Practical RIS Gain without the Pain via Randomization and Opportunistic Scheduling in 5G~NR Wireless Systems: Theory and Experiments

Technical Summary: Practical RIS Gain without the Pain via Randomization and Opportunistic Scheduling in 5G NR Wireless Systems

Problem Statement
Reconfigurable Intelligent Surfaces (RISs) offer a promising avenue for enhancing signal strength, coverage, and throughput in beyond-5G systems by controlling wireless propagation environments. However, conventional RIS-assisted systems typically rely on explicit Channel State Information (CSI) estimation followed by complex optimization of RIS phase configurations to maximize Signal-to-Noise Ratio (SNR). In practical multi-user 5G New Radio (NR) environments, this approach incurs significant overhead in channel estimation, computational complexity, and control signaling. Furthermore, integrating optimization-based RIS designs into the 5G NR protocol stack—specifically accounting for CSI reporting, Modulation and Coding Scheme (MCS) selection, and Medium Access Control (MAC) scheduling—can substantially diminish the theoretical gains of such designs. There is a critical need for RIS strategies that exploit reflectivity properties without requiring explicit CSI acquisition or real-time phase optimization.

Methodology
This work proposes and validates a low-complexity approach where RIS phase states are randomly switched among predefined configurations, independent of instantaneous CSI. The resulting channel fluctuations are opportunistically exploited by the inherent Proportional Fair (PF) scheduling mechanism native to 5G NR.

1. System Architecture: The authors developed an end-to-end 5G NR testbed integrating the OpenAirInterface (OAI) software stack with a custom-built, in-house RIS. The RIS is a 32×32 uniform planar array (1024 elements) operating at 5 GHz with a 300 MHz bandwidth, utilizing one-bit PIN diodes for phase switching. The testbed uses USRP B210s as radio units and operates in a Time-Division Duplex (TDD) mode with a 40 MHz bandwidth.
2. Randomization Strategy: Instead of optimizing phases for specific users, the RIS randomly selects phase vectors from a predefined sampling distribution derived from the angular directions of the User Equipments (UEs). The RIS configuration remains fixed for a switching interval $T_s$ before switching to a new random state.
3. Opportunistic Scheduling: The 5G Base Station (gNB) utilizes a PF scheduler. The scheduler selects the UE with the highest PF metric (instantaneous throughput divided by long-term average throughput) for transmission. The long-term average is updated using an Exponentially Weighted Moving Average (EWMA) with a window size $T_c$. The core hypothesis is that random RIS switching creates channel variations that the PF scheduler can exploit to select users with favorable instantaneous conditions, thereby harvesting RIS gains without explicit coordination.
4. Theoretical Framework: The authors developed a theoretical model characterizing the interaction between RIS switching dynamics and PF scheduling. They analyzed the convergence of system throughput as the PF window $T_c$ increases and derived design guidelines for selecting $T_s$ and $T_c$ to maximize throughput.
5. Experimental Validation: Experiments were conducted with two UEs in an indoor environment where the direct line-of-sight path was blocked. The study evaluated Reference Signal Received Power (RSRP), Block Error Rate (BLER), MCS index, and throughput under various RIS switching intervals ($T_s$) and PF averaging windows ($T_c$).

Key Contributions

1. End-to-End Testbed: The development and demonstration of a real-time, over-the-air (OTA) 5G NR system integrating an OAI-based gNB and a physically built RIS, quantifying performance gains in a realistic protocol stack environment.
2. Theoretical Convergence: Analytical proof (Theorem 1) demonstrating that as the PF scheduling window $T_c$ becomes large, the long-term throughput of a system with randomly configured RIS converges to the throughput achieved by an optimized RIS under Round-Robin (RR) scheduling.
3. Design Guidelines: Analytical derivation (Theorem 2) establishing the relationship between the PF window $T_c$, the RIS switching interval $T_s$, the number of UEs, and the number of RIS states. The analysis shows that $T_c$ needs to scale only logarithmically with the number of UEs and RIS states to achieve near-optimal throughput.
4. Experimental Verification: Empirical demonstration that a randomly tuned RIS, when paired with PF scheduling, achieves performance (RSRP, BLER, MCS, throughput) comparable to optimized RIS designs, without the overhead of CSI acquisition or phase optimization.
5. Robustness to UE Location: Experimental evidence showing that even with arbitrary UE locations (not necessarily aligned with RIS beamforming directions), the randomized RIS system significantly outperforms a system without an RIS.

Results

• Performance Gains: In single-UE scenarios, the RIS provided a 10–13 dB increase in RSRP (a >10× power increase), resulting in a throughput boost from ~26 Mbps (without RIS) to ~53 Mbps (with RIS).
• Throughput Comparison: In multi-UE scheduling scenarios, the randomized RIS with PF scheduling achieved throughput levels that approached the "genie-aided" benchmark (optimized RIS with RR scheduling). Specifically, with a PF window $T_c = 10$ s and RIS switching interval $T_s = 5$ s, the system achieved an aggregate throughput of 52 Mbps, significantly outperforming the no-RIS baseline (26 Mbps) and the Round-Robin scheduling with random RIS (~40-48 Mbps).
• Parameter Sensitivity: The experiments confirmed that an intermediate RIS switching interval ($T_s \approx 5-9$ s) yields the best throughput. If $T_s$ is too large, the system fails to explore RIS states within the averaging window; if $T_s$ is too small, there is insufficient time for MCS adaptation to stabilize before the RIS reconfigures.
• Fairness: Despite the opportunistic nature of the scheduling, the system maintained long-term fairness, with UEs receiving roughly equal time slots over the duration of the experiment.

Significance and Claims
The paper claims that integrating an RIS into 5G systems does not necessarily require substantial complexity or explicit CSI optimization. The key takeaway is that randomized RIS operation, combined with native 5G NR opportunistic scheduling mechanisms, is sufficient to deliver most of the performance gains commonly attributed to optimization-based RIS designs.

This approach offers a practical and scalable path for real-world RIS deployment because:

• It eliminates the overhead of explicit CSI acquisition and phase optimization.
• It requires no coordination between the RIS and the 5G NR system.
• It leverages existing 5G NR scheduling infrastructure (PF scheduler) to harvest RIS benefits.
• It provides significant performance improvements even when UE locations are arbitrary and not perfectly aligned with RIS beamforming directions.

The authors conclude that while their experiments were conducted in a specific indoor, two-user setup, the concept extends to practical multi-antenna systems with larger numbers of users, presenting a viable alternative to complex optimization-based RIS strategies.