Source-linked AI summary
LLM-Assisted Hierarchical Interference Control for Satellite-Terrestrial Integrated Networks
Heying Cao, Hao Yuan, Xing Zhang
TL;DR
LEO interference varies over time and space, creating a tradeoff between protecting terrestrial receivers and avoiding overly conservative controls that waste LEO resources. A hierarchical controller combines QoS-constrained power control and adaptive sensing with LLM ranking of screened bands and deterministic safety verification. The proposed method achieves 320.97 Mbps with 100% QoS satisfaction and keeps INR below the −12.2 dB threshold; power-only and muting achieve 159.97/146.89 Mbps and 50.8%/46.4% QoS satisfaction, respectively.
Problem
LEO interference varies over time and space, creating a tradeoff between protecting terrestrial receivers and avoiding overly conservative controls that waste LEO resources.
Source: §I · p. 1Method
A hierarchical controller combines QoS-constrained power control and adaptive sensing with LLM ranking of screened bands and deterministic safety verification.
Source: §I · p. 1Results
The proposed method achieves 320.97 Mbps with 100% QoS satisfaction and keeps INR below the −12.2 dB threshold; power-only and muting achieve 159.97/146.89 Mbps and 50.8%/46.4% QoS satisfaction, respectively.
Source: B. Aggregate INR Performance · p. 4Takeaways & Limitations
Power reduction or satellite muting can protect the TN receiver but significantly sacrifices LEO service quality compared with the proposed method.
Source: B. Aggregate INR Performance · p. 4
Abstract
from arXiv · showhide
Time-varying aggregate interference from low-Earth-orbit (LEO) satellites challenges terrestrial-network (TN) protection, whereas conservative control degrades LEO quality of service (QoS). This letter proposes a hierarchical framework combining geometry-based interference estimation, QoS-constrained power control, adaptive energy detection, large language model (LLM)-assisted candidate ranking, stay-or-switch decisions, and target-band selection, with deterministic verification. The current band is retained only when both spectrum-hole and power-feasibility conditions hold, with periodic reevaluation. Over a 300-s horizon with 0.5-s sampling, the proposed method keeps the aggregate INR from exceeding the $-12.2$~dB threshold and achieves a time-averaged mean LEO-link rate of $320.97$~Mbps with $100\%$ QoS satisfaction. It markedly outperforms power-only and satellite-muting in jointly protecting the TN and preserving LEO QoS.
I. INTRODUCTION
LEO–TN spectrum sharing must manage time-varying aggregate interference without sacrificing LEO resources through overly conservative protection. The proposed framework combines hierarchical power control and adaptive sensing with LLM-assisted band decisions and deterministic verification.
I. INTRODUCTION
- Orbital mobility makes aggregate LEO interference time-varying and spatially non-uniform, while protection must balance TN requirements against LEO resource use.Prior approaches do not fully integrate deterministic orbital geometry with slot-level assessment or physical-layer constraints for intelligent control.Source: §I · p. 1
- The framework prioritizes current-band QoS-constrained power control, screens alternative bands when needed or periodically, then uses LLM ranking and deterministic safety verification.The LLM jointly selects whether to stay or switch and the target band using sensing, INR/ACIR, and LEO QoS information.Source: §I · p. 1
A. Terrestrial Network
- The TN is modeled as clustered base stations uniformly distributed in a disk, with an offset protected receiver and nearest serving base station.The link uses path-loss exponent 3 and Nakagami-m fading with m = 1, corresponding to Rayleigh fading.Source: A. Terrestrial Network · p. 1
B. LEO Satellite System
- LEO satellites follow a Walker constellation, and visibility is determined by elevation; their beams point nadir and use the ITU-R S.1528 antenna pattern.The protected TN receiver is omnidirectional; at 2 GHz with 35.16 dBi peak satellite gain, the stated 3 dB beamwidth is approximately 3°.Source: B. LEO Satellite System · pp. 1–2
C. Aggregate Interference
Aggregate interference is calculated for a selected band from satellite transmission and propagation terms, with band coupling and an INR-based protection limit.
C. Aggregate Interference
- Satellite transmit power and free-space channel gain determine each satellite’s contribution toward the protected TN receiver.Satellite power is bounded by Pmax, and the channel gain accounts for propagation to the receiver.Source: C. Aggregate Interference · p. 2
- The applicable ACIR is converted to a linear band-coupling factor for the band-dependent aggregate interference.The stated factor is 10^(−ACIR/10) and lies in [0, 1].Source: C. Aggregate Interference · p. 2
- The maximum allowable aggregate interference is set from the INR threshold, using the TN receiver noise power and threshold in dB.Source: C. Aggregate Interference · p. 2
D. Problem Formulation
Problem (6) jointly balances aggregate-rate, INR, spectrum-hole-quality, and band-change terms while enforcing TN protection, LEO QoS, power, and spectrum-hole constraints. The proposed solution is a low-complexity heuristic without a claim of global optimality.
D. Problem Formulation
- The joint control problem balances normalized aggregate-rate, INR-cost, and spectrum-hole-quality terms against a frequency-distance band-change cost.The band-change cost is zero when retaining the current band and increases with the frequency distance of a reconfiguration.Source: D. Problem Formulation · p. 2
- Constraints (6b)–(6e) enforce TN protection, LEO QoS, transmit-power limits, and spectrum-hole selection, excluding bands detected as occupied.The spectrum-hole-quality utility distinguishes sensing quality among the remaining holes.Source: D. Problem Formulation · p. 2
- The proposed method uses a low-complexity heuristic to solve the joint power–spectrum problem without claiming global optimality.The nonnegative weights ωR, ωI, ωH, and ωS set the objective-term weights.Source: D. Problem Formulation · p. 2
III. PROPOSED METHOD
The hierarchical controller checks the current band, screens alternatives when needed, and uses an LLM to select whether to stay or switch and which band to target. Deterministic power control and safety verification precede execution.
III. PROPOSED METHOD
- The algorithm takes the current band, band set, and epoch observations as input and returns a feasible power-band pair or BLOCKED.The high-level sequence is summarized as current-band evaluation, alternative screening, LLM selection, and deterministic verification.Source: §III · pp. 2–3
- The controller returns the current-band configuration when its power and hole checks pass and periodic reevaluation is not due.Otherwise, it screens alternative bands, has the LLM rank candidates and select a stay-or-switch action and target, then verifies power feasibility before execution.Source: §III · pp. 2–3
- A proposed stay is overridden when current-band power is infeasible, while a periodic switch may be suppressed by the stated hysteresis condition.The algorithm tries the selected candidate first, then the remaining screened candidates, and returns BLOCKED if none is feasible and the current band is infeasible.Source: §III · p. 3
A. Current-Band Power Control
Current-band control first tests aggregate interference under maximum satellite power, then applies QoS-aware power control if needed. Dominant interferers are reduced first, and sensing is activated if TN protection cannot be met within QoS-aware power bounds.
A. Current-Band Power Control
- The controller retains the current band when maximum-power aggregate interference is within Icap and the maximum-power configuration satisfies LEO QoS.Otherwise, it performs QoS-aware power control on the current band.Source: A. Current-Band Power Control · p. 3
- Minimum transmit power is derived from the Shannon rate constraint to satisfy the LEO QoS requirement Rmin.Power control is constrained by the resulting QoS-aware lower bounds.Source: A. Current-Band Power Control · p. 3
- Visible satellites are ordered by interference contribution, and dominant interferers are reduced first; sensing starts if TN protection remains infeasible without violating QoS-aware lower bounds.Source: A. Current-Band Power Control · p. 3
B. Adaptive Spectrum Sensing
Adaptive spectrum sensing examines alternatives when current-band power control is infeasible, using received-signal energy and an adaptive threshold. The detector distinguishes TN activity from noise and updates its background estimate using trusted noise-only samples.
B. Adaptive Spectrum Sensing
- When current-band power control is infeasible, alternative bands are examined using a received sensing signal modeled as TN signal plus AWGN.The binary hypotheses distinguish noise-only H0 from TN-signal-plus-noise H1.Source: B. Adaptive Spectrum Sensing · p. 3
- The sensing SNR characterizes TN-activity detectability at the sensing receiver and is distinct from the LEO-to-TN INR constraint.An occupied band with prescribed sensing SNR γsen has PTN,j = σ^2_w,j10^(γsen/10).Source: B. Adaptive Spectrum Sensing · p. 3
- Each band’s background energy is estimated from a trusted H0 sliding window, with estimated background mean and standard deviation used in the adaptive threshold.The threshold is set given a target false-alarm probability.Source: B. Adaptive Spectrum Sensing · p. 3
- Only H0 samples update the background window, while H1 samples are excluded to avoid threshold chasing; the detector then forms the available spectrum-hole set.Source: B. Adaptive Spectrum Sensing · p. 3
C. Prompt-Engineered LLM-Assisted Spectrum Decision
The prompt-engineered spectrum-selection design gives the LLM a structured decision task, then sends its structured output for deterministic safety verification.
C. Prompt-Engineered LLM-Assisted Spectrum Decision
- The prompt has four parts, including domain constraints and a task to choose whether to stay on the current band or switch to a spectrum hole.Occupied bands are prohibited, and staying is disallowed when power control is infeasible and the current band is unsafe.Source: C. Prompt-Engineered LLM-Assisted… · p. 3
- Slot-level observations supply interference estimates, sensing results, QoS and power feasibility, candidate bands, and switching cost to the decision prompt.The prompt also declares the TN protection threshold and ACIR-based interference estimates.Source: C. Prompt-Engineered LLM-Assisted… · p. 3
- The LLM returns an action, target band, risk label, and explanation, which are passed to deterministic safety verification before execution.Fig. 3 depicts the proposed interference-control procedure as a flowchart.Source: C. Prompt-Engineered LLM-Assisted… · p. 3, Figure 3 · p. 4
D. Safety Verification
Deterministic verification accepts a ranked candidate only when it jointly satisfies the constraints in Problem (6).
D. Safety Verification
- If the selected band fails verification, other candidates are tested; if none passes, the slot is marked blocked.Acceptance requires all constraints in Problem (6) to be jointly satisfied.Source: D. Safety Verification · p. 4
A. Experimental Setup
The experiments compare the proposed method with fixed-band power control and satellite muting, evaluating TN protection and LEO QoS. The proposed method stays below the INR threshold while preserving higher LEO rates and QoS satisfaction.
A. Experimental Setup
- The setup uses DeepSeek-R1:7B with deterministic safety verification and compares against no control, fixed-band power-only control, and fixed-band satellite muting.No control uses Pmax on the reference band; the two controlled baselines reduce power or successively mute strong interferers.Source: A. Experimental Setup · p. 4
B. Aggregate INR Performance
- 320.97 Mbps with 100% QoS satisfaction accompanies INR below −12.2 dB; power-only and muting yield 159.97 and 146.89 Mbps, with 50.8% and 46.4% QoS satisfaction.Power-only and muting suppress INR by reducing power or disabling satellites, but significantly sacrifice LEO service quality.Source: Figure 5 · p. 4, Figure 4 · p. 4, B. Aggregate INR Performance · p. 4
C. Spectrum Sensing and Frequency Selection
- The proposed method selects bands that satisfy TN protection while accounting for sensing availability and switching cost, rather than always choosing the minimum-INR band.The time-frequency INR distribution changes as orbital movement changes the dominant interfering satellite and ACIR reference band.Source: C. Spectrum Sensing and Frequency… · pp. 4–5, Figure 6 · p. 5
D. Sensitivity Analysis
Sensitivity tests show that the method keeps INR below the protection threshold across the tested QoS and transmit-power settings. Detector performance improves with sensing SNR and target false-alarm probability, while the conclusion frames the approach as balancing TN protection, LEO QoS, and switching overhead.
D. Sensitivity Analysis
- The method keeps INR below the protection threshold as Rmin or Pmax varies; larger Rmin raises the QoS-aware power lower bound and may increase switching need.Higher Pmax makes potential aggregate interference stronger and TN protection more challenging.Source: D. Sensitivity Analysis · p. 5
E. Energy-Detection Characterization
- At P_fa=0.01, detection probability rises from 1.71% at γ_sen=−16 dB to 52.81% at −8 dB, illustrating the low-SNR sensing tradeoff and need for deterministic verification.Detection probability increases with both sensing SNR and target false-alarm probability.Source: E. Energy-Detection Characterization · p. 5
V. CONCLUSION
- The framework combines geometry-based assessment, QoS-aware power control, adaptive sensing, constrained spectrum selection, and deterministic verification to balance TN protection, LEO QoS, and switching overhead.Source: §V · p. 5