Unmanned aerial vehicle cluster service awareness clustering routing method based on wide-narrow band fusion

Abstract

The invention discloses an unmanned aerial vehicle cluster service awareness clustering routing method based on wide-narrow band fusion, and the method comprises the steps: carrying out the role and link division of unmanned aerial vehicle nodes, and carrying out the initial clustering, and obtaining an initial layering clustering model based on wide-narrow band fusion; executing an intra-cluster optimal next hop discovery algorithm on the initial layered clustering model, and establishing an intra-cluster routing list; performing topology stability maintenance on the initial hierarchical clustering model in the unmanned aerial vehicle moving scene; performing cross-cluster routing discovery on the maintained layered clustering model, executing cross-cluster topology sensing and accurate addressing based on a compressed bloom filter, and determining a target cluster; executing gateway election based on link quality weighting and boundary prediction, and determining gateway nodes of cross-cluster transmission; and on the basis of the gateway node and the intra-cluster routing list, performing global perception cross-cluster dual-path routing discovery and dynamic service shunting, and realizing intra-cluster stable networking, cross-cluster rapid addressing, reliable gateway switching and service on-demand shunting transmission in the high-dynamic unmanned aerial vehicle cluster.

Description

A UAV swarm service-aware clustering routing method based on broadband and narrowband fusion Technical Field

[0001] This invention belongs to the field of UAV ad hoc network communication and routing protocol technology, and particularly relates to a UAV cluster service-aware clustering routing method based on broadband and narrowband convergence. Background Technology

[0002] A Flying Ad Hoc Network (FANET) is a decentralized, multi-hop, self-organizing network composed of multiple unmanned aerial vehicles (UAVs) connected by wireless communication links. Early research on FANETs primarily focused on the military field, applying them to military reconnaissance, communication relay, and combat support. With the rapid development of civilian UAV technology, FANETs have attracted wider attention, leading to a series of related research and applications. In recent years, FANETs have not only been applied and explored to some extent in traditional fields such as agricultural plant protection, environmental monitoring, logistics distribution, and areas with poor communication coverage, but are also gradually expanding into cutting-edge scenarios such as disaster relief, smart city 3D perception, low-altitude economy, and SAGIN (space-air-ground integrated network).

[0003] As the scale of UAV swarms continues to expand and the complexity of collaborative tasks increases, the communication bottleneck of FANET systems becomes increasingly apparent. FANET is characterized by high node movement speed, high degrees of freedom in three-dimensional space, and frequent topology changes. This leads to problems such as slow route convergence and excessive control overhead when dealing with such highly dynamic networks, making it unsuitable for the needs of highly dynamic mobile nodes. Therefore, a protocol solution more suitable for the FANET operating environment needs to be designed. For example, Chinese invention patent CN111556550A proposes a routing method for UAV network communication. This method uses the exchange of link information and node information between nodes for link awareness and neighbor discovery, further completing cluster head election, intra-cluster topology establishment, inter-cluster network topology establishment, and routing table generation, thus realizing clustered networking and route construction in UAV networks. Chinese invention patent CN115396975A proposes a networked, low-latency, high-speed data transmission method for unmanned aerial vehicle (UAV) swarm systems. This method involves clustering UAVs according to their task types, with cluster head nodes facilitating inter-cluster communication. Within each cluster, an improved DSDV routing protocol is employed. Furthermore, the network bandwidth of each cluster is dynamically allocated based on task-type information transmission requirements and channel environment awareness results to reduce routing overhead and processing latency as the network scales up. Chinese invention patent CN117202308A proposes a UAV swarm networking optimization and intra-cluster coordinated channel allocation method. This method uses a game-theoretic alliance algorithm to cluster UAVs, selecting the UAV with the highest current battery power as the cluster head UAV within each cluster. Optimal communication routes from the cluster head UAV to the ground station and other cluster head UAVs are planned. Simultaneously, a convex optimization algorithm is used to optimize flight paths, and intra-cluster channel resources are allocated in a coordinated manner.

[0004] However, the challenges currently faced by FANET at the Medium Access (MAC) layer and network layer are mainly concentrated in three aspects: underlying link metrics, network architecture scalability, and cross-cluster heterogeneous interaction. 1. (CN111556550A) Due to the high-speed movement of UAV nodes, physical links will frequently experience sudden deterioration and asymmetry, making it difficult for routing metrics based on historical link status to reflect the current real link quality in a timely manner. This results in delayed route switching, continuous selection of invalid links, and inefficient use of channel bandwidth, affecting the sensitivity and reliability of route discovery between nodes.

[0005] 2. (CN115396975A) As the scale of UAV swarms expands, the scalability of traditional planar network architectures is poor. A large amount of control signaling needs to be propagated throughout the network and competes with service data for limited broadband communication resources, leading to a rapid increase in network management overhead. Under highly dynamic topology conditions, this method struggles to track changes in node locations and topology relationships in a timely manner, easily causing local routing oscillations, frequent reconfigurations, and a decline in overall network operational stability, affecting the continuous networking capability and topology maintenance efficiency of large-scale UAV swarms.

[0006] 3. (CN117202308A) Traditional cross-cluster communication methods rely heavily on multi-hop forwarding. Cross-cluster data needs to be transmitted through multiple relay nodes, which can lead to latency accumulation and load concentration. When faced with sudden business growth or high-speed movement of gateway nodes, traffic convergence, local congestion, and false link interruption phenomena may occur at the cluster head or on a single path, even if the link is not actually failed. This affects the continuity, timeliness, and transmission capacity of cross-cluster communication and high-bandwidth services. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, the present invention aims to provide a broadband and narrowband converged UAV swarm service-aware clustering routing method. This method involves constructing an initial hierarchical clustering model based on broadband and narrowband convergence, establishing an enhanced BATMAN transmission quality metric model, employing a broadband and narrowband dual-timing maintenance strategy and a boundary hysteresis decision mechanism, executing cross-cluster topology awareness and precise addressing based on compressed Bloom filters, combining a gateway election mechanism based on link quality weighting and boundary prediction, and implementing globally aware cross-cluster dual-path route discovery and dynamic service offloading. This enables stable intra-cluster networking, fast cross-cluster addressing, reliable gateway switching, and on-demand service offloading in highly dynamic UAV swarms.

[0008] To achieve the above objectives, the technical solution adopted in this invention is as follows: a UAV cluster service-aware clustering routing method based on broadband and narrowband fusion, comprising the following steps: Step 1: Assigning roles and links to UAV nodes and initially forming clusters to obtain an initial hierarchical clustering model based on broadband and narrowband fusion; wherein, UAV nodes are divided into cluster head nodes, backup cluster head nodes, cluster member nodes, and gateway nodes; link types are divided into intra-cluster broadband links, intra-cluster narrowband links, inter-cluster narrowband links, and cross-cluster gateway broadband links; Step 2: Executing an intra-cluster best next-hop discovery algorithm on the initial hierarchical clustering model to establish intra-cluster routes. Step 3: Maintain the topology stability of the initial hierarchical clustering model under the UAV mobile scenario to obtain the maintained hierarchical clustering model; Step 4: Perform cross-cluster route discovery on the maintained hierarchical clustering model, and perform cross-cluster topology awareness and precise addressing based on compressed Bloom filters to determine the target cluster where the destination node is located; Step 5: Based on the determined target cluster, perform gateway election based on link quality weighting and boundary prediction to determine the gateway node for cross-cluster transmission; Step 6: Based on the gateway node for cross-cluster transmission and the intra-cluster routing list, perform globally aware cross-cluster dual-path route discovery and dynamic service diversion.

[0009] In step 1, the initial clustering process is as follows: Initially, a discovery window timer and a contention window timer are set in the programming module of each UAV; within the preset time of the discovery window timer, each node to be added to the cluster listens to the OGM messages periodically broadcast by neighboring nodes through the broadband link, parses and collects neighbor parameters; when the preset time of the discovery window timer expires, the node to be added to the cluster calculates its own comprehensive score based on its local state and the state of neighboring nodes, and compares it with the comprehensive scores of neighboring nodes. If its own comprehensive score is the highest in the neighborhood, it promotes itself to a candidate cluster head node; when multiple candidate cluster head nodes appear simultaneously in the same area, within the node's preset contention window, if a candidate cluster head node receives an OGM contention message with a higher comprehensive score within its own broadband range, it abandons its candidate cluster head status, is downgraded to a cluster member node, and initiates a clustering request; if a candidate cluster head node is in a conflict scenario where its own broadband cannot cover it, each candidate cluster head node exchanges OGM contention messages through narrowband beacons, and the candidate cluster head node with the lower comprehensive score performs a fallback and... The process of sending election flags is stopped. When the competition window ends and the candidate cluster head node does not receive any rebuttal messages from other candidate cluster head nodes with higher overall scores, the candidate cluster head node becomes the official cluster head node. Subsequently, surrounding candidate cluster head nodes send cluster entry request messages to the official cluster head node. After verifying the validity of the messages, the cluster head node sets the corresponding node as a cluster member node of its own cluster and adds it to the cluster's member list, completing the initial cluster entry. The cluster head node selects the second-highest ranked cluster member node from its own member list as the backup cluster head node and... In the broadcast signaling, the cluster member node is designated through the backup cluster head identifier field. Subsequently, when a cluster member node at the cluster edge receives OGM packets from different cluster nodes, it determines itself as a potential gateway node. After the potential gateway node completes the link symmetry verification based on the OGM packet, it reports to the cluster head node of its own cluster via intra-cluster unicast. The unicast packet includes the target cluster, neighboring cluster access points, and cross-cluster gateway broadband link quality. The cluster head node updates and maintains the inter-cluster gateway list, counts the transmission quality of each link from its own cluster to the target cluster, and completes the initial appointment of the gateway node.

[0010] In step 1, the formula for calculating the comprehensive score is: in, Indicates the remaining energy factor. Indicates the link quality factor. Represents the relative mobility factor. This represents the degree deviation factor; all election indicators have been normalized. , , , As a weighting factor, and In step 2, the intra-cluster best next-hop discovery algorithm is as follows: the destination node periodically broadcasts its OGM through the intra-cluster broadband link, and the neighboring nodes calculate the comprehensive link metric to the destination node after receiving the OGM. The source node writes the destination node, its corresponding best next hop, and the overall link metric to the cluster's routing list. If the neighboring node has already determined the destination node as the best next hop, it carries the updated overall link metric in the forwarded OGM and propagates it outward. When the source node receives OGMs forwarded from multiple candidate next hop nodes with the same destination node, it compares the overall link metric of each candidate path, selects the one with the highest value as the best next hop, and updates the corresponding cluster's routing list. If the source node has a direct connection with the destination node, it records the direct path first. If the direct path is blocked or its quality degrades, it automatically switches to a multi-hop path with a higher overall link metric.

[0011] In step 2, the integrated link metric value The calculation formula is: in, This indicates the quality of the OGM received by the local node from its neighboring nodes. Indicates echo quality, Indicates the jump penalty factor; The formula for calculating prior link quality is: in, As a physical layer penalty factor, Indicates the attenuation coefficient. Indicates the signal-to-noise ratio at the current moment. This indicates the demodulation sensitivity threshold of the receiver. The weight coefficients for different segments satisfy the following conditions: ,and ; Indicates the first The actual number of OGM packets received within each segment. Indicates the first The expected number of OGM packets to be received within each segment.

[0012] In step 3, the topology stability maintenance includes: Maintenance 1: The cluster head node periodically broadcasts heartbeat packets via narrowband links within the cluster to keep the topology alive. The heartbeat packets include the cluster head node identifier, message sequence number, timestamp, and member status summary; Maintenance 2: In the programming module of the cluster member nodes, a broadband aging timer, a narrowband keep-alive timer, and a broadband timeout threshold are set; if the broadband aging timer times out but the narrowband keep-alive timer does not, the cluster member node is only marked as unstable; if the narrowband keep-alive timer times out or the broadband keep-alive timer interruption duration exceeds the broadband timeout threshold, the cluster head node determines that the cluster member node has left its own cluster domain and notifies all cluster member nodes to update the intra-cluster topology; Maintenance 3: In the cluster member nodes... In the programming module of the node, a hysteresis threshold is set. If the edge cluster member node detects that the comprehensive score of the neighboring cluster head node is greater than the sum of the comprehensive score of the current cluster head node and the set hysteresis threshold, a cross-cluster switch is triggered, that is, the node exits the current cluster and requests to join the neighboring cluster. If the difference in comprehensive scores is less than the set hysteresis threshold, the original cluster affiliation status remains unchanged. Maintenance 4: In the programming module of the backup cluster head node, a heartbeat loss threshold is set. The backup cluster head node periodically synchronizes the member list and inter-cluster gateway list data managed by the cluster head node through the intra-cluster broadband link, and continuously listens for the narrowband heartbeat packets of the cluster head node. Once the loss time exceeds the set heartbeat loss threshold, the backup cluster head node immediately broadcasts a takeover announcement on the narrowband channel.

[0013] In step 4, the cross-cluster topology awareness and precise addressing process based on compressed Bloom filters is as follows: Each cluster head node performs a CRC8 hash operation based on its own MAC address to generate a unique cluster identifier across the entire network, and periodically exchanges heartbeat packets carrying compressed Bloom filter digests through inter-cluster narrowband links; when cross-cluster communication occurs, the source cluster head node first extracts the IP address feature bits of the destination node, and uses hash matching to search in the maintained Bloom filter digest to quickly locate the target cluster where the destination node is located; if the hash matching result only hits one cluster, the source cluster head node directly identifies that cluster as the target cluster; if multiple clusters are hit simultaneously due to hash collisions of the Bloom filter, the source cluster head node temporarily suspends the data packet and sends node verification messages to multiple candidate cluster head nodes through inter-cluster narrowband links; the cluster head node that actually contains the target node returns a verification confirmation response after checking the member list of its own cluster, and the source cluster head node locks the unique target cluster accordingly; the generation formula of the cluster identifier is as follows: in, This represents the MAC address of the cluster head node. This represents an 8-bit cyclic redundancy check hash function.

[0014] In step 4, the Bloom filter digest is mapped globally using the member list managed by the cluster head node. Specifically, the cluster head node reads the member list of its cluster, extracts the IP addresses of all cluster member nodes, initializes a bit array of length 128, and clears all bits. For each cluster member node IP address, the cluster head node sequentially inputs k hash functions to perform calculations, obtains k mapping positions, and sets the bits corresponding to the k mapping positions to 1. The cluster head node repeats the above mapping operation for all cluster member nodes in the member list. After all processing is completed, the Bloom filter digest of the cluster is obtained.

[0015] In step 5, the gateway election process based on link quality weighting and boundary prediction is as follows: After a cluster member node at the cluster edge detects a broadband signal from a neighboring cluster, it evaluates the quality of the cross-cluster gateway broadband link. That is, when the signal-to-noise ratio of the broadband signal from the neighboring cluster is consistently not less than the cross-cluster gateway discovery threshold, the cluster member node marks itself as a candidate gateway node. When the quality of the cross-cluster gateway broadband link meets the evaluation requirements and the link is a bidirectional symmetrical link, the candidate gateway node reports the target cluster identifier and a comprehensive score of the cross-cluster gateway broadband link quality to the cluster head node. After receiving reports from multiple candidate gateway nodes, the cluster head node combines the intra-cluster broadband link quality from the cluster head node to the candidate gateway node to perform a comprehensive score on the multiple candidate gateway nodes and establish a logical gateway forwarding mechanism. The forwarding facet table records the IP address, target cluster identifier, and overall score of each gateway node, and assigns a strict lifetime to each entry. Subsequently, the cluster head node activates the node with the highest overall score as the primary gateway node. The primary gateway node periodically reports its physical location and smoothing speed information to the cluster head node. Based on this information, the cluster head node uses a prediction model to predict the remaining dwell time of the primary gateway node within the current cluster coverage area. When the prediction result is lower than a preset security threshold, the current primary gateway node is determined to be a failed gateway node. The cluster head node then removes the failed gateway node from the logical gateway forwarding facet table and triggers alternative gateway nodes to take over according to the overall score order, completing the primary gateway switchover. The formula for calculating the overall score is as follows: in, For cross-cluster external quality; For intra-cluster backhaul quality; The weighting coefficients are represented; the prediction model is as follows: the cluster head node establishes a boundary circle model with its own physical location as the origin and the maximum effective communication distance in narrowband as the safety radius R; it is assumed that the main gateway node maintains [the following parameters] in a short period of time. Calculate the remaining dwell time of the main gateway node after it flies out of the boundary, given uniform linear motion. : in, The relative spatial coordinates of the primary gateway node at the current moment. For a time difference, This indicates the smooth speed of the main gateway node. The instantaneous relative velocity measured at the current moment. These are the smoothing weighting coefficients.

[0016] In step 6, the globally aware cross-cluster dual-path route discovery and dynamic service offloading process is as follows: When a source node within a cluster initiates communication to an external target node and the intra-cluster routing list is not matched, the source node first sends a cross-cluster routing request containing the destination node's IP address and service type to the source cluster head node. Upon receiving the request, the source cluster head node quickly retrieves the destination cluster to which the target node belongs based on the global Bloom filter mapping. Subsequently, the source cluster head node selects a specific cross-cluster transmission path according to the service type. For services with low bandwidth requirements, narrowband inter-cluster links between cluster head nodes are used for transmission. The target cluster head node then forwards the service to the target node. For services with high bandwidth requirements, the source cluster head node selects an activated primary gateway node. The activated primary gateway node generates a host network reachability announcement message and broadcasts its ability to reach the target cluster to its own cluster. The source cluster head node updates its intra-cluster routing list accordingly. Afterward, services with high bandwidth requirements bypass the cluster head node and are forwarded to the primary gateway node via intra-cluster multi-hop forwarding. The primary gateway node then establishes a connection with the target cluster gateway node through the cross-cluster gateway broadband link and completes the cross-cluster transmission. After the data transmission is completed, the receiver returns an acknowledgment message.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing an initial hierarchical clustering model based on broadband and narrowband convergence, the present invention can decouple the roles of cluster head nodes, backup cluster head nodes, cluster member nodes and gateway nodes, and carry out hierarchical collaborative design of intra-cluster broadband links, intra-cluster narrowband links, inter-cluster narrowband links and cross-cluster gateway broadband links, thereby improving the separation degree of control plane and data plane in UAV swarm network, significantly reducing network maintenance overhead, and improving the continuous and stable networking capability of large-scale UAV swarms.

[0018] 2. This invention completes cluster head election by calculating a comprehensive score based on the remaining energy factor, link quality factor, relative mobility factor, and degree deviation factor, and combining it with a broadband and narrowband collaborative conflict resolution method. This approach can take into account node energy, link quality, motion stability, and cluster size adaptability, thereby improving the stability and rationality of the cluster head election results.

[0019] 3. This invention optimizes the TQ metric model of native BATMAN, and can incorporate physical layer real-time signal-to-noise ratio, segmented weighted sliding window, echo quality detection and hop count penalty factor into the intra-cluster routing metric, thereby improving the intra-cluster best next hop discovery response to highly dynamic link changes.

[0020] 4. This invention, through a dual-timed maintenance strategy for both wide and narrow bands and a boundary hysteresis decision-making mechanism, can distinguish between two states: short-term broadband interruption and actual node departure from the cluster. It also suppresses frequent switching of edge nodes between adjacent clusters, thereby improving the stability of cluster domain topology maintenance.

[0021] 5. By setting up a backup cluster head node within the cluster and adopting a smooth switching strategy, the present invention enables the backup cluster head node to quickly take over cluster management when the cluster head node fails or becomes disconnected, thereby improving the continuity and fault recovery capability of the cluster domain control plane and enhancing the sensitivity and reliability of route discovery between nodes.

[0022] 6. This invention establishes a cross-cluster topology awareness and precise addressing mechanism based on compressed Bloom filters among the cluster head nodes, which can complete the target cluster location without transmitting the complete member list. It also reduces the impact of false judgments caused by hash collisions by combining node verification messages, thereby improving the lightweight nature and accuracy of cross-cluster addressing.

[0023] 7. This invention, through a gateway election mechanism based on link quality weighting and boundary prediction, can comprehensively consider the cross-cluster link quality, intra-cluster backhaul quality, and boundary dwell time of candidate gateway nodes, and trigger a switchover in advance before the primary gateway is about to cross the boundary, thereby improving the continuity and reliability of cross-cluster broadband data channels.

[0024] 8. This invention improves routing adaptability and overall transmission efficiency in different service scenarios by performing globally aware cross-cluster dual-path routing discovery and dynamic service diversion, and by combining HNA to achieve route redirection. It enables low-bandwidth services such as control commands and state synchronization to be transmitted via inter-cluster narrowband links, while high-bandwidth services such as video streams and sensing data are transmitted via cross-cluster gateway broadband links.

[0025] In summary, compared with existing technologies, this invention incorporates real-time signal-to-noise ratio at the physical layer, segmented weighted sliding window, echo quality detection, and hop count penalty factor into the intra-cluster routing metric, optimizing the TQ metric model of BATMAN and improving the high sensitivity and reliability of inter-node route discovery. By constructing an initial hierarchical clustering model based on broadband and narrowband fusion, physical isolation between control flow and data flow is achieved, significantly reducing network maintenance overhead and improving the stability and scalability of large-scale cluster networking. By adopting a gateway election mechanism based on link quality weighting and boundary prediction, and implementing a globally aware cross-cluster dual-path route discovery and dynamic service offloading strategy, the advantages of heterogeneous links are fully utilized, reducing end-to-end latency and avoiding congestion bottlenecks. In conclusion, the UAV cluster service-aware clustering routing method of this invention achieves stable intra-cluster networking, fast cross-cluster addressing, reliable gateway switching, and on-demand service offloading in highly dynamic UAV clusters. Attached Figure Description

[0026] Figure 1 is a communication diagram of a UAV swarm service perception clustering model based on broadband and narrowband fusion.

[0027] Figure 2 shows the distribution of the cross-layer sensing weighted sliding window.

[0028] Figure 3 shows the election evolution model of the cluster head node.

[0029] Figure 4 is a schematic diagram of the dual timer management mechanism of edge node hysteresis anti-oscillation mechanism and node passive off-cluster.

[0030] Figure 5 shows the mapping relationship based on the compressed Bloom filter.

[0031] Figure 6 is a schematic diagram of the interaction of the multi-hop forwarding protocol within a drone swarm.

[0032] Figure 7 shows the optimal mechanism for inter-cluster gateway discovery and state maintenance.

[0033] Figure 8 is a diagram of the gateway preemptive handover mechanism based on boundary dwell time prediction.

[0034] Figure 9 shows the interaction between global awareness cross-cluster dual-path route discovery and service traffic splitting. Detailed Implementation

[0035] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0036] In this invention, each UAV is equipped with a positioning device and an inertial navigation system to obtain its own physical position and speed, as well as a wideband radio (WBR) and two narrowband radios (NBRs), labeled NBR1 and NBR2. Both the wideband and narrowband radios can be frequency-modulated to adjust their frequency band range, and dynamic frequency point allocation can be performed on the wireless channels subsequently.

[0037] A service-aware clustering routing method for UAV swarms based on broadband and narrowband convergence includes the following steps: Step 1: Assign roles and links to UAV swarm nodes and initially cluster them to obtain an initial hierarchical clustering model based on broadband and narrowband convergence; as shown in Figure 1, the hierarchical clustering model includes UAV node role assignment and link type assignment; wherein, UAV nodes are functionally divided into Cluster Head (CH), Backup Cluster Head (BCH), Cluster Member (CM), and Cluster Gateway Node (CGN); links are functionally divided into intra-cluster broadband links, intra-cluster narrowband links, inter-cluster narrowband links, and cross-cluster gateway broadband links. This division of labor ensures that high-speed intra-cluster services, control and maintenance, inter-cluster control coordination, and cross-cluster forwarding fall on different links, avoiding the problem of control signaling and service data competing for the same broadband resources in traditional planar networks.

[0038] For the intra-cluster wideband data link (WBDL), it serves as the foundation of the data plane and is mainly used for high-bandwidth service data transmission between intra-cluster members as well as the broadcasting and forwarding of source node messages (OGM).

[0039] For the Cluster Control Link (CCL), it serves as the intra-cluster control link, used for heartbeat keep-alive between the cluster head node and cluster member nodes, topology updates, node entry and exit management, and management status synchronization between the cluster head node and the standby cluster head node.

[0040] Inter-Cluster Link (ICL) is mainly used as a management channel for multi-cluster collaboration across the entire network, for exchanging topology summaries, performing global routing, and handling cluster identifier conflicts between adjacent cluster head nodes.

[0041] For cross-cluster gateway broadband links, cluster member nodes located at the cluster edge and with stable broadband reachability to neighboring clusters are activated on demand to carry high-bandwidth traffic offloading across clusters.

[0042] As shown in Figure 3, after completing the role and link division, the initial clustering process begins. Before clustering starts, a discovery window timer and a contention window timer are set in the programming module of each UAV. Initially, each UAV node is in a state of waiting to join the cluster.

[0043] Discovery Window Phase (T0-T1): Within the preset time of the discovery window timer, each candidate cluster node (IDN) listens to the OGM messages periodically broadcast by neighboring nodes through the broadband link, parses and collects neighbor parameters, including election indicators, comprehensive scores and topology feature information carried by neighboring nodes; Competition and Conflict Resolution Phase (T1-T2): When the preset time of the discovery window timer expires, the candidate cluster node calculates its own comprehensive score based on its local state and the state of neighboring nodes, and compares it with the comprehensive scores of neighboring nodes. If its own comprehensive score is the highest in the neighborhood, it promotes itself to candidate cluster head node.

[0044] The comprehensive score of a node is calculated based on the election index parameters of its neighboring nodes. The formula is as follows: in, Indicates the remaining energy factor. Indicates the link quality factor. Represents the relative mobility factor. This represents the degree deviation factor; all election indicators have been normalized. , , , As a weighting factor, and .

[0045] Then, when multiple candidate cluster head nodes appear simultaneously in the same area, the conflict resolution process begins. Specifically, within the pre-defined contention window of each node, if a candidate cluster head node receives an OGM contention message with a higher overall score within its respective bandwidth, it relinquishes its candidate cluster head status, is demoted to a cluster member node, and initiates a cluster entry request. If a candidate cluster head node is in a conflict scenario where its bandwidth cannot cover it, the candidate cluster head nodes exchange OGM contention messages via narrowband beacons. The candidate cluster head node with the lower overall score yields and stops sending election flags, thus ensuring that only one official cluster head node is ultimately retained. This broadband-narrowband cooperative conflict resolution method utilizes broadband to improve the efficiency of scoring comparison within the neighborhood, while using narrowband to ensure the reliability of long-distance control decisions.

[0046] Role stabilization and cluster domain confirmation phase (T2-T3): When the contention window ends and the pending cluster head node does not receive any rebuttal information from other pending cluster head nodes with higher overall scores, the pending cluster head node becomes the official cluster head node. Subsequently, surrounding pending cluster head nodes send cluster entry request messages to the official cluster head node. After the cluster head node verifies the validity of the message, it sets the corresponding node as a cluster member node of its own cluster and adds it to the cluster's member list, completing the initial cluster entry.

[0047] Final Topology Phase (T3): To further improve the fault tolerance of the cluster domain, the cluster head node selects the second-ranked cluster member node from its member list as a backup cluster head node, and designates this member node in its broadcast signaling through the backup cluster head identifier field. Subsequently, a gateway election strategy is used to determine the cross-cluster gateway node. The election strategy is as follows: when a cluster member node at the cluster edge receives an OGM packet from a different cluster node, it determines itself as a potential gateway node. After the potential gateway node completes the link symmetry verification based on the OGM packet, it reports to the cluster head node of its own cluster via intra-cluster unicast. The unicast packet contains information such as the target cluster, neighboring cluster access points, and cross-cluster gateway broadband link quality. The cluster head node updates and maintains the inter-cluster gateway list, calculates the transmission quality (TQ) of each link from its own cluster to the target cluster, and completes the initial appointment of the gateway node. At this point, the initial hierarchical clustering model based on broadband and narrowband convergence is established.

[0048] Step 2: After completing the intra-cluster role division and initial clustering in Step 1, in order to achieve stable data forwarding, the intra-cluster best next-hop discovery algorithm is further executed on the initial hierarchical clustering model to establish an intra-cluster routing list. In the intra-cluster UAV network, the Better Approach To Mobile Ad-hoc Networking (BATMAN) distributed protocol is adopted. The TQ metric of native BATMAN is obtained by statistically analyzing the acceptance rate of the average window, which suffers from singleness and lag issues in highly dynamic scenarios. This invention optimizes the TQ metric model of native BATMAN to obtain an enhanced BATMAN TQ metric model. The comprehensive link metric value is calculated based on the enhanced BATMAN TQ metric model. The expression for the native BATMAN TQ metric is: Here, Received_OGMs represents the number of packets received, and Total_Window_Size represents the total window size.

[0049] The specific process of model optimization is as follows: the UAV node obtains the real-time signal-to-noise ratio at the physical layer, performs a priori evaluation of the broadband link quality, and calculates the prior TQ using a segmented weighted sliding window, denoted as... The echo quality between nodes is checked, and the echo quality is obtained by neighboring nodes calculating the OGM quality received from themselves, thereby eliminating unidirectional asymmetric links; based on the above optimization parameters, a hop count penalty factor is introduced, and finally, the comprehensive link metric is calculated. The calculation formula is as follows: in, This indicates the quality of the OGM received by the local node from its neighboring nodes. Indicates echo quality, This represents the jump penalty factor.

[0050] The formula for calculating the prior link quality is as follows: in, As a physical layer penalty factor, Indicates the attenuation coefficient. Indicates the signal-to-noise ratio at the current moment. This indicates the demodulation sensitivity threshold of the receiver, in dB. The weight coefficients for different segments satisfy the following conditions: ,and ; Indicates the first The actual number of OGM packets received within each segment. Indicates the first The expected number of OGM packets to be received within each segment.

[0051] During high-speed movement, the OGM information received by the UAV node in the early stages can no longer fully reflect the current link status. Therefore, the original single sliding window is divided into multiple consecutive time segments according to time sequence, and different weights are assigned to them. The most recent time segment has a higher weight, and the earlier time segments have a lower weight, so that the node pays more attention to the latest link status changes when calculating the current link quality. As shown in Figure 2, it is divided into 4 segments with different weights. The latest Segment 1 has a weight of 50%, and the weights decrease sequentially, with the oldest Segment 4 having a weight of only 5%.

[0052] Based on this, the intra-cluster best next-hop discovery algorithm can be specifically described as follows: the destination node periodically broadcasts its OGM (Optimal Next Hop) through the intra-cluster broadband link, and neighboring nodes calculate the comprehensive link metric to the destination node after receiving the OGM. And the destination node, its corresponding best next hop, and the overall link metric value are then used. Write it to the cluster's routing list; if the neighboring node has already determined the destination node as the best next hop, then continue to include the updated composite link metric in the forwarded OGM. And propagate outwards; the source node simultaneously receives OGMs forwarded from multiple candidate next-hop nodes, all targeting the same destination node, and at this point compares the comprehensive link metric values ​​of each candidate path. Select the one with the highest value as the best next hop and update the corresponding intra-cluster route list.

[0053] If the source node forms a high-quality direct connection with the destination node, the direct connection path is recorded first. If the direct connection path is blocked or its quality degrades, the system automatically switches to a multi-hop path with a higher overall link metric, thus enabling dynamic route updates within the cluster from direct connection priority to multi-hop alternatives.

[0054] This step has provided a foundation for stable intra-cluster routing and node connectivity for subsequent cluster domain maintenance. Therefore, we proceed to step 3 to maintain the topology and cluster domain stability in highly dynamic scenarios.

[0055] Step 3: For the initial hierarchical clustering model, topology stability maintenance is performed in the UAV mobile scenario to ensure the continuous and stable operation of the clustered network in a high-dynamic scenario. The present invention maintains topology stability in the following four aspects: Maintenance 1: The cluster head node periodically broadcasts heartbeat packets through the narrowband link within the cluster to keep the topology alive. Even if the broadband link of the cluster member node is interrupted due to short-term spatial occlusion or changes in maneuvering attitude, as long as the narrowband heartbeat can be received, the basic topology hierarchy can be maintained. The heartbeat packet includes information such as the cluster head node identifier, message sequence number, timestamp, and member status summary.

[0056] Maintenance 2: In the programming module of the cluster member node, a broadband aging timer, a narrowband keep-alive timer, and a broadband timeout threshold are set; the cluster head node maintains the broadband aging timer and narrowband keep-alive timer of the cluster member node according to the broadband and narrowband dual-timekeeping maintenance strategy, which solves the problem of frequent link disconnection caused by high dynamism; the broadband and narrowband dual-timekeeping maintenance strategy is as follows: if the broadband aging timer expires but the narrowband keep-alive timer does not expire, the cluster member node is only marked as unstable and is not removed from the member list managed by the cluster head node; if the narrowband keep-alive timer expires or the broadband keep-alive timer interruption duration exceeds the broadband timeout threshold, the cluster head node determines that the cluster member node has left its own cluster domain and notifies all cluster member nodes to update the intra-cluster topology.

[0057] Maintenance 3: Set a hysteresis threshold H in the programming module of the cluster member node, and perform anti-local oscillation processing on each cluster edge node according to the boundary hysteresis decision mechanism. The boundary hysteresis decision mechanism is as follows: if the edge cluster member node detects that the comprehensive score of the neighboring cluster head node is greater than the sum of the comprehensive score of the current cluster head node and the set hysteresis threshold, it triggers a cross-cluster switch, that is, it exits the current cluster and requests to join the neighboring cluster; if the difference in comprehensive scores is less than the set hysteresis threshold, it maintains the original cluster affiliation status.

[0058] Maintenance 4: For the backup cluster head node obtained in step 1, be prepared to execute a smooth handover strategy at any time. The smooth handover strategy is as follows: set a heartbeat loss threshold in the programming module of the backup cluster head node; the backup cluster head node periodically synchronizes the member list and inter-cluster gateway list data managed by the cluster head node through the intra-cluster broadband link; and continuously monitors the narrowband heartbeat packets of the cluster head node. Once the heartbeat loss time exceeds the set heartbeat loss threshold, the backup cluster head node immediately broadcasts a takeover announcement on the narrowband channel to ensure high-reliability connectivity of the intra-cluster topology.

[0059] Figure 4 illustrates the scenario of multiple drone nodes moving in the cluster edge region. The current cluster member node maintains a certain control connection with the cluster head node of its own cluster, while gradually approaching the cluster head node of the adjacent cluster. When the conditions for entering a new cluster are met, the node is allowed to join the new cluster; otherwise, it retains its original cluster affiliation under the effect of the hysteresis threshold, thereby avoiding the edge node from repeatedly switching between two adjacent clusters.

[0060] After completing step 3, the network can maintain a stable cluster structure and available intra-cluster routes in highly dynamic drone scenarios, providing a stable prerequisite for subsequent cross-cluster topology awareness, cross-cluster gateway selection, and service offloading.

[0061] Step 4: Perform cross-cluster route discovery on the maintained hierarchical clustering model, executing cross-cluster topology awareness and precise addressing based on compressed Bloom filters to determine the target cluster where the destination node resides. After completing the cluster domain topology maintenance in Step 3, the structure of each cluster is basically stable, and the affiliation relationships of nodes within the cluster and intra-cluster routes have been established. If the source node and the destination node belong to different clusters, end-to-end communication cannot be completed solely by intra-cluster routes. Therefore, further cross-cluster route discovery is required to determine the target cluster where the destination node resides. To this end, Step 4 establishes a cross-cluster topology awareness and precise addressing mechanism based on compressed Bloom filters (BF) between the cluster head nodes.

[0062] The cluster identifier, denoted as Cluster Identifier (CID), is used to distinguish different clusters. The compressed BF digest contains the IP addresses of all cluster member nodes in the cluster. Specifically, each cluster head node performs a CRC8 hash operation based on its own MAC address to generate a unique cluster ID across the entire network, and periodically exchanges heartbeat packets carrying the compressed BF digest through narrowband links between clusters. As shown in Figure 5, the BF digest is mapped using a global Bloom filter through the member list managed by the cluster head node. That is, the cluster head node reads the member list of its cluster, extracts the IP addresses of all cluster member nodes, initializes a 128-bit array, and clears all bits. For each cluster member node's IP address, the cluster head node sequentially inputs k hash functions to perform calculations, obtains k mapping positions, and sets the bits corresponding to the k mapping positions to 1. The cluster head node repeats the above mapping operation for all cluster member nodes in the member list. After all processing is completed, the Bloom filter digest of the cluster is obtained. To ensure dynamic reliability, the cluster head node refreshes according to a preset period. The Bloom filter is periodically rebuilt based on the latest list of members within the cluster.

[0063] This global Bloom filter mapping method enables lightweight synchronization of cluster-level topology summary information across the entire network without directly transmitting the complete member list. The formula for generating the cluster ID is as follows: in, This represents the MAC address of the cluster head node. This represents an 8-bit cyclic redundancy check hash function.

[0064] After completing cluster-level topology summary synchronization, when cross-cluster communication occurs, the source cluster head node (SN) _ CH first extracts the IP address feature bits of the destination node (DN) and then uses hash matching to search in the maintained BF digest to quickly locate the target cluster where the destination node is located.

[0065] If the hash match results in a match with only one cluster, the source cluster head node directly identifies that cluster as the target cluster, as shown in Figure 6. If multiple clusters match simultaneously due to hash collisions in the Bloom filter, the source cluster head node temporarily suspends the data packet and sends node verification messages to multiple candidate cluster head nodes via the inter-cluster narrowband link. The cluster head node that actually contains the target node returns a verification confirmation response after checking its own cluster's member list, and the source cluster head node thus locks onto the unique target cluster. This method allows for relatively accurate cross-cluster positioning with limited narrowband control resources.

[0066] Therefore, in step 4, based on the stable cluster domains constructed and maintained in steps 1 to 3, each cluster head node forms a lightweight cluster-level topology-aware view facing the entire network, enabling the source cluster head node to quickly and with low overhead locate the target cluster to which the destination node belongs, providing an addressing basis for subsequent service routing distribution.

[0067] Step 5: During the cross-cluster data channel establishment and high-dynamic topology maintenance phase, based on the determined target cluster, a gateway election based on link quality weighting and boundary prediction is performed to determine the gateway node for cross-cluster transmission. After the target cluster location is completed in Step 4, the network already knows which cluster the data should be sent to. However, it is still necessary to further resolve the issue of which node the data should be forwarded through across the cluster. Since cross-cluster services are usually not suitable for the cluster head node to directly handle for a long time, a suitable gateway node as the cross-cluster forwarding entry point needs to be selected from the cluster edge nodes. For this purpose, Step 5 executes a gateway election mechanism based on link quality weighting and boundary prediction. The boundary prediction refers to predicting whether a node is about to leave the communication coverage area of ​​the current cluster based on the node's location and speed information.

[0068] Specifically, cluster member nodes at the cluster edge acquire the signal-to-noise ratio (SNR) of external broadband signals in real time through the underlying interface. Upon detecting broadband signals from neighboring clusters, they evaluate the quality of cross-cluster gateway broadband links. When the quality of the cross-cluster gateway broadband link meets the evaluation requirements and the link is a bidirectional symmetrical link, the candidate gateway node reports the target CID and a comprehensive score of the cross-cluster gateway broadband link quality to the cluster head node. After receiving reports from multiple candidate gateway nodes, the cluster head node, considering the intra-cluster broadband link quality from the cluster head node to the candidate gateway node, performs a comprehensive score on the multiple candidate gateway nodes and establishes a logical gateway forwarding surface table. This table records the IP address, target CID, and comprehensive score of each gateway node, and assigns a strict Time-to-Live (TTL) to each entry. Subsequently, the cluster head node activates the node with the highest comprehensive score as the primary gateway node.

[0069] The formula for calculating the overall score is as follows: in, Cross-cluster external quality represents the cross-cluster gateway broadband link quality from the adjacent cluster head node to the candidate gateway node; Intra-cluster backhaul quality refers to the intra-cluster broadband link quality from the cluster head node to the candidate gateway node. This represents the weighting coefficient, which is usually set to 0.6 to prioritize cross-cluster quality.

[0070] The quality assessment of the cross-cluster gateway broadband link is as follows: when the signal-to-noise ratio of the broadband signals of adjacent clusters is consistently not less than the cross-cluster gateway discovery threshold. At that time, the cluster member node will mark its own state as a candidate gateway node.

[0071] As shown in Figure 7, the reachability relationships between multiple candidate gateway nodes near the cluster boundary and adjacent clusters are illustrated. The cluster head node compares multiple candidate gateway nodes based on the cross-cluster gateway broadband link quality and the connectivity within its own cluster, and then activates the main gateway node that is currently more suitable to undertake cross-cluster forwarding tasks.

[0072] Furthermore, considering the high mobility of drone nodes, the main gateway node needs to periodically report its physical location and smoothed velocity information to the cluster head node. A time-scaled gateway boundary dwell time prior prediction model is adopted, and the cluster head node predicts the remaining dwell time of the main gateway node within the current cluster coverage area based on this information; when the prediction result is lower than a preset safety threshold... If the current primary gateway node is determined to be a failed gateway node, it indicates that the current primary gateway node is about to break the connection due to exceeding its boundaries. The cluster head node then removes the failed gateway node from the logical gateway forwarding plane table and triggers the alternative gateway nodes to take over according to the comprehensive score order, thus completing the primary gateway switchover and ensuring the continuity and stability of the cross-cluster data channel.

[0073] Figure 8 illustrates the gateway preemptive switching mechanism based on boundary dwell time prediction. To address the data packet loss issue caused by edge gateways frequently flying out of cluster coverage in highly dynamic UAV swarms, a time-scaled gateway boundary dwell time prior prediction model is introduced to replace the traditional reconnection mode. This model eliminates instantaneous spikes and physical measurement noise from UAVs in complex environments. When the gateway node periodically reports its status, it does not directly use instantaneous relative velocity, but instead uses the exponentially weighted moving average (EWMA) algorithm to calculate a smoothed velocity vector. The formula is as follows: in, The instantaneous relative velocity measured at the current moment. To smooth out the weighting coefficients, the true motion trend of the gateway can be predicted, preventing the prediction model from being erroneously triggered due to sudden changes in instantaneous velocity.

[0074] For the gateway boundary dwell time prior prediction model, it is mainly used to quantitatively evaluate the remaining dwell time of the current master gateway node. The cluster head node establishes a boundary circle model with its own physical location as the origin and the maximum effective narrowband communication distance as the safety radius R. It is assumed that the master gateway node maintains a certain dwell time within a short period of time. The motion is uniform in a straight line. The trajectory of the main gateway node is combined with the equation of the boundary circle. By solving the quadratic equation, the remaining dwell time of the main gateway node after flying out of the boundary can be calculated. : in, The relative spatial coordinates of the primary gateway node at the current moment. It is a time difference; These are spatial geometric coefficients derived from the current relative coordinates and smooth velocity vector of the main gateway node. They can reduce the dimension of positional changes in three-dimensional space to the time dimension, enabling the cluster head to have the ability to predict global topology deterioration.

[0075] Therefore, step 5, based on the target cluster determined in step 4, selects the currently available primary gateway node for cross-cluster transmission and promptly replaces it when the gateway node is about to leave the cluster, thereby providing a stable data channel for cross-cluster service transmission in step 6.

[0076] Step 6: In the cross-cluster multi-hop transmission and service distribution phase, based on the gateway nodes and intra-cluster routing lists for cross-cluster transmission, globally aware cross-cluster dual-path route discovery and dynamic service offloading are performed. After determining the target cluster in Step 4 and completing gateway election in Step 5, the network can enter the cross-cluster multi-hop transmission and service distribution phase. Step 6 performs globally aware cross-cluster dual-path route discovery and dynamic service offloading to select the appropriate transmission method according to different service requirements. The dual-path refers to the fact that cross-cluster communication can use two different transmission paths: one is an inter-cluster narrowband link between cluster head nodes, and the other is a cross-cluster gateway broadband link between gateway nodes. By allocating different types of services to different paths, the efficiency of cross-cluster communication can be improved, and the data forwarding pressure on the cluster head nodes can be reduced.

[0077] Specifically, when a source node within a cluster initiates communication to an external target node and the cluster's routing list is not matched, the source node first sends a cross-cluster routing request containing the destination node's IP address and service type to the source cluster head node. Upon receiving the request, the source cluster head node quickly retrieves the target cluster to which the target node belongs using the global Bloom filter mapping relationship established in step 4.

[0078] Subsequently, the source cluster head node selects a specific cross-cluster transmission path based on the service type. Specifically, as shown in Figure 9, for services with low bandwidth requirements such as control commands, state synchronization, collaborative scheduling, and target indication, narrowband links between cluster head nodes can be used for transmission, and then the target cluster head node forwards the service to the target node. For services with high bandwidth requirements such as video streams, sensing data, mapping data, file return, and image data, the source cluster head node selects the activated master gateway node from step 5. The activated master gateway node generates a Host Network Announcement (HNA) and broadcasts its ability to reach the target cluster to its local cluster. The source cluster head node updates its intra-cluster routing list accordingly. Afterward, services with high bandwidth requirements bypass the cluster head node, undergo multi-hop forwarding within the cluster to reach the master gateway node, and then the master gateway node establishes a connection with the target cluster gateway node via a cross-cluster gateway broadband link to complete the cross-cluster transmission. After data transmission is complete, the receiver returns an acknowledgment message.

[0079] Through the aforementioned dual-path mechanism, the high reliability control capability of inter-cluster narrowband links and the high throughput transmission capability of cross-cluster gateway broadband links can be utilized respectively according to the service type, thereby improving the transmission efficiency and service adaptability of cross-cluster communication.

Claims

1. A UAV swarm service-aware clustering routing method based on broadband and narrowband fusion, characterized in that, The process includes the following steps: Step 1: Assign roles and links to UAV nodes and initially cluster them to obtain an initial hierarchical clustering model based on broadband and narrowband fusion; where UAV nodes are divided into cluster head nodes, backup cluster head nodes, cluster member nodes, and gateway nodes; link types are divided into intra-cluster broadband links, intra-cluster narrowband links, inter-cluster narrowband links, and cross-cluster gateway broadband links; Step 2: Execute the intra-cluster best next-hop discovery algorithm on the initial hierarchical clustering model to establish an intra-cluster routing list; Step 3: Maintain the topology stability of the initial hierarchical clustering model under UAV mobile scenarios to obtain a maintained hierarchical clustering model; Step 4: Perform cross-cluster route discovery on the maintained hierarchical clustering model, execute cross-cluster topology awareness and precise addressing based on compressed Bloom filters, and determine the target cluster where the destination node is located; Step 5: Based on the determined target cluster, perform gateway election based on link quality weighting and boundary prediction to determine the gateway node for cross-cluster transmission; Step 6: Based on the gateway node for cross-cluster transmission and the intra-cluster routing list, perform globally aware cross-cluster dual-path route discovery and dynamic service diversion.

2. The UAV swarm service-aware clustering routing method according to claim 1, characterized in that: In step 1, the initial clustering process is as follows: Initially, a discovery window timer and a contention window timer are set in the programming module of each UAV; within the preset time of the discovery window timer, each node to be added to the cluster listens to the OGM messages periodically broadcast by neighboring nodes through the broadband link, and parses and collects neighbor parameters. When the window timer expires, the node to be added to the cluster calculates its own comprehensive score based on its local state and the state of its neighboring nodes, and compares it with the comprehensive scores of its neighboring nodes. If its own comprehensive score is the highest in the neighborhood, it will be promoted to a node to be a candidate cluster head. When multiple candidate cluster head nodes appear simultaneously in the same area, within the node's preset competition window, if a candidate cluster head node receives an OGM competition message with a higher comprehensive score within its respective bandwidth range, it will relinquish its candidate cluster head status, be downgraded to a cluster member node, and initiate a cluster entry request. If the candidate cluster head nodes are in a conflict scenario where their respective broadband coverage is insufficient, the candidate cluster head nodes exchange OGM contention messages through narrowband beacons. The candidate cluster head node with the lower overall score will back off and stop sending election flags. When the competition window ends and the undetermined cluster head node does not receive any rebuttal information from other undetermined cluster head nodes with higher overall scores, the undetermined cluster head node becomes the official cluster head node. Subsequently, surrounding nodes awaiting cluster entry send cluster entry request messages to the official cluster head node. After verifying the validity of the message, the cluster head node sets the corresponding node as a cluster member node of its own cluster and adds it to the cluster's member list, completing the initial cluster entry. The cluster head node selects the second-ranked cluster member node from its own member list as a backup cluster head node and designates this cluster member node in its broadcast signaling through the backup cluster head identifier field. Subsequently, when a cluster member node at the cluster edge receives OGM packets from different cluster nodes, it determines itself as a potential gateway node. After the potential gateway node completes the link symmetry verification based on the OGM packet, it reports to the cluster head node of its own cluster via intra-cluster unicast. The unicast packet includes the target cluster, neighboring cluster access points, and cross-cluster gateway broadband link quality. The cluster head node updates and maintains the inter-cluster gateway list, calculates the transmission quality of each link from its own cluster to the target cluster, and completes the initial appointment of the gateway node.

3. The UAV swarm service-aware clustering routing method according to claim 2, characterized in that: In step 1, the formula for calculating the comprehensive score is: in, Indicates the remaining energy factor. Indicates the link quality factor. Represents the relative mobility factor. This represents the degree deviation factor; all election indicators have been normalized. 、 、 、 As a weighting factor, and 。 4. The UAV swarm service-aware clustering routing method according to claim 1, characterized in that: In step 2, the intra-cluster best next hop discovery algorithm is as follows: The destination node periodically broadcasts its OGM through the intra-cluster broadband link. After receiving the OGM, the neighboring node calculates the comprehensive link metric value to the destination node and writes the destination node and its corresponding best next hop and comprehensive link metric value into the intra-cluster routing list. If the neighboring node has already determined the destination node as the best next hop, it will carry the updated composite link metric in the forwarded OGM and propagate it outwards; if the source node simultaneously receives OGMs forwarded from multiple candidate next hop nodes with the same destination node, it will compare the composite link metric values ​​of each candidate path. The highest value is selected as the best next hop, and the corresponding intra-cluster routing list is updated; if the source node is directly connected to the destination node, the direct connection path is recorded first. If the direct path is blocked or its quality degrades, it will automatically switch to a multi-hop path with a higher overall link metric.

5. The UAV swarm service-aware clustering routing method according to claim 4, characterized in that: In step 2, the integrated link metric value The calculation formula is: in, This indicates the quality of the OGM received by the local node from its neighboring nodes. Indicates echo quality, Indicates the jump penalty factor; The formula for calculating prior link quality is: in, As a physical layer penalty factor, Indicates the attenuation coefficient. Indicates the signal-to-noise ratio at the current moment. This indicates the demodulation sensitivity threshold of the receiver. The weight coefficients for different segments satisfy the following conditions: ,and ； Indicates the first The actual number of OGM packets received within each segment. Indicates the first The expected number of OGM packets to be received within each segment.

6. The UAV swarm service-aware clustering routing method according to claim 1, characterized in that: In step 3, the topology stability maintenance includes: Maintenance 1: The cluster head node periodically broadcasts heartbeat packets via narrowband links within the cluster to keep the topology alive. The heartbeat packets include the cluster head node identifier, message sequence number, timestamp, and member status summary; Maintenance 2: In the programming module of the cluster member nodes, a broadband aging timer, a narrowband keep-alive timer, and a broadband timeout threshold are set; if the broadband aging timer times out but the narrowband keep-alive timer does not, the cluster member node is only marked as unstable; if the narrowband keep-alive timer times out or the broadband keep-alive timer interruption duration exceeds the broadband timeout threshold, the cluster head node determines that the cluster member node has left its own cluster domain and notifies all cluster member nodes to update the intra-cluster topology; Maintenance 3: In the cluster member nodes... In the programming module of the node, a hysteresis threshold is set. If the edge cluster member node detects that the comprehensive score of the neighboring cluster head node is greater than the sum of the comprehensive score of the current cluster head node and the set hysteresis threshold, a cross-cluster switch is triggered, that is, the node exits the current cluster and requests to join the neighboring cluster. If the difference in comprehensive scores is less than the set hysteresis threshold, the original cluster affiliation status remains unchanged. Maintenance 4: In the programming module of the backup cluster head node, a heartbeat loss threshold is set. The backup cluster head node periodically synchronizes the member list and inter-cluster gateway list data managed by the cluster head node through the intra-cluster broadband link, and continuously listens for the narrowband heartbeat packets of the cluster head node. Once the loss time exceeds the set heartbeat loss threshold, the backup cluster head node immediately broadcasts a takeover announcement on the narrowband channel.

7. The UAV swarm service-aware clustering routing method according to claim 1, characterized in that: In step 4, the cross-cluster topology awareness and precise addressing process based on compressed Bloom filters is as follows: Each cluster head node performs a CRC8 hash operation based on its own MAC address to generate a unique cluster identifier across the entire network, and periodically exchanges heartbeat packets carrying compressed Bloom filter digests through inter-cluster narrowband links; when cross-cluster communication occurs, the source cluster head node first extracts the IP address feature bits of the destination node, and uses hash matching to search in the maintained Bloom filter digest to quickly locate the target cluster where the destination node is located; if the hash matching result only hits one cluster, the source cluster head node directly determines that cluster as the target cluster; if multiple clusters are hit simultaneously due to hash collisions of the Bloom filter, the source cluster head node temporarily suspends the data packet and sends node verification messages to multiple candidate cluster head nodes through inter-cluster narrowband links; The cluster head node that actually contains the target node returns a verification confirmation response after checking its own cluster's member list, and the source cluster head node uses this to locate the unique target cluster; the formula for generating the cluster identifier is as follows: in, This represents the MAC address of the cluster head node. This represents an 8-bit cyclic redundancy check hash function.

8. The UAV swarm service-aware clustering routing method according to claim 7, characterized in that: In step 4, the Bloom filter digest is mapped globally using the member list managed by the cluster head node. Specifically, the cluster head node reads the member list of its cluster, extracts the IP addresses of all cluster member nodes, initializes a bit array of length 128, and clears all bits. For each cluster member node IP address, the cluster head node sequentially inputs k hash functions to perform calculations, obtains k mapping positions, and sets the bits corresponding to the k mapping positions to 1. The cluster head node repeats the above mapping operation for all cluster member nodes in the member list. After all processing is completed, the Bloom filter digest of the cluster is obtained.

9. The UAV swarm service-aware clustering routing method according to claim 1, characterized in that: In step 5, the gateway election process based on link quality weighting and boundary prediction is as follows: After a cluster member node at the cluster edge detects a broadband signal from an adjacent cluster, it evaluates the quality of the cross-cluster gateway broadband link. That is, when the signal-to-noise ratio of the broadband signal from the adjacent cluster is consistently not less than the cross-cluster gateway discovery threshold, the cluster member node marks itself as a candidate gateway node. When the quality of the cross-cluster gateway broadband link meets the evaluation requirements and the link is a bidirectional symmetrical link, the candidate gateway node reports the target cluster identifier and the comprehensive score of the cross-cluster gateway broadband link quality to the cluster head node. After receiving reports from multiple candidate gateway nodes, the cluster head node combines the intra-cluster broadband link quality from the cluster head node to the candidate gateway node to perform a comprehensive score on the multiple candidate gateway nodes and establishes a logical gateway forwarding surface table. This table records the IP address, target cluster identifier, and comprehensive score of the gateway node, and assigns a strict lifetime to each entry. Subsequently, the cluster head node activates the node with the best score as the main gateway node based on the comprehensive score of the candidate gateway nodes; The primary gateway node periodically reports its physical location and smoothing speed information to the cluster head node. Based on this information, the cluster head node uses a prediction model to predict the remaining dwell time of the primary gateway node within the current cluster coverage area. When the prediction result is lower than a preset security threshold, the current primary gateway node is determined to be a failed gateway node. The cluster head node then removes the failed gateway node from the logical gateway forwarding plane table and triggers alternative gateway nodes to take over according to the comprehensive score, completing the primary gateway switchover. The formula for calculating the comprehensive score is as follows: in, For cross-cluster external quality; For intra-cluster backhaul quality; The weighting coefficients are represented; the prediction model is as follows: the cluster head node establishes a boundary circle model with its own physical location as the origin and the maximum effective communication distance in narrowband as the safety radius R; it is assumed that the main gateway node maintains [the following parameters] in a short period of time. Calculate the remaining dwell time of the main gateway node after it flies out of the boundary, given uniform linear motion. ： in, The relative spatial coordinates of the primary gateway node at the current moment. For a time difference, This indicates the smooth speed of the main gateway node. The instantaneous relative velocity measured at the current moment. These are the smoothing weighting coefficients.

10. The UAV swarm service-aware clustering routing method according to claim 1, characterized in that: In step 6, the globally aware cross-cluster dual-path route discovery and dynamic service diversion process is as follows: When the source node in the cluster initiates communication to the external target node and the intra-cluster route list is not matched, the source node first sends a cross-cluster route request containing the destination node's IP address and service type to the source cluster head node. Upon receiving the request, the source cluster head node quickly retrieves the destination cluster to which the target node belongs based on the global Bloom filter mapping. Subsequently, the source cluster head node selects a specific cross-cluster transmission path according to the service type. For services with low bandwidth requirements, transmission is performed using narrowband links between cluster head nodes, and then the target cluster head node forwards the service to the target node. For services with high bandwidth requirements, the source cluster head node selects an activated master gateway node. The activated master gateway node generates a host network reachability announcement message and broadcasts its ability to reach the target cluster to its own cluster. The source cluster head node updates its intra-cluster routing list accordingly. Afterward, services with high bandwidth requirements bypass the cluster head node, are forwarded via multiple hops within the cluster to the master gateway node, and then the master gateway node establishes a connection with the target cluster gateway node through a cross-cluster gateway broadband link to complete the cross-cluster transmission. After the data transmission is completed, the receiver will return an acknowledgment message.

Images