RIS-Assisted Rank Enhancement With Commodity WiFi Transceivers: Real-World Experiments

Reconfigurable intelligent surfaces (RISs) are a promising enabling technology for the sixth-generation ($6$G) of wireless communications. RISs, thanks to their intelligent design, can reshape the wireless channel to provide favorable propagation conditions for information transfer. In this work, we experimentally investigate the potential of RISs to enhance the effective rank of multiple-input multiple-output (MIMO) channels, thereby improving spatial multiplexing capabilities. In our experiment, commodity WiFi transceivers are used, representing a practical MIMO system. In this context, we propose a passive beam-focusing technique to manipulate the propagation channel between each transmit-receive antenna pair and achieve a favorable propagation condition for rank improvement. The proposed algorithm is tested in two different channel scenarios: low and medium ranks. Experimental results show that, when the channel is rank-deficient, the RIS can significantly increase the rank by $112%$ from its default value without the RIS, providing a rank increment of $1.5$. When the rank has a medium value, a maximum of $61%$ enhancement can be achieved, corresponding to a rank increment of $1$. These results provide the first experimental evidence of RIS-driven rank manipulation with off-the-shelf WiFi hardware, offering practical insights into RIS deployment for spatial multiplexing gains.

💡 Research Summary

**
This paper presents the first real‑world experimental validation of using reconfigurable intelligent surfaces (RIS) to enhance the effective rank of a multiple‑input multiple‑output (MIMO) channel when the transceiver hardware is a commodity Wi‑Fi system. The authors focus on a 3 × 3 MIMO link operating at the 5 GHz Wi‑Fi band (channel 48, 20 MHz bandwidth) and investigate how a passive RIS can be programmed to increase the number of usable spatial streams, i.e., the channel’s degrees of freedom, which is quantified by the effective rank (Rₑ) derived from the normalized singular values of the channel matrix.

System model and rank metric
The received signal is modeled as y = H V x + n, where H = H_d + H₂ Ψ H₁ combines the direct Tx‑Rx path (H_d) and the reflected path through the RIS. Ψ is a diagonal matrix of element‑wise reflection coefficients βₙe^{jψₙ} with βₙ≈1 and ψₙ∈{0,π} (1‑bit phase control). The effective rank is defined as
Rₑ = exp(−∑_{i=1}^{rank(H)} \bar q_i ln \bar q_i),
with \bar q_i being the normalized singular values of H. Maximizing Rₑ directly improves the conditioning of H and enables more parallel data streams.

Proposed passive beam‑focusing algorithm
Algorithm 1 (Passive Beam Focusing) searches over all possible Tx‑Rx antenna pairs (n_T, n_R). For each pair, the RIS phases are toggled element‑by‑element from 0 to π; after each toggle the reflected power G = |h_{T2,n_R} Ψ h_{1,n_T}|² is measured. If G increases, the new phase is kept; otherwise it is reverted. After scanning all N elements, the SVD of the resulting H is computed and Rₑ is evaluated. The antenna pair that yields the largest Rₑ and its associated phase configuration ψ* are stored. The overall computational complexity is O(N_T N_R N + N_T N_R · min(N_T,N_R)² max(N_T,N_R)), which is modest for the considered 3 × 3 system and allows frequent re‑optimization when the environment changes.

Experimental platform
Two TP‑Link N750 routers equipped with Atheros AR9k chipsets are flashed with the OpenWrt firmware and the CSI extraction tool. One router operates in injection mode, sending packets; the other receives them and extracts per‑subcarrier channel state information (CSI). A host PC collects the CSI and runs MATLAB scripts to build the 3 × 3 channel matrix, perform SVD, and compute Rₑ. The RIS hardware consists of 1‑bit programmable metasurface modules (256 elements each). Experiments are carried out with a single RIS module and with four modules placed side‑by‑side (total 1024 elements). For comparison, a large copper sheet (≈9 × RIS area) and a “Fixed Phase” configuration (all elements set to 0 or π) are also tested.

Results – low‑rank scenario
In the baseline (no RIS) the effective rank is ≈1.27. With a single RIS, the Fixed‑Phase and Passive Beam Focusing methods achieve similar Rₑ values (≈1.96–1.97), representing a modest ≈52 % increase. The copper sheet yields Rₑ≈1.92, slightly lower than the RIS because the specular reflection concentrates energy into a dominant singular mode, leaving secondary modes weak. The Passive Beam Focusing algorithm, however, creates a Fresnel‑like field with several partially coherent lobes, boosting multiple entries of H and improving the smallest singular values, which leads to a more robust rank increase.

Results – enlarged RIS (four modules)
When four RIS modules are used, Passive Beam Focusing dramatically raises the effective rank to ≈2.81, a 112 % improvement over the baseline and an absolute increase of ΔRₑ≈1.5. The Fixed‑Phase configuration, in contrast, only reaches Rₑ≈1.73, because the larger aperture narrows the array factor, making the reflected field more specular and concentrating power into a single spatial mode. This demonstrates that intelligent phase optimization is essential to exploit the additional aperture.

Results – medium‑rank scenario
For a richer scattering environment where the baseline Rₑ≈1.77, the four‑RIS Passive Beam Focusing still provides a noticeable gain (Rₑ≈2.59, ≈46 % increase). The Fixed‑Phase case performs similarly, but the gain is smaller than in the low‑rank case because the channel already contains many independent paths, limiting the RIS’s ability to create new distinct propagation modes.

Key insights and contributions

1. Practical hardware – The study uses off‑the‑shelf Wi‑Fi routers rather than laboratory USRPs, showing that RIS‑assisted rank enhancement is feasible with existing consumer devices.
2. Passive beam focusing – A low‑complexity, 1‑bit phase‑only algorithm can significantly reshape the channel matrix, increasing both the largest and several smaller singular values.
3. Scalability – Adding more RIS elements improves the potential gain, but only when the phase pattern is optimized; a naïve fixed‑phase configuration can even degrade performance.
4. Benchmarking against metal – The RIS outperforms a simple copper reflector because the latter provides a purely specular bounce, whereas the RIS creates a richer field distribution.
5. Near‑field operation – The Tx and Rx are within the Fresnel distance of the RIS, which is realistic for indoor deployments and explains why the algorithm can exploit element‑wise phase control effectively.

Limitations and future work
The experiments are limited to 1‑bit phase control and near‑field distances; extending to multi‑bit or continuous phase shifters, far‑field scenarios, and dynamic traffic conditions would be valuable. Moreover, the current study focuses on a single 3 × 3 link; investigating multi‑user MIMO, uplink/downlink asymmetry, and joint RIS‑precoder design are promising directions.

Conclusion
The paper convincingly demonstrates that a reconfigurable intelligent surface, even with coarse 1‑bit phase resolution, can be programmed to reshape the propagation environment and substantially increase the effective rank of a practical Wi‑Fi MIMO channel. This translates into higher spatial multiplexing capability without modifying the legacy Wi‑Fi protocol, highlighting RIS as a low‑cost, backward‑compatible upgrade path for future 6G networks.