Integrated Sensing and Communication Signal Processing Based on Compressed Sensing Over Unlicensed Spectrum Bands

Haotian Liu, Zhiqing Wei, Fengyun Li, Yuewei Lin, Hanyang Qu, Huici Wu, Zhiyong FengIEEE Trans. Cogn. Commun. Netw 2024 · DOI

TL;DR

This paper proposes a novel Integrated Sensing and Communication (ISAC) signal processing algorithm that leverages Compressed Sensing (CS) and machine learning to achieve high-precision sensing over dynamic, non-continuous unlicensed spectrum bands. The method significantly improves range and velocity estimation accuracy and reduces sidelobes, addressing key challenges in future 6G systems.

What Problem Does This Paper Address?

Future mobile communication systems, particularly those aiming for 6G, are increasingly integrating sensing capabilities alongside communication. This integration, known as ISAC, promises enhanced spectrum and hardware efficiency. However, a significant challenge arises when these systems operate in unlicensed spectrum bands, which are often fragmented and dynamically changing. Conventional sensing algorithms struggle in such non-continuous environments, leading to degraded performance, including reduced signal-to-noise ratio (SNR) and increased sidelobes in power spectrum estimations. This paper tackles this problem by developing a robust ISAC signal processing technique specifically designed for these challenging unlicensed spectrum conditions.

Methods

The core of the proposed solution is the Joint CS and Machine Learning ISAC Signal Processing Algorithm (JCMSA). It combines several key techniques:

  • Compressed Sensing (CS): Instead of traditional Fourier Transform-based methods, JCMSA frames the sensing problem as a sparse signal reconstruction task. This allows for accurate estimation even with undersampled data, which is crucial for dynamic spectrum environments.
  • Resource Block Group (RBG) Information: The algorithm leverages 5G NR's RBG configuration information to identify available spectrum subcarriers. This information is used to create a "spectrum occupancy sequence" that helps in processing the channel information matrix.
  • Fast Iterative Shrinkage-Thresholding Algorithm (FISTA): This algorithm is employed to efficiently solve the convex optimization problem arising from the CS reconstruction, enabling accurate estimation of range and velocity.
  • K-fold Cross-Validation (KCV): Machine learning, specifically KCV, is used to find the optimal regularization parameter ($\lambda$) for the FISTA algorithm. This ensures optimal reconstruction performance and faster convergence.
  • Dynamic Spectrum Handling: The algorithm processes the channel information matrix by zeroing out unavailable subcarriers based on the spectrum occupancy sequence, making it adaptable to time-varying spectrum conditions.

Datasets

The paper does not explicitly mention the use of pre-existing datasets for training. Instead, it relies on simulated data generated based on the described ISAC signal model and the parameters provided in Table III. The evaluation is performed using these simulated datasets under different Signal-to-Noise Ratio (SNR) conditions and two distinct spectrum occupancy scenarios.

Key Insights and Ideas

  • CS is a natural fit for dynamic spectrum sensing: The paper demonstrates that CS's ability to reconstruct sparse signals from undersampled data is highly effective in non-continuous spectrum bands, outperforming traditional methods.
  • Leveraging 5G NR information is practical: Utilizing RBG configuration information from existing 5G standards provides a practical way to adapt to dynamic spectrum availability without requiring entirely new signal designs.
  • Machine learning for parameter optimization is crucial: KCV is shown to be an effective method for finding the optimal regularization parameter for CS, leading to improved performance and robustness.
  • Sidelobe reduction is achievable even with spectrum holes: The proposed JCMSA algorithm successfully mitigates sidelobe deterioration, a common issue in non-continuous spectrum sensing, by employing CS reconstruction.

Key Results

The JCMSA algorithm demonstrates significant improvements over conventional 2D FFT and the improved algorithm from [19]:

  • Range Estimation Accuracy: Achieved a maximum improvement of 54.66% over the improved 2D FFT algorithm and 84.36% over the conventional 2D FFT algorithm.
  • Velocity Estimation Accuracy: Achieved a maximum improvement of 41.54% over the improved 2D FFT algorithm and 97.09% over the conventional 2D FFT algorithm.
  • Sidelobe Performance: The proposed algorithm consistently maintains zero sidelobes, whereas conventional methods suffer from significant sidelobe deterioration.
  • Anti-Noise Performance: JCMSA shows comparable or better anti-noise performance than the improved algorithm in [19], especially at low SNRs, and significantly outperforms the conventional 2D FFT.

Limitations and Future Work

  • Computational Complexity: The paper acknowledges that JCMSA has the highest computational complexity among the compared methods due to the iterative nature of FISTA. This is presented as a trade-off for achieving superior sensing performance.
  • Future Work: While not explicitly detailed as future work, the paper's focus on unlicensed spectrum bands and the demonstrated performance improvements suggest that further research could explore more complex dynamic spectrum scenarios, multi-target detection, and integration with advanced communication techniques for even more comprehensive ISAC systems. The paper also implicitly suggests that optimizing the computational complexity of JCMSA would be a valuable direction.

Summary updated

BibTeX
@article{liu2024integrated,
  author = {Haotian Liu and Zhiqing Wei and Fengyun Li and Yuewei Lin and Hanyang Qu and Huici Wu and Zhiyong Feng},
  title = {{Integrated Sensing and Communication Signal Processing Based on Compressed Sensing Over Unlicensed Spectrum Bands}},
  journal = {IEEE Trans. Cogn. Commun. Netw.},
  year = {2024},
  doi = {10.1109/tccn.2024.3391307},
}

Loading related content...

Figure