Hybrid On-Board/On-Ground Precoding for HTS

• Hybrid on-board/on-ground precoding is a satellite architecture that splits the overall precoding between a fixed on-board beamforming stage and an adaptive ground precoding stage to balance interference mitigation and feeder bandwidth use.
• The approach employs a two-stage optimization where the satellite implements a low-complexity, slowly updated transformation while the gateway dynamically manages interference using accurate CSI.
• This architecture reduces payload complexity and feeder-link demands while achieving near-optimal performance in interference-limited, aggressive frequency reuse high-throughput satellite systems.

Searching arXiv for the cited papers to ground the article in current literature. {"^^^^^^^^7query7 OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^^^^^"," OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7query7^^^^} I’ll also search by title in case identifier lookup is not supported directly. {"^^^^^^^^7query7 High Throughput Satellite: Hardware Foundation, Resource Allocation, and Precoding^^^^^^^^7\7 OR ^^^^^^^^7\7 Multiuser Hybrid Precoding for Coexistence with LEO Satellite Communication^^^^^^^^7\7 OR ^^^^^^^^7\7 to Rapidly and Robustly Optimize Hybrid Precoding^^^^^^^^7\7 OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7query7^^^^} Hybrid on-board/on-ground precoding is a deployment architecture for multibeam high throughput satellite forward links in which the overall precoder is split between the gateway and the satellite. In the multibeam HTS setting, the satellite typically generates ^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7query7query7^^^^–^^^^7query7query7query7query7^^^^+ spot beams, and aggressive or full frequency reuse causes each user terminal to observe not only its intended beam but also strong co-channel interference from adjacent beams; the forward link therefore behaves as a multiuser MIMO broadcast channel in an interference-limited regime. Hybrid precoding addresses this regime by decomposing the transmit transformation into an on-board beamforming matrix and a ground precoding matrix, so that computationally intensive, adaptive interference mitigation remains on the ground while a simpler, often fixed or slowly varying beamforming stage is implemented on the satellite (&&&^^^^7query7^^^^&&&).

^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^. Multibeam HTS context and the motivation for a split precoder

A standard multibeam HTS architecture comprises a space segment, a ground segment, and a user segment. The space segment includes the multibeam antenna, analog or digital or regenerative transponders, beamforming networks, TWTAs, digital channelizers, and possibly on-board DSP. The ground segment comprises one or several gateway stations, with RF/IF/baseband chains; the baseband is the natural location for complex precoding algorithms. The user segment consists of fixed and mobile user terminals connected through user links, typically in Ku/Ka band, while feeder links between gateways and the satellite are typically in Ka or Q/V band (&&&^^^^7query7^^^^&&&).

The need for precoding follows directly from frequency reuse. HTS systems pursue Tbps-class capacity through many narrow spot beams and aggressive reuse of the same carrier across neighboring beams. Under full reuse, the system becomes interference-limited rather than noise-limited. Precoding is therefore required to shape transmitted signals so that user channel vectors are as orthogonal as possible and to cancel or strongly reduce inter-beam interference at user terminals, thereby enabling acceptable SINR under full reuse (&&&^^^^7query7^^^^&&&).

Within this architecture, hybrid on-board/on-ground precoding occupies an intermediate position between two extremes. Pure on-ground precoding retains a largely bent-pipe satellite and places the full precoder computation on Earth, which preserves algorithmic flexibility but places very high demands on feeder-link bandwidth and, in multi-gateway systems, on calibration and coordination. Pure on-board precoding minimizes feeder-link bandwidth and inter-gateway CSI exchange, but it is constrained by payload mass, power, memory, and digital processing capability. Hybridization arises from the attempt to preserve the advantages of both while softening their respective bottlenecks (&&&^^^^7query7^^^^&&&).

^^^^7max_results7^^^^. Architectural decomposition and deployment taxonomy

The defining mathematical decomposition of hybrid on-board/on-ground precoding is

PRESERVED_PLACEHOLDER_^^^^7query7^^^^

where PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^ denotes the on-board beamforming or combining matrix and PRESERVED_PLACEHOLDER_^^^^7max_results7^^^^ denotes the ground precoder. In this split, the gateway side performs more complex and adaptive processing, while the satellite implements a structured, lower-dimensional, fixed, or slowly updated transformation that supports beam generation, feeder-bandwidth compaction, and coarse interference control (&&&^^^^7query7^^^^&&&).

The survey literature classifies multibeam HTS precoding into single-gateway on-ground precoding, multiple-gateway on-ground precoding, on-board precoding, and hybrid on-board/on-ground precoding. The hybrid case is distinguished not by a specific algorithm, but by this division of labor across Earth and space. The on-board stage may be analog, as in an RF beamforming network, or digital in advanced digital bent-pipe or regenerative payloads. Accordingly, hybrid on-board/on-ground precoding should not be conflated with the separate literature on hybrid analog/digital precoding; the former is a deployment split, although the on-board part may itself use mixed analog/digital hardware (&&&^^^^7query7^^^^&&&).

Deployment	Main processing location	Main stated trade-off
Single-gateway on-ground	Gateway baseband	Full CSI and global precoder, but very high feeder bandwidth demand
Multiple-gateway on-ground	Per-gateway baseband	Reduced feeder load per gateway, but inter-cluster interference and coordination burden
On-board precoding	Satellite payload	Minimal feeder bandwidth and little inter-GW exchange, but severe payload constraints
Hybrid on-board/on-ground	Split across gateway and satellite	Trade-off between feeder load, payload complexity, and near-terrestrial precoding performance

This taxonomy clarifies why hybridization is attractive in GEO HTS. The satellite need not carry the full burden of per-frame optimization, yet the gateway no longer needs to transport all fully precoded beam signals over the feeder link. The survey characterizes hybrid on-board/on-ground precoding as reducing feeder-link load and payload complexity compared with fully terrestrial interference-aware beamforming while achieving performance close to full on-ground precoding (&&&^^^^7query7^^^^&&&).

^^^^7query7^^^^. Signal model, CSI structure, and optimization criteria

The surveyed literature is organized around the multibeam MIMO broadcast model

PRESERVED_PLACEHOLDER_^^^^7query7^^^^

with PRESERVED_PLACEHOLDER_^^^^7\7^^^^ representing the overall downlink channel matrix, PRESERVED_PLACEHOLDER_^^^^7 OR \7^^^^ the data symbol vector, and PRESERVED_PLACEHOLDER_^^^^7 OR \7^^^^ additive noise plus unmitigated interference. In the hybrid case, $\mathbf{W}$ is factorized as $$\mathbf{B}_{\text{onboard}}\mathbf{P}_{\text{ground}}$$. The effective channel encapsulates feeder-link effects, feeder antennas and impairments, on-board feed-beam mapping, and user-link propagation. In multiple-gateway settings, $\mathbf{H}$ becomes block-structured, with inter-cluster terms representing inter-cluster interference (&&&^^^^7query7^^^^&&&).

This model immediately exposes two technical points that are central to hybrid designs. First, feeder-link interference must be modeled jointly with user-link interference in multi-gateway and hybrid systems; the survey explicitly notes M-MMSE and leakage-aware MMSE designs in this context. Second, per-antenna or per-feed power constraints are essential because the precoder must respect TWTA limits on each feed chain. Hybrid designs therefore do not optimize only user-link SINR; they also operate under feeder-link, payload, and amplifier constraints that reshape the effective channel seen by the ground precoder (&&&^^^^7query7^^^^&&&).

CSI acquisition is similarly asymmetric. User-link CSI is obtained at the gateway through feedback, including pilots and return-link measurements following the DVB-S^^^^7max_results7^^^^/S^^^^7max_results7 superframe structure. In single-gateway systems the gateway is typically assumed to have almost global CSI. In multi-gateway systems each gateway has local CSI for its own beams, with partial CSI exchange in some hyper-cluster designs. In hybrid architectures, most CSI is used on the ground to design PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7query7^^^^ and often also PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^, while the satellite does not estimate the full user channel per frame; rather, the on-board matrix is fixed or updated slowly through control signaling (&&&^^^^7query7^^^^&&&).

The principal optimization criteria surveyed for HTS precoding also underpin the hybrid case. These include maximum sum rate,

PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7max_results7^^^^

rate balancing,

PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7query7^^^^

and rate matching,

PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7\7^^^^

SINR constraints in multicast settings and power-efficiency objectives are likewise standard. Hybrid precoding is described as a two-stage optimization: a slow or offline design of PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7 OR \7^^^^ using channel statistics and robustness considerations, followed by fast frame-by-frame adaptation of PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7 OR \7^^^^ to current CSI, user grouping, and per-antenna power constraints (&&&^^^^7query7^^^^&&&).

^^^^7\7^^^^. Design families and representative hybrid schemes

The surveyed design space includes several distinct families. In single-gateway hybrid systems, one line of work uses a fixed on-board beamforming matrix for signal compression and an adaptive ground precoder such as R-ZF or UpConst-MMSE; this reduces feeder bandwidth while leaving most interference mitigation to the gateway. A second line jointly optimizes the on-board matrix and an MMSE ground precoder. Joroughi and co-authors design PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)77^^^^ to minimize SMSE under eigenvalue perturbations, and the survey reports that the resulting hybrid arrangement approaches the performance of full on-ground precoding. Thibault and co-authors evaluate SVD, DFT, and PCA designs for the on-board matrix, with SVD identified as the most robust. A third line emphasizes low-complexity beamspace or sparse constructions, including DPSS beamspace, Bartlett, GSR, and FAPI, the latter being described as yielding low complexity and good orthogonality (&&&^^^^7query7^^^^&&&).

In multiple-gateway hybrid systems, the principal challenge shifts to inter-cluster interference, feeder-link interference, and coordination. One surveyed design combines on-board SVD beamforming with ZF terrestrial precoding, but it requires full inter-gateway CSI. Other surveyed works combine on-board SVD beamforming with MMSE ground precoding to cancel intra- and inter-cluster interference without full CSI exchange, and use a two-level frame-precoding structure in which on-board and ground matrices are updated at different rates. Further examples include a control-theoretic gateway PD-I controller coupled to on-board beamforming for cross-cluster leakage reduction, and a two-level leakage-aware design in which on-board leakage-aware MMSE beamforming mitigates feeder-link interference while an SCA-based multicast precoder on the ground handles user-link interference (&&&^^^^7query7^^^^&&&).

These families share a consistent architectural logic. The on-board matrix is tasked with dimension reduction, beam shaping, feeder-bandwidth compaction, or suppression of cross-cluster leakage, all under severe complexity limits. The ground matrix exploits richer CSI and more abundant compute to realize MMSE-, ZF-, or SCA-type interference mitigation, multicast adaptation, and scheduling. The practical significance of the hybrid split is therefore not merely computational; it is a structural decomposition aligned with the natural asymmetry between gateway baseband resources and payload constraints (&&&^^^^7query7^^^^&&&).

^^^^7 OR \7^^^^. Interaction with resource allocation and interference-aware coexistence

Hybrid on-board/on-ground precoding is tightly coupled to resource allocation. The survey states that bandwidth allocation, power allocation, time-slot assignment, and beam hopping determine the interference pattern and power budget that the precoder must resolve. Frequency-reuse decisions shape the interference matrix; feed and TWTA power budgets define the feasible transmit region; beam-hopping patterns alter the temporal interference structure. Precoding then feeds back into these resource dimensions by reducing effective inter-beam interference, enabling more aggressive frequency reuse, and improving SINR at a given power level so that power or bandwidth can be reallocated more efficiently (&&&^^^^7query7^^^^&&&).

For hybrid schemes specifically, the design of PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^8 is coupled to payload resources such as the number of feeds, the beamforming-network type, and the TWTA map. The survey notes that some hybrid works effectively perform joint beamforming and power control under feeder-bandwidth constraints, even though explicit joint optimization equations are not provided. A plausible implication is that hybrid precoding should be interpreted as part of a broader HTS control stack rather than as an isolated PHY-layer primitive (&&&^^^^7query7^^^^&&&).

A related but distinct strand of literature strengthens this interpretation. In a terrestrial multiuser MIMO coexistence scenario with LEO satellites, an interference-aware hybrid analog/digital precoder is formulated as a sum-rate objective penalized by satellite interference power, with unit-modulus analog constraints and a total transmit power constraint. Although all active precoding in that setting resides on the ground, simulations report that the method improves interference-to-noise power by ^^^^7max_results7max_results7^^^^.^^^^7\7^^^^ dB relative to the best hybrid baseline in interference terms, keeps the INR-threshold violation probability at ^^^^7query7^^^^.^^^^7query7 OR \7^^^^%, and maintains average UE sum-rate within ^^^^7max_results7^^^^.^^^^7 OR \7^^^^9–^^^^7query7^^^^% of the highest-rate hybrid baseline (&&&^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^&&&). This suggests that penalty-based objectives of the form

PRESERVED_PLACEHOLDER_^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^9

can be transplanted into hybrid on-board/on-ground architectures when protected receivers, cross-links, or feeder links must be incorporated directly into the precoder design.

The coexistence result also sharpens a broader point: exact nulling is not the only meaningful target in hybrid hardware. The soft-penalty method in the terrestrial/LEO case is presented as more compatible with hybrid constraints than strict nulling, because hybrid architectures may not realize exact zeros under unit-modulus, limited-RF-chain, or distributed-control restrictions (&&&^^^^7id:(Chen et al., 1 Aug 2025) OR id:(Razavi et al., 3 Jun 2025) OR id:(Lavi et al., 2023)7^^^^&&&). In the satellite setting, the same reasoning aligns naturally with leakage-aware MMSE and two-level hybrid designs surveyed for HTS (&&&^^^^7query7^^^^&&&).

^^^^7 OR \7^^^^. Practical constraints, adjacent learning methods, and open directions

The implementation bottlenecks for hybrid on-board/on-ground precoding are primarily physical rather than purely algorithmic. The survey emphasizes payload mass, power consumption, limited on-board memory bandwidth, TWTA nonlinearities, and beamforming-network complexity. Fully digital on-board precoding for hundreds of beams is described as not yet practical, which motivates low-complexity on-board matrices such as SVD, DPSS, GSR, and FAPI. In multi-gateway systems, time and frequency synchronization across gateways is critical because on-board beamforming may combine signals from several gateways; imperfect synchronization degrades interference cancellation and leaves residual inter-cluster interference. Feeder links in Q/V band provide additional bandwidth but suffer severe rain attenuation and outages, so hybrid schemes that reduce feeder load mitigate rather than eliminate feeder vulnerability (&&&^^^^7query7^^^^&&&).

CSI accuracy is another limiting factor. Long GEO RTT creates channel aging, and hybrid architectures are explicitly designed so that the satellite does not require exact per-frame CSI. Fixed or slowly updated on-board matrices are favored because they tolerate statistical CSI and reduce signaling overhead. The survey also notes robust multicast precoding under CSI errors and identifies channel aging as a motivation for AI-based channel prediction and low-latency beamforming, while also stating that AI/ML has not yet been practically integrated in operational satellite systems (&&&^^^^7query7^^^^&&&).

Adjacent hybrid-precoding work on learnable optimization provides one concrete methodological direction. In a wideband multiuser hybrid analog/digital setting, unfolded projected-gradient and projected conceptual mirror-prox optimizers are trained by learning iteration-dependent step sizes, while preserving the optimizer’s interpretable structure. Numerical results show that this approach can use over ten times less iterations than conventional optimization with shared hyperparameters while achieving similar and even improved sum-rate performance, and the same framework is extended to a robust minimax design under bounded CSI errors (&&&^^^^7max_results7^^^^&&&). The same source explicitly maps the frequency-flat analog precoder to an on-board RF beamforming stage and the per-subcarrier digital precoders to ground-side or more capable on-board baseband processing, which makes the method directly relevant as a fast and robust optimizer for split architectures (&&&^^^^7max_results7^^^^&&&).

The open directions identified for HTS are correspondingly system-level. They include better CSI acquisition and prediction under GEO RTT, scalable multi-gateway coordination with reduced inter-gateway CSI exchange, further development of sparse, beamspace, and low-rank on-board processing, tighter integration with bandwidth, power, and beam-hopping allocation, and reliable operation over Q/V-band feeder links through gateway diversity, switching, outage prediction, and intelligent routing (&&&^^^^7query7^^^^&&&). Taken together, these directions position hybrid on-board/on-ground precoding as an architectural compromise in which beamforming, feeder design, payload flexibility, gateway deployment, and interference management are jointly optimized rather than treated as separable subproblems.