Unmanned aerial vehicle dynamic spectrum sharing method and device based on edge federated learning

Abstract

The invention provides an unmanned aerial vehicle dynamic spectrum sharing method and device based on edge federated learning, and the method comprises the steps: carrying out the frequency spectrum scanning of a target frequency range through scanning equipment carried on each unmanned aerial vehicle, and obtaining the frequency spectrum information of each frequency band collected by each unmanned aerial vehicle end; training a local model deployed by the corresponding unmanned aerial vehicle side based on the frequency spectrum information of each frequency band collected by each unmanned aerial vehicle side, and calculating and uploading a local model gradient of each unmanned aerial vehicle side when a global update time node is reached; the central server side updates the global model of the central server side based on the received local model gradient of each unmanned aerial vehicle side, and issues the updated global model parameters to each unmanned aerial vehicle side; the unmanned aerial vehicle ends update respective local models based on the updated global model parameters; and each unmanned aerial vehicle end performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, so that the communication quality is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular, to a method and device for dynamic spectrum sharing of unmanned aerial vehicles based on edge federated learning. Background Art

[0002] With the rapid growth in the number of low-altitude unmanned aerial vehicle clusters, the demand for the real-time performance, reliability, and security of communication spectra by unmanned aerial vehicles in complex airspace environments has become more difficult to meet. In scenarios of high-density flight and frequent communication, unmanned aerial vehicle clusters often face problems such as communication frequency band conflicts, high transmission delays, and poor anti-interference capabilities, severely restricting the application capabilities of unmanned aerial vehicle systems in fields such as military reconnaissance, emergency communication, and intelligent logistics.

[0003] In the prior art, there are mainly the following types of spectrum allocation methods: First, the method of using a centralized control center for unified spectrum allocation is relatively common. However, this type of solution relies on the central node for spectrum status collection and scheduling decision-making, resulting in a relatively high overall communication delay of the system, often exceeding 500 milliseconds, and there is a risk of single-point failure where the entire communication system collapses when the central node fails. Second, some solutions adopt a fixed frequency band pre-configuration method, which assigns exclusive frequency bands to each flight unit through a static mapping method. However, this type of method cannot dynamically sense the usage status of spectra and interference situations in the airspace, resulting in generally low spectrum utilization rates, and may cause a large amount of spectrum resources to be idle or in conflict. Third, in recent years, existing technologies have attempted to improve the intelligence level of spectrum allocation by introducing traditional machine learning methods. However, such methods usually rely on a centralized training mode, and the data collected by each unmanned aerial vehicle needs to be uploaded to the central server for unified modeling and optimization. This not only poses a significant risk of data privacy leakage but also, due to the long model parameter update period (often exceeding 24 hours), it is difficult to adapt to the dynamic changing environmental requirements and rapid decision-making requirements.

[0004] Therefore, the existing spectrum allocation and management methods still have significant deficiencies in terms of real-time performance, reliability, robustness, and privacy protection. There is an urgent need for a new spectrum sharing method with both dynamic intelligent scheduling capabilities, a distributed secure communication mechanism, and rapid response capabilities to meet the collaborative communication requirements of future high-density unmanned aerial vehicle systems in dynamic and complex electromagnetic environments. Summary of the Invention

[0005] The present invention provides a method and device for dynamic spectrum sharing of unmanned aerial vehicles based on edge federated learning to solve the deficiencies in real-time performance, reliability, robustness, and privacy protection in the prior art.

[0006] The present invention provides a method for dynamic spectrum sharing of unmanned aerial vehicles based on edge federated learning, including: Based on the scanning devices carried on each unmanned aerial vehicle (UAV) to perform spectrum scanning on the target frequency range, spectrum information of each frequency band collected at each UAV side is obtained, and a local model deployed at the corresponding UAV side is trained based on the spectrum information of each frequency band collected at each UAV side; If the global update time node is reached, then based on the local models at each UAV side and the spectrum information of each frequency band collected at the corresponding UAV side, the local model gradients of each UAV side are calculated and uploaded for the central server side to update the global model at the central server side based on the received local model gradients of each UAV side, and the updated global model parameters are sent down to each UAV side; each UAV side updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; Each UAV side performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

[0007] According to a UAV dynamic spectrum sharing method based on edge federated learning provided by the present invention, the spectrum scanning of the target frequency range based on the scanning devices carried on each UAV to obtain the spectrum information of each frequency band collected at each UAV side includes: Set the spectrum scanning parameters of each UAV side; Based on the scanning devices carried on each UAV, spectrum scanning is performed according to the spectrum scanning parameters of the corresponding UAV side to obtain a three-dimensional spectrum heat map including the distribution of the spectrum energy of each frequency band changing with time and frequency; Each UAV side extracts features from the three-dimensional spectrum heat map to obtain the interference feature vectors of each frequency band; Obtain the historical channel quality records of each frequency band; Combine the three-dimensional spectrum heat map, the interference feature vectors of each frequency band, and the historical signal quality records into the spectrum information of each frequency band.

[0008] According to a UAV dynamic spectrum sharing method based on edge federated learning provided by the present invention, the calculation and upload of the local model gradients of each UAV side based on the local models at each UAV side and the spectrum information of each frequency band collected at the corresponding UAV side for the central server side to update the global model at the central server side based on the received local model gradients of each UAV side includes: Based on the local models deployed on each UAV and the spectrum information of each frequency band collected at the corresponding UAV side, calculate the local model gradients of each UAV side; At the UAV side, encrypt the local model gradients of the corresponding UAV side based on the homomorphic encryption algorithm to obtain the encrypted local model gradients; Each drone terminal uploads the encrypted local model gradient to the central server, so that the central server decrypts the encrypted local model gradient based on the decryption algorithm to obtain the original local model gradient, and updates the global model of the central server based on the original local model gradient.

[0009] According to a method for dynamic spectrum sharing of drones based on edge federated learning provided by the present invention, each drone terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, including: For any drone, obtain the channel capacity of the communication channels of each frequency band and the task importance level of the task undertaken by the any drone, and use the local model of the any drone terminal to predict the channel quality of each frequency band according to the spectrum information of each frequency band collected by the any drone terminal; Based on the channel capacity of the communication channels of each frequency band, the task importance level of the task undertaken by the any drone, and the channel quality of each frequency band, construct a state space; Based on the currently used frequency band and a preset candidate frequency band set, construct an action space; Based on the state space, the action space, and a preset reward function, use the Q-Learning algorithm to perform dynamic frequency band selection to determine the optimal communication frequency band corresponding to the any drone currently.

[0010] According to a method for dynamic spectrum sharing of drones based on edge federated learning provided by the present invention, the reward function is specifically: R = α×C+β / I-γ×Δf Wherein, C is the channel capacity of the communication channel of any frequency band, I is the channel quality of the any frequency band, Δf is the switching overhead from the currently used frequency band to the any frequency band, and α, β, and γ are preset weights.

[0011] According to a method for dynamic spectrum sharing of drones based on edge federated learning provided by the present invention, after each drone terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, it further includes: After any drone terminal performs dynamic frequency band selection to obtain the optimal communication frequency band, elect 21 drones from all drones as verification nodes; Send the optimal communication frequency band to the verification nodes, so that each verification node performs consensus on the optimal communication frequency band based on the Byzantine fault-tolerant consensus algorithm to obtain the consensus conclusion of each verification node; If the consensus conclusions of more than 2 / 3 of the verification nodes are confirmations, broadcast the optimal communication frequency band to the adjacent drones of the any drone.

[0012] The present invention also provides a drone dynamic spectrum sharing device based on edge federated learning, including: A spectrum scanning unit, configured to perform spectrum scanning on a target frequency range based on scanning devices carried on each drone, obtain spectrum information of each frequency band collected by each drone end, and train a local model deployed on the corresponding drone end based on the spectrum information of each frequency band collected by each drone end; A federated learning unit, configured to, if the global update time node is reached, calculate and upload the local model gradients of each drone end based on the local models of each drone end and the spectrum information of each frequency band collected by the corresponding drone end, so that the central server end updates the global model of the central server end based on the received local model gradients of each drone end, and send the updated global model parameters to each drone end; each drone end updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; A dynamic frequency band selection unit, configured to perform dynamic frequency band selection on each drone end based on the channel quality of each frequency band output by the local model.

[0013] According to a drone dynamic spectrum sharing device based on edge federated learning provided by the present invention, calculating and uploading the local model gradients of each drone end based on the local models of each drone end and the spectrum information of each frequency band collected by the corresponding drone end, so that the central server end updates the global model of the central server end based on the received local model gradients of each drone end, includes: Calculating the local model gradients of each drone end based on the local models deployed on each drone and the spectrum information of each frequency band collected by the corresponding drone end; Encrypting the local model gradients of the corresponding drone end on the drone end based on a homomorphic encryption algorithm to obtain encrypted local model gradients; Each drone end uploads the encrypted local model gradients to the central server end, so that the central server end decrypts the encrypted local model gradients based on a decryption algorithm to obtain the original local model gradients, and updates the global model of the central server end based on the original local model gradients.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the drone dynamic spectrum sharing method based on edge federated learning as described in any one of the above when executing the program.

[0015] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for dynamic spectrum sharing of drones based on edge federated learning as described in any one of the above.

[0016] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the method for dynamic spectrum sharing of drones based on edge federated learning as described in any one of the above.

[0017] The method and device for dynamic spectrum sharing of drones based on edge federated learning provided by the present invention perform spectrum scanning on a target frequency range through scanning devices carried on each drone to obtain spectrum information of each frequency band collected at each drone end, and train local models deployed at the corresponding drone ends based on the spectrum information of each frequency band collected at each drone end. When reaching the global update time node, based on the local models at each drone end and the spectrum information of each frequency band collected at the corresponding drone end, calculate and upload the local model gradients of each drone end for the central server to update the global model at the central server end based on the received local model gradients of each drone end, and send the updated global model parameters to each drone end; each drone end updates its own local model based on the updated global model parameters to achieve real-time spectrum prediction driven by federated learning; each drone end performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, constructs dual protection through interference feature extraction and reinforcement learning to improve communication quality, and most of the calculations are completed at the edge, greatly reducing communication overhead. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of the method for dynamic spectrum sharing of drones based on edge federated learning provided by the present invention; Figure 2 It is a schematic architecture diagram of the drone dynamic spectrum sharing system provided by the present invention; Figure 3 It is a comparison diagram of the effects of the method provided by the present invention and the traditional method; Figure 4 It is a schematic structural diagram of the device for dynamic spectrum sharing of drones based on edge federated learning provided by the present invention; Figure 5It is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0020] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0021] Figure 1 It is a schematic flowchart of the method for dynamic spectrum sharing of unmanned aerial vehicles (UAVs) based on edge federated learning provided by the present invention. As Figure 1 shown, the method includes: Step 110: Perform spectrum scanning on a target frequency range based on the scanning devices carried on each UAV to obtain the spectrum information of each frequency band collected by each UAV terminal, and train the local models deployed on the corresponding UAV terminals based on the spectrum information of each frequency band collected by each UAV terminal; Step 120: If the global update time node is reached, calculate and upload the local model gradients of each UAV terminal based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal, so that the central server terminal can update the global model of the central server terminal based on the received local model gradients of each UAV terminal, and send the updated global model parameters to each UAV terminal; each UAV terminal updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; Step 130: Each UAV terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

[0022] Here, as Figure 2 shown, the architecture of the entire UAV dynamic spectrum sharing system includes: (1) Physical layer Responsible for real-time environmental perception and data collection, including the scanning devices (airborne software-defined radio (SDR) modules) carried on the UAVs and the edge computing units. The SDR module scans the 20 MHz - 6 GHz frequency band at a millisecond level to generate a spectrum heat map containing three-dimensional information of frequency, time, and space; the edge computing unit (such as NVIDIA Jetson) completes signal preprocessing and feature extraction to provide high-precision input data for the upper layer.

[0023] (2) Intelligent layer Implement distributed intelligent decision-making based on the federated learning framework. Each drone trains a lightweight AI model locally to analyze spectrum availability, predict interference risks, and determine the channel quality of each frequency band. The model parameters of multiple drones are aggregated through encrypted communication to generate a globally optimized strategy, which not only protects data privacy but also improves the accuracy of channel quality prediction.

[0024] (3)Control layer Execute dynamic spectrum allocation and multi-drone collaboration. Use reinforcement learning algorithms to real-time select the optimal communication frequency band, and ensure the rapid synchronization of the strategy through the blockchain consensus protocol. When malicious interference is detected, automatically switch to the backup frequency band and mark the location of the interference source to form a "perception - decision - execution" closed-loop control.

[0025] The three levels cooperate efficiently through 5G / self-organizing network. While ensuring communication security, it greatly improves the spectrum utilization rate and provides a reliable spectrum resource management solution for high-density drone swarms.

[0026] Specifically, to achieve dynamic spectrum sharing, spectrum scanning can be performed on the target frequency range (such as the 20MHz - 6GHz band) based on the scanning devices carried on each drone to obtain the spectrum information of each frequency band collected at each drone end. In some embodiments, the spectrum scanning parameters of each drone end can be set so that, based on the scanning devices (SDR modules) carried on each drone, spectrum scanning is performed according to the spectrum scanning parameters of the corresponding drone end, and the spectrum scanning data of multiple drone ends is fused using Kriging interpolation. Then, it is updated with a sliding window at intervals of 0.1 seconds to obtain a three-dimensional spectrum heat map showing the distribution of the spectrum energy of each frequency band changing with time and frequency. Subsequently, feature extraction can be performed on the above three-dimensional spectrum heat map at each drone end to obtain the interference feature vectors of each frequency band. In addition, the historical channel quality records of each frequency band are obtained, and the above three-dimensional spectrum heat map, the interference feature vectors of each frequency band, and the historical signal quality records are combined into the spectrum information of each frequency band.

[0027] Among them, the SDR spectrum scanning parameters can be configured based on the following method: def sdr_scan_config(): frequency_range = (20e6, 6e9) # Scanning frequency band 20MHz - 6GHz scan_step = 1e6 # 1MHz step dwell_time = 100e-3 # Dwell time per frequency point 100ms gain = 30 # RF front-end gain 30dB return generate_sweep_table(frequency_range, scan_step, dwell_time,gain) The configuration table generated here is passed to the SDR hardware control module to configure the real-time spectrum scanning parameters and serves as the basic data structure for subsequent spectrum heatmap generation.

[0028] The following function can be used to extract features from the three-dimensional spectrum heatmap to obtain the interference feature vectors for each frequency band: def wavelet_feature_extract(raw_signal): # The input raw_signal is the original I / Q signal data collected by the SDR receiver in the three-dimensional spectrum heatmap, with a length of 1024 sampling points, containing the electromagnetic environment information within a specific frequency band.

[0029] # The output is the interference feature vector (12-dimensional) coeffs = pywt.wavedec(raw_signal, 'db4', level=5) # 5-layer wavelet decomposition energy_ratio = [np.linalg.norm(c) / N for c in coeffs] entropy = [shannon_entropy(c) for c in coeffs[:3]] return normalize(energy_ratio + entropy) Based on the spectrum information of each frequency band collected by each drone, the local models deployed on the corresponding drones can be trained. Among them, the local model can be a convolutional neural network for predicting the channel quality of each frequency band based on the input spectrum information of each frequency band.

[0030] In some embodiments, the local model can be constructed in the following manner: class SpectrumCNN(nn.Module): # The input of the CNN model includes the two-dimensional heatmap slices (1×F×S) of the three-dimensional spectrum heatmap, where F is the size of the frequency dimension and S is the number of spatial grids, as well as the interference feature vectors and historical signal quality records of the corresponding frequency bands def __init__(self): super().__init__() self.layers = nn.Sequential( nn.Conv2d(1, 16, kernel_size=(3,5)),# Input channel 1 (spectrum intensity) nn.ReLU(), nn.MaxPool2d((2,3)), nn.Conv2d(16, 32, kernel_size=(3,3)), nn.ReLU(), nn.AdaptiveAvgPool2d((4,4)), nn.Flatten(), nn.Linear(32 × 4 × 4, 64), nn.Linear(64, 3))# Output: Channel quality scores [excellent / medium / poor] for communication channels in any frequency band, used to guide subsequent channel selection decisions If the global update time node is reached, based on the local models of each drone terminal and the spectrum information of each frequency band collected by the corresponding drone terminal, calculate and upload the local model gradients of each drone terminal for the central server to update the global model of the central server based on the received local model gradients of each drone terminal, and then send the updated global model parameters to each drone terminal. Among them, the structure and function of the global model are the same as those of the local models of each drone terminal, and it is also used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band.

[0031] In some embodiments, the local model gradients of each drone terminal can be calculated based on the local models deployed on each drone and the spectrum information of each frequency band collected by the corresponding drone terminal. Subsequently, at the drone terminal, the local model gradients of the corresponding drone terminal are encrypted based on the homomorphic encryption algorithm to obtain the encrypted local model gradients, and then each drone terminal uploads the encrypted local model gradients to the central server. At the central server, the encrypted local model gradients can be decrypted based on the decryption algorithm corresponding to the homomorphic encryption algorithm to obtain the original local model gradients, and the global model of the central server is updated based on the original local model gradients.

[0032] After the central server sends the updated global model parameters to each drone terminal, each drone terminal can update its own local model based on the updated global model parameters. Each drone terminal can perform dynamic frequency band selection based on the channel quality of each frequency band output by its own local model to achieve the best communication effect.

[0033] In some embodiments, for any unmanned aerial vehicle (UAV), the channel capacity of communication channels in each frequency band (the theoretical maximum transmission rate bps of the currently used channel) and the task importance level of the task undertaken by the UAV can be obtained, and the local model at the UAV end can be used to predict the channel quality of each frequency band based on the spectrum information of each frequency band collected by the UAV end in the manner given in the above embodiments. When performing dynamic frequency band selection, a state space is constructed based on the channel capacity of communication channels in each frequency band, the task importance level of the task undertaken by the UAV, and the channel quality of each frequency band. Based on the currently used frequency band and the preset candidate frequency band set, an action space is constructed. Based on the state space, action space, and a preset reward function, the Q-Learning algorithm is used for dynamic frequency band selection to determine the optimal communication frequency band corresponding to the UAV currently.

[0034] Among them, the update mechanism of the Q-Learning algorithm is as follows: Q[state][action]+= learning_rate × (reward+gamma×np.max(Q[next_state])-Q[state][action]) return Q In some embodiments, the reward function is specifically: R = α×C+β / I-γ×Δf Among them, C is the channel capacity of the communication channel in any frequency band, I is the channel quality of this frequency band, Δf is the switching overhead from the currently used frequency band to this frequency band, and α, β, and γ are preset weights.

[0035] After dynamic frequency band selection is performed at each UAV end based on the channel quality of each frequency band output by the local model, the currently corresponding optimal communication frequency band can also be synchronized to adjacent UAV nodes through a distributed consensus protocol. Specifically, after a UAV end performs dynamic frequency band selection to obtain the optimal communication frequency band, several (for example, 21) UAVs can be elected from all UAVs as verification nodes, and the optimal communication frequency band is sent to each verification node for each verification node to perform consensus on the optimal communication frequency band based on the Byzantine fault-tolerant consensus algorithm to obtain the consensus conclusion of each verification node. If the consensus conclusion of more than 2 / 3 of the verification nodes is confirmation, the optimal communication frequency band is broadcast to the adjacent UAVs of this UAV to achieve dynamic spectrum sharing.

[0036] In summary, the method provided by the embodiments of the present invention performs spectrum scanning on the target frequency range through the scanning devices carried on each unmanned aerial vehicle (UAV), obtains the spectrum information of each frequency band collected by each UAV terminal, and trains the local model deployed on the corresponding UAV terminal based on the spectrum information of each frequency band collected by each UAV terminal. When the global update time node is reached, based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal, calculate and upload the local model gradients of each UAV terminal for the central server to update the global model of the central server based on the received local model gradients of each UAV terminal, and send the updated global model parameters to each UAV terminal; each UAV terminal updates its local model based on the updated global model parameters to achieve real-time spectrum prediction driven by federated learning; each UAV terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, constructs double protection through interference feature extraction and reinforcement learning to improve communication quality, and most of the calculations are completed at the edge, greatly reducing the communication overhead.

[0037] The comparison diagram of the effects of the method provided by the present invention and the traditional method is as Figure 3 shown. Among them, the spectrum utilization rate is increased to more than 80% (measured data: 83.7% in urban environment, 89.2% in field environment), the communication interruption rate is reduced to less than 3% (comparative experiment: 15.6% for the traditional method), and the anti-wideband interference ability is increased by 8.2% (maintaining QoS at an interference intensity of -90 dBm).

[0038] Next, the UAV dynamic spectrum sharing device based on edge federated learning provided by the present invention will be described. The UAV dynamic spectrum sharing device based on edge federated learning described below can be correspondingly referred to the UAV dynamic spectrum sharing method based on edge federated learning described above.

[0039] Based on any of the above embodiments, Figure 4 is the structural schematic diagram of the UAV dynamic spectrum sharing device based on edge federated learning provided by the present invention, as Figure 4 shown. The device includes: A spectrum scanning unit 410, configured to perform spectrum scanning on the target frequency range through the scanning devices carried on each UAV, obtain the spectrum information of each frequency band collected by each UAV terminal, and train the local model deployed on the corresponding UAV terminal based on the spectrum information of each frequency band collected by each UAV terminal; A federated learning unit 420, which, when reaching the global update time node, calculates and uploads the local model gradients of each UAV terminal based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal, so that the central server terminal updates the global model of the central server terminal based on the received local model gradients of each UAV terminal, and distributes the updated global model parameters to each UAV terminal; each UAV terminal updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band. A dynamic frequency band selection unit 430, which is used for each UAV terminal to perform dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

[0040] The device provided by the embodiment of the present invention scans the spectrum of the target frequency range through the scanning devices carried on each UAV to obtain the spectrum information of each frequency band collected by each UAV terminal, and trains the local model deployed on the corresponding UAV terminal based on the spectrum information of each frequency band collected by each UAV terminal. When reaching the global update time node, based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal, calculate and upload the local model gradients of each UAV terminal, so that the central server terminal updates the global model of the central server terminal based on the received local model gradients of each UAV terminal, and distributes the updated global model parameters to each UAV terminal; each UAV terminal updates its own local model based on the updated global model parameters to achieve real-time spectrum prediction driven by federated learning; each UAV terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, constructs double protection through interference feature extraction and reinforcement learning to improve communication quality, and most of the calculations are completed at the edge, greatly reducing the communication overhead.

[0041] Based on any of the above embodiments, the spectrum scanning of the target frequency range through the scanning devices carried on each UAV to obtain the spectrum information of each frequency band collected by each UAV terminal includes: Set the spectrum scanning parameters of each UAV terminal; Based on the scanning devices carried on each UAV, perform spectrum scanning according to the spectrum scanning parameters of the corresponding UAV terminal to obtain a three-dimensional spectrum heat map including the distribution of the spectrum energy of each frequency band changing with time and frequency; Each UAV terminal extracts features from the three-dimensional spectrum heat map to obtain the interference feature vectors of each frequency band; Obtain the historical channel quality records of each frequency band; Combine the three-dimensional spectrum heat map, the interference feature vectors of each frequency band, and the historical signal quality records into the spectrum information of each frequency band.

[0042] Based on any of the above embodiments, calculating and uploading the local model gradients of each UAV terminal based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal, so that the central server terminal updates the global model of the central server terminal based on the received local model gradients of each UAV terminal, including: Calculating the local model gradients of each UAV terminal based on the local models deployed on each UAV and the spectrum information of each frequency band collected by the corresponding UAV terminal; Encrypting the local model gradients of the corresponding UAV terminal on the UAV terminal based on the homomorphic encryption algorithm to obtain the encrypted local model gradients; Each UAV terminal uploads the encrypted local model gradients to the central server terminal for the central server terminal to decrypt the encrypted local model gradients based on the decryption algorithm to obtain the original local model gradients, and update the global model of the central server terminal based on the original local model gradients.

[0043] Based on any of the above embodiments, the dynamic frequency band selection by each UAV terminal based on the channel quality of each frequency band output by the local model includes: For any UAV, obtaining the channel capacity of the communication channels of each frequency band and the task importance level of the task undertaken by the any UAV, and using the local model of the any UAV terminal to predict the channel quality of each frequency band according to the spectrum information of each frequency band collected by the any UAV terminal; Constructing a state space based on the channel capacity of the communication channels of each frequency band, the task importance level of the task undertaken by the any UAV, and the channel quality of each frequency band; Constructing an action space based on the currently used frequency band and a preset candidate frequency band set; Based on the state space, the action space, and a preset reward function, using the Q-Learning algorithm for dynamic frequency band selection to determine the optimal communication frequency band corresponding to the any UAV currently.

[0044] Based on any of the above embodiments, the reward function is specifically: R = α×C+β / I-γ×Δf Wherein, C is the channel capacity of the communication channel of any frequency band, I is the channel quality of the any frequency band, Δf is the switching overhead from the currently used frequency band to the any frequency band, and α, β, and γ are preset weights.

[0045] Based on any of the above embodiments, after the dynamic frequency band selection by each UAV terminal based on the channel quality of each frequency band output by the local model, it further includes: After any drone terminal performs dynamic frequency band selection to obtain the optimal communication frequency band, 21 drones are elected from each drone as verification nodes; The optimal communication frequency band is sent to the verification nodes, so that each verification node can reach a consensus on the optimal communication frequency band based on the Byzantine fault tolerance consensus algorithm, and obtain the consensus conclusion of each verification node; If the consensus conclusions of more than 2 / 3 of the verification nodes are confirmations, the optimal communication frequency band is broadcast to the adjacent drones of the any drone.

[0046] Figure 5 It is a schematic structural diagram of the electronic device provided by the present invention. As Figure 5 shown, the electronic device may include: a processor 510, a memory 520, a communication interface 530, and a communication bus 540. Among them, the processor 510, the memory 520, and the communication interface 530 communicate with each other through the communication bus 540. The processor 510 can call the logical instructions in the memory 520 to execute the drone dynamic spectrum sharing method based on edge federated learning. The method includes: performing spectrum scanning on the target frequency range based on the scanning devices carried on each drone to obtain the spectrum information of each frequency band collected by each drone terminal, and training the local model deployed on the corresponding drone terminal based on the spectrum information of each frequency band collected by each drone terminal; if the global update time node is reached, based on the local models of each drone terminal and the spectrum information of each frequency band collected by the corresponding drone terminal, calculate and upload the local model gradients of each drone terminal, so that the central server terminal can update the global model of the central server terminal based on the received local model gradients of each drone terminal, and send the updated global model parameters to each drone terminal; each drone terminal updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; each drone terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

[0047] In addition, when the logical instructions in the above-mentioned memory 520 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0048] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the method for dynamic spectrum sharing of drones based on edge federated learning provided by the above-mentioned various methods. The method includes: performing spectrum scanning on a target frequency range based on the scanning devices carried on each drone to obtain the spectrum information of each frequency band collected at each drone end, and training the local models deployed at the corresponding drone ends based on the spectrum information of each frequency band collected at each drone end; if the global update time node is reached, based on the local models at each drone end and the spectrum information of each frequency band collected at the corresponding drone end, calculating and uploading the local model gradients of each drone end for the central server to update the global model at the central server end based on the received local model gradients of each drone end, and sending the updated global model parameters to each drone end; each drone end updating its own local model based on the updated global model parameters; the local models and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; each drone end performing dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

[0049] In another aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for dynamic spectrum sharing of drones based on edge federated learning provided above. The method includes: performing spectrum scanning on a target frequency range based on scanning devices carried on each drone to obtain spectrum information of each frequency band collected at each drone end, and training a local model deployed at the corresponding drone end based on the spectrum information of each frequency band collected at each drone end; if the global update time node is reached, calculating and uploading the local model gradients of each drone end based on the local models of each drone end and the spectrum information of each frequency band collected at the corresponding drone end for the central server to update the global model at the central server end based on the received local model gradients of each drone end, and sending the updated global model parameters to each drone end; each drone end updating its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; each drone end performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

[0050] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0051] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for dynamic spectrum sharing of unmanned aerial vehicles based on edge federated learning, characterized in that, Including: Performing spectrum scanning on a target frequency range based on scanning devices carried on each unmanned aerial vehicle (UAV) to obtain spectrum information of each frequency band collected at each UAV end, and training a local model deployed at the corresponding UAV end based on the spectrum information of each frequency band collected at each UAV end; If the global update time node is reached, based on the local models at each UAV end and the spectrum information of each frequency band collected at the corresponding UAV end, calculating and uploading the local model gradients of each UAV end for the central server end to update the global model at the central server end based on the received local model gradients of each UAV end, and sending the updated global model parameters to each UAV end; each UAV end updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band; Each UAV end performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

2. The method for dynamic spectrum sharing of unmanned aerial vehicles based on edge federated learning according to claim 1, wherein The performing spectrum scanning on a target frequency range based on scanning devices carried on each UAV to obtain spectrum information of each frequency band collected at each UAV end includes: Setting spectrum scanning parameters for each UAV end; Based on the scanning devices carried on each UAV, performing spectrum scanning according to the spectrum scanning parameters of the corresponding UAV end to obtain a three-dimensional spectrum heat map including the distribution of the spectrum energy of each frequency band changing with time and frequency; Each UAV end extracts features from the three-dimensional spectrum heat map to obtain interference feature vectors of each frequency band; Obtaining historical channel quality records of each frequency band; Combining the three-dimensional spectrum heat map, the interference feature vectors of each frequency band, and the historical signal quality records into the spectrum information of each frequency band.

3. The method for dynamic spectrum sharing of unmanned aerial vehicles based on edge federated learning according to claim 1, wherein The calculating and uploading the local model gradients of each UAV end based on the local models at each UAV end and the spectrum information of each frequency band collected at the corresponding UAV end for the central server end to update the global model at the central server end based on the received local model gradients of each UAV end includes: Calculating the local model gradients of each UAV end based on the local models deployed on each UAV and the spectrum information of each frequency band collected at the corresponding UAV end; Encrypting the local model gradients of the corresponding UAV end at the UAV end based on the homomorphic encryption algorithm to obtain the encrypted local model gradients; Each UAV end uploads the encrypted local model gradients to the central server end for the central server end to decrypt the encrypted local model gradients based on the decryption algorithm to obtain the original local model gradients, and updating the global model at the central server end based on the original local model gradients.

4. The method for dynamic spectrum sharing of drones based on edge federated learning according to any one of claims 1 to 3, characterized in that, The each UAV end performing dynamic frequency band selection based on the channel quality of each frequency band output by the local model includes: For any UAV, obtaining the channel capacity of the communication channels of each frequency band and the task importance level of the task undertaken by the any UAV, and using the local model at the any UAV end to predict the channel quality of each frequency band based on the spectrum information of each frequency band collected at the any UAV end; Construct a state space based on the channel capacity of communication channels in each frequency band, the task importance level of the task undertaken by any one of the unmanned aerial vehicles (UAVs), and the channel quality of each frequency band. Construct an action space based on the currently used frequency band and a preset set of candidate frequency bands. Based on the state space, the action space, and a preset reward function, use the Q-Learning algorithm for dynamic frequency band selection to determine the optimal communication frequency band corresponding to any one of the UAVs currently.

5. The method for dynamic spectrum sharing of drones based on edge federated learning according to claim 4, wherein The reward function is specifically: R = α×C+β / I-γ×Δf where C is the channel capacity of the communication channel in any frequency band, I is the channel quality of any frequency band, Δf is the switching overhead from the currently used frequency band to any frequency band, and α, β, and γ are preset weights.

6. The method for dynamic spectrum sharing of drones based on edge federated learning according to any one of claims 1 to 3, characterized in that After each UAV terminal performs dynamic frequency band selection based on the channel quality of each frequency band output by the local model, it further includes: After any UAV terminal performs dynamic frequency band selection to obtain the optimal communication frequency band, elect 21 UAVs from all UAVs as verification nodes. Send the optimal communication frequency band to the verification nodes for each verification node to reach a consensus on the optimal communication frequency band based on the Byzantine fault-tolerant consensus algorithm, and obtain the consensus conclusion of each verification node. If the consensus conclusion of more than 2 / 3 of the verification nodes is confirmation, broadcast the optimal communication frequency band to the adjacent UAVs of any UAV.

7. An unmanned aerial vehicle dynamic spectrum sharing device based on edge federated learning, characterized in that, It includes: A spectrum scanning unit, which is used to perform spectrum scanning on the target frequency range based on the scanning devices carried on each UAV to obtain the spectrum information of each frequency band collected by each UAV terminal, and train the local model deployed on the corresponding UAV terminal based on the spectrum information of each frequency band collected by each UAV terminal. A federated learning unit, which is used to, if the global update time node is reached, calculate and upload the local model gradients of each UAV terminal based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal, so that the central server terminal can update the global model of the central server terminal based on the received local model gradients of each UAV terminal, and send the updated global model parameters to each UAV terminal; each UAV terminal updates its own local model based on the updated global model parameters; the local model and the global model are used to predict the channel quality of each frequency band based on the input spectrum information of each frequency band. A dynamic frequency band selection unit, which is used for each UAV terminal to perform dynamic frequency band selection based on the channel quality of each frequency band output by the local model.

8. The drone dynamic spectrum sharing device based on edge federated learning according to claim 7, characterized in that, The calculation and upload of the local model gradients of each UAV terminal based on the local models of each UAV terminal and the spectrum information of each frequency band collected by the corresponding UAV terminal for the central server terminal to update the global model of the central server terminal based on the received local model gradients of each UAV terminal includes: Calculate the local model gradients of each UAV terminal based on the local models deployed on each UAV and the spectrum information of each frequency band collected by the corresponding UAV terminal. Encrypt the local model gradients of the corresponding UAV terminal on the UAV terminal based on the homomorphic encryption algorithm to obtain the encrypted local model gradients. Each drone terminal uploads the encrypted local model gradient to the central server, so that the central server decrypts the encrypted local model gradient based on the decryption algorithm to obtain the original local model gradient, and updates the global model of the central server based on the original local model gradient.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for dynamic spectrum sharing of drones based on edge federated learning according to any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for dynamic spectrum sharing of drones based on edge federated learning according to any one of claims 1 to 6.

Images