A target communication network topology restoration method for a perception network

Abstract

The application belongs to the technical field of communication network topology reasoning, and particularly relates to a target communication network topology restoration method for a perception network. The application comprises continuously receiving signals radiated by a target communication network by using the perception network; extracting discrete communication events from the continuously collected received signals at fixed time intervals; modeling all the communication events as an improved multi-dimensional Hawkes process, and arranging a likelihood function; solving the likelihood function by using an expectation maximization algorithm, and extracting the target communication network topology from a solution graph structure. The application models the communication events of the target communication network received by the perception network as a unified model, can comprehensively consider the influence of the perception network nodes and the communication network nodes on the communication, and outputs a graph structure expression of the comprehensive perception network and target communication network, which is more expressive than a general matrix. The final graph structure of the perception network and the target communication network can be calculated through multiple iterations of the expectation maximization algorithm.

Description

Technical Field

[0001] This invention belongs to the field of communication network topology reasoning technology, specifically relating to a method for restoring the topology of a target communication network for sensing networks. Background Technology

[0002] In the field of communication network topology reasoning, topology reasoning for non-cooperative wireless communication networks is a very important research direction. In recent years, many experts and scholars have proposed many efficient and reliable methods for topology reasoning in non-cooperative wireless communication networks.

[0003] In 2019, Wu Qihui et al. disclosed a single-sensor topology sensing method and device for unreliable information (Chinese Patent Application No.: 201911042150.8). This method models the communication process as a multidimensional Hawkes process, then uses the method of minimizing the negative log-likelihood to obtain the preliminary adjacency matrix, and then filters the adjacency matrix by setting a threshold. The final adjacency matrix obtained is the topology information of the target network. This method uses only one sensor to receive all signals from the target network, which places high demands on the sensor. The sensor needs to have a low noise figure, and the scale of the target network it can sense is still limited, lacking scalability.

[0004] In 2021, Liu Zitong et al. disclosed a topology-aware method, device, and system for non-cooperative wireless networks (Chinese Patent Application No.: 202110576245.9). This method compares Granger causality between communication nodes pairwise and uses the mean of the Granger causality zero distribution to filter causal relationships to obtain the final target communication network topology. This method, which relies on pairwise judgment of connectivity to reconstruct the final topology, is inefficient and cannot consider the combined effects of multiple nodes.

[0005] In 2022, Song Yehui et al. published an article titled "Communication Topology Inference Technology for Non-cooperative UAV Communication Networks" in the Journal of Electronics and Information Technology, revealing their research results on capturing the communication topology of UAV communication networks. This article does not limit the sensing system to specific forms such as centralized or distributed systems. It reconstructs the target network topology by analyzing the correlation, transfer entropy, and Granger causality inference between signal sources among multiple signals, or by modeling communication events in all signals as multidimensional Hawkes processes and then reconstructing the target network topology through a solution process. The communication topology inference technology in this article only considers centralized or distributed sensing systems as enhanced single sensors with a larger sensing range. However, in addition to extending the sensing range of a single sensor, the sensing network can obtain more information about the target communication network and its relationship with the sensing network that a single sensor cannot obtain at a single location. This information is beneficial for more accurately reconstructing the target network topology. Summary of the Invention

[0006] The purpose of this invention is to provide a method for restoring the topology of a target communication network for sensing networks.

[0007] A method for restoring the topology of a target communication network for sensing networks includes the following steps:

[0008] Step 1: Use a sensing network to continuously receive signals radiated by the target communication network;

[0009] Step 2: Extract discrete communication events from continuously acquired received signals at fixed time intervals;

[0010] Step 3: Model all communication events as an improved multidimensional Hawkes process and derive the likelihood function;

[0011] Step 4: Use the expectation-maximization algorithm to solve the likelihood function and extract the target communication network topology from the solution graph structure.

[0012] Furthermore, in step 1, the sensing network includes sensing network nodes and a main unit; the sensing network nodes collect the time-duration signal T and then transmit the collected signal S. ij (t) and the start time of the record. Send to the switchboard; signal S ij (t) represents the signal received by sensing network node i from target communication network node j; the main unit traverses all start collection times, sets the latest start collection time t1 as the overall start time, and sets the signal S collected by each sensing network node as... ij The portion of (t) earlier than t1 is removed.

[0013] Furthermore, step 2 specifically involves: using t1 as the first sampling point, and using a fixed time interval T... s Equal intervals and equal number of times for all signals S ij (t) Sampling; T s <T w T w This refers to the duration of the communication frame data.

[0014] Furthermore, the improved multidimensional Hawkes process in step 3 is expressed as follows:

[0015]

[0016] Where, λ ij (t) represents the intensity of the communication event observed by the i-th node of the sensing network at time t, which is the intensity of the communication event observed by the j-th node of the target communication network. ij This corresponds to the basic event intensity; G C It is a graph structure representation of the target communication network; E CDIt is a graph representation of the relationship between the perception network and the target communication network; N ij (t) represents the number of communication events observed by the i-th node of the sensing network at time t and before; t k Represents the k-th timestamp; φ(t) is the influence function, φ(t) = Gγ(t); G = {G C E CD G D}, G is the graph structure representation shared by the perception network and the target communication network, G D It is a graph structure representation of the perceptual network, and γ(t) is a kernel function that reflects the influence of historical events on the current event;

[0017] The likelihood function of the improved multidimensional Hawkes process is:

[0018]

[0019] in,

[0020] Furthermore, step 4 specifically involves: based on the timestamp t n-1 and all communication events S previously observed by the perception network ij (n-1) Estimate the timestamp t n The graph structure representation of the perception network and the target communication network G(t) n This process is called the E-step; then, according to G(t) n Estimate t n Communication event S i ' j (n), calculate S i ' j (n) and the observed S ij (n) Error, then add the previously observed data to the error and re-estimate the graph structure. This process is called the M-step. Iterate the E-step and M-step repeatedly until they converge to obtain the final optimal graph structure. Filter out the nodes and edges of the communication target network in the graph structure and organize them into a matrix. Binarize the topology matrix to obtain the final network topology.

[0021] Furthermore, step 4 specifically includes:

[0022] Step 4.1: Initialize parameter N true =0, t=t m ;

[0023] Step 4.2: Calculate λ according to the improved multidimensional Hawkes process formula. ij (t);

[0024] Step 4.3: If λ ij(t)≥λ T Then determine S i ′ j (t) is a communication event; otherwise, determine S. i ′ j (t) is not a communication event;

[0025] Step 4.4: If S i ′ j (t)==S ij (t), then let N t =N t +1;

[0026] Step 4.5: If t m If ≤t≤T, then let t=t+1 and return to step 4.1.2; otherwise, proceed to step 4.1.6.

[0027] Step 4.6: Convert the error between the predicted event and the actual event into the posterior probability P(Θ) of Θ;

[0028]

[0029] Step 4.7: Calculation Partial derivatives with respect to each node x

[0030] Step 4.8: Calculate the parameter increment of P(Θ) at each node.

[0031] Step 4.9: Add the increments of each partial derivative to the implicit parameter Θ to obtain the new implicit parameter Θ′. x ;

[0032]

[0033] Step 4.10: Repeat steps 4.1 to 4.9 multiple times until the implicit parameters finally converge, obtaining the final result.

[0034] Step 4.11: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] Medium variables The target network communication nodes are organized into a matrix. The final predicted target network topology.

[0035] The beneficial effects of this invention are as follows:

[0036] This invention models the communication events received by the sensing network from the target communication network into a unified model, comprehensively considering the impact of sensing network nodes and communication network nodes on communication. This invention can account for scenarios where a sensing network node receives no signal, receives a signal from one communication network node, or receives signals from multiple communication network nodes. All communication events observed by the sensing network can participate in the calculation of network topology reconstruction. The output of this invention is a graph structure representation of the sensing network and the target communication network, which is richer in representation than a typical matrix. It does not require rigorous analytical solutions; the final graph structure of the sensing network and the target communication network can be calculated through multiple iterations using the expectation-maximization algorithm. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of a scenario according to the present invention.

[0038] Figure 2 This is the overall flowchart of the present invention.

[0039] Figure 3 This is a diagram of the execution content architecture of the sensing network PBX in this invention.

[0040] Figure 4 This is the E-step flowchart of the maximum expectation algorithm for the improved multidimensional Hawkes process in this invention.

[0041] Figure 5 This is the M-step flowchart of the maximum expectation algorithm for the improved multidimensional Hawkes process in this invention. Detailed Implementation

[0042] The present invention will now be further described with reference to the accompanying drawings.

[0043] This invention provides a method for reconstructing the topology of a target communication network using a sensing network, relating to the field of communication network topology reasoning. The method includes: continuously receiving signals radiated by the target communication network using a sensing network; extracting communication events from the received signals at fixed time intervals; modeling all communication events as an improved multidimensional Hawkes process; deriving a likelihood function; solving the likelihood function using an expectation-maximization algorithm to obtain a graph structure of the sensing network and the target communication network; and extracting the topology of the target communication network from the graph structure. This invention is specifically designed to reconstruct the target network topology using wireless communication target network signal data received by the sensing network. The reconstruction process does not require the deletion of signal data received by the sensing network, thus fully utilizing the signals received by the sensing network.

[0044] (1) Use a sensing network to continuously receive signals radiated by the target communication network;

[0045] The target communication network transmits data wirelessly and has no cooperative relationship with the sensing network. The network nodes use omnidirectional antennas to transmit signals, so the signals can be received not only by the target communication nodes but also by nodes in the sensing network.

[0046] The number of nodes in the sensing network and the communication network can be the same or different. The sensing network can be concentrated on one side of the communication network or surround the communication network.

[0047] The sensing network has a central control unit responsible for running algorithms for extracting communication events and restoring topology structures.

[0048] The relationship between a sensing network node and a target communication network node can be categorized into three types: a sensing network node cannot receive signals from a target communication network node, a sensing network node can receive signals from a single node in the target communication network, and a sensing network node can receive signals from multiple nodes in the target communication network.

[0049] The number of signal sequences received by the sensing network is not directly related to the number of sensing network nodes or the number of target communication network nodes. Only when the number of sensing network nodes equals the number of target communication network nodes and the sensing network nodes can receive signals from each node of the target communication network without repetition or omission, is the number of signal sequences received by the sensing network equal to the number of sensing network nodes and the number of target communication network nodes.

[0050] When multiple nodes in a sensing network receive signals from the same target communication network node, the signals from the same source received by each sensing network node need to be preserved.

[0051] When the sensing network continuously receives signals radiated by the target communication network, the communication relationship of the target communication network must remain unchanged.

[0052] The target communication network transmits data wirelessly and has no cooperative relationship with the sensing network. Nodes of the sensing network may not receive signals from the target communication network, or they may receive signals from one or more nodes of the target communication network.

[0053] (2) Extract communication events from the received signals at fixed time intervals;

[0054] All sensing nodes send signals and their corresponding start times to the central control unit, which then aligns all signals and sets intervals T. s Extract communication events, T s Less than the communication frame data duration T w .

[0055] Continuously acquired signals cannot be directly used in the modeling method of this invention; it is necessary to extract discrete communication events S(n) from the continuously acquired signals S(t).

[0056] In this invention, a communication event refers to a target communication network node sending a frame of data.

[0057] Sampling duration T s Less than the communication frame data length T w This ensures that no communication events are missed.

[0058] Because the time it takes for the signal acquisition command to be transmitted to each sensing network node and the device delay of each sensing network node are different, the start time for each sensing network node to acquire signals is different. If each node extracts communication events independently, there will be time deviations between the communication events.

[0059] Therefore, after the sensing network nodes complete signal acquisition, they send the signals to the main control unit. The main control unit sets the latest time to start signal acquisition as the signal acquisition start time t1 for the entire sensing network, and assigns the portion of the signals acquired by each node earlier than t1 to the main control unit. ij (t), t < t1 is removed.

[0060] Assuming t1 starts with a timestamp interval T s Extract communication time.

[0061] (3) Model all communication events as improved multidimensional Hawkes processes and derive the likelihood function;

[0062] The improved multidimensional Hawkes process is a stochastic process model proposed in this invention. Compared with the general multidimensional Hawkes process, its influence function φ uu' The matrix w in (t) and the likelihood function parameter Θ uu' The model was replaced with a graph structure G, and all communication events observed by the perceptual network were included in the improved multidimensional Hawkes process model.

[0063] The method of this invention models all communication events observed by the sensing network as an improved multidimensional Hawkes process, which is expressed as follows:

[0064]

[0065] Where, λ ij (t) represents the intensity of the communication event observed by the i-th node of the sensing network at time t, which is the intensity of the communication event observed by the j-th node of the target communication network. ij It corresponds to the basic event strength, G. C It is a graph structure representation of the target communication network, E CD It is a graph structure representation of the relationship between the perception network and the target communication network, N ij (t) represents the number of communication events observed by the i-th node of the sensing network at time t and before, where t is the number of such events. k This represents the k-th timestamp.

[0066] φ(t) is the influence function, expressed in the following form:

[0067] φ(t)=Gγ(t)

[0068] Where, G = {G C E CD G D}, G is the graph structure representation shared by the perception network and the target communication network, see Appendix Figure 1 G D It is a graph structure representation of the perceptual network, and γ(t) is the kernel function that reflects the influence of historical events on the current event, which is usually an exponential function.

[0069] The improved multidimensional Hawkes process is a point process model proposed in this invention specifically for expressing communication events observed by a sensing network in a target communication network. Compared to the general multidimensional Hawkes process, the improved multidimensional Hawkes process also considers the influence of sensing network nodes with the same field of view.

[0070] The same field of view is represented by the communication event S observed by node i in the perceptual network. ij (n) and the communication events S observed by the sensing network node i' i'j (n) come from the same target network communication node j.

[0071] The improved multidimensional Hawkes process can utilize all communication events observed by the sensor network, i.e., all S ij (n) can all be used as inputs for solving the improved multidimensional Hawkes process.

[0072] The likelihood function of the improved multidimensional Hawkes process is expressed as follows:

[0073]

[0074] in, As can be seen, unlike the general multidimensional Hawkes process likelihood function Θ which contains a matrix, the improved multidimensional Hawkes process likelihood function Θ contains a graph.

[0075] (4) Solve the likelihood function using the expectation maximization algorithm, and extract the target communication network topology from the solution graph structure;

[0076] The graph structure G consists of three parts: the perceptual network graph G. D Target communication network diagram G C And the relationship between the sensing network and the target communication network (Diagram E) CD , will G C By rearranging the data into a matrix, we can obtain the matrix representation W of the target network topology.

[0077] The expectation-maximization algorithm is a well-known algorithm for solving non-convex optimization functions. In this invention, the expectation-maximization algorithm is applied to solve the improved multidimensional Hawkes process, and its steps are as follows:

[0078] Step 4.1: Based on timestamp t n-1 and all communication events S previously observed by the perception network ij (n-1) Estimate the timestamp t n The graph structure representation of the perception network and the target communication network G(t) n This process is called the E-step;

[0079] Step 4.2: Then, according to G(t) n Estimate t n Communication event S i ' j (n), calculate S i ' j (n) and the observed S ij (n) error, then add the previously observed data to the error and re-estimate the graph structure. This process is called M-step.

[0080] Step 4.3: Iterate this process repeatedly until it converges to obtain the final optimal graph structure;

[0081] Step 4.4: Filter out the nodes and edges of the communication target network in the graph structure and organize them into a matrix form;

[0082] Step 4.5: Binarize the topology matrix to obtain the final network topology.

[0083] The advantages of this invention are: 1. It models the communication events of the target communication network received by the sensing network as a unified model, which can comprehensively consider the impact of sensing network nodes and communication network nodes on communication; 2. It can consider the cases where the sensing network node does not receive a signal, receives a signal from one communication network node, and receives signals from multiple communication network nodes; 3. All communication events observed by the sensing network can participate in the calculation of network topology reconstruction; 4. The method outputs a graph structure representation of the sensing network and the target communication network, which is richer in expressive power than a general matrix; 5. It does not require strict analytical solution, and the final graph structure of the sensing network and the target communication network can be calculated by multiple iterations of the expectation-maximization algorithm.

[0084] Example 1:

[0085] Step 1. Multiple computers form a wireless self-organizing network through a router, with each computer equipped with a general software radio peripheral as a receiver for the sensing network;

[0086] Step 2. Frequency division multiplexing is used at each node of the communication network, with one frequency point corresponding to one node;

[0087] Step 3. Each node in the communication network sends data to each other according to the pre-set communication relationship. After sending one frame of data, a node waits for the destination node to return a frame of data before sending the next frame of data.

[0088] Step 4. Arrange the nodes of the sensing network so that all nodes can receive signals from the target communication network;

[0089] Step 5. Observe the length of the communication node frame data, and denot it as T. w ;

[0090] Step 6. After the sensing network node collects the signal for duration T, it transmits the collected signal S. ij (t) and the start time of the record. Send to the main switchboard, see attachment Figure 3 S211;

[0091] Step 7. The switchboard iterates through all start times. Find the latest start time Set as the overall start time, see Figure 3 S212;

[0092] Step 8. Extract all signals and remove S. ij (t), See the part. Figure 3 S213;

[0093] Step 9. For the first sampling point, T s All pairs of numbers S with equal intervals ij (t), Sampling is performed to obtain the complete target communication network events observed, where T s Less than T w ,See Figure 3 S214;

[0094] Step 10. Set the kernel function γ(t) as a monotonically decreasing exponential function;

[0095] Step 11. Randomly initialize the implicit parameter Θ;

[0096] Step 12. Specify event S ij A portion of S (t) ij (t), t≥t m Used to verify the estimate;

[0097] Step 13. Based on the improved multidimensional Hawkes process intensity function formula, and based on S... ij(t), t≥t m The prediction of the event at the current timestamp is made based on all events prior to each timestamp. (See appendix.) Figure 4 S421 to the attached Figure 4 S425;

[0098] Step 14. Convert the error between the predicted event and the actual event into the posterior probability P(Θ) of Θ, see Appendix. Figure 4 S426;

[0099] Step 15. Adjust Θ according to the P(Θ) value, see appendix. Figure 5 ;

[0100] Step 16. Repeat steps 13 to 15 multiple times to obtain the final Θ, denoted as Θ.

[0101] Step 17. Medium variables The target network communication nodes are organized into a matrix. The final predicted target network topology.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for restoring the topology of a target communication network for sensing networks, characterized in that, Includes the following steps: Step 1: Use a sensing network to continuously receive signals radiated by the target communication network; Step 2: Extract discrete communication events from continuously acquired received signals at fixed time intervals; Step 3: Model all communication events as an improved multidimensional Hawkes process and derive the likelihood function; The improved multidimensional Hawkes process is represented as: in, Represents the perceptual network's first... Each node Observe the target communication network at all times. The intensity of communication events occurring at each node, This corresponds to the basic event intensity; It is a graph structure representation of the target communication network; It is a graph representation of the relationship between the perception network and the target communication network; It is the first perceptual network Each node The target communication network was observed at time and prior. The number of communication events occurring at each node; Indicates the first A timestamp; For the influence function, ; , It is a graph structure representation shared by both the perception network and the target communication network. It is a graph structure representation of the perceptual network. The kernel function reflects the impact of historical events on the current event; Step 4: Solve the likelihood function using the expectation-maximization algorithm to extract the target communication network topology from the solution graph structure; Based on timestamp and all communication events previously observed by the perception network. Estimate the timestamp Graph structure representation of the perception network and the target communication network This process is called the E-step; then according to Estimate Communication events ,calculate Compared with what has been observed The error is then added to the previously observed data to re-estimate the graph structure; this process is called the M-step. The E-step and M-step are iterated repeatedly until convergence is achieved, and the final optimal graph structure is obtained. The nodes and edges of the communication target network in the graph structure are filtered out and arranged into a matrix. The topology matrix is ​​binarized to obtain the final network topology.

2. The method for restoring the topology of a target communication network for a sensing network according to claim 1, characterized in that: In step 1, the sensing network includes sensing network nodes and a main unit; the sensing network nodes collect data. The signal will be acquired after the duration signal. and the recorded start time Send to switchboard; signal For sensing network nodes Received from the target communication network node The signal; the switchboard traverses all start collection times and selects the latest start collection time. Set the overall start time, and collect the signals from each sensing network node. earlier than Partial removal.

3. The method for restoring the topology of a target communication network for a sensing network according to claim 2, characterized in that: Step 2 specifically involves: using For the first sampling point, at a fixed time interval All signals at equal intervals and for equal number of times sampling; , This refers to the duration of the communication frame data.

4. The method for restoring the topology of a target communication network for a sensing network according to claim 2, characterized in that: In step 3, the likelihood function of the improved multidimensional Hawkes process is: in, .

5. The method for restoring the topology of a target communication network for a sensing network according to claim 1, characterized in that: Step 4 specifically involves: Step 4.1: Initialize parameters , ; Step 4.2: Calculate based on the improved multidimensional Hawkes process formula. ; Step 4.3: If Then determine It is a communication event; otherwise, determine This is not a communication event; Step 4.4: If Then let ; Step 4.5: If Then let Return to step 4.1.2; Otherwise, proceed to step 4.1.6; Step 4.6: Convert the error between the predicted event and the actual event into... posterior probability ; Step 4.7: Calculation For each node partial derivatives ; Step 4.8: Calculation Increment of parameters at each node ; Step 4.9: Implicit Parameters Adding the increments of each partial derivative yields new implicit parameters. ; Step 4.10: Repeat steps 4.1 to 4.9 multiple times until the implicit parameters finally converge, obtaining the final result. ; Step 4.11: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] Medium variables The target network communication nodes are organized into a matrix. , The final predicted target network topology.

Images