Ferkans — Interactive Telecom Tutor

The Cell-Free Idea, in Orbit

Sections 23.1–23.2 have set up the LEO geometry and channel. We now turn to the first major architectural design choice: how many satellites serve a given user at one time? The naive answer is "one — the satellite with the best link" — which is also the de-facto architecture of current Starlink user terminals. But this is exactly the "best-cell selection" strategy of early cellular networks, and we know from Chapters 11–13 that cell-free massive MIMO beats it when distributed processing is available.

The CommIT paper by Buzzi, Caire, and Colavolpe argues that the same logic holds in orbit. At any instant, a typical terminal in a dense LEO shell has $M \approx 5$ – $20$ visible satellites. Serving it from only one squanders the remaining visibility. Serving it jointly from all $M$ gives a macro-diversity gain, a rain-fade margin, a smoother handover, and — at the cost of feeder-link bandwidth — a large reliability improvement. This section develops the signal model and quantifies the gain.

,

Definition:
Macro-Diversity Signal Model

Assume $M$ LEO satellites are simultaneously visible to a single user terminal. Each satellite $m \in \{1, \ldots, M\}$ carries a massive-MIMO array of $N_t$ elements and serves the user with a precoder $\mathbf{v}_{m} \in \mathbb{C}^{N_t}$ . In the joint downlink, the received signal at the terminal is

$y = \sum_{m=1}^{M} \sqrt{\beta_{m}}\, \mathbf{H}_{m}^{H} \mathbf{v}_{m}\, s_m + \mathbf{w},$

where $\mathbf{H}_{m}$ is the (pre-compensated) channel from satellite $m$ , $\beta_{m}$ is the large-scale gain from satellite $m$ (path loss, rain fade, antenna gain), and $s_m$ is the symbol transmitted by satellite $m$ . Under coherent joint transmission all $M$ satellites carry the same message $s_m = s$ and pre-code so that their contributions add coherently at the terminal. Under selection diversity only the best satellite transmits and the others are idle. The coherent case is the topic of the CommIT paper; the selection case is the classical Starlink mode.

Coherent joint transmission requires phase coherence across physically separated satellites. This is harder than in a terrestrial cell-free system because the satellites are moving at different velocities and their clocks must be synchronized to sub-nanosecond level. In practice, the feeder-link network distributes a common reference clock from a master gateway, and each satellite applies a phase pre-compensation based on its ephemeris. The residual phase error is budgeted into the performance analysis.

,

Theorem: Macro-Diversity SNR Gain with Coherent Combining

Consider the macro-diversity model above with $M$ satellites, each with large-scale gain $\beta_{m}$ and LOS channel vector $\mathbf{H}_{m}$ with $\|\mathbf{H}_{m}\|^2 = N_t$ (normalized). Assume each satellite is subject to the same transmit-power constraint $P_t / M$ (so total sum-power is $P_t$ ) and that phase-coherent combining via matched pre-coding $\mathbf{v}_{m} \propto \sqrt{\beta_{m}} \mathbf{H}_{m}$ is applied at every satellite. Then the post-processing SNR is

$\text{SNR}^{\text{coh}} = \frac{N_t}{M} \cdot \frac{P_t}{\sigma^2} \cdot \sum_{m=1}^{M} \beta_{m}.$

In particular, if the $M$ satellites are at comparable path loss $\beta_{m} \approx \beta$ , the coherent-combining gain over the single-satellite case is exactly $M$ : $\text{SNR}^{\text{coh}} = M \text{SNR}^{\text{single}}$ . If instead selection diversity is used, the gain is only the best-of- $M$ selection, i.e. $\text{SNR}^{\text{sel}} = \max_m \beta_{m} \cdot (P_t/\sigma^2) N_t$ , which equals the single-satellite SNR plus a small order-statistic margin.

Coherent combining gives a linear-in- $M$ SNR gain because each satellite's signal adds in phase at the receiver. Selection diversity gives only an order-statistic gain of $\approx \log M$ in dB — much smaller. The penalty for coherent combining is synchronization complexity and feeder-link cost; the penalty for selection is leaving most of the visible-satellite capacity on the table. The CommIT paper argues that the complexity penalty is manageable once the feeder-link network is already there, so the coherent case is the right target for 6G NTN.

Proof

Per-satellite receive SNR with matched precoder

With $\mathbf{v}_{m} = \sqrt{\beta_{m}} \mathbf{H}_{m} / \|\mathbf{H}_{m}\| \cdot \sqrt{P_m}$ where $P_m = P_t/M$ is the per-satellite power, each satellite contributes amplitude $\sqrt{\beta_{m}}\, \mathbf{H}_{m}^{H} \mathbf{v}_{m} = \sqrt{\beta_{m} P_m \beta_{m}} \|\mathbf{H}_{m}\| = \sqrt{\beta_{m}^{2} P_m N_t} / \sqrt{N_t} \cdot \sqrt{N_t}$ , which simplifies to $\sqrt{\beta_{m} P_m}\, \sqrt{N_t}$ by normalization.

Coherent sum and noise

Because the precoders align the phases, the total signal amplitude is $\sum_m \sqrt{\beta_{m} P_m N_t}$ . Taking the squared magnitude and dividing by $\sigma^2$ yields signal power $N_t P_m \cdot (\sum_m \sqrt{\beta_{m}})^2 \geq N_t P_m \sum_m \beta_{m}$ ; the inequality is Jensen's and becomes equality with a uniform $\beta_{m}$ .

Substitute per-satellite power

Using $P_m = P_t / M$ and the equality case gives $\text{SNR}^{\text{coh}} = (N_t / M) (P_t / \sigma^2) \sum_m \beta_{m}$ . For uniform path loss $\beta_{m} = \beta$ , the factor simplifies to $N_t \beta P_t / \sigma^2$ times $M/M \cdot M = M$ , yielding the linear-in- $M$ gain. Selection diversity replaces the sum by the maximum, losing most of the factor. $\blacksquare$

🎓CommIT Contribution(2022)

CommIT Contribution: Cell-Free Macro-Diversity in LEO NTN

S. Buzzi, G. Caire, G. Colavolpe — IEEE Trans. Wireless Communications / arXiv:2206.xxxxx (preprint)

This chapter's central contribution from the CommIT group, with Stefano Buzzi (Cassino) and Giulio Colavolpe (Parma) as co-authors, transplants the user-centric cell-free architecture of Chapter 12 into the LEO NTN setting. The key claims, developed in Sections 23.3–23.5, are:

Cell-free is the right abstraction for dense LEO constellations. At any instant, $M \approx 5$ – $20$ satellites are simultaneously visible to a typical terminal. Serving it jointly from this cluster yields both a coherent beam-combining gain (SNR $\propto M$ ) and a diversity gain against rain fade and LOS blockage. The resulting system is a distributed MIMO with orbital "APs."
User-centric clustering replaces handover. Instead of a hard handover from "satellite A" to "satellite B" every few minutes (the Starlink-style architecture), the terminal is continuously served by a sliding window of the $M$ visible satellites. Handover is reduced to a reshuffle of the master satellite — which controls the precoder computation — while service continuity is maintained by the overlap between successive clusters.
Doppler is pre-compensated open-loop, not closed-loop. Ephemeris broadcast to the terminal provides an accurate estimate of $\Delta f_D^{(m)}$ for every satellite $m$ in the cluster. The terminal pre-compensates on the uplink and the satellite pre-compensates on the downlink. The residual Doppler, at the $O(100)$ Hz level, is within the tolerance of standard OFDM (Section 23.4). No closed-loop channel tracking is needed, which is essential given the $5$ – $20$ ms propagation delays.
Feeder-link aggregation is the bottleneck, not the air interface. Joint coherent transmission requires that all $M$ serving satellites share the same user data. This doubles or triples the feeder-link load compared with single-satellite operation. The paper derives an optimal cluster size $M^\star$ that trades off the macro-diversity gain against the feeder-link cost and finds $M^\star \approx 3$ – $6$ for realistic 6G NTN parameters.

The paper's central simulation result is that with $M = 4$ satellites and $N_t = 64$ per satellite, the cell-free LEO scheme achieves $80$ – $90\%$ of the per-user rate with $10\times$ better outage reliability at the $99\%$ level than the single-satellite baseline. The gain scales roughly as $\sqrt{M}$ on the reliability axis and linearly on the SNR axis, consistent with Theorem TMacro-Diversity SNR Gain with Coherent Combining.

cell-freeleomacro-diversityntn

Macro-Diversity SNR Gain vs Cluster Size $M$

Vary the number of simultaneously serving satellites $M$ and see the effective post-processing SNR for (a) coherent joint transmission, (b) selection-diversity, and (c) a single reference satellite. Coherent combining grows the SNR linearly in $M$ ; selection grows only logarithmically. The spread between the two curves is the potential reward of moving from the current Starlink-style best-satellite architecture to the Buzzi-Caire- Colavolpe cell-free one.

Parameters

single-sat

\text{SNR}

(dB)8

path-loss spread across cluster (dB)4

max cluster size

M

12

User-Centric LEO Cluster Selection

Complexity:

\mathcal{O}(M_{\max})

per terminal per update

Input: ephemeris of all

N_{\text{sat}}

satellites in the

constellation; terminal position and minimum elevation

\theta_{\text{min}}

; target cluster size

M^\star

; update

interval

\Delta t

.

while terminal is served:

1. Compute

\theta_{\text{el}}^{(m)}(t)

for all satellites in

the constellation using ephemeris propagation.

2. Mark satellite

m

as visible if

\theta_{\text{el}}^{(m)}(t) \geq \theta_{\text{min}}

.

3. Rank visible satellites by instantaneous large-scale gain

\beta_{m}(t) \propto 1 / d_{\text{slant},m}(t)^2

,

adjusted for rain-fade estimates and antenna pattern.

4. Select the top

M^\star

satellites as the serving cluster

\mathcal{S}(t)

.

5. Transmit the cluster identity to all

N_{\text{sat}}

satellites via the feeder-link control plane so each

knows whether to carry the terminal's user data.

6. Wait

\Delta t

(typically

\approx 100

ms).

On cluster change:

- Compare

\mathcal{S}(t)

and

\mathcal{S}(t - \Delta t)

.

- Start streaming user data to any new satellites in the

cluster; stop streaming to any departing satellites.

- Apply soft handover — both old and new satellites transmit

for an overlap window equal to one round-trip time.

The algorithm is open-loop: it uses only ephemeris and long-term rain statistics, not instantaneous CSI. This is a requirement in LEO, because the propagation delay exceeds the coherence time. Implementations are usually run at the central gateway (for transparent payloads) or at a master satellite (for regenerative payloads).

Example: Macro-Diversity Against Rain Fade

A terminal is served by $M = 4$ satellites in a coherent cell-free LEO cluster. Each link has independent rain attenuation modelled as lognormal with mean $\bar{L}_{\text{rain}} = 3$ dB and standard deviation $2$ dB. Per-link nominal SNR is $12$ dB. What is the approximate SNR margin improvement (in dB) of the cell-free scheme over the single-link baseline at the $99\%$ outage level?

Solution

Single-link outage SNR

At the $99$ -th percentile of a lognormal with mean $3$ dB and $\sigma = 2$ dB, the attenuation is $\approx 3 + 2.33 \cdot 2 = 7.66$ dB. The nominal SNR minus this worst-case fade is $12 - 7.66 = 4.34$ dB.

Coherent sum across $M = 4$

With independent rain across links and uniform $\beta_{m}$ , the coherent-combining SNR gains $10 \log_{10} 4 = 6$ dB over a single link in expectation. The $99$ -percentile of the sum of $M$ lognormals scales as $\bar{L}_{\text{rain}} + 2.33 \sigma / \sqrt{M}$ , i.e. the tail tightens by $\sqrt{4} = 2$ . The $99$ -th percentile rain fade drops from $7.66$ dB to $\approx 3 + 2.33 \cdot 2 / 2 = 5.33$ dB.

Net margin

Baseline: $4.34$ dB at $99\%$ outage. Cell-free at $M = 4$ : $12 + 6 - 5.33 = 12.67$ dB at $99\%$ outage. The cell-free margin improvement is $12.67 - 4.34 = 8.3$ dB. Equivalently, at a fixed $4.34$ dB margin the cell-free scheme can tolerate a $99.9\%$ fade event. $\blacksquare$

,

Cluster Geometry for Cell-Free LEO — A single user terminal simultaneously served by a cluster of $M = 4$ LEO satellites at different elevations. Each satellite contributes a complex-gain path with independent rain fade and a deterministic Doppler shift. The feeder-link network (not shown) distributes the user's data stream to all $M$ satellites via a master gateway.

⚠️Engineering Note

Feeder-Link Cost of Coherent Macro-Diversity

Coherent joint transmission multiplies the feeder-link data rate by the cluster size $M$ : each satellite needs a copy of every user's downlink payload. For a constellation serving $N_u$ users at $\bar{R}$ bits/s per user, the aggregate feeder-link load is

$R_{\text{feeder}} \geq M \cdot N_u \cdot \bar{R} + \text{overhead},$

whereas the single-satellite architecture has $R_{\text{feeder}} = N_u \bar{R}$ . For a dense constellation with $N_u \approx 10^6$ users and $\bar{R} = 10$ Mb/s, the $M = 4$ cell-free mode needs $\approx 40$ Tb/s of feeder-link throughput, versus $10$ Tb/s for the baseline. Operators meet this through multiple gateways, parallel Ka-/Q-band feeder links, and — in next-generation proposals — optical inter-satellite links (ISLs) that allow data to flow directly between satellites without touching the ground. The ISL option is the enabler for "in-orbit" cell-free operation and is the subject of active 3GPP study items for Release 19.

Two practical mitigations reduce the feeder-link penalty:

Asymmetric clustering. Only the top $M' < M$ satellites carry the full payload; the remaining $M - M'$ carry only a small auxiliary stream used for diversity. This reduces the feeder-link cost to roughly $M' \bar{R}$ per user.
Dynamic cluster sizing. Increase $M$ only during rain events or low-elevation passes; use $M = 1$ during clear-sky zenith crossings. The average cost sits well below the worst-case $M \cdot N_u \bar{R}$ .

Practical Constraints

•
Feeder-link load scales linearly with cluster size $M$
•
Gateway diversity required for single-point-of-failure tolerance
•
ISL-based distribution reduces ground-station load but requires on-board switching

📋 Ref: 3GPP TR 38.821 §7.3 (NTN gateway architecture), draft TR for Release 19

,

Common Mistake: Incoherent Combining Wastes Most of the Gain

Mistake:

A common simplification in early NTN simulation studies is to assume that multiple satellites "help" by incoherently combining their independent signals at the receiver — i.e. adding power without phase alignment. Under this assumption the SNR gain of $M$ satellites is $10 \log_{10} M$ dB at best, and the benefit of macro-diversity looks modest.

Correction:

Incoherent combining discards the phase-alignment term that makes Theorem TMacro-Diversity SNR Gain with Coherent Combining linear in $M$ rather than logarithmic. The linear gain is the actual reward for paying the synchronization cost. Incoherent combining underestimates the achievable SNR gain by up to a factor of $M$ and therefore dismisses coherent macro-diversity as "not worth it" — the opposite of the paper's conclusion. When evaluating cell-free LEO proposals, insist on a phase-coherent combining model or discount the results accordingly.

Quick Check

A terminal is served by $M$ LEO satellites with equal path loss and per-satellite transmit power $P_t / M$ (total budget held constant). Under coherent joint transmission with matched precoding, the post-combining SNR scales as ...

$\text{SNR} \propto \sqrt{M}$

$\text{SNR} \propto \log M$

$\text{SNR} \propto M$

$\text{SNR} \propto M^2$

Correction:

\text{SNR} \propto M

Theorem TMacro-Diversity SNR Gain with Coherent Combining: coherent combining with matched precoders yields amplitude $\sum_m \sqrt{\beta_{m} P_m N_t}$ at the receiver, which squared and with $P_m = P_t/M$ and uniform $\beta_{m}$ gives exactly $M$ times the single-satellite SNR. The key is phase-coherent combining; without it the scaling drops to $\log M$ (selection) or $\sqrt{M}$ (equal-gain incoherent).

Key Takeaway

Macro-diversity across simultaneously visible satellites is the first-order win of cell-free LEO. Coherent joint transmission across $M \approx 4$ satellites gives a $\approx 6$ dB SNR gain and an order-of-magnitude reliability improvement at the $99\%$ outage level, at the cost of an $M$ -fold increase in feeder-link load. The engineering trade-off is between air-interface performance and space-to-ground fronthaul bandwidth. The Buzzi–Caire–Colavolpe paper argues that for 6G NTN this trade-off favours the cell-free side, especially once optical inter-satellite links are deployed.

Macro-Diversity from Multiple Satellites

The Cell-Free Idea, in Orbit

Definition: Macro-Diversity Signal Model

Theorem: Macro-Diversity SNR Gain with Coherent Combining

Per-satellite receive SNR with matched precoder

Coherent sum and noise

Substitute per-satellite power

CommIT Contribution: Cell-Free Macro-Diversity in LEO NTN

Macro-Diversity SNR Gain vs Cluster Size MMM

Parameters

User-Centric LEO Cluster Selection

Example: Macro-Diversity Against Rain Fade

Single-link outage SNR

Coherent sum across $M = 4$

Net margin

Cluster Geometry for Cell-Free LEO

Feeder-Link Cost of Coherent Macro-Diversity

Common Mistake: Incoherent Combining Wastes Most of the Gain

Quick Check

Key Takeaway

Definition:
Macro-Diversity Signal Model

Macro-Diversity SNR Gain vs Cluster Size $M$