Ferkans — Interactive Telecom Tutor

From Cellular to Cell-Free

Conventional (co-located) massive MIMO concentrates all $M$ antennas at a single base station, creating a cell with sharp boundaries. Cell-edge users suffer from large path loss to their serving BS and strong interference from neighbouring cells. An alternative paradigm — cell-free massive MIMO — distributes the antennas across the coverage area as a large number of access points (APs), all connected to a central processing unit (CPU) via a fronthaul network. Every AP coherently serves every user: there are no cell boundaries, no handovers, and no cell-edge users. This user-centric architecture promises to eliminate the inter-cell interference problem and provide uniformly high service quality across the entire coverage area.

Definition:
Cell-Free Massive MIMO Architecture

A cell-free massive MIMO system consists of $N$ access points (APs), each with $L_{\text{AP}}$ antennas (often $L_{\text{AP}} = 1$ ), distributed over a wide area. The total number of distributed antennas is $M_{\text{total}} = N \cdot L_{\text{AP}}$ . All APs are connected to a central processing unit (CPU) and jointly serve $K$ single-antenna users, where $M_{\text{total}} \gg K$ .

The uplink received signal at AP $n$ from user $k$ is:

$y_n = \sum_{k=1}^{K}\sqrt{p_k}\,g_{nk}\,x_k + w_n$

where $g_{nk} = \sqrt{\beta_{nk}}\,\tilde{g}_{nk}$ is the channel from user $k$ to AP $n$ , with large-scale fading $\beta_{nk}$ that depends on the distance between AP $n$ and user $k$ .

The key distinction from co-located massive MIMO: each user is close to at least some APs (macro diversity), ensuring that no user experiences uniformly weak channels.

,

Definition:
MR Processing in Cell-Free Massive MIMO

With MR combining in cell-free massive MIMO, each AP $n$ locally multiplies its received signal by the conjugate of its channel estimate and forwards the result to the CPU:

$\hat{x}_k = \sum_{n=1}^{N}\hat{g}_{nk}^* y_n$

The resulting SINR for user $k$ under the UatF bound is:

$\text{SINR}_k^{\text{CF}} = \frac{p_k\!\left(\sum_{n=1}^{N}\gamma_{nk}\right)^2} {\sum_{j=1}^{K}p_j\sum_{n=1}^{N}\gamma_{nk}\beta_{nj} + \sigma^2\sum_{n=1}^{N}\gamma_{nk}}$

where $\gamma_{nk} = \mathbb{E}[|\hat{g}_{nk}|^2]$ is the mean square of the channel estimate at AP $n$ for user $k$ .

The key feature is that the desired signal sums coherently across all $N$ APs, while interference sums incoherently due to the distributed geometry.

Cell-Free vs. Cellular Topology

Compare the two massive MIMO deployment paradigms side by side. In the cellular (co-located) topology, a cell-edge user suffers from large path loss. In the cell-free (distributed) topology, every user is close to at least a few access points, providing uniformly good service through macro diversity.

Cellular vs. cell-free: distributing antennas eliminates the cell-edge problem.

Cell-Free vs. Cellular Massive MIMO

Compare the CDF of per-user rates for co-located (cellular) and cell-free massive MIMO with the same total number of antennas. The cell-free architecture provides dramatically better 95%-likely rate (5th percentile of the CDF) due to macro diversity.

Parameters

M_{\mathrm{total}}

(total antennas)64

K

(number of users)10

SNR [dB]0

Number of APs

N

16

Theorem: Cell-Free Achieves Better 95%-Likely Rate

Consider a fixed total number of antennas $M_{\text{total}}$ and $K$ users uniformly distributed over a coverage area. Let $R_{\text{co-loc}}^{(0.05)}$ and $R_{\text{CF}}^{(0.05)}$ denote the 5th-percentile per-user rates for co-located and cell-free deployments, respectively, both using MR combining. Then:

$R_{\text{CF}}^{(0.05)} \geq R_{\text{co-loc}}^{(0.05)}$

with equality only in degenerate cases (all APs co-located or single user at the BS location).

Moreover, the cell-free 95%-likely rate scales as:

$R_{\text{CF}}^{(0.05)} \sim \log_2\!\left(1 + \frac{M_{\text{total}}}{K}\cdot c_{\text{macro}}\right)$

where $c_{\text{macro}} > 0$ is a macro-diversity constant that depends on the AP density and path-loss exponent, and is bounded away from zero for any user location (unlike co-located systems where cell-edge users have $\beta_k \to 0$ ).

In a co-located system, the worst-case user sits at the cell edge with a very small $\beta_k$ , making its SINR proportional to $M\beta_{\min}$ . In the cell-free architecture, every user is near at least a few APs, so the effective channel gain $\sum_n \gamma_{nk}$ has a much smaller dynamic range. The worst-case user in cell-free has a sum gain that is orders of magnitude larger than $\beta_{\min}$ in the co-located case. Cell-free sacrifices some peak rate (the best user is farther from its nearest antenna cluster) to dramatically improve the worst-case rate.

Proof

Path-loss comparison

In co-located massive MIMO with a single BS, the 5th-percentile user has path loss:

$\beta_{\min} \propto r_{\max}^{-\alpha}$

where $r_{\max}$ is the cell radius and $\alpha$ is the path-loss exponent. For a hexagonal cell with $\alpha = 3.8$ and cell radius 500 m, $\beta_{\min}$ can be 30--40 dB below $\beta$ at the cell centre.

Distributed macro-diversity gain

In cell-free with $N$ APs uniformly distributed, the 5th-percentile user has effective channel gain:

$\sum_{n=1}^{N}\gamma_{nk} \geq \gamma_{\text{near}} \cdot N_{\text{near}}$

where $N_{\text{near}}$ is the number of nearby APs (within, say, 100 m) and $\gamma_{\text{near}}$ is their average channel quality. Even the worst-located user is near at least a few APs when $N$ is large.

Formal comparison

By Jensen's inequality on the sum of log terms and the concavity of the rate function:

$R_{\text{CF}}^{(0.05)} \geq R_{\text{co-loc}}^{(0.05)}$

because the distributed channel gains $\{\beta_{nk}\}$ have smaller variance across users than the co-located gains $\{\beta_k\}$ . $\blacksquare$

Cell-Free vs. Cellular Massive MIMO

Property	Co-located (Cellular)	Cell-Free
Antenna deployment	All $M$ at one BS	$N$ APs with $M/N$ each, distributed
Cell boundaries	Hard cell edges	No cell boundaries
Worst-case user	Cell-edge: $\beta \propto r_{\max}^{-\alpha}$	Always near some APs
95%-likely rate	Low (limited by cell-edge)	High (macro diversity)
Peak rate	High for cell-centre	Slightly lower (antennas distributed)
Fronthaul requirement	Minimal (single site)	High (AP-to-CPU links)
Channel estimation	Centralised at BS	Local at each AP + CPU fusion
Handover	Required at cell edges	Not needed (user-centric)
Power control	Per-cell optimisation	Network-wide optimisation
Pilot contamination	Inter-cell (severe at edges)	Reduced (geographic separation)
Scalability	Good (single BS)	Requires scalable fronthaul
Standard support	5G NR (Release 15+)	Under study for 6G

Example: Distributed vs. Co-Located System

Consider $M_{\text{total}} = 64$ antennas and $K = 4$ users in a $1 \times 1$ km area. Compare two deployments: (a) Co-located: all 64 antennas at the centre. (b) Cell-free: $N = 16$ APs with 4 antennas each, placed on a regular grid.

Assume path-loss exponent $\alpha = 3.8$ , SNR = 0 dB at 100 m reference distance, and MR combining. Compute the approximate 5th-percentile rate for each deployment.

Solution

Co-located deployment

The cell-edge user is approximately 707 m from the centre (corner of the square). At reference distance 100 m:

$\beta_{\text{edge}} = \left(\frac{100}{707}\right)^{3.8} = 7.07^{-3.8} \approx 0.00047$

$\text{SINR}_{\text{edge}}^{\text{MR}} \approx \frac{64 \times 0.00047}{3 \times 0.00047 + 1} \approx \frac{0.030}{1.0014} \approx 0.030$

$R_{\text{edge}} \approx \log_2(1.030) \approx 0.04\;\text{bits/s/Hz}$

Cell-free deployment

With 16 APs on a $4 \times 4$ grid (spacing 250 m), the worst-case user is at most $\sim 177$ m from the nearest AP. The 4 nearest APs are within 250 m:

$\sum_{n \in \text{near}}\beta_{nk} \approx 4 \times \left(\frac{100}{177}\right)^{3.8} \approx 4 \times 0.114 = 0.456$

Including the remaining 12 APs (farther, contributing $\sim 0.05$ total):

$\sum_{n=1}^{16}\gamma_{nk} \approx 0.506$

$\text{SINR}_{\text{worst}}^{\text{CF}} \approx \frac{0.506^2 \times 4}{3 \times 0.506 + 1} \approx \frac{1.024}{2.518} \approx 0.41$

$R_{\text{worst}} \approx \log_2(1.41) \approx 0.50\;\text{bits/s/Hz}$

Comparison

The cell-free architecture provides a $12\times$ improvement in 5th-percentile rate (0.50 vs. 0.04 bits/s/Hz) for the same total antenna count. This illustrates the fundamental advantage of macro diversity: eliminating the cell edge. $\blacksquare$

Quick Check

What is the main advantage of cell-free massive MIMO over co-located massive MIMO?

Higher peak rate for the best user

Lower computational complexity

Uniformly good service with much better worst-case rate

No need for channel estimation

Correction:

Uniformly good service with much better worst-case rate

By distributing antennas, every user is close to at least some APs, providing macro diversity that eliminates the cell-edge problem. The 95%-likely rate improves dramatically.

🚨Critical Engineering Note

Fronthaul Capacity and Latency for Cell-Free Massive MIMO

The practical feasibility of cell-free massive MIMO hinges on the fronthaul network connecting APs to the CPU. Each AP must forward either raw baseband samples or processed statistics to the CPU, creating stringent bandwidth and latency requirements.

Fronthaul capacity:

Centralised processing (CPU has raw samples): Each AP forwards its received signal at the full baseband rate. For 100 MHz bandwidth, 16-bit I/Q, this is $100\text{M} \times 2 \times 16 = 3.2$ Gbps per AP. With $N = 64$ APs, total fronthaul demand is $\sim 200$ Gbps.
Local processing (AP forwards soft estimates): Only $K$ complex scalars per coherence block per AP, reducing fronthaul by $\sim M/K$ . This is the approach used in scalable cell-free implementations.

Latency constraint: The round-trip fronthaul latency must be within the HARQ timing budget. In 5G NR with 0.5 ms slot duration, one-way fronthaul latency must be $< 100\;\mu$ s for 1 ms HARQ round-trip, requiring fibre or high-quality Ethernet fronthaul.

Scalable implementations (Bjornson and Sanguinetti, 2020) address this by having each AP serve only nearby users and each user be served by only nearby APs, reducing the effective fronthaul load to a manageable level while retaining most of the macro-diversity benefit.

Practical Constraints

•
Centralised processing: ~3.2 Gbps per AP per 100 MHz carrier
•
Local processing reduces fronthaul by factor M/K
•
One-way fronthaul latency < 100 us for 5G NR HARQ

Historical Note: The Cell-Free Massive MIMO Concept

2017

The cell-free massive MIMO paradigm was formulated by Hien Quoc Ngo, Alexei Ashikhmin, Hong Yang, Erik G. Larsson, and Thomas L. Marzetta in their 2017 paper "Cell-Free Massive MIMO Versus Small Cells." Building on earlier work on network MIMO and distributed antenna systems, they showed that a large number of single-antenna APs connected to a CPU and using simple conjugate beamforming can provide 5 $\times$ to 10 $\times$ better 95%-likely throughput than small-cell systems with the same total number of antennas. The key insight was combining the benefits of massive MIMO (channel hardening, simple processing) with distributed antennas (macro diversity, no cell edges), while keeping the processing simple enough for practical implementation. The concept has since been extended to multi-antenna APs, local MMSE processing, and scalable implementations, and is a leading candidate architecture for 6G.

🎓CommIT Contribution(2017)

User-Centric Cell-Free Massive MIMO

H. Q. Ngo, A. Ashikhmin, H. Yang, E. G. Larsson, T. L. Marzetta, G. Caire — IEEE Transactions on Wireless Communications

This paper formulated the cell-free massive MIMO concept and provided the first rigorous performance analysis comparing it to small-cell deployments. The key findings relevant to this chapter:

Conjugate beamforming with $N$ single-antenna APs achieves $5$ -- $10\times$ better 95%-likely per-user throughput than small cells with the same total antenna count.
A closed-form max-min power control algorithm was derived for the cell-free setting.
Scalable implementations by Bjornson, Sanguinetti, and Caire later showed how to reduce fronthaul requirements through dynamic cooperation clustering while retaining macro-diversity benefits.

The cell-free concept has since become a leading architecture candidate for 6G, with Caire and collaborators continuing to develop practical algorithms for multi-antenna APs, local MMSE processing, and hybrid centralised-distributed architectures.

cell-freemassive-mimomacro-diversityCommIT

Cell-Free Massive MIMO

A network architecture where a large number of geographically distributed access points (APs) coherently serve all users in a coverage area without cell boundaries. All APs connect to a central processing unit (CPU) and jointly process signals, providing macro diversity and uniformly good coverage.

Related: Massive MIMO, Macro Diversity

Why This Matters: Reconfigurable Intelligent Surfaces and Cell-Free MIMO

Reconfigurable intelligent surfaces (RIS) are a complementary technology to cell-free massive MIMO. While cell-free distributes active antenna elements (APs with RF chains, ADCs, amplifiers), RIS deploys passive reflecting elements that reshape the propagation environment without any active RF hardware.

The synergy is natural: in a cell-free architecture, RIS panels can be deployed alongside APs to:

Create virtual line-of-sight paths for users in deep shadow
Enhance the channel rank for users with rank-deficient channels
Reduce the number of active APs needed for coverage

The RIS specialised book develops the electromagnetic theory, beamforming optimisation, and system-level integration of RIS with massive MIMO in detail. The CommIT group has contributed to array-fed RIS architectures for sub-THz multiuser multibeam communication.

Macro Diversity

Diversity gain achieved by receiving (or transmitting) signals from multiple geographically separated access points. In cell-free massive MIMO, macro diversity ensures that every user has at least some nearby APs with strong channel gains, even if others are far away.

Fronthaul

The communication link connecting distributed access points to the central processing unit in a cell-free architecture. Fronthaul capacity and latency are critical constraints that determine the level of cooperation possible among APs.

Related: Cell-Free Massive MIMO

Cell-Free Massive MIMO

From Cellular to Cell-Free

Definition: Cell-Free Massive MIMO Architecture

Definition: MR Processing in Cell-Free Massive MIMO

Cell-Free vs. Cellular Topology

Cell-Free vs. Cellular Massive MIMO

Parameters

Theorem: Cell-Free Achieves Better 95%-Likely Rate

Path-loss comparison

Distributed macro-diversity gain

Formal comparison

Cell-Free vs. Cellular Massive MIMO

Example: Distributed vs. Co-Located System

Co-located deployment

Cell-free deployment

Comparison

Quick Check

Fronthaul Capacity and Latency for Cell-Free Massive MIMO

Historical Note: The Cell-Free Massive MIMO Concept

User-Centric Cell-Free Massive MIMO

Cell-Free Massive MIMO

Why This Matters: Reconfigurable Intelligent Surfaces and Cell-Free MIMO

Macro Diversity

Fronthaul

Definition:
Cell-Free Massive MIMO Architecture

Definition:
MR Processing in Cell-Free Massive MIMO