Hybrid Constellation Modulation for Symbol-Level Precoding in RIS-Enhanced MU-MISO Systems

Yupeng Zheng, Yi Ma, Rahim TafazolliSPAWC 2025 · DOI

TL;DR

This paper introduces a novel "hybrid-constellation modulation" (HCM) scheme designed to improve symbol-level precoding (SLP) in Reconfigurable Intelligent Surface (RIS)-enhanced multi-user MIMO (MU-MISO) systems. By cleverly combining Quadrature Amplitude Modulation (QAM) and Amplitude Shift Keying (ASK) constellations, HCM expands the regions where constructive interference can be exploited, leading to significant performance gains over traditional QAM.

What Problem Does This Paper Address?

Modern wireless systems are increasingly relying on Reconfigurable Intelligent Surfaces (RIS) to boost signal quality and spectral efficiency in multi-user, multi-input, single-output (MU-MISO) scenarios. A key technique for maximizing performance in these systems is Symbol-Level Precoding (SLP), which goes beyond traditional methods by actively exploiting constructive interference (CI) between users. However, applying SLP in RIS-enhanced systems faces two major hurdles:

  1. Scalability of Optimization: Existing methods for optimizing both the RIS configuration and SLP require exhaustively checking all possible transmit symbol combinations. This becomes computationally infeasible as the modulation order (how much data each symbol carries) and the number of users increase.
  2. Limitations of QAM: Conventional QAM constellations have very narrow regions where CI can be effectively exploited. This limits the potential gains from SLP, especially when compared to simpler modulation schemes like Phase Shift Keying (PSK).

This paper aims to overcome these limitations by proposing a new modulation scheme and a more efficient optimization approach.

Methods

The core innovation of this paper is the Hybrid-Constellation Modulation (HCM). HCM combines two sub-constellations (SCs):

  • QAM SCs: Standard QAM symbols.
  • ASK SCs: Amplitude Shift Keying symbols.

The key idea is that ASK symbols are primarily defined by their real part. This leaves the imaginary part "free" to be manipulated for SLP. By superposing ASK symbols onto QAM symbols, HCM effectively widens the "constructive interference" (CI) regions compared to pure QAM. The paper categorizes HCM symbols into three types based on their fixed and variable parts for SLP optimization.

To address the scalability issue, the authors propose a two-stage optimization method:

  1. RIS Phase Shift Optimization: Instead of brute-forcing all RIS configurations, they use a successive refinement method, which is computationally efficient and achieves near-optimal results.
  2. SLP Optimization: They reformulate the SLP problem to maximize the Signal-to-Noise Ratio (SNR) of the received "fixed" symbol parts while ensuring the "variable" parts stay within the newly expanded CI regions. This is achieved through a series of optimization steps, including a sub-optimal heuristic algorithm that significantly reduces computational complexity compared to exhaustive search.

Datasets

Not discussed in the paper. The evaluation is based on simulations using a defined system model and channel characteristics.

Key Insights and Ideas

  • HCM's Clever CI Expansion: The genius of HCM lies in leveraging the inherent structure of ASK. By allowing the imaginary part of ASK symbols to be manipulated, it creates larger "sweet spots" for SLP to exploit constructive interference, a capability QAM lacks.
  • Two-Stage Optimization is Key: The paper demonstrates that decoupling the RIS phase shift optimization from the SLP optimization is crucial for scalability. This sequential approach avoids the combinatorial explosion of checking all symbol combinations simultaneously.
  • Sub-Optimal SLP Algorithm: The proposed sub-optimal SLP algorithm offers a practical trade-off between performance and computational complexity, making SLP viable for higher-order modulations and more users.
  • Robustness to RIS Resolution: Interestingly, the HCM scheme shows similar performance gains even with lower RIS phase shift resolutions (e.g., 1-bit vs. 2-bit), suggesting its resilience to practical hardware limitations.

Key Results

The paper presents simulation results showing the Symbol Error Rate (SER) performance of their proposed HCM-SLP scheme against traditional QAM-SLP and QAM-ZF (Zero-Forcing precoding).

  • Significant SER Gains:
    • For 16-QAM, HCM-SLP achieves up to 1.5 dB gain over QAM-SLP.
  • For 64-QAM, HCM-SLP achieves up to 1 dB gain over QAM-SLP.
  • RIS Design Impact: The optimized RIS phase shifts provide a substantial performance improvement (over 8 dB gain) even before SLP is applied, highlighting the importance of RIS configuration.
  • Scalability Improvement: The two-stage optimization method significantly improves scalability compared to prior state-of-the-art joint optimization algorithms.

Limitations and Future Work

  • Computational Complexity: While the proposed two-stage method is more scalable, the sub-optimal SLP algorithm for HCM still has a complexity that scales exponentially with the number of users. The authors acknowledge this and suggest that further research could focus on reducing this complexity.
  • Practical Implementation: The paper assumes perfect Channel State Information (CSI) at the BS, which is an idealization. Future work could explore the performance under imperfect CSI.
  • HCM Design Space: The paper focuses on specific HCM configurations (16-HCM and 64-HCM). Exploring other combinations of ASK and QAM sub-constellations or different ways to superpose them could yield further improvements.
  • Real-World Deployment: Investigating the performance of HCM-SLP in more realistic channel models and considering hardware impairments would be valuable for practical deployment.

Summary updated

BibTeX
@inproceedings{zheng2025hybrid,
  author = {Yupeng Zheng and Yi Ma and Rahim Tafazolli},
  title = {{Hybrid Constellation Modulation for Symbol-Level Precoding in RIS-Enhanced MU-MISO Systems}},
  booktitle = {SPAWC},
  year = {2025},
  doi = {10.1109/spawc66079.2025.11143452},
}
Figure