Distributed collaborative waveform switching method and system for networked computing

Abstract

The invention discloses a distributed collaborative waveform switching method and system oriented to networked computing, and mainly solves the problems of overlarge time delay and overhigh complexity when distributed collaborative waveform switching achieves the consistency of the whole network in a multi-hop ad hoc network environment in the prior art. The implementation scheme comprises a distributed collaboration stage and a distributed decision consistency stage. Wherein in the distributed collaboration stage, each node periodically exchanges a state packet and a collaboration data packet, and in the exchange process, distributed waveform decision making and time delay information updating are respectively executed in the nodes; in the distributed decision consistency stage, a unique central node is selected through the collaborative data packet stored in the distributed collaborative stage, and the central node performs majority decision through the internally stored waveform decision information of each node to make an optimal waveform switching decision. According to the method, the calculation complexity can be reduced, the decision timeliness and accuracy can be improved, the distributed consistency of the whole network can be ensured, the communication time delay can be remarkably reduced, the system performance can be improved, and the method can be used for networked calculation in a wireless ad hoc network.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and specifically relates to a distributed collaborative waveform switching method and system, which can be used for networked computing in wireless ad hoc networks to improve the system's anti-interference ability and communication performance. Background Art

[0002] Distributed collaboration can be divided into two types of algorithms: strong consistency and weak consistency. Strong consistency algorithms include Paxos algorithm, Raft algorithm, etc., while weak consistency algorithms include final consistency algorithm, temporary consistency algorithm, etc.

[0003] The distributed collaborative strong consistency algorithm uses synchronization mechanisms such as log replication and election protocols to synchronize and coordinate the data of each node so that all participating nodes in the network can maintain a consistent state at any time.

[0004] The distributed collaborative weak consistency algorithm focuses more on the high availability and scalability of the system. It allows temporary inconsistencies between some nodes in the system, but the system design ensures that consistency can be achieved in the end. It is suitable for scenarios with high requirements on response time and tolerance for a certain degree of inconsistency.

[0005] When performing distributed collaborative waveform switching in a multi-hop ad hoc network, nodes must be in the same waveform to collaborate and communicate, and the decision consistency of all nodes in the cluster must be guaranteed during the waveform switching process to avoid communication failure or data loss caused by inconsistent waveforms. Therefore, a strong consistency mechanism needs to be implemented to ensure the waveform consistency of the entire network during the waveform switching process.

[0006] At present, in the field of distributed collaborative consistency, the most widely used strong consistency distributed algorithm is the Raft algorithm. The Raft algorithm is divided into five stages: random timeout election, initiating voting, waiting for response, half election and decision issuance. Among them, in the random timeout election stage, the node starts the election after the random time expires, becomes a candidate node and competes for the leader position; then, the node votes for the latest candidate node according to the term number; the candidate node waits for the voting response of other nodes, and if it obtains more than half of the support, it will be elected as the leader; finally, the leader node issues the waveform switching decision according to the majority opinion to ensure the consistency of the whole network. Although this distributed collaborative consistency algorithm can realize the consistency decision of waveform switching through information exchange and collaborative calculation between nodes, improve the network communication quality and anti-interference ability, the algorithm process is relatively cumbersome, involving multiple stages, resulting in long delays. The randomness of its election mechanism also makes the selection of leader nodes not accurate enough, further increases the delay of decision information issuance, and affects the system performance. Therefore, in a multi-hop self-organizing network environment, how to control the delay and reduce the complexity while ensuring the consistency of the waveform is still a technical problem that needs to be solved urgently.

[0007] The patent document with application number CN202410461259.X discloses a "elastic network networking method and device based on distributed collaborative nodes". It synchronizes time and measures channel quality of multiple self-organizing network forwarding nodes, builds a collaborative cloud based on these measurement results, and then forms an elastic network; by realizing time synchronization of multiple self-organizing network forwarding nodes, the time base of all nodes in the network is ensured to be consistent; through channel quality measurement, the communication status between nodes is obtained, so as to carry out intelligent resource allocation, routing selection and communication strategy formulation; the elastic network built based on multiple collaborative clouds has high flexibility and scalability, and can dynamically adjust the network structure and resource configuration according to actual needs to cope with various emergencies and communication needs. This method realizes network resource allocation and adaptive adjustment, and improves the flexibility of the network. However, the disadvantage of this scheme is that the process overhead of time synchronization and channel quality measurement is large, especially in a multi-hop self-organizing network environment, as the number of nodes increases, the delay increases accordingly, affecting real-time performance; at the same time, when the number of nodes increases, the computational complexity increases, and the reliability of the system cannot be guaranteed.

[0008] The patent document with application number CN201911170210.4 discloses a "waveform design method for broadband frequency hopping clustered multi-level self-organizing network". It combines distributed centerless multi-hop frequency hopping synchronization, initial network establishment, dynamic gateway selection, physical layer modulation and demodulation with encoding and decoding, and adaptive routing selection based on link learning to realize the waveform design of broadband frequency hopping clustered multi-level self-organizing network. Multi-hop frequency hopping synchronization does not require a preset synchronization center node and a special frequency hopping frequency set and pattern. The synchronization time is related to factors such as the number of frequency points, the frequency hopping rate, and the number of relay hops. Each node probabilistically transmits a synchronization frame based on the local clock and the preset frequency hopping pattern. After receiving the frame header, other nodes maintain the synchronization frequency to complete the frequency hopping synchronization. The link performance is evaluated through the learning strategy to achieve adaptive routing selection, ensuring that dynamic adjustment can be made when the link quality changes in the multi-hop link. Although this method achieves flexible multi-hop synchronization and efficient data transmission path selection, improves the network's adaptability and communication efficiency, and avoids centralized synchronization of central nodes, it lacks strong waveform consistency guarantees and may lead to waveform inconsistencies between different nodes, thus affecting communication stability. Summary of the invention

[0009] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and propose a distributed collaborative waveform switching method and system for networked computing, so as to further improve network reliability, reduce computational complexity, reduce latency, improve decision-making timeliness and accuracy, and ensure distributed consistency across the entire network.

[0010] The technical principle for realizing the purpose of the present invention is: to realize the design of distributed collaborative waveform switching method by coordinating the consensus algorithm and the OODA loop concept in a multi-hop ad hoc network, and its technical solution includes:

[0011] 1. A distributed collaborative waveform switching method for networked computing, characterized by comprising two stages: a distributed collaborative stage and a distributed decision consistency stage;

[0012] In the distributed collaboration stage, each node in the network periodically exchanges status packets and collaboration data packets to achieve real-time information collaboration and storage between nodes. During this exchange process, distributed waveform decisions and delay information updates are performed in the nodes respectively, and waveform decisions and delay data are dynamically adjusted to optimize communication efficiency and improve anti-interference capabilities.

[0013] In the distributed decision consistency stage, a unique central node is selected through the collaborative data packet information stored in the distributed collaboration stage; the central node makes the best waveform switching decision through majority judgment based on the waveform decision information of each node stored internally, and transmits it to other nodes in the entire network through broadcasting; after receiving the decision information, other nodes adjust their own waveforms according to the decision of the central node to ensure that the entire network reaches a consensus on waveform switching, thereby improving the collaborative efficiency and reliability of the network.

[0014] Further, the state packet: consists of a state packet header + state information, wherein the state packet header includes a state packet type, a sequence number and a state information offset; the state information includes a network state, an interference state and an environment state;

[0015] Furthermore, the collaborative data packet is composed of a collaborative packet header + collaborative data, wherein the collaborative packet header includes the collaborative data packet type, sequence number and collaborative information offset; the collaborative data includes node address, location information, average delay, maximum delay, waveform decision information and current role of the node.

[0016] 2. A distributed collaborative waveform switching system for networked computing, comprising:

[0017] The node sets up a perception state interaction module, which is used to periodically interact with the neighbor node state packets and the coordination data packets, that is, to update the storage of the neighbor node state packets by comparing the sequence numbers, to collect the node state information and exchange coordination data packets with the neighbor node in a period of 1s, and to exchange state packets with the neighbor node in a period of 2s;

[0018] The node sets a cognitive processing module to build an efficient status packet and collaborative data packet processing system, that is, to process the status packets and collaborative data packets of neighboring nodes by removing the status packet header and the collaborative packet header, and distribute them to the set shared buffer;

[0019] The node sets up an anti-interference networking decision module, which uses the state information in the shared buffer as input, makes distributed waveform decisions through the reinforcement learning model DQN, and makes the waveform switching strategy that best suits the current network environment;

[0020] The node sets a dynamic control module for the election of the central node and the control of the final waveform decision, that is, the only central node is selected from the collaborative data packet information stored by the node in the distributed collaboration stage, and the central node makes the best waveform switching decision by majority judgment in the distributed decision consistency stage.

[0021] Furthermore, the node is provided with a dynamic control module, including:

[0022] The node sets a central node election submodule, which is used to elect the node with the shortest communication delay as the central node, that is, the average delay of each node is compared through the collaborative data in the collaborative data packet of each node in the whole network stored in the node, and the node with the smallest average delay is preferentially selected; if there are multiple nodes with the same average delay, the maximum delay of each node is further compared, and the node with the smallest maximum delay is preferentially selected; if there are multiple nodes with the same maximum delay, the node with the smallest node address is selected as the only central node;

[0023] The node sets the final waveform decision submodule, which is used to make the best waveform switching decision by majority judgment. That is, after the central node is elected, the central node traverses the waveform decision information of each node stored internally, calculates the frequency of each waveform being selected, and finds the waveform with the highest selection frequency as the candidate waveform. If the candidate waveform is unique, the waveform is selected as the final waveform decision; if not unique, the waveform with stronger anti-interference ability is preferentially selected as the final waveform decision.

[0024] Compared with the prior art, the present invention has the following advantages:

[0025] First, since the nodes of the present invention periodically exchange status packets and coordination data packets through the perception status interaction module in the distributed coordination stage, an efficient coordination mechanism can be implemented in a multi-hop self-organizing network, which significantly improves coordination efficiency and adaptability, reduces communication overhead and system complexity, and enhances network reliability and responsiveness.

[0026] Second, since the nodes of the present invention can make waveform decisions independently and share decision information with other nodes in the distributed decision consistency stage, centralized control and excessive synchronization in traditional methods are avoided, so that the system can make decisions quickly and improve the timeliness and accuracy of waveform decisions.

[0027] Third, the nodes of the present invention elect the node with the shortest communication delay as the central node, and the central node sends decision information to other nodes in the entire network, thereby ensuring the distributed consistency of the entire network while minimizing the delay.

[0028] Fourthly, the node of the present invention makes the best waveform switching decision through majority judgment, which reduces the computational complexity and significantly improves the response speed of the system in making waveform switching decisions when facing changes in the external environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a general flow chart of the distributed collaborative waveform switching method for networked computing of the present invention;

[0030] Figure 2 yes Figure 1 The distributed collaboration phase in the implementation sub-flow chart;

[0031] Figure 3 yes Figure 1 The distributed decision consistency phase in the implementation sub-flow chart;

[0032] Figure 4 It is a structural block diagram of a distributed collaborative waveform switching system for networked computing of the present invention;

[0033] Figure 5 It is a simulation diagram comparing the communication delay of sending decision information to the whole network using the distributed collaborative waveform switching method of the present invention and the existing distributed collaborative Raft algorithm. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work should all fall within the scope of protection of the present invention.

[0035] Example 1: Distributed collaborative waveform switching method for networked computing

[0036] Reference Figure 1 ,The implementation of this instance includes two stages: the distributed ,collaboration stage and the distributed decision consistency stage.

[0037] 1. In the distributed collaboration stage, each node coordinates information between nodes and executes distributed waveform decision and delay information update.

[0038] Reference Figure 2In the distributed collaboration stage, each node in the network periodically exchanges status packets and collaboration packets to achieve real-time information collaboration and storage between nodes. During this exchange process, the nodes perform distributed waveform decision-making and delay information updates respectively, and dynamically adjust waveform decision-making and delay data to optimize communication efficiency and improve anti-interference capabilities. The implementation steps include:

[0039] Step 1: Each node performs information coordination among nodes, that is, periodically exchanges status packets and coordination data packets.

[0040] The state packet is composed of a state packet header + state information, wherein the state packet header includes a state packet type, a sequence number and a state information offset; the state information includes a network state, an interference state and an environment state;

[0041] The collaborative data packet consists of a collaborative packet header + collaborative data, wherein the collaborative packet header includes the collaborative data packet type, sequence number and collaborative information offset; the collaborative data includes node address, location information, average delay, maximum delay, waveform decision information and current role of the node.

[0042] The implementation of this step includes the following:

[0043] 1.1) Status packet coordination between nodes:

[0044] 1.1.1) Each node collects its internal state information at the network layer with a period of 1 second, and merges it with the state information in the state packets of neighboring nodes within three hops stored in the node, and adds a state packet header to the merged information to generate a state packet to be sent;

[0045] 1.1.1) Each node sends status packets to neighboring nodes through the sending end network layer-MAC layer-baseband waveform BB path with a period of 2s, and receives the status packets of neighboring nodes through the baseband waveform BB-MAC layer-receiving end network layer path, and stores them in the storage space of the node;

[0046] Through the above-mentioned periodic state information coordination process between nodes, each node can obtain the state information of the node and neighboring nodes within three hops.

[0047] 1.2) Collaboration between nodes Data packet collaboration:

[0048] 1.2.1) Each node exchanges collaborative data packets in a period of 1 second. In the collaborative data, the average delay reflects the overall communication efficiency of the node and other nodes, the maximum delay indicates the longest transmission delay from the node to the farthest node, and the current role of the node reflects whether the node is a central node or an ordinary node;

[0049] 1.2.2) Each node integrates the collaborative data of all nodes in each cycle, adds it to the collaborative packet header, and exchanges collaborative data packets between nodes according to the sending network layer-MAC layer-baseband waveform BB packet sending path and the baseband waveform BB-MAC layer-receiving network layer packet receiving path.

[0050] Step 2: The node performs distributed waveform decision-making based on the collected status information of neighboring nodes within three hops.

[0051] The distributed waveform decision is to collect and store the state packets of neighboring nodes within three hops, and then use the reinforcement learning model DQN including an input layer, a hidden layer and an output layer in the node, and use the state information in the state packets of each node stored in the node as the input layer input to obtain the distributed waveform decision result of the node in the output layer.

[0052] The specific implementation of this step includes the following:

[0053] 2.1) A two-dimensional structure array is set in the node as a storage space for collecting and storing neighbor node status packets. The first dimension of the two-dimensional structure array represents the node number, and the second dimension represents the time;

[0054] 2.2) When a node receives a neighbor node status packet, it updates the storage based on whether the neighbor node status packet exists in the two-dimensional structure array:

[0055] If the two-dimensional structure array does not have a neighbor node status package, add the neighbor node status package to the two-dimensional structure array;

[0056] If the two-dimensional structure array has a neighbor node status packet, compare the sequence numbers of the neighbor node status packet A in the two-dimensional structure array and the received neighbor node status packet B:

[0057] If the sequence number of B is greater than that of A, the neighbor node state packet is added to the two-dimensional structure array;

[0058] Otherwise, ignore the neighbor node status packet;

[0059] The value range of the sequence number is set to [0,255]. As a special case, if the sequence number count starts to cycle, that is, the sequence number of the last neighbor node status packet received is 255, and the sequence number of the currently received neighbor node status packet is 0, then the sequence number comparison rule is 1>0>255. According to this rule, the current neighbor node status packet is added to the two-dimensional structure array.

[0060] 2.3) Remove the header of the neighbor node status packet received in the two-dimensional structure array, and distribute the status information to the shared buffer in the node according to the node number, so as to maximize the concurrent performance of reading and writing the status information through the high efficiency of storing and accessing data in the shared buffer;

[0061] 2.4) Taking the state information of each node stored in the shared buffer in the node as input, the waveform decision calculation is performed based on the DQN reinforcement learning model to select the waveform switching strategy that best suits the current network environment:

[0062] 2.4.1) Obtain the status information in the shared buffer within the node, including the frame error rate in the network status, the interference frequency domain occupancy in the interference status, and the transmission rate in the environment status, so as to respectively reflect the reliability of data transmission, the interference degree of the spectrum, and the transmission rate of each node;

[0063] 2.4.2) Construct a reinforcement learning model DQN including an input layer, a hidden layer and an output layer, wherein the input layer takes the state information of each node stored in the shared buffer within the node as input, and the output layer outputs the predicted Q value of each waveform;

[0064] 2.4.3) Define the reward function in the reinforcement learning model DQN: sending rate * (1-frame error rate) + interference penalty. This reward function is used to guide the DQN model to learn the optimal waveform switching strategy to adapt to the dynamically changing interference environment;

[0065] 2.4.4) Obtain the input required by the reward function through the network status of the state information in the shared buffer within the node, including the current waveform, transmission rate and frame error rate, and obtain the reward value calculated by the reward function under the current waveform;

[0066] 2.4.5) Use the reward value as the input of the reinforcement learning model DQN and use the Q-learning algorithm to update its target Q value:

[0067] The Q-learning algorithm is a reinforcement learning algorithm based on value iteration. It makes decisions by learning the expected future rewards of each state-action pair, that is, calculating the immediate reward based on the immediate return obtained by taking a certain action in the current state, and updating the target Q value based on the immediate reward, gradually optimizing the waveform selection strategy, and the update formula is:

[0068]

[0069] Where Q(s,a) is the current Q value of taking waveform selection action a in state s; α is the learning rate, which is used to control the speed of Q value update; R is the immediate reward after taking the current waveform selection action; γ is the discount factor, which indicates the impact of future rewards; is the maximum Q-value of all possible actions a' in the new state s'.

[0070] 2.4.6) After updating the target Q value through the Q-learning algorithm update formula, calculate the error between the predicted Q value output by the output layer of the reinforcement learning model DQN and the target Q value output by the Q-learning algorithm update formula. Adjust the weight of the model through the calculated error to obtain the optimal waveform selection under different state information inputs, that is, the waveform decision result of the reinforcement learning model DQN.

[0071] 2.5) After storing the waveform decision result of the reinforcement learning model DQN in the waveform decision information in the collaborative data of this node, the waveform switching control is performed according to the current node role in the collaborative data of this node:

[0072] If the current node role is not the central node, it will adjust its own waveform after receiving the waveform decision information sent by the central node;

[0073] If the current node role is a central node, a majority decision is executed to make the best waveform switching decision and broadcast it to the entire network.

[0074] Step 3: Update the delay information within the node.

[0075] 3.1) Obtain the maximum delay and average delay of all node cooperation data packets stored in the node. The maximum delay represents the longest transmission delay from the node to the farthest node, and the average delay represents the overall communication efficiency between the node and other nodes;

[0076] 3.2) Traverse the maximum delay of all nodes stored in the node, compare the maximum delay of each node with the maximum delay of this node, check whether there is a larger delay value, and update the maximum delay of this node:

[0077] If a larger delay value is found, the maximum delay of this node is updated to the larger value;

[0078] If no larger delay value is found, no change is made;

[0079] 3.3) In the process of traversing the maximum delays of all nodes stored in the node, the maximum delay information of all nodes stored in the node is accumulated, the average delay is calculated by weighted average, and the average delay value of this node is updated to the average delay after weighted average calculation.

[0080] 2. In the distributed decision consistency stage, each node elects a central node and the central node makes the best waveform decision.

[0081] Reference Figure 3In the distributed decision consistency stage, a unique central node is selected through the collaborative data packet information stored in the distributed collaboration stage; the central node makes the best waveform switching decision through the waveform decision information of each node stored internally, and transmits it to other nodes in the entire network through broadcasting; after receiving the decision information, other nodes adjust their own waveforms according to the decision of the central node to ensure that the entire network reaches a consensus on waveform switching, thereby improving the collaborative efficiency and reliability of the network. The implementation steps include:

[0082] Step 4: The central node is elected through the collaborative data packet information stored in the distributed collaborative stage.

[0083] In order to effectively avoid efficiency problems caused by excessive delay and improve the overall response speed and decision-making timeliness of the system, it is necessary to select the central node with the shortest communication delay in the whole network through the maximum delay, average delay and node address in the coordinated data packets of each node stored in the node. Its implementation includes:

[0084] 4.1) Find the node with the smallest average delay among the node collaboration data stored in the node:

[0085] If the average delay of this node is the smallest and unique, then this node becomes the central node and the process ends;

[0086] If the average delay of this node is not the minimum, the process ends;

[0087] If there are multiple nodes with the same average delay, execute step 4.2).

[0088] 4.2) Compare the maximum delay in the coordination data of each node stored in the node:

[0089] If the maximum delay of this node is the smallest and unique, then this node becomes the central node and the process ends;

[0090] If the maximum delay of this node is not the minimum, the process ends;

[0091] If there are multiple nodes with the same maximum delay, execute step 4.3).

[0092] 4.3) Compare the node addresses in the node coordination data stored in the node:

[0093] If the node address of this node is the smallest, then this node becomes the central node and the process ends;

[0094] Otherwise, it means that this node cannot become the central node, and no changes are made and the process ends.

[0095] Step 5: The central node uses the waveform decision information of each node stored internally to make a majority judgment to make the best waveform switching decision.

[0096] 5.1) The central node traverses the waveform decision information of each node stored internally, counts the frequency of each waveform being selected, and finds the waveform with the highest selection frequency as the candidate waveform;

[0097] 5.2) Determine the waveform switching decision based on the attributes of the candidate waveform:

[0098] If the candidate waveform is unique, then the waveform is used as the final result of the majority decision waveform switching decision;

[0099] If the candidate waveform is not unique, the waveform with stronger anti-interference ability is preferentially selected as the final result of the majority judgment waveform switching decision.

[0100] Example 2: Distributed collaborative waveform switching system for networked computing

[0101] Reference Figure 4 The system of this example includes: a node setting perception state interaction module 1, a cognitive processing module 2, an anti-interference networking decision module 3 and a dynamic control module 4, wherein the dynamic control module 4 includes a central node election submodule 41 and a final waveform decision submodule 42.

[0102] The perception status interaction module 1 is used for periodically interacting with the status packets and collaborative data packets of neighboring nodes, that is, collecting the status information of the node and exchanging collaborative data packets with neighboring nodes in a period of 1 second, and exchanging status packets with neighboring nodes in a period of 2 seconds. It uses the two-dimensional structure array set in the node as the storage space for the status packets and collaborative data packets, and updates the storage by comparing the serial numbers of the status packets and the collaborative data packets.

[0103] The cognitive processing module 2 is used to build an efficient status packet and collaborative data packet processing system, which processes the neighbor node status packets and collaborative data packets stored in the two-dimensional structure array by removing the status packet header and the collaborative packet header, and allocates the neighbor node status packets and collaborative data packets to the set shared buffer.

[0104] The anti-interference networking decision module 3 uses the state information in the shared buffer as input, performs distributed waveform decision-making through the reinforcement learning model DQN, makes the waveform switching strategy that best suits the current network environment, and stores the decision results in the collaborative data packet.

[0105] The dynamic control module 4 is used for the election of the central node and the control of the final waveform decision. It selects a unique central node through the information of the collaborative data stored in the two-dimensional structure array by the node in the distributed collaboration stage, and the central node makes a majority decision to make the best waveform switching decision in the distributed decision consistency stage. Among them:

[0106] The node with the shortest communication delay is selected as the central node through the node setting central node election submodule 41, that is, the average delay of each node is compared with the collaborative data in the collaborative data packet of each node in the whole network stored in the node, and the node with the smallest average delay is preferentially selected; if there are multiple nodes with the same average delay, the maximum delay of each node is further compared, and the node with the smallest maximum delay is preferentially selected; if there are multiple nodes with the same maximum delay, the node with the smallest node address is selected as the only central node;

[0107] The final waveform decision submodule is set through the node, and the majority judgment is made to make the best waveform switching decision. That is, after the central node is elected, the central node traverses the waveform decision information of each node stored internally, calculates the frequency of each waveform being selected, and finds the waveform with the highest selection frequency as the candidate waveform. If the candidate waveform is unique, the waveform is selected as the final waveform decision; if not unique, the waveform with stronger anti-interference ability is preferentially selected as the final waveform decision.

[0108] The effect of the present invention can be further illustrated by the following simulation results:

[0109] 1. Simulation conditions

[0110] Build a container platform through Docker container in the virtual machine to simulate multi-hop network, and write relevant code in the container platform;

[0111] The settings include network size, network topology, and interference environment simulation parameters.

[0112] By testing the effect of majority judgment waveform switching decision in the face of dynamic environment changes under different network scales, and comparing it with the existing distributed collaborative Raft algorithm, the difference in communication delay performance between the two comparison methods is evaluated.

[0113] 2. Simulation content and results

[0114] Under the above simulation conditions, the present invention and the existing distributed cooperative Raft algorithm are used to perform waveform switching in a dynamic environment, and the communication delay performance of the two is compared. The results are as follows: Figure 5 shown.

[0115] from Figure 5 It can be seen that the existing distributed collaborative Raft algorithm has a high communication delay for the central node to send waveform decisions to the entire network due to the randomness of the election stage. However, the present invention can select the central node with the shortest communication delay for decision information to be sent to the entire network, thus significantly reducing the communication delay for the central node to send decision information to the entire network, improving the overall efficiency and stability of the network, and ensuring the consistency of the entire network distribution.

[0116] It should be noted that the step numbers in the specification and claims of the present invention are only for a clear description of the implementation scheme of the present invention to facilitate understanding, and the order of the step numbers is not limited.

Claims

1. A distributed collaborative waveform switching method for networked computing, characterized in that: include: There are two stages: distributed collaboration stage and distributed decision consistency stage; In the distributed collaboration stage, each node in the network periodically exchanges status packets and collaboration data packets to achieve real-time information collaboration and storage between nodes. During this exchange process, distributed waveform decisions and delay information updates are performed in the nodes respectively, and waveform decisions and delay data are dynamically adjusted to optimize communication efficiency and improve anti-interference capabilities. In the distributed decision consistency stage, a unique central node is selected through the collaborative data packet information stored in the distributed collaboration stage; the central node makes the best waveform switching decision by majority judgment based on the waveform decision information of each node stored internally, and transmits it to other nodes in the entire network through broadcasting; After receiving the decision information, other nodes adjust their own waveforms according to the decision of the central node to ensure that the entire network reaches a consensus on 1 waveform switching, thereby improving the network's collaborative efficiency and reliability.

2. The method according to claim 1, characterized in that: The state packet: consists of a state packet header + state information, wherein the state packet header includes a state packet type, a sequence number and a state information offset; the state information includes a network state, an interference state and an environment state; The collaborative data packet consists of a collaborative packet header and collaborative data, wherein the collaborative packet header includes the collaborative data packet type, sequence number and collaborative information offset; the collaborative data includes the node address, location information, average delay, maximum delay, waveform decision information and the current role of the node.

3. The method according to claim 1, characterized in that: Each node in the network periodically exchanges status packets and coordination data packets, which include: (3a) Nodes exchange status packets by periodically sending status packets: (3a1) Each node collects its internal state information at the network layer with a period of 1 second, and merges it with the state information in the state packets of neighboring nodes within three hops stored in the node, and then adds a state packet header to generate a state packet to be sent; (3a2) Each node sends a status packet to a neighboring node via the path of the sending network layer-MAC layer-baseband waveform BB with a period of 2 seconds, and receives the status packet of the neighboring node via the path of the baseband waveform BB-MAC layer-receiving network layer, and stores it in the storage space of the node; (3b) Nodes exchange collaborative data packets by periodically sending collaborative data packets: (3b1) Each node exchanges collaborative data packets at a period of 1 second; (3b2) Each node integrates the collaborative data of all nodes in each cycle, adds a collaborative packet header, and exchanges collaborative data packets between nodes according to the packet sending and receiving paths consistent with (3a2).

4. The method according to claim 1, characterized in that: During the exchange process, distributed waveform decision making is performed within the node, the implementation of which includes: (4a) The node collects and stores the perceived status of the status packet, and when receiving the neighbor node status packet, it updates the storage according to whether the neighbor node status packet exists in the storage space: If the storage space does not have a neighbor node status package, add the neighbor node status package to the storage space; If the storage space has a neighbor node status packet, compare the sequence numbers of the neighbor node status packet A in the storage space and the received neighbor node status packet B: If B's ​​sequence number is greater than A's, the neighbor node status packet is added to the storage space; Otherwise, ignore the neighbor node status packet; (4b) removing the header of the neighbor node status packet received in (4a), and distributing the status information to the set shared buffer according to the node number, so as to maximize the concurrent performance of data reading and writing; (4c) Using the state information in the shared buffer in (4b) as input, waveform decision calculation is performed based on the DQN reinforcement learning model to select the waveform switching strategy that best suits the current network environment; (4d) The decision result of (4c) is stored in the waveform decision information in the collaborative data, and the waveform switching control is performed according to the current node role in the collaborative data: If the current node role is not the central node, it will adjust its own waveform after receiving the waveform decision information sent by the central node; If the current node role is a central node, a majority decision is executed to make the best waveform switching decision and broadcast it to the entire network.

5. The method according to claim 4, characterized in that: In step (4c), waveform decision calculation is performed based on the reinforcement learning model DQN, and its implementation includes: (4c1) obtaining the state information in the shared buffer in (4b), including the frame error rate in the network state, the interference frequency domain occupancy in the interference state, and the transmission rate in the environment state, so as to respectively reflect the reliability of data transmission, the interference degree of the spectrum, and the transmission rate of each node; (4c2) constructing a reinforcement learning model DQN whose input is the state information in step (4c1) and whose output is the predicted Q value for each waveform; (4c3) The reward function is defined as effective throughput: sending rate*(1-frame error rate)+interference penalty, which is used to guide the DQN model to learn the optimal waveform switching strategy to maximize the effective throughput; (4c4) Based on the currently selected waveform and the calculated reward, the DQN model updates its target Q value through the Q-learning algorithm. The update formula is: Where Q(s,a) is the current Q value of taking action a in state s; α is the learning rate, which is used to control the speed of Q value update; R is the immediate reward after taking the current action, that is, the effective throughput; γ is the discount factor, which indicates the impact of future rewards; is the maximum Q value of all possible actions a' in the new state s'; (4c5) After updating the target Q value in (4c4), the error between the output of the reinforcement learning model DQN in (4c2) and the target Q value is calculated, and the weight of the model is adjusted through back propagation to obtain the optimal waveform selection strategy under different state information inputs, thereby improving the effective throughput of the system.

6. The method according to claim 1, characterized in that: During the exchange process, the node performs delay information update, which includes: (6a) Obtain the maximum delay and average delay of all node cooperation data packets stored in the node, which represent the longest transmission delay from the node to the farthest node and the overall communication efficiency between the node and other nodes respectively; (6b) Traverse the maximum delays of all nodes stored in the node, compare them with the maximum delay of this node, check whether there is a larger delay value, and update the maximum delay of this node: If a larger delay value is found, the maximum delay of this node is updated to the larger value; If no larger delay value is found, no change is made; (6c) During the traversal process of (6b), the maximum delay information of all nodes stored in the node is accumulated, the average delay is calculated by weighted average, and the average delay value of this node is updated.

7. The method according to claim 1, characterized in that: In the distributed decision consistency stage, the unique central node is selected through the collaborative data packet information stored in the distributed collaboration stage, and its implementation includes: (8a) Find the node with the minimum average delay stored in the node: If the average delay of this node is the smallest and unique, then this node becomes the central node and the process ends; If the average delay of this node is not the minimum, the process ends; If there are multiple nodes with the same average delay, execute (8b); (8b) Compare the maximum delays of each node stored in the node: If the maximum delay of this node is the smallest and unique, then this node becomes the central node and the process ends; If the maximum delay of this node is not the minimum, the process ends; If there are multiple nodes with the same maximum delay, execute (8c); (8c) Compare the node addresses of each node stored in the node: If the node address of this node is the smallest, then this node becomes the central node and the process ends; Otherwise, end the process.

8. The method according to claim 1, characterized in that: The central node makes the best waveform switching decision by majority judgment based on the waveform decision information of each node stored internally, including: (8a) The central node traverses the waveform decision information of each node stored internally, calculates the frequency of each waveform being selected, and finds the waveform with the highest selection frequency as the candidate waveform; (8b) Determine the waveform switching decision based on the candidate waveform attributes in (8a): If the candidate waveform is unique, then the waveform is used as the waveform switching decision result; If the candidate waveform is not unique, the waveform with stronger anti-interference ability is preferentially selected as the waveform switching decision result.

9. A distributed collaborative waveform switching system for networked computing, characterized in that: include: The node sets up a perception state interaction module, which is used to periodically interact with the neighbor node state packets and the coordination data packets, that is, to update the storage of the neighbor node state packets by comparing the sequence numbers, to collect the node state information and exchange coordination data packets with the neighbor node in a period of 1s, and to exchange state packets with the neighbor node in a period of 2s; The node sets a cognitive processing module to build an efficient status packet and collaborative data packet processing system, that is, to process the status packets and collaborative data packets of neighboring nodes by removing the status packet header and the collaborative packet header, and distribute them to the set shared buffer; The node sets up an anti-interference networking decision module, which uses the state information in the shared buffer as input, makes distributed waveform decisions through the reinforcement learning model DQN, and makes the waveform switching strategy that best suits the current network environment; The node sets a dynamic control module for the election of the central node and the control of the final waveform decision, that is, the only central node is selected from the collaborative data packet information stored by the node in the distributed collaboration stage, and the central node makes the best waveform switching decision by majority judgment in the distributed decision consistency stage.

10. The system according to claim 10, characterized in that The node setting dynamic control module includes: The node sets a central node election submodule, which is used to elect the node with the shortest communication delay as the central node, that is, the average delay of each node is compared through the collaborative data in the collaborative data packet of each node in the whole network stored in the node, and the node with the smallest average delay is preferentially selected; if there are multiple nodes with the same average delay, the maximum delay of each node is further compared, and the node with the smallest maximum delay is preferentially selected; if there are multiple nodes with the same maximum delay, the node with the smallest node address is selected as the only central node; The node sets the final waveform decision submodule, which is used to make the best waveform switching decision by majority judgment. That is, after the central node is elected, the central node traverses the waveform decision information of each node stored internally, calculates the frequency of each waveform being selected, and finds the waveform with the highest selection frequency as the candidate waveform. If the candidate waveform is unique, the waveform is selected as the final waveform decision; if not unique, the waveform with stronger anti-interference ability is preferentially selected as the final waveform decision.

Images