⚡ electrical engineering

RIS-Assisted Cell-Free Massive MIMO: RIS-MS Selection in FR1 and FR3

This paper proposes a novel RIS-user association algorithm and analyzes the integration of reconfigurable intelligent surfaces into cell-free massive MIMO networks across FR1 and FR3 bands, demonstrating that optimizing RIS-to-user pairing based on line-of-sight connectivity significantly enhances spectral efficiency while highlighting critical trade-offs between training overhead and performance gains.

Original authors: Alejandro de la Fuente, Fernando Galindo, Uriel García-Bárbulo, Sandra-Noemy Arana-Alegre, Jan García-Morales

Published 2026-04-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Alejandro de la Fuente, Fernando Galindo, Uriel García-Bárbulo, Sandra-Noemy Arana-Alegre, Jan García-Morales

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to have a conversation in a huge, noisy, crowded room (like a busy airport terminal or a packed concert). In the old way of doing things, you'd have a few loudspeakers (cell towers) trying to shout over the crowd. But if you are standing behind a pillar or in a corner, the sound gets blocked, and you can't hear a thing.

This paper proposes a new, smarter way to fix that problem using two main ideas: Cell-Free Massive MIMO and Reconfigurable Intelligent Surfaces (RIS).

Here is the breakdown in simple terms:

1. The Setup: The "Cell-Free" Room

Instead of having a few big loudspeakers, imagine hundreds of tiny, friendly microphones and speakers scattered all over the room. This is Cell-Free Massive MIMO.

How it works: Everyone in the room is connected to all the speakers at once. If you are in a corner, the speakers right next to you help you out, while the ones far away help too. It creates a blanket of sound that covers everyone evenly, so no one is left in a "dead zone."

2. The Problem: The "Wall"

Even with hundreds of speakers, sometimes there is a giant concrete wall or a thick metal door between you and the speakers. The sound bounces off the wall and gets lost. This is the Non-Line-of-Sight (NLoS) problem.

3. The Solution: The "Magic Mirror" (RIS)

Enter the RIS. Think of an RIS as a giant, smart mirror made of thousands of tiny, adjustable tiles.

What it does: If a sound wave hits the mirror, the mirror can instantly change the angle of every single tile to bounce the sound exactly where you need it to go, bypassing the wall.
The Catch: To know how to angle the tiles, the mirror needs to "listen" to the room first. This takes time and energy (called "pilot overhead"). If you have too many mirrors or too many tiny tiles, the time spent "listening" and "figuring it out" might actually waste more time than it saves.

4. The Big Question: Which Mirror Helps Whom?

The paper tackles a tricky question: If you have 20 mirrors and 10 people, who gets which mirror?

The Bad Way: You could try to connect every mirror to every person, or just pick mirrors randomly. This is chaotic and wastes a lot of time setting things up.
The Paper's Smart Way: The authors propose a "LoS-First" Strategy.
- They look at who is having the worst signal (the people behind the most walls).
- They check which mirrors have a direct, clear line of sight to those struggling people.
- They assign the mirror to that person only if the mirror can see them clearly.
- Why? A mirror is useless if it's looking at a wall. It only works if it has a clear path to bounce the signal. If a mirror can't see anyone clearly, it just stays in a "default mode" (like a regular mirror) so it doesn't waste time trying to figure out a connection that doesn't exist.

5. The Frequency Twist: FR1 vs. FR3

The paper also tests this in two different "types of air":

FR1 (Lower Frequency): Like a deep, rumbling bass. It travels far and goes through walls easily, but it doesn't carry as much data (like a slow internet connection).
FR3 (Higher Frequency): Like a high-pitched whistle. It carries huge amounts of data (super fast internet), but it gets blocked by walls easily.
The Discovery: The "Magic Mirrors" are actually more useful in the FR3 range. Why? Because the higher frequency waves are so easily blocked that the mirrors are needed to bounce them around. Plus, because the waves are smaller, you can fit more tiny tiles on the same size mirror, making the mirror much more powerful.

6. The Secret Sauce: "Chunking" the Mirrors

The paper found a clever trick to handle the "time wasted listening" problem.

Instead of adjusting every single tiny tile on a mirror individually (which takes forever), they group the tiles into blocks (like a 4x4 grid).
They adjust the whole block at once.
The Result: You get 90% of the benefit but with 10% of the setup time. This makes the system fast enough to be practical for real-world use.

The Bottom Line

This paper says: "Don't try to connect everything to everything."
Instead, be smart. Find the people with the worst connections, find the mirrors that can see them clearly, and assign them. If a mirror can't see anyone, leave it alone. And if you want super-fast speeds (using the new FR3 bands), use these smart mirrors, but group the tiles together so you don't waste time setting them up.

In a nutshell: It's about using smart mirrors to bounce signals around obstacles, but doing it in a way that doesn't waste time figuring out who needs help. It's the difference between shouting instructions to a whole crowd versus whispering the right direction to the one person who is lost.

1. Problem Statement

The paper addresses the integration of Reconfigurable Intelligent Surfaces (RISs) into Cell-Free Massive MIMO (CF-mMIMO) networks, specifically focusing on the challenges of RIS-to-Mobile Station (MS) selection and channel estimation overhead across different frequency bands.

Key challenges identified include:

RIS-MS Association: Determining which RIS should assist which user. Inefficient association (e.g., assigning RISs without Line-of-Sight (LoS) links) leads to negligible performance gains while consuming valuable resources.
Training Overhead: Configuring RISs typically requires estimating cascaded channels (MS-RIS-AP). The pilot overhead scales linearly with the number of RIS elements ( $N_R$ ) and the number of assigned RISs. In dense networks or high-frequency bands (like FR3), this overhead can consume the coherence block, negating spectral efficiency (SE) gains.
Frequency Band Differences: The paper investigates the trade-offs between FR1 (sub-6 GHz, robust coverage, lower bandwidth) and FR3 (upper midband, 7.125–24.25 GHz, "golden band" for 6G). While FR3 allows for denser antenna arrays and higher potential gains, it is more susceptible to blockage and requires more complex channel estimation.

2. Methodology and System Model

System Architecture:

A TDD-based CF-mMIMO system with $L$ Access Points (APs) and $K$ single-antenna users.
$R$ RISs are deployed, each with $N_R$ reflective elements.
The system uses Spatially Correlated Ricean Fading to model both LoS and Non-Line-of-Sight (NLoS) components, crucial for FR1 and FR3 analysis.
Combining Scheme: Scalable Partial-Minimum Mean Square Error (P-MMSE) combining is employed to reduce fronthaul signaling complexity compared to centralized MMSE.

Proposed Framework:
The authors propose a two-stage approach to manage RIS configuration and overhead:

LoS-Aware RIS-MS Selection Algorithm:
- Principle: A RIS is only assigned to a user if a LoS link exists between them.
- Rationale: Without LoS, the RIS-Rx channel is dominated by diffuse scattering, making phase-shift optimization ineffective and the training overhead unjustifiable.
- Process:
  - Users are sorted by the number of APs with LoS links (prioritizing "weak" users with fewer LoS connections).
  - Unassigned RISs (those with NLoS to all users) are configured with a default low-overhead pattern (pseudo-random or orthogonal codebook) without requiring separation channel estimation.
  - Assigned RISs undergo specific phase-shift optimization.
Channel Estimation Strategy:
- Separation Stage: Estimates direct MS-AP and cascaded MS-RIS-AP links. Overhead $\tau_{p1} = R_A(1 + N_R)$ , where $R_A$ is the number of assigned RISs.
- Aggregated Stage: Standard uplink training for data combining.
- Subarray-Based Semi-Blind Sketching (Method 5): To mitigate overhead for large RISs, the paper proposes partitioning RIS elements into $N_B$ blocks. Instead of estimating $N_R$ states, only $1 + N_B$ states are required. This captures the dominant channel structure while drastically reducing pilot length.
Phase-Shift Optimization:
- For assigned RISs, a sequential update algorithm optimizes the phase of each element to maximize the equivalent channel gain for the specific assigned user.

3. Key Contributions

Novel Selection Algorithm: A LoS-centric RIS-user association policy that dynamically assigns RISs only when they can provide a deterministic channel boost, leaving NLoS RISs in a low-overhead default state.
Overhead-Aware Framework: Explicit modeling of the trade-off between RIS density/element count and pilot overhead. The paper demonstrates that unassigned RISs do not consume separation pilots, preserving coherence block resources.
Scalable Block-Level Grouping: Introduction of a subarray-based semi-blind sketching technique (Method 5) that reduces the separation overhead from $O(N_R)$ to $O(N_B)$ , enabling the use of large-aperture RISs in FR3 without saturating the coherence block.
FR1 vs. FR3 Comparative Analysis: A comprehensive evaluation of how frequency bands impact RIS-assisted CF-mMIMO, highlighting FR3's potential for high SE if overhead is managed correctly.

4. Numerical Results

The simulations were conducted with $L=20$ APs, $K=10$ (and $30$ for stress tests) users, and varying numbers of RISs ( $R$ ).

Spectral Efficiency (SE) vs. RIS Count:
- The proposed LoS-aware selection (Method 1) significantly outperforms random assignment or exhaustive optimization, particularly for moderate RIS counts ( $R=5$ to $15$).
- Saturation Point: As $R$ increases, the training overhead eventually outweighs the channel gain benefits. For example, increasing from $R=15$ to $R=20$ yields diminishing returns due to pilot exhaustion.
- Subarray Grouping (Method 5): This method consistently outperforms full-element configuration (Method 1) when $R \ge 5$ . It achieves a superior SE-overhead trade-off by reducing the pilot burden.
User Fairness (CDF of SE):
- The proposed strategy significantly improves the 5th-percentile SE (cell-edge users) by directing RIS resources to users with the weakest direct links.
- The system remains robust even in high-density scenarios ( $K=30$ ).
Frequency Band Comparison (FR1 vs. FR3):
- Physical Size Constraint: When comparing RIS panels of equal physical size ( $30 \times 30$ cm), the number of elements ( $N_R$ ) scales with frequency (e.g., $N_R=64$ at 3.5 GHz vs. $N_R=1024$ at 15 GHz).
- FR3 Challenges: At 15 GHz, the massive increase in $N_R$ causes the training overhead of Method 1 to degrade performance significantly (Method 4, random configuration, sometimes outperforms Method 1).
- FR3 Solution: Method 5 (block grouping) successfully mitigates this, allowing FR3 to leverage its higher element density for substantial SE gains without the overhead penalty.

5. Significance and Conclusion

This paper establishes that RIS-assisted CF-mMIMO is a viable and powerful architecture for 6G, provided that intelligent association and scalable estimation are employed.

Strategic Insight: Blindly deploying RISs or assigning them without LoS verification is counterproductive. The "LoS-first" selection policy ensures that RIS resources are only spent where they yield deterministic gains.
FR3 Potential: The FR3 band is identified as a "golden band" for these systems. While it offers higher capacity due to denser element packing, it requires advanced techniques (like block-level sketching) to manage the resulting channel estimation overhead.
Future Outlook: The work emphasizes that future 6G deployments must move beyond simple RIS integration toward overhead-aware, scalable architectures that balance pilot requirements with performance gains, particularly in dense urban environments utilizing FR3 frequencies.