. This enabled ConvAE-Advanced to adaptively transmit information according to channel conditions. Simulations for a 2-by-2 multiple-input multiple-output system under the WINNER2 A1 scenario shows that ConvAE-Advanced outperforms ConvAE across all SNR ranges. Most importantly, ConvAE-Advanced can achieve a better BER and achievable rate performance without additional wireless resource usage compared to ConvAE.

中文翻译：

---

ConvAE-Advanced：跨多个时隙的自适应传输以实现容错操作

通信机器学习的最新进展表明，通道自动编码器可以通过端到端优化彻底改变传统通信系统。为了使信道自动编码器可靠地通过空中传输，需要一种能够自适应使用资源的方案。因此，我们提出了 ConvAE-Advanced，一种改进的信道自动编码器结构，可以自适应地跨多个时隙传输。ConvAE-Advanced 通过使用资源感知残差块和整个资源功率归一化，在 ConvAE 中利用未开发的输入维度。这使得 ConvAE-Advanced 能够根据信道条件自适应地传输信息。在 WINNER2 A1 场景下对 2×2 多输入多输出系统的仿真表明，ConvAE-Advanced 在所有 SNR 范围内的性能都优于 ConvAE。最重要的是，