RIS-Aided Cell-Free Massive MIMO

The 6G Architecture: Distributed Everything

Cell-free massive MIMO (CFmMIMO) replaces the single BS with a network of distributed access points (APs), all connected to a central processing unit (CPU) via fronthaul. Each user sees many APs; each AP serves many users; the cell boundary disappears. For 6G, CFmMIMO is a leading candidate for ubiquitous high-rate coverage.

Adding RIS to CFmMIMO is a natural next step: distributed RIS panels complement the distributed APs. The RIS panels provide passive aperture while the APs provide active transmission. Coverage: ubiquitous. Rate: very high. Cost: careful engineering required.

Definition:
RIS-Aided Cell-Free Massive MIMO

A RIS-aided cell-free architecture consists of:

$L$ access points (APs), each with $N_t$ antennas, connected via high-speed fronthaul to a CPU.
$M$ passive RIS panels, distributed across the coverage area.
$K$ users.

The channel from user $k$ to AP $\ell$ via RIS panel $m$ is

$\mathbf{g}_{\ell k m}^H = \mathbf{h}_{\ell k, d}^H + \mathbf{h}_{\ell k, 2, m}^H \boldsymbol{\Phi}_m \mathbf{H}_{1, \ell m},$

where $\mathbf{H}_{1,\ell m}$ is the AP- $\ell$ -to-panel- $m$ channel and $\mathbf{h}_{\ell k, 2, m}$ is the panel- $m$ -to-user- $k$ channel seen from AP $\ell$ .

The CPU jointly optimizes: the AP precoders $\{\mathbf{W}_\ell\}_{\ell=1}^L$ and the RIS phase matrices $\{\boldsymbol{\Phi}_m\}_{m=1}^M$ .

Theorem: CFmMIMO + RIS Capacity Scaling

Under favorable propagation and coherent AP + RIS combining, the per-user rate in a RIS-aided CFmMIMO system scales as

$R_k \approx \log_2\!\left(1 + \frac{P_t\,L N_t + M N^2}{\sigma^2}\right) \ \text{bits/s/Hz},$

where the first term is the AP aperture (AP gain) and the second is the RIS aperture (coherent multi-panel gain). Both components contribute; neither dominates at typical deployment sizes.

For $L N_t \sim 64, M N^2 \sim 10^4$ : RIS dominates at moderate SNR. The RIS contribution is typically $1$ - $2$ orders of magnitude larger than the active-aperture contribution in coverage-limited scenarios.

The Cell-Free + RIS Synergy

CFmMIMO and RIS are natural partners:

CFmMIMO provides uniform active-aperture coverage via distributed APs. No cell boundaries.
RIS provides passive aperture amplification, especially in coverage gaps that APs cannot reach.
Combined: ubiquitous coverage with high per-user rate.

The architecture is symmetric in a deep sense: both APs and RIS panels are distributed infrastructure, centrally coordinated by the CPU. The distinction between "transmitter" and "channel shaper" blurs — both are manipulated by the same optimization.

This is the vision for 6G networks: thousands of cheap APs

passive RIS panels, distributed across a city, jointly optimized to deliver quasi-optimal coverage and capacity.

CFmMIMO + RIS Rate Scaling

Plot per-user rate as $L$ (AP count) and $M$ (RIS panel count) vary, at fixed total coverage area. Shows the tradeoff between active and passive aperture in cell-free deployment.

Parameters

Number of APs

L

Number of RIS panels

M

Antennas per AP4

Elements per panel128

Number of users

K

Distributed Joint Optimization for CFmMIMO + RIS

Complexity: Centralized:

O(L^3 N_t^{3} K + M N^3)

. Distributed:

O(L N_t^{3} K + M N^3)

per AP per round.

Input: channels from all

K

users to all

L

APs through all

M

panels.

Output: AP precoders

\{\mathbf{W}_\ell\}

, RIS phases

\{\boldsymbol{\Phi}_m\}

Centralized version (CPU does everything):

1. Aggregate all CSI at the CPU.

2. Run multi-panel AO (Section 12.1's Algorithm) with

L N_t

aggregated active dims.

3. Distribute precoders to APs via fronthaul.

Distributed version (APs cooperate via fronthaul):

1. Each AP locally estimates its own CSI.

2. APs exchange compressed CSI via fronthaul.

3. Distributed BCD: each AP and each RIS panel updates locally given neighbors' state.

4. Converges to a near-centralized optimum with

\sim 2

3\times

more iterations.

Distributed implementation is crucial at scale: for $L = 64, M = 32$ , centralized optimization is prohibitively expensive. Distributed BCD trades some optimality for dramatic scalability.

🎓CommIT Contribution(2024)

Scalable Joint Scheduling for RIS-Aided Cell-Free MIMO

G. Caire, I. Atzeni — IEEE Trans. Wireless Commun. (preprint)

Caire and collaborators (2024) tackle the scalability challenge in RIS-aided cell-free networks. The CommIT contribution is a hierarchical scheduling framework:

Macro-scheduling (slow): user-to-cluster assignment (which APs and RIS panels serve which users).
Micro-scheduling (fast): within each cluster, joint AP and RIS optimization.

The hierarchy decouples a nominally O( $(LMK)^3$ ) problem into parallel smaller O( $L_c M_c K_c^3$ ) problems per cluster. For realistic 6G parameters ( $L = 100, M = 32, K = 200$ ), this reduces optimization time from hours to milliseconds. Enables practical deployment of RIS-aided cell-free at city scale. The paper also formalizes the pilot reuse across clusters and the graceful degradation under cluster handovers.

This is the CommIT contribution for the multi-RIS chapter: practical algorithms at 6G deployment scale.

cell-freecf-risschedulingscalabilitycaire-2024

⚠️Engineering Note

Practical CFmMIMO + RIS Deployment

Scaling CFmMIMO + RIS to 6G deployment:

Fronthaul capacity: APs need $\sim 1$ - $10$ Gbps fronthaul to the CPU. Passive RIS only needs $\sim 10$ - $100$ kbps control-link.
Synchronization: all APs and RIS panels need common phase reference for coherent coordination. Optical fronthaul handles APs; RIS needs separate (but simpler) sync.
Pilot scaling: pilot overhead grows with number of users + RIS panels. Use pilot reuse (coherence-block-level) and compressed sensing.
Failure modes: one AP or panel failing doesn't kill the network — it degrades gracefully. The large $L, M$ give redundancy.
Upgrade path: start with a few APs + RIS panels; incrementally add more without re-planning the whole network.

Practical Constraints

•
Typical 6G cell-free network: $L \sim 100$ APs, $M \sim 10$ - $50$ RIS panels per sq-km.
•
Fronthaul bandwidth: $\sim 10$ Gbps per AP to CPU.
•
RIS control bandwidth: $\sim 10$ kbps per panel (much lower).
•
Total optimization time per coherence block: $\sim 100$ - $500$ ms at 6G coherence ( $\sim 1$ - $10$ ms).

Quick Check

Under perfect synchronization across $M$ panels, the coherent multi-RIS SNR scales as:

$M N^2$

$M^2 N^2$

$N^{2M}$

$M N^{2M}$

Correction:

M^2 N^2

Coherent combining across M panels yields M² in power gain, multiplied by the per-panel N² coherent gain. Total: M² N². Without synchronization, the scaling falls to M N² (incoherent).

Common Mistake: Don't Underestimate Multi-Panel Complexity

Mistake:

"RIS is passive; adding more panels is free."