Beamforming-based satellite-ground integrated communication antenna system

Abstract

The application discloses a satellite-ground integrated communication antenna system based on beamforming, belongs to the technical field of satellite and ground communication integration, and solves the problems of CSI time lag deviation, mismatch between computing power and real-time performance, and insufficient beamforming robustness in existing high-dynamic satellite-ground communication. The system constructs a satellite-ground-cloud three-level cooperative beamforming architecture, adopts a hierarchical deployment mechanism of operation tasks: the cloud is responsible for global data aggregation and model offline training, the ground solves the benchmark beamforming weight matrix based on minimum-maximum robust optimization, and the satellite terminal completes real-time phase compensation of beam weight through lightweight time lag prediction. The system realizes periodic + event double-trigger full-link closed-loop optimization based on the PPO algorithm, can accurately correct the time lag deviation, avoids beam main lobe deviation and gain loss, significantly improves the link anti-interference ability, and is suitable for high-dynamic, high-speed and large-capacity global communication scenes such as low-orbit satellite constellation.

Description

Technical Field

[0001] This invention relates to the field of satellite-ground communication convergence technology, and in particular to a satellite-ground integrated communication antenna system based on beamforming, which is suitable for high-dynamic, high-speed, and high-capacity satellite-ground communication scenarios such as low-Earth orbit satellite constellations and space-ground integrated information networks. Background Technology

[0002] Satellite-terrestrial integrated communication, as a core component of next-generation mobile communication networks, can overcome the geographical limitations of terrestrial communication and achieve seamless coverage across all regions. Beamforming technology is a key technology for improving the gain of satellite-terrestrial communication links, suppressing interference, and increasing spectrum utilization. Currently, the beamforming schemes of existing satellite-terrestrial integrated communication antenna systems still have many technical shortcomings, making it difficult to meet the real-time, robustness, and stability requirements of high-dynamic satellite-terrestrial communication. Specific problems are reflected in the following aspects: CSI time lag bias leads to beamforming failure: The high dynamic motion of low-orbit satellites and ground terminals causes significant time lag bias in the acquisition of channel status information (CSI) and beam generation in satellite-to-ground communication. The beamforming weights cannot match the real-time channel status, resulting in beam main lobe deviation, gain loss, and increased multi-user interference, which can lead to communication link interruption in severe cases.

[0003] The computing power is not matched with the real-time requirements: the existing system does not adopt a hierarchical deployment mechanism for computing tasks, and the global high-complexity computing and beam-level real-time computing are mixed together, making it difficult to balance global resource scheduling, regional channel processing and on-board rapid response. As a result, the system has low operating efficiency and large response latency.

[0004] Insufficient robustness of beamforming: Traditional solutions lack robust optimization design for time delay, Doppler frequency offset and interference fluctuations, and have not built a full-link closed-loop adaptive optimization mechanism, which cannot dynamically adapt to high dynamic channel changes and has poor anti-interference capability of the link.

[0005] The hardware structure is not sufficiently disclosed and lacks systematic protection: Existing technologies only vaguely mention phased array antenna modules, without disclosing the specific hardware structure, interface specifications and implementation methods of spaceborne and ground antennas. At the same time, there is no targeted protection for the core spaceborne hardware, resulting in insufficient disclosure and a lack of protection scope. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies, such as CSI time delay deviation, mismatch between computing power and real-time performance, insufficient robustness, and insufficient hardware disclosure. It provides a beamforming-based satellite-ground integrated communication antenna system, which fully meets the needs of high-dynamic satellite-ground communication by constructing a three-level collaborative architecture of satellite-ground-cloud, designing a lightweight time delay prediction and real-time compensation module, building a full-link closed-loop optimization mechanism, and clarifying the hardware implementation scheme.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A beamforming-based satellite-ground integrated communication antenna system includes a spaceborne active phased array antenna subsystem, a ground gateway station antenna and processing subsystem, and a cloud intelligent control center. The three are bidirectionally connected through a communication link to construct a three-level collaborative beamforming architecture of satellite-ground-cloud. A hierarchical deployment mechanism for computing tasks is adopted to allocate computing tasks with different real-time requirements to the corresponding levels for execution. The system receives channel measurement data fed back from the terminal and forms a closed-loop optimization mechanism for the entire link to achieve adaptive adjustment of beamforming under high dynamic satellite-ground link.

[0008] This invention adopts a hierarchical deployment mechanism for computing tasks, dividing the system's computing tasks into three levels based on real-time and complexity requirements, thereby achieving optimal matching of computing resources and business needs. The boundaries of each level of tasks are shown in the table below: The cloud-based intelligent control center comprises a data aggregation module, a model offline training module, and a global beam scheduling module. The data aggregation module performs global aggregation and preprocessing of satellite orbit data, ground gateway station channel monitoring data, and terminal multi-dimensional measurement data for the entire constellation. The model offline training module performs offline training and parameter iteration updates of the time-delay prediction model and closed-loop optimization model, and distributes the trained model parameters to ground gateway stations and satellite terminals. The global beam scheduling module dynamically allocates and coordinates beam resources across the entire constellation based on network service requirements and channel conditions.

[0009] The ground gateway station antenna and processing subsystem includes a gateway station antenna array and a regional baseband processing unit. The regional baseband processing unit incorporates a joint robust optimization unit based on the minimum-maximum robust optimization criterion, with the core objective of maximizing the system signal-to-interference-plus-noise ratio (SINR) in the worst case. Simultaneously, it constrains the time delay deviation fluctuation range to not exceed 10% of the round-trip time (RTT) of the satellite-to-ground link and the Doppler frequency offset fluctuation range to not exceed 15% of the terminal's maximum Doppler frequency offset. The reference beamforming weight matrix is ​​then obtained and sent to the satellite-borne terminal. The weight coefficients of the constraints are dynamically allocated by the cloud intelligent control center based on historical SINR feedback.

[0010] The spaceborne active phased array antenna subsystem includes a multi-band common-aperture antenna array, a multi-channel transceiver module, and a spaceborne baseband processing unit. The multi-band common-aperture antenna array employs a multiplexed and nested structure, supporting at least one of the Ka / Q / V / Ku / C bands, with 16~128 array elements. The multi-channel transceiver module incorporates an amplitude and phase calibration unit with a calibration accuracy ≤0.5° phase error and ≤0.3dB amplitude error. The spaceborne baseband processing unit incorporates a lightweight time-delay prediction module and a real-time weight compensation module. It receives the reference beamforming weight matrix from the ground, outputs short-term predictions of the time-delay deviation through the lightweight time-delay prediction model, calculates the phase compensation factor for each antenna channel, performs phase rotation compensation on the reference weight matrix, and drives the antenna array to generate the target shaped beam.

[0011] The system constructs a full-link closed-loop optimization mechanism based on the near-end policy optimization (PPO) algorithm. The input is multi-dimensional measurement data such as signal-to-interference-plus-noise ratio (SINR), channel quality index (CQI), Doppler frequency offset, and signal-to-interference ratio (SIR) fed back by the terminal.

[0012] The state space of the PPO algorithm is a vector composed of multidimensional measurement data [SINR, CQI, SIR, Doppler frequency offset], the action space is [iteration step size, compensation coefficient adjustment amount, feedback weight allocation coefficient], and the reward function is: in, The signal-to-interference-plus-noise ratio (SINR) weighting coefficient. To calculate the complexity weighting coefficients, satisfying ; The difference in average signal-to-interference-plus-noise ratio (SIR) of the terminal before and after this optimization (unit: dB). This is the normalized value of the computational complexity corresponding to this beam optimization, with a value range of [0,1].

[0013] The closed-loop optimization adopts a dual-trigger mode that combines periodic triggering and event triggering: the periodic triggering interval is 0.8 to 1.5 times the round-trip time (RTT) of the satellite-to-ground link; the event triggering is triggered immediately when the SINR fed back by the terminal is lower than the preset threshold or the beam pointing error exceeds the preset threshold.

[0014] RTT (Round Trip Time): The total transmission delay of a signal from a ground gateway station to a satellite and back to the ground gateway station; Time delay bias: the time difference between the acquisition time of channel state information (CSI) and the beam generation time; Doppler frequency shift: The frequency shift of the received signal caused by the relative motion between the satellite and the terminal; SINR threshold: The minimum signal-to-interference-plus-noise ratio (SINR) required to determine the degradation of communication link quality; Beam pointing error: The angle between the actual beam main lobe pointing and the target pointing.

[0015] System hardware structure and interface specifications Spaceborne active phased array hardware structure Array structure: It adopts a multi-layer PCB stack-up design, which consists of a radiating unit layer, a first dielectric substrate layer, a power supply network layer, a second dielectric substrate layer, a metal ground layer, a digital control layer, and a power supply layer from top to bottom. The total thickness is ≤5mm, which meets the requirements of spaceborne lightweighting and integration.

[0016] Dual-frequency common aperture realization: The design adopts unit multiplexing + nested layer design. The same radiating unit is a nested microstrip patch structure. The inner patch operates in the high frequency band (Ka / Q) and the outer patch operates in the low frequency band (Ku / C). Dual-frequency independent beamforming is achieved through independent feeding network without the need to increase the array area.

[0017] Multi-channel transceiver topology: It adopts a topology structure of "1-to-N power divider network → digitally controlled amplitude and phase modulation chip → power amplifier (transmit channel) / low noise amplifier (receive channel) → calibration network → antenna unit"; each channel integrates an independent 6-digit digitally controlled phase shifter (0.5° step) and a 6-digit digitally controlled attenuator (0.1dB step) to achieve high-precision amplitude and phase modulation.

[0018] Interface definition: RF interface: SMP type blind mating interface, operating frequency coverage 10-40GHz, VSWR ≤1.2; Digital control interface: LVDS differential interface, transmission rate ≥1Gbps, supports serial command transmission; Power supply interface: DC-DC isolated power supply interface, input voltage 28V, output voltage ±5V, 3.3V, conversion efficiency ≥90%.

[0019] Ground gateway station antenna structure Array configuration: planar microstrip phased array, with 64-256 array elements, arranged in a square grid, with an element spacing of 0.5 times the operating wavelength; Polarization: Dual polarization (horizontal + vertical), supports polarization diversity reception, polarization isolation ≥25dB; Duplex mode: Frequency division duplex (FDD), transmit / receive frequency band interval ≥ 500MHz, transmit / receive isolation ≥ 60dB; Receiving channel: Each unit has one independent receiving channel, integrating a low-noise amplifier (noise figure ≤ 1.5dB) and a downconverter; Calibration port: One common calibration port is set at the center of the array for full-channel amplitude and phase consistency calibration, with a calibration accuracy of ≤0.3° phase error and ≤0.2dB amplitude error.

[0020] Satellite-to-ground wireless link: Operating frequency band: Ka band (uplink 27.5-30GHz, downlink 17.7-20.2GHz); Modulation method: Adaptive modulation and coding (QPSK / 8PSK / 16QAM / 64QAM), dynamically switching according to channel quality; Synchronization mechanism: Network-wide time synchronization based on GPS second pulse, with a synchronization accuracy of ≤1μs; Heartbeat mechanism: The satellite-borne terminal sends a heartbeat packet to the ground every 100ms. If the heartbeat packet times out after 3 times, the link is considered to be interrupted, triggering the beam switching process.

[0021] Ground-to-cloud link: Adopts a dedicated encrypted fiber optic network with a transmission rate of ≥10Gbps, supports IPSec encrypted transmission, and has a data transmission latency of ≤10ms.

[0022] Data frame format: It adopts a fixed-length frame structure (1024 bytes), which includes a frame header (16 bytes, including timestamp, frame sequence number, and data type), a data payload (992 bytes), and a check bit (16 bytes, CRC32 check).

[0023] System data flow and closed-loop trigger flow Complete data flow path: terminal channel perception → ground gateway station data acquisition → cloud intelligent control center global data aggregation and preprocessing → cloud model offline training and parameter distribution → ground gateway station solves the reference beam weight → satellite-borne end time delay prediction and phase compensation → antenna array generates target beam → terminal receives signal and feeds back measurement data → enters the next round of closed-loop optimization.

[0024] Closed-loop trigger flow: Periodic trigger flow: Arrival at a fixed period → Cloud intelligent control center updates global optimization strategy → Ground gateway station re-solves reference beam weights → Satellite terminal updates compensation coefficients → Beam parameter adjustment; Event trigger flow: Terminal reports SINR below threshold / beam pointing error exceeds limit → Ground gateway immediately reports to cloud → Cloud triggers emergency optimization process → Ground quickly updates baseline weights → Satellite-based terminal provides real-time compensation → Beam parameter correction.

[0025] The beneficial effects of this invention are as follows: 1. Accurately solve the time delay deviation problem: Through onboard lightweight time delay prediction and real-time weight compensation, the beamforming weight matches the real channel state, avoiding beam main lobe deviation and gain loss, and fundamentally solving the beamforming failure problem caused by time delay in high dynamic satellite-to-ground communication.

[0026] 2. Achieve optimal matching of computing power and real-time performance: Adopting a three-level collaborative architecture of satellite-ground-cloud and hierarchical deployment of computing tasks, tasks with different complexities and real-time requirements are carried out in layers, which greatly improves the system's operating efficiency and response speed.

[0027] 3. Significantly improve beamforming robustness: Relying on ground minimum-maximum robustness optimization and PPO dual-trigger closed-loop optimization, it can dynamically adapt to time delay, Doppler frequency offset and interference fluctuations, and the link anti-interference capability and stability are greatly enhanced.

[0028] 4. Sufficient hardware disclosure and comprehensive protection: The specific hardware structure, interface specifications and implementation methods of the spaceborne and ground antennas are clearly disclosed, and the core hardware features are incorporated into the scope of system protection, which meets the requirements of sufficient patent disclosure and achieves comprehensive protection. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the beamforming optimization method for a satellite-terrestrial integrated communication antenna system based on beamforming, as described in this invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The beamforming-based satellite-ground integrated communication antenna system of the present invention is applicable to high-speed, high-dynamic, and high-capacity satellite-ground communication scenarios such as low-orbit satellite constellations and space-ground integrated information networks. The following provides two specific embodiments in combination with different application scenarios and hardware parameter configurations. The technical means not described in detail in each embodiment are all conventional means in the field, and non-core points of the invention will not be elaborated.

[0032] Example 1 See Figure 1 This embodiment is adapted to civilian broadband communication scenarios using low-Earth orbit satellite constellations, and is designed for ground-based low- and medium-speed mobile terminals (vehicle-mounted terminals, fixed broadband terminals) and portable terminals to achieve integrated satellite-ground broadband communication.

[0033] Application scenario parameters: Low Earth orbit satellite speed 7.8km / s, satellite-to-ground link round-trip time (RTT) 6ms; ground terminal moving speed ≤60km / h, mainly distributed in urban and suburban areas, single terminal downlink rate requirement ≥100Mbps.

[0034] System hardware and functional module configuration: Cloud Intelligent Control Center: Deployed on a distributed server cluster in the cloud, it is equipped with a data aggregation module, an offline model training module, and a global beam scheduling module. It completes the global multi-dimensional channel data aggregation of 50 low-orbit satellites in the entire constellation and the ground terminals of the entire network, as well as the offline training and parameter distribution of lightweight LSTM time-delay prediction models and PPO closed-loop optimization models.

[0035] Ground gateway station antenna and processing subsystem: The gateway station antenna array adopts a 128-element Ku-band planar microstrip phased array with dual polarization design and frequency division duplex; the regional baseband processing unit has a built-in joint robust optimization unit based on the minimum-maximum robust optimization criterion, which is responsible for regional channel data processing within a service radius of 200km. The weighting coefficients of the cloud center allocation constraints are: time delay deviation 0.35, Doppler frequency offset 0.3, and interference signal 0.35.

[0036] The spaceborne active phased array antenna subsystem adopts a 32-element Ku / C dual-band common aperture design, with the array surface being a 7-layer PCB stacked structure and a total thickness of 4.2mm. The multi-channel transceiver component has a built-in amplitude and phase calibration unit, with an actual calibration accuracy of 0.3° phase error and 0.2dB amplitude error. The spaceborne baseband processing unit has a built-in lightweight LSTM time delay prediction model and a weight real-time compensation module. The LSTM model parameters are: model size 28KB, single hidden layer structure, number of hidden elements 16, single-step inference time 25μs, and the time delay prediction error is controlled to 50% of the peak-to-peak value of the time delay deviation within the historical 10 sampling periods.

[0037] Communication interface: The satellite-to-ground link uses the Ka band with adaptive modulation and coding; the ground-to-cloud uses an encrypted fiber optic network with a transmission rate of 10Gbps; the data frame uses a fixed-length format of 1024 bytes and is based on GPS second pulse synchronization.

[0038] Beamforming optimization method execution flow: Global multidimensional data acquisition and preprocessing: The cloud intelligent control center collects orbital parameters of the entire constellation of satellites, Ku / C band channel monitoring data of ground gateway stations, and multidimensional measurement data of terminal SINR / CQI / Doppler frequency offset / SIR. After mean filtering for noise reduction and normalization, a standardized dataset is obtained.

[0039] Cloud-Ground Collaborative Benchmark Weight Pre-Optimization: The cloud center completes offline training of a lightweight LSTM model and sends the parameters to the ground gateway station; the gateway station, together with the robust optimization unit, aims to maximize the system SINR in the worst case, constraining the time delay deviation fluctuation ≤ 0.6ms (10% of RTT) and the Doppler frequency offset fluctuation ≤ 15% of the terminal's maximum Doppler frequency offset, and solves the regional benchmark beamforming weight matrix and sends it to the spaceborne unit.

[0040] Onboard time-delay adaptive compensation and beam generation: The onboard LSTM model takes the satellite's real-time orbital parameters, terminal movement speed, and historical time-delay data as inputs and outputs short-term predicted values ​​of time-delay deviation; the weight real-time compensation module calculates the phase compensation factors of 32 antenna channels and performs phase rotation compensation on the reference weight matrix; the amplitude and phase calibration unit eliminates hardware amplitude and phase errors in real time and drives the antenna array to generate target-shaped beams.

[0041] Multi-dimensional sensing closed-loop optimization: The terminal feeds back multi-dimensional measurement data to the gateway station and cloud center. The cloud center optimizes the beam parameters based on the PPO algorithm, setting α=0.65, β=0.35, and the reward function is... The system adopts a dual-trigger update mode of periodic + event, with a periodic trigger interval of 8ms (1.3 times the RTT) and event trigger thresholds of SINR≤12dB and beam pointing error≥4°. After triggering, the system immediately iterates and optimizes and updates the model parameters and weight solution strategy.

[0042] Implementation Results: In this embodiment, the measured beam main lobe pointing deviation was ≤2°, the main lobe gain loss was controlled within 0.8dB, the average SINR of the system was improved by 10dB, the spectrum utilization rate of civilian terminals was improved by 40%, the downlink rate of a single terminal was stable at over 120Mbps, and no link interruption occurred.

[0043] Example 2 This embodiment is adapted to high-speed emergency communication scenarios in integrated space-ground information networks, targeting ground-based high-speed mobile terminals (high-speed rail, low-altitude drones) and emergency communication terminals to achieve integrated space-ground emergency data transmission in highly dynamic scenarios.

[0044] Application scenario parameters: Low-orbit satellite movement speed 7.8km / s, satellite-to-ground link round-trip time (RTT) 4ms; Ground high-speed terminal movement speed: high-speed rail ≤350km / h, low-altitude drone ≤100km / h, single terminal downlink rate requirement ≥500Mbps.

[0045] System hardware and functional module configuration: Cloud Intelligent Control Center: Deployed on a high-performance cloud server cluster, it is equipped with an enhanced data aggregation module and a high-speed model training module. It can complete the high-speed aggregation of global multi-dimensional channel data from 80 low-orbit satellites and emergency communication terminals in the entire constellation. The training iteration efficiency of LSTM and PPO models is improved by 50%, and it supports the second-level delivery of model parameters.

[0046] Ground gateway station antenna and processing subsystem: The gateway station antenna array adopts a 256-element Ka-band high-gain phased array with dual polarization design and frequency division duplex; the regional baseband processing unit has a built-in high-performance joint robust optimization unit, adopts a parallel computing architecture, and the time to solve the reference weight matrix is ​​≤1ms. It is responsible for emergency communication area channel processing within a service radius of 300km. The weight coefficients of the cloud center allocation constraints are: time delay deviation 0.4, Doppler frequency deviation 0.35, and interference signal 0.25.

[0047] The spaceborne active phased array antenna subsystem adopts a 64-element Ka / Q dual-band common aperture design, with a 7-layer PCB stacked structure and a total thickness of 4.5mm. The multi-channel transceiver component has a built-in high-precision amplitude and phase calibration unit, with an actual calibration accuracy of 0.2° phase error and 0.1dB amplitude error. The lightweight LSTM time delay prediction model built into the spaceborne baseband processing unit has the following parameters: model size 45KB, single hidden layer structure, 32 hidden elements, single-step inference time 40μs, and time delay prediction error controlled to 55% of the peak-to-peak time delay deviation within the historical 10 sampling periods.

[0048] Communication interface: The satellite-to-ground link uses the Ka band with adaptive modulation and coding, supporting up to 64QAM; the ground-to-cloud uses an encrypted fiber optic network with a transmission rate of 40Gbps; the data frame uses a fixed-length format of 1024 bytes with a synchronization accuracy of ≤0.5μs.

[0049] Beamforming optimization method execution flow: Global multidimensional data acquisition and preprocessing: The cloud intelligent control center acquires satellite orbit parameters of the entire constellation, Ka / Q band channel monitoring data of ground gateway stations, and multidimensional measurement data of high-speed terminals through high-speed communication links between the satellite and the ground. Wavelet transform denoising and standardization are used for data preprocessing, and the data preprocessing time is ≤2ms.

[0050] Cloud-Ground Collaborative Baseline Weight Pre-Optimization: The cloud center completes offline training of the LSTM model and rapidly distributes the parameters to the ground gateway station; the gateway station, in conjunction with the robust optimization unit, takes maximizing the system SINR in the worst case as the core objective, constraining the time delay deviation fluctuation ≤ 0.4ms (10% of RTT) and the Doppler frequency offset fluctuation ≤ 15% of the terminal's maximum Doppler frequency offset, and quickly solves the baseline beamforming weight matrix through parallel computation and distributes it to the spaceborne unit.

[0051] Onboard time-delay adaptive compensation and beamforming: The onboard LSTM model acquires satellite orbital parameters, high-speed terminal moving speed, and historical time-delay data in real time, and outputs short-term predicted values ​​and confidence intervals of time-delay deviation within 40μs; the weight real-time compensation module quickly calculates the phase compensation factors of 64 antenna channels and performs phase rotation compensation on the reference weight matrix; the amplitude and phase calibration unit performs real-time calibration at the microsecond level to eliminate the impact of hardware errors on beamforming.

[0052] Multi-dimensional sensing closed-loop optimization: The high-speed terminal feeds back multi-dimensional measurement data to the gateway station and cloud center in real time. The cloud center optimizes the beam parameters based on the PPO algorithm, setting α=0.7, β=0.3, and the reward function is... The dual-trigger update mode is set as follows: the periodic trigger interval is 5ms (1.25 times the RTT), and the event trigger threshold is: SINR≤15dB, beam pointing error≥2°. After triggering, the cloud center and the gateway station work together to complete millisecond-level iterative optimization.

[0053] Implementation Results: In this embodiment, for high-speed trains traveling at 350km / h and low-altitude UAVs traveling at 100km / h, the measured beam main lobe pointing deviation is ≤1°, the main lobe gain loss is controlled within 0.5dB, the average SINR of the system is improved by 15dB, the downlink rate of a single terminal is stable at over 600Mbps, and the beamforming parameter update response time is ≤3ms, which fully meets the needs of high-dynamic emergency communication scenarios.

[0054] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A satellite-to-ground integrated communication antenna system based on beamforming, characterized in that, This includes a spaceborne active phased array antenna subsystem, a ground gateway station antenna and processing subsystem, and a cloud-based intelligent control center; The spaceborne active phased array antenna subsystem, the ground gateway station antenna and processing subsystem, and the cloud intelligent control center are bidirectionally connected through communication links to construct a three-level collaborative beamforming architecture of space-ground-cloud. The three-level collaborative architecture adopts a hierarchical deployment mechanism for computing tasks, which allocates computing tasks with different real-time requirements to the corresponding levels for execution. The system receives channel measurement data fed back from the terminal and forms a full-link closed-loop optimization mechanism to achieve adaptive adjustment of beamforming under high dynamic satellite-to-ground links.

2. The system according to claim 1, characterized in that, The hierarchical deployment mechanism for computing tasks is as follows: The cloud-based intelligent management and control center is responsible for global non-real-time high-complexity calculations and full constellation resource scheduling; The ground gateway station antenna and processing subsystem is responsible for regional-level medium real-time calculations and solving for reference beam parameters. The spaceborne active phased array antenna subsystem is responsible for beam-level high real-time lightweight computation and real-time beam adjustment. After completing the offline training of the time-delay prediction model and the closed-loop optimization model, the cloud intelligent control center sends the model parameters to the ground gateway station and the satellite terminal.

3. The system according to claim 2, characterized in that, The cloud-based intelligent control center includes a data aggregation module, a model offline training module, and a global beam scheduling module. The data aggregation module is used to complete the global multi-dimensional channel data aggregation of the entire constellation of satellites and the entire network of ground terminals; The offline training module is used to complete the offline training and parameter updates of the time delay prediction model and the closed-loop optimization model. The global beam scheduling module is used to complete the global coordinated scheduling of beam resources across the entire constellation.

4. The system according to claim 2, characterized in that, The ground gateway station antenna and processing subsystem includes a gateway station antenna array and a regional baseband processing unit; The regional baseband processing unit incorporates a joint robust optimization unit based on the minimum-maximum robust optimization criterion, which is used to solve the reference beamforming weight matrix with the core objective of maximizing the system signal-to-interference-plus-noise ratio (SINR) in the worst case. When solving the problem, the constraint time delay deviation fluctuation range shall not exceed 10% of the round-trip time (RTT) of the satellite-to-ground link, and the Doppler frequency offset fluctuation range shall not exceed 15% of the maximum Doppler frequency offset of the terminal. The weight coefficients of the constraint conditions shall be dynamically allocated by the cloud intelligent management and control center based on historical SINR feedback.

5. The system according to claim 2, characterized in that, The spaceborne active phased array antenna subsystem includes a multi-band common aperture antenna array, a multi-channel transceiver assembly, and a spaceborne baseband processing unit. The multi-band common aperture antenna array adopts a unit multiplexing + stacked nested structure, supports at least one of the Ka / Q / V / Ku / C bands, and has 16 to 128 array elements. The multi-channel transceiver unit has a built-in amplitude and phase calibration unit with a calibration accuracy of ≤0.5° phase error and ≤0.3dB amplitude error. The spaceborne baseband processing unit has a built-in lightweight time-delay prediction module and a real-time weight compensation module, which are used to receive the reference beamforming weight matrix, output the short-term prediction value of the time-delay deviation, and complete the real-time adaptive compensation of the beam weight.

6. The system according to claim 5, characterized in that, The lightweight time-delay prediction module is a single-hidden-layer LSTM network optimized by pruning, 8-bit integer quantization and knowledge distillation. The model size is ≤50KB, the number of hidden units is 16~32, and the single-step inference time is ≤50μs. Its inputs include real-time satellite orbital parameters, terminal moving speed, and historical time delay data. The output is a short-term predicted value of the time delay deviation and the corresponding confidence interval. The time delay prediction error is less than 60% of the peak-to-peak value of the time delay deviation in the past 10 sampling periods. The real-time weight compensation module calculates the phase compensation factor for each antenna channel based on the predicted time delay deviation value and performs phase rotation compensation on the reference beamforming weight matrix.

7. The system according to claim 1, characterized in that, The full-link closed-loop optimization mechanism is implemented based on the PPO algorithm, with the inputs being the signal-to-interference-plus-noise ratio (SINR), channel quality index (CQI), Doppler frequency offset, and signal-to-interference ratio (SIR) fed back by the terminal. A dual-trigger mode combining periodic triggering and event triggering is adopted: the periodic triggering interval is 0.8 to 1.5 times the round-trip time (RTT) of the satellite-to-ground link; the event triggering is triggered immediately when the SINR fed back by the terminal is lower than the preset threshold or the beam pointing error exceeds the preset threshold. The reward function of the PPO algorithm is: , where α is the signal-to-interference-plus-noise ratio weighting coefficient, β is the computational complexity weighting coefficient, α>β and α+β=1.

8. A beamforming optimization method for a satellite-terrestrial integrated communication antenna system based on beamforming, characterized in that, Based on a three-tiered collaborative architecture of satellite-ground-cloud, the execution includes the following steps: The S1 cloud intelligent control center collects global multi-dimensional data and completes offline training of the time-delay prediction model and the closed-loop optimization model. Based on regional channel data, the S2 ground gateway station solves the reference beamforming weight matrix using a robust optimization algorithm and sends it to the satellite-borne terminal. Based on the time-delay prediction results, the S3 satellite-borne terminal performs real-time phase compensation on the reference beam weights to generate target-shaped beams. The S4 terminal feeds back channel measurement data, triggering full-link closed-loop optimization and dynamically adjusting beamforming parameters and model parameters.

9. The method according to claim 8, characterized in that: In S2, robust optimization takes maximizing the system SINR in the worst case as its core objective, and constrains the time delay deviation fluctuation range to not exceed 10% of the RTT and the Doppler frequency offset fluctuation range to not exceed 15% of the maximum Doppler frequency offset at the terminal. In S3, time delay prediction uses a single hidden layer LSTM network optimized by pruning, 8-bit integer quantization and knowledge distillation, with 16~32 hidden units and a single-step inference time of ≤50μs; In S4, a dual triggering mode combining periodic triggering and event triggering is adopted. The periodic triggering interval is 0.8 to 1.5 times the RTT, and the event triggering is triggered immediately when the SINR is lower than the preset threshold or the beam pointing error exceeds the preset threshold.

Images