Distributed Joint Multi-cell Optimization of IRS Parameters with Linear Precoders

Reinhard Wiesmayr, Michael L. Honig, Michael Joham, Wolfgang Utschick

Introduction

As wireless communication moves toward 6G, Intelligent Reflecting Surfaces (IRS) have emerged as a transformative technology. By adjusting the phase shifts of passive reflecting elements, an IRS can "program" the wireless environment to enhance signal strength or suppress interference. However, the practical deployment of IRS in multi-cell networks is hindered by the immense overhead of acquiring accurate Channel State Information (CSI). In a multi-cell scenario, base stations (BTS) must account for both local channels and cross-channels to neighboring users to manage interference, a task that typically requires massive training sequences and tight inter-cell coordination.

This paper proposes a distributed framework that bypasses explicit channel estimation entirely. Instead, it optimizes IRS phase shifts and beamformers directly from received training signals, significantly reducing overhead while maintaining high performance in interference-heavy environments.

System Model and the CSI Challenge

The researchers consider a multi-cell network where each cell is equipped with a BTS and an IRS. The primary challenge in such a setup is the "CSI nightmare": to set the IRS parameters optimally, the system traditionally needs to know every channel link, including those causing inter-cell interference.

Multi-cell IRS-assisted network model illustrating the interaction between base stations, reflecting surfaces, and users across cell boundaries.

As shown in the system model above, users in one cell (e.g., UE $k$) are subject to interference from neighboring base stations (BTS2) and reflections from neighboring IRS units (IRS2). Estimating all these individual links is often computationally and logistically prohibitive.

The Direct Optimization Framework

To solve this, the authors introduce a "Sample $\rightarrow$ Optimize" approach. This framework consists of four key components:

  1. Synchronized Training: All users across the network send pilot signals simultaneously. During this period, the IRS units cycle through a predefined set of orthogonal phase-shift patterns.
  2. Direct SINR Estimation: Rather than decomposing the received signal into individual channel components, the BTS estimates the Signal-to-Interference-plus-Noise Ratio (SINR) directly from the sample means of the received pilots. This implicitly captures the aggregate effect of all interference sources.
  3. Least Squares (LS) Filtering: For multi-antenna (MIMO) systems, the BTS uses a bi-directional training scheme to optimize receive filters. These filters are then used as downlink beamformers, leveraging channel reciprocity.
  4. Distributed Execution: Each cell operates independently. A BTS only needs to know the training sequences of its own users, eliminating the need for high-speed backhaul sharing of raw CSI data between cells.

Performance and Interference Mitigation

The proposed direct optimization method demonstrates remarkable robustness against inter-cell interference. In traditional schemes that ignore neighboring cells (Partial CSI), performance often saturates as the signal power increases because the interference power grows at the same rate.

Downlink sum rate performance versus SNR, comparing the proposed direct optimization methods against perfect CSI and partial estimation benchmarks.

As illustrated in the results above, the "Direct Decentralized" and "Direct Centralized" methods track the "Perfect CSI" benchmark closely. Crucially, as the noise floor drops (moving to the right on the x-axis), these methods continue to scale their sum-rate, whereas the "Partial Chan. Est." method hits a performance ceiling. This confirms that the direct optimization approach successfully "steers" the IRS to suppress interferers without ever explicitly measuring their channels.

Conclusion

The distributed direct optimization framework offers a scalable and practical path for IRS integration in 6G networks. By shifting from explicit channel estimation to direct signal-based optimization, the method reduces coordination overhead and provides inherent resilience to inter-cell interference. Future work will explore extending this framework to dynamic power control and multi-stream data transmission to further push the boundaries of spectral efficiency.

Summary updated

BibTeX
@inproceedings{wiesmayr2022distributed,
  author = {Reinhard Wiesmayr and Michael L. Honig and Michael Joham and Wolfgang Utschick},
  title = {{Distributed Joint Multi-cell Optimization of IRS Parameters with Linear Precoders}},
  booktitle = {ICC},
  year = {2022},
  doi = {10.1109/icc45855.2022.9839115},
}
Figure