Digital Twin Aided Compressive Sensing: Enabling Site-Specific MIMO Hybrid Precoding

TL;DR

This work tackles channel estimation for site-specific MIMO systems with hybrid precoding by leveraging a site-specific digital twin to generate synthetic data for end-to-end deep learning of compressive sensing measurement vectors and RF codebook prediction. By training on synthetic data and optionally refining with a small amount of real data, the approach enables accurate RF precoder/combiner selection while drastically reducing the data collection burden. The authors demonstrate that models trained on digital twin data transfer well to real deployments, and that model refinement can match real-data performance with substantially fewer real-world samples. Overall, the digital twin framework offers a practical pathway to deploy DL-based CS for hybrid precoding in realistic environments with limited data collection overhead.

Abstract

Compressive sensing is a promising solution for the channel estimation in multiple-input multiple-output (MIMO) systems with large antenna arrays and constrained hardware. Utilizing site-specific channel data from real-world systems, deep learning can be employed to learn the compressive sensing measurement vectors with minimum redundancy, thereby focusing sensing power on promising spatial directions of the channel. Collecting real-world channel data, however, is challenging due to the high overhead resulting from the large number of antennas and hardware constraints. In this paper, we propose leveraging a site-specific digital twin to generate synthetic channel data, which shares a similar distribution with real-world data. The synthetic data is then used to train the deep learning models for learning measurement vectors and hybrid precoder/combiner design in an end-to-end manner. We further propose a model refinement approach to fine-tune the model pre-trained on the digital twin data with a small amount of real-world data. The evaluation results show that, by training the model on the digital twin data, the learned measurement vectors can be efficiently adapted to the environment geometry, leading to high performance of hybrid precoding for real-world deployments. Moreover, the model refinement approach can enable the digital twin aided model to achieve comparable performance to the model trained on the real-world dataset with a significantly reduced amount of real-world data.

Figures (5)

• Figure 1: This diagram illustrates the adopted system model and the key idea of the proposed digital twin aided compressive sensing for hybrid precoding. The digital twin generates synthetic channel data, which shares a similar distribution as the real-world data. The channel encoder mimics the channel sensing process and learns the transmit and receive measurement vectors that capture the promising spatial directions of the channel. Based on the received channnel measurements, the RF precoder/combiner predictor predicts the codebook indices for the RF precoder/combiner. The RF precoder/combiner predictor, for instance, can be implemented in the baseband processor at the receiver.
• Figure 2: The figure shows the adopted target scenario, which is built based on a section of downtown Boston. The base station is placed along a vertical street, while the foliage is represented by the dark green objects in the layout. The user grid is highlighted by the red box.
• Figure 3: This figure presents the prediction accuracy of the RF precoder with different numbers of measurement vectors. We consider the number of measurement vectors to be set to $\{1,2,4,8,16,32\}$.
• Figure 4: This figure shows the beam patterns of the measurement vectors learned by the channel encoder. The number of measurement vectors is set to $\{1,8\}$. Fig. \ref{['fig:beam_real']} shows the learned beam patterns by the model trained on the target data, while Fig. \ref{['fig:beam_DT']} presents the learned beam patterns by the model trained on the digital twin data.
• Figure 5: This figure presents the prediction accuracy of the RF precoder with different numbers of real-world data points. The model is pretrained on $10240$ synthetic data points and then fine-tuned on real-world data.