Uplink Assisted Joint Channel Estimation and CSI Feedback: An Approach Based on Deep Joint Source-Channel Coding

TL;DR

An uplink assisted joint CE and and CSI feedback approach via deep learning for downlink CSI acquisition is proposed, which mitigates performance degradation caused by distribution bias across separately trained modules in traditional modular communication frameworks.

Abstract

In frequency division duplex (FDD) multiple-input multiple-output (MIMO) wireless communication systems, the acquisition of downlink channel state information (CSI) is essential for maximizing spatial resource utilization and improving system spectral efficiency. The separate design of modules in AI-based CSI feedback architectures under traditional modular communication frameworks, including channel estimation (CE), CSI compression and feedback, leads to sub-optimal performance. In this paper, we propose an uplink assisted joint CE and and CSI feedback approach via deep learning for downlink CSI acquisition, which mitigates performance degradation caused by distribution bias across separately trained modules in traditional modular communication frameworks. The proposed network adopts a deep joint source-channel coding (DJSCC) architecture to mitigate the cliff effect encountered in the conventional separate source-channel coding. Furthermore, we exploit the uplink CSI as auxiliary information to enhance CSI reconstruction accuracy by leveraging the partial reciprocity between the uplink and downlink channels in FDD systems, without introducing additional overhead. The effectiveness of uplink CSI as assisted information and the necessity of an end-toend multi-module joint training architecture is validated through comprehensive ablation and scalability experiments.

Figures (11)

• Figure 1: The construction of AI-based CE and CSI feedback: \ref{['SEFNet']} separate CE and CSI feedback; \ref{['JEFNet_SSCC']} joint CE and CSI feedback with SSCC; \ref{['JEFNet_JSCC']} joint CE and CSI feedback with DJSCC.
• Figure 2: The network architecture of UJEFNet is illustrated with detailed figures of key modules. For clarity, some operations that do not involve trainable parameters are omitted from the figure, and parameter settings for certain network layers are summarized in Table \ref{['net_config']}. For convolutional and transposed convolutional layers, respectively denoted as "$\mathrm{Conv}$" and "$\mathrm{TransConv}$", we use the notation "$F|(W_1, W_2)|(S_1, S_2)$", where $F$ represents the number of filters, $W_1 \times W_2$ denotes the window size, and $S_1 \times S_2$ indicates the step size. For fully connected layers, denoted as "$\mathrm{Dense}$", Table \ref{['net_config']} specifies the number of neurons.
• Figure 3: Architecture of the AI-based uplink CE network framework. "SeLU" in the figure refers to the Scaled Exponential Linear Unit (SeLU) activation function.
• Figure 4: The confidence intervals for Pearson correlation coefficients with $r_{\mathrm{hg}}$, $r_{\mathrm{hh}}$, $r_{\mathrm{gg}}$.
• Figure 5: We outline the training and testing strategies for UJEFNet under non-ideal uplink CE. The networks at the transmitter and receiver are simplified using "Encoder" and "Decoder" representations, respectively. To indicate parameter updates during training, we use the fire flag, signifying that the corresponding module parameters are trainable at that stage. Conversely, the snowflake flag is used to denote modules with frozen parameters.