Multi-Satellite MIMO Overview

• Multi-satellite MIMO is a distributed satellite system that uses cooperative clusters of antennas to achieve spatial diversity and multiplexing gains.
• It employs dynamic clustering, joint beamforming, and robust precoding strategies to improve spectral efficiency, link capacity, and service reliability.
• The approach integrates centralized and decentralized processing, along with learning-based optimization, to effectively manage synchronization, channel estimation, and handover challenges.

Multi-satellite MIMO (multiple-input multiple-output), also known as distributed satellite MIMO or cooperative satellite MIMO, refers to the use of a cluster or network of spatially separated satellites—each equipped with transmit/receive antenna arrays or a single antenna—to jointly serve user terminals, relay multiple data streams, and achieve spatial diversity and multiplexing gains analogous to terrestrial MIMO but in the satellite domain. The concept addresses the scaling of capacity, link reliability, and spectral efficiency in ultra-dense low Earth orbit (LEO) constellations, GEO multi-satellite deployments, and ground–satellite feeder links, exploiting inter-satellite links (ISLs) and distributed processing for coordinated transmission.

1. Architectures and Channel Models

Multi-satellite MIMO architectures span a range from tightly clustered, centralized schemes (with a local “network controller” or “super-satellite node” for coordination), to user-centric, dynamically clustered, and even fully decentralized arrays. In LEO constellations, satellites are grouped into clusters of $M$ “satellite access points” (SAPs), all jointly serving the same set of $K$ user terminals (UTs). In each cluster, SAPs are interconnected—typically via high-speed ISLs—to a central processing unit (CPU) that orchestrates pilot assignment, beamforming, handover, and resource allocation (Abdelsadek et al., 2022, Humadi et al., 2024).

The core channel model is often Rician and block-fading: for user $k$ and satellite $m$,

$$h_{m,k} = \sqrt{L_{m,k}} \left( \sqrt{\kappa_{m,k}/(\kappa_{m,k}+1)}\,e^{j\phi_{m,k}} + \sqrt{1/(\kappa_{m,k}+1)}\,\tilde h_{m,k} \right),$$

where $L_{m,k}$ captures distance loss, shadowing, and antenna pattern; $\kappa_{m,k}$ is the Rician K-factor; phase $\phi_{m,k}$ models LoS coherence; and $\tilde h_{m,k}$ is the small-scale NLoS fade (Abdelsadek et al., 2022). The aggregate MIMO channel $H$ is constructed by stacking these elements for all user–satellite pairs, with large-scale (path-loss, K-factor) and small-scale components.

In ground feeder links, “near-field” MIMO effects emerge for kilometer-scale distributed arrays with satellites at hundreds to thousands of kilometers, giving rise to spherical-wave and radiative near-field models (Vennam et al., 12 Aug 2025). For multi-stream downlink and uplink, channels are typically modeled as rank-1 (or structured rank-deficient in the presence of synchronization errors or delay spread), with steering vectors reflecting the geometry of each link (Wang et al., 26 Dec 2025, Ramezani et al., 13 Mar 2026).

2. Distributed Transmission, Cooperation, and Clustering

Effective exploitation of multi-satellite MIMO relies on cooperative architectures capable of sharing data and CSI between satellites through ISLs. Centralized schemes collect CSI (instantaneous or statistical) and user data at a master node (“super-satellite,” gateway, or ground hub) for joint beamforming/precoding and user scheduling (Abdelsadek et al., 2022, Wang et al., 12 May 2025). Distributed and partially decentralized approaches offload computation to “edge” satellites or allow each node to compute or infer precoders based on periodic state exchange (e.g., position, attitude, power budgets) and shared slow-varying CSI (Cao et al., 21 Mar 2026).

User-centric clustering constitutes a scalable approach: each user is dynamically assigned a cluster of satellites within its visibility set (by elevation cutoff or path strength), with further clustering updates as satellite motion or visibility changes. Initial cluster selection can be based on maximum channel gain, maximum service time, or other heuristics. The average serving cluster size is kept small—e.g., 3–7 satellites per user in practical configurations—yielding most of the spectral efficiency gains of fully cooperative multi-satellite systems, but at much reduced backhaul and computational overhead (Humadi et al., 2024). The “full-cooperative” mode corresponds to all satellites jointly serving all users, and is typically used as a theoretical upper bound in performance studies.

Decentralized architectures leverage local inference at each satellite, supported by dual-branch tensor-equivariant neural networks that aggregate both local and global context from low-rate state exchange, reducing inter-satellite signaling overhead by over 90% compared to fully centralized WMMSE solvers (Cao et al., 21 Mar 2026).

3. Precoding, Resource Allocation, and Robust Transmission

Central to multi-satellite MIMO is the design of coherent joint precoding across distributed, moving satellites. With accurate CSI and synchronization, classical linear precoders are extended: zero-forcing (ZF), regularized ZF (RZF), and weighted minimum mean square error (WMMSE) approaches are all adapted for the distributed setting, subject to per-satellite power constraints. Each satellite $K$0 applies a transmit vector $K$1 to its symbol $K$2, with beamforming based on steering vectors and long-term or instantaneous CSI (Bakhsh et al., 2024, Wang et al., 26 Dec 2025, Ramezani et al., 13 Mar 2026).

When only statistical CSI is available (due to latency, Doppler, or CSI feedback constraints), statistical-CSI-aware designs leverage long-term channel statistics (path loss, angular spread, Rician factor, etc.) for robust precoding (Jo et al., 15 Aug 2025, Cao et al., 21 Mar 2026). Weighted sum-rate optimization problems are then recast via deterministic upper bounds, covariance decomposition (CDWMMSE), and iterative block-coordinate methods, often yielding closed-form per-iteration updates (Wang et al., 12 May 2025, Wang et al., 26 Dec 2025, Ramezani et al., 13 Mar 2026).

Robustness to synchronization errors (excess delays, Doppler uncertainty, phase misalignments) is addressed via phase shift-aware or compensation-aware precoders and robust MSE minimization that incorporates the distribution of the CSI error, particularly critical in distributed massive MIMO LEO clusters given satellite motion and finite ISL delays (Abdelsadek et al., 2022, Omid et al., 2024).

Power allocation and handover management present additional challenges. The D-JPAHM framework jointly optimizes power allocation and handover in a cross-layer MINLP to maximize both network throughput and service continuity, implemented by metaheuristic (genetic algorithm) search and deep-learning-based surrogate models for real-time deployment (Abdelsadek et al., 2022).

4. Multiplexing Gains, Spectral Efficiency, and Capacity

Multi-satellite MIMO delivers substantial improvements in link capacity, reliability, and spatial multiplexing. In downlink, properly designed distributed massive MIMO clusters can achieve 2–3× spectral efficiency and 50–100% increases in service time compared to single-satellite handover schemes (Abdelsadek et al., 2022). User-centric dynamic clustering with phase-aware precoding offers spectral efficiency gains close (within 5%) to full-cooperative baselines while dramatically reducing signaling (Humadi et al., 2024).

Capacity increases nearly linearly with the minimum of total satellite transmit antennas and UT receive antennas. For Rayleigh/Rician channels,

$K$3

with capacity scaling as $K$4 at high SNR for $K$5 dual-polarized satellites and $K$6 receive antennas (Adeogun, 2014). In uplink, clusters of $K$7 cooperating LEO satellites can support $K$8 the conventional single-satellite capacity for handheld devices, with sub-1% error rates when $K$9 (Omid et al., 2023).

Beamspace MIMO extends these benefits, combining codebook-based earth-moving beamforming with low-dimension digital precoding: iterative and closed-form CDWMMSE beamspace solutions achieve up to 95–99% of “full” MIMO capacity at a fraction of computational complexity (Wang et al., 26 Dec 2025).

5. Synchronization, Channel Estimation, and Implementation

Synchronization among distributed satellites is mandatory for coherent joint transmission. Solutions include closed-loop carrier frequency and phase tracking via reference tones, adaptive control loops, and out-of-band common reference signaling. In a field trial with two co-located GEO satellites, sub-10-degree phase error after 250 ms round-trip compensation was found sufficient for practical ZF precoding, with off-the-shelf DVB-S2x receivers (Storek et al., 2020).

Channel estimation may utilize out-of-band common reference signals (CRS), orthogonal training, and large-scale message-passing for grant-free access (as in MIMO-OTFS) (Shen et al., 2024). The trade-off between pilot overhead and channel sparsity is managed by parameterizing channels with basis expansion models tuned to the structured angular–delay sparsity of satellite links. In distributed settings, block-sparse Bayesian learning, structured expectation propagation (AEP), and message-passing algorithms enable effective per-satellite and joint estimation, with centralized and distributed AEP modes typically matching in performance after one or two soft symbol exchanges (Shen et al., 2024).

6. Algorithmic Scalability, Learning-Based Approaches, and Complexity

Multi-satellite MIMO demands scalable, real-time optimization across potentially hundreds of nodes and users. Algorithmic advances include tensor-equivariant neural networks and dense transformer networks that preserve permutation symmetry under user and satellite indices, generalizing across numbers of users, satellites, and array sizes (Cao et al., 21 Mar 2026, Wang et al., 12 May 2025). Surrogate learning-based WMMSE and CDWMMSE models, trained on synthetic/channel-traced data, achieve near-optimal precoding and allocation at dramatically reduced runtime and communication cost, with full 3D scalability and robust performance in dynamic LEO settings.

Heuristic, closed-form, and model-driven designs (e.g., location-informed beam assignment, non-iterative beam domain precoding) remain attractive for their low computational footprint and good (within 10–20%) performance compared to more complex iterative or learning-based algorithms (Wang et al., 26 Dec 2025).

7. Experimental Results, Design Insights, and Applications

Over-the-air field demonstrations, e.g., streaming two video channels via ZF-precode dual-GEO satellites, validate the feasibility of real-world satellite MIMO (Storek et al., 2020). Distributed massive MIMO clusters in LEO simulation yield service time and spectral efficiency gains of 2–3× over single-satellite and best-channel switching baselines (Abdelsadek et al., 2022). Near-field ground feeder arrays with kilometer-scale apertures and 16 panel arrays (ArrayLink) achieve dish-class gain ($k$048 dBi) while supporting up to four spatial streams at 500 km under rigorous real-world measurement (Vennam et al., 12 Aug 2025).

Design recommendations include prioritizing dual-polarization and ensuring the number of ground receive antennas matches the total transmit degrees of freedom; employing user-centric clustering to minimize power and backhaul; limiting cluster size to control overhead/latency; and exploiting robust compensation for phase/delay errors. Applications range from LEO mega-constellation broadband, mobile direct-to-device, IoT random access, backhaul feeder links, to high-fidelity semantic communication and relaying.

---

References

• (Abdelsadek et al., 2022) Distributed Massive MIMO for LEO Satellite Networks
• (Vennam et al., 12 Aug 2025) Satellites are closer than you think: A near field MIMO approach for Ground stations
• (Humadi et al., 2024) Distributed Massive MIMO System with Dynamic Clustering in LEO Satellite Networks
• (Cao et al., 21 Mar 2026) Deep Learning-Based Multi-Satellite Massive MIMO Transmission: Centralized or Decentralized?
• (Storek et al., 2020) Multi-Satellite Multi-User MIMO Precoding: Testbed and Field Trial
• (Adeogun, 2014) Capacity and Error Rate Analysis of MIMO Satellite Communication Systems in Fading Scenarios
• (Wang et al., 26 Dec 2025) Multi-Satellite Multi-Stream Beamspace Massive MIMO Transmission
• (Omid et al., 2023) Space MIMO: Direct Unmodified Handheld to Multi-Satellite Communication
• (Shen et al., 2024) Massive MIMO-OTFS-Based Random Access for Cooperative LEO Satellite Constellations
• (Ramezani et al., 13 Mar 2026) Joint and Streamwise Distributed MIMO Satellite Communications with Multi-Antenna Ground Users
• (Omid et al., 2024) Tackling Delayed CSI in a Distributed Multi-Satellite MIMO Communication System
• (Jo et al., 15 Aug 2025) Multi-Satellite Cooperative MIMO Transmission: Statistical CSI-Aware RSMA Precoding Design
• (Wang et al., 12 May 2025) Statistical CSI-Based Distributed Precoding Design for OFDM-Cooperative Multi-Satellite Systems
• (Wang et al., 9 May 2026) Semantic Communication for Multi-Satellite Massive MIMO Transmission: A Mixture of Cooperative Modes Framework
• (Bakhsh et al., 2024) Multi-Satellite MIMO Systems for Direct User-Satellite Communications: A Survey