Cross-zone cooperation adaptive handover decision method in ultra-dense heterogeneous wireless network

Abstract

The application claims a cross-zone cooperation adaptive handover decision method in an ultra-dense heterogeneous wireless network, belonging to the field of mobile communication, and specifically comprising the following steps: first, an improved Kalman filter position prediction model is obtained according to the historical trajectory information of a vehicle, and the positions of the terminal at two time points are predicted. Second, the candidate network set for handover and hopping is generated in advance according to the predicted positions. Then, by defining a hopping factor and using an interval number multi-attribute decision algorithm of modified Jaccard similarity, a cross-zone cooperation adaptive handover decision scheme is proposed to generate an optimal handover strategy for the terminal. Finally, experimental simulation shows that the algorithm can reduce the handover times of the vehicle-mounted terminal, reduce the handover failure rate and improve the transmission efficiency of the network.

Description

Technical Field

[0001] This invention pertains to vertical handover methods in ultra-dense heterogeneous wireless networks and falls under the field of mobile communications. Specifically, it relates to a cross-regional cooperative adaptive handover decision-making method. Background Technology

[0002] With the evolution of 5G technology, deploying small cell base stations on infrastructure such as utility poles, streetlights, and buses in urban areas has become possible. The dense deployment of small cell networks can improve spectrum efficiency and network access capacity, creating conditions for the explosive growth of data transmission in vehicle-to-everything (V2X) networks. However, in vehicle-organized heterogeneous wireless networks, the high dynamic movement of vehicles and the miniaturization of cell structures mean that onboard terminals face the predicament of constantly switching between networks. This inevitably increases signaling overhead and the risk of link disconnection, thus affecting user experience. Therefore, how to minimize the number of handovers while ensuring the quality of service for onboard terminals through vertical handover algorithms, addressing the frequent handover problem caused by highly dynamic onboard terminals continuously traversing 5G microcells and WiFi networks, has become a hot research topic in this field.

[0003] Currently, many studies have addressed the frequent handover problem in heterogeneous wireless networks, and all have achieved certain results. For example, the literature [Palas MR, Islam R., Roy P., et al. Multi-criteria handovermobility management in 5G cellular network[J]. Computer Communications, 2021, 174(8): 81-91] proposes a multi-attribute vertical handover algorithm based on mobility trend quantization. By considering the mobility trend quantization parameters of the terminal to predict the target area of ​​the terminal, it alleviates the problem of excessive handover in ordinary multi-attribute decision algorithms. The literature [Yang Mingji, Wu Ye, Fan Huafeng. Vertical handover algorithm for heterogeneous vehicle-to-everything network based on motion trend prediction[J]. Microelectronics & Computer, 2018, 35(4): 119-123, 129.] calculates the duration of vehicle terminal access to the base station by predicting the vehicle's motion trend, and divides the terminal into narrow mobility nodes and wide mobility nodes according to this time, and then executes the corresponding handover strategy, thereby reducing handover latency. The literature [Tokuyama K., Kimura T., Miyoshi N. Data rate and handoff rate analysis for user mobility in cellular networks[C] / / 2018IEEE Wireless Communications and Networking Conference(WCNC). Barcelona, ​​Spain:IEEE Press 2018:1-6.] proposes a time-based hop handover algorithm, which controls the handover frequency of mobile users by setting a hop time threshold, thereby reducing the handover rate of terminals. The literature [Al-Naffouri, Tareq Y., ElSawy, et al. Velocity-aware handover management in two-tier cellular networks[J].IEEE Transactions on Wireless Communications, 2017, 16(3):1851-1867.] proposes a speed-aware handover scheme, which establishes a speed-aware model based on stochastic geometry theory, thereby determining the base stations that need to be skipped on the terminal's movement trajectory, thus reducing the handover failure rate.The paper [Costa A, Pacheco L, D Rosário, et al. Skipping-based handover algorithm for video distribution over ultra-dense VANET[J]. Computer Networks, 2020, 176: 1-12.] proposes a multi-attribute-based hopping handover algorithm by designing a hopping mechanism that integrates mobility prediction, quality of service, and quality of experience awareness. This improves the reliability of handover and alleviates ping-pong handover.

[0004] While the aforementioned literature can alleviate frequent handovers to some extent, in vehicle-organized heterogeneous wireless network environments, the problem becomes more severe if only motion trends are predicted without accurate analysis of the specific location changes of the vehicular terminals. Furthermore, although the hop-based handover algorithms mentioned in these studies reduce the number of handovers to some extent, they fail to determine the target network the terminal can access after a hop. To address these issues, this paper proposes a vertical handover algorithm based on Location Prediction and Cross Region Cooperation (LPCRC). This algorithm first introduces an improved Kalman Filter (IKF) model to predict the terminal's location at the next two time steps, generating a set of candidate networks for handover and hops in advance. Then, by defining a hop factor and employing a multi-attribute decision algorithm with modified Jaccard similarity, an adaptive cross-region cooperation handover decision scheme is proposed to generate the optimal handover strategy for the terminal. Summary of the Invention

[0005] This invention aims to solve the problems of the prior art mentioned above. It proposes a cross-regional cooperative adaptive handover decision method in ultra-dense heterogeneous wireless networks. The technical solution of this invention is as follows:

[0006] A cross-area cooperative adaptive handover decision method in ultra-dense heterogeneous wireless networks includes the following steps:

[0007] 101. Handover Triggering Steps: Periodically collect the signal strength and network bandwidth of the vehicle terminal in the current network, calculate the handover triggering factor α, and if the handover triggering factor α of terminal i in network j is... ij If (t) = 1, a switch is triggered; otherwise, no switch is triggered.

[0008] 102. Mobility prediction steps: The vehicle's motion data is collected by the positioning device on the vehicle and recorded in the historical trajectory database. When the vehicle terminal triggers a handover, the vehicle's historical trajectory information is used to predict the terminal's position at the next two time points through the IKF model.

[0009] 103. Handover Decision Steps: First, based on the predicted location, a candidate network set CNS_1 for handover and a cooperative network set CNS_2 for hopping are generated in advance. Then, the hopping factor is defined using network topology and terminal motion state, and the score of each network in CNS_1 and CNS_2 is calculated using an interval number multi-attribute decision algorithm that corrects Jaccard similarity. Finally, based on the hopping factor and network score results, the optimal handover strategy is generated for the terminal.

[0010] Furthermore, step 101 involves periodically collecting the signal strength and network bandwidth of the vehicle-mounted terminal in the current network and calculating the switching trigger factor α, specifically including:

[0011] 201. Received signal strength: At time t, the received signal strength of terminal i accessing network j is expressed as:

[0012] RSS ij (t)=P j -ηlgdis ij (t)+μ (0,σ) (1)

[0013] Among them, P j Let η represent the wireless signal transmit power of network j, and η represent the path loss factor. ij (t) represents the distance from terminal i to network j at time t, μ (0,σ) It is a Gaussian random variable with a mean of 0 and a variance of σ.

[0014] 202. Network bandwidth: The network bandwidth obtained by terminal i when accessing network j at time t can be expressed as:

[0015]

[0016] Where u represents the number of vehicle terminals accessing network j at time t, and B r The prb represents the bandwidth of each resource block. ij (t) represents the number of resource blocks allocated to vehicle terminal i in network j, PRB j Let represent the maximum number of resource blocks that network j can provide. Therefore, the handover triggering factor α of vehicle terminal i in its original access network j at time t can be expressed as:

[0017]

[0018] Among them, when RSS is lower than the set threshold RSS th The sum of the hysteresis margin (HM) and the bandwidth is lower than the minimum bandwidth requirement for terminal i to run its services. At that time, α ij (t) = 1, a switch needs to be triggered; otherwise, α ij (t) = 0, no trigger.

[0019] Furthermore, step 102 involves collecting vehicle motion data using the vehicle's positioning device and recording it in a historical trajectory database. When the vehicle terminal triggers a switch, the historical trajectory information is used to predict the terminal's position at the next two time points using an IKF model. Specifically, this includes:

[0020] 301. Theoretical prediction: Assume the vehicle's state at time t is S(t) = [la(t), lo(t)] T Where la(t) is the dimension data at time t, and lo(t) is the longitude data; if the best estimated state at time t-1 is S′(t-1), then the predicted state at time t is estimated according to the theoretical model as follows:

[0021]

[0022] Where F represents the state transition matrix, which describes how the state transitions from the previous time step to the next state. To represent the prediction noise, the error is expressed using the covariance matrix:

[0023]

[0024] Use Q p To represent the noise in the prediction model, substituting equation (4) into equation (5) yields the following process for the propagation of the error covariance matrix between adjacent time steps:

[0025] P′(t)=FP(t-1)F T +Q p (6)

[0026] 302. Let Z(t) be the state observed by the GPS positioning device at time t, S(t) be the observation matrix, and Q be the observation noise. g Then, the transformation process from the predicted state to the observed state of the vehicle at time t can be expressed as:

[0027] Z(t)=HS(t)+Q g (7)

[0028] 303. State Update: The predicted state and observed state at time t are obtained in formulas (4) and (7), respectively. The predicted value is corrected by the observed value, so that the corrected best estimated state is obtained as follows:

[0029]

[0030] in Let K(t) be the residual between the actual and expected observations, and K(t) be the Kalman gain at time t. The calculation process is as follows:

[0031] K(t)=P′(t)H T (HP′(t) T H T +Q g ) -1 (9)

[0032] The role of Kalman gain is to balance the predicted state covariance P and the observed state covariance Q. g The size of the Kalman gain is used to determine the proportion of the roles of the prediction model and the observation model in the prediction process. After obtaining the Kalman gain, the noise covariance matrix P(t) of the best estimated state needs to be updated for the next prediction, where E is the identity matrix.

[0033] P(t)=(Ε-K(t)H)P′(t) (10)

[0034] 304. Improved Prediction Model: An improved Kalman filter position prediction model is obtained by introducing an attenuation memory filtering method.

[0035] Furthermore, step 103 generates a candidate network set CNS_1 for switching and a cooperative network set CNS_2 for hopping, specifically including:

[0036] Assuming that vehicle terminal i triggers a handover at time t, its position at time t+1 can be predicted using the historical motion trajectory of the vehicle terminal at the previous t times according to the IKF model. When the user's handover request arrives, the backend discovers all networks within the connection range based on the predicted position and uses the network obtained at this position as the candidate network set after triggering the handover, denoted as CNS_1. Similarly, based on the historical motion trajectory at the previous t+1 times, the position at time t+2 can be predicted, and the network obtained at this position is used as the cooperative network set for skip handover after the handover is triggered, denoted as CNS_2. CNS_1 and CNS_2 are collectively referred to as the candidate network set.

[0037] Furthermore, the handover decision parameters specifically include: when the vehicle terminal triggers a handover, a new network needs to be selected for the terminal to access from the candidate network sets CNS_1 and CNS_2; since data transmission rate, network latency, and packet loss rate are key indicators for measuring the performance of the access network during the vehicle terminal's movement, these three parameters are used to evaluate the network performance; due to the influence of network topology and terminal movement state, accessing a new target network may cause frequent handovers of the terminal, a skip factor is defined, and networks in the candidate network set that are prone to causing frequent handovers are marked as networks that need to be skipped.

[0038] Furthermore, the calculation formulas for the data transmission rate, network latency, and packet loss rate specifically include:

[0039] Data transmission rate: According to Shannon's formula, the data transmission rate of a terminal accessing the network is related to bandwidth and signal-to-noise ratio parameters. The data transmission rate e obtained by terminal i accessing network j at time t is... ij (t) can be expressed as:

[0040] e ij (t)=B ij (t)×log2(1+SNR ij (t)) (11)

[0041] Among them, B ij (t) represents the bandwidth resources allocated by network j to terminal i at time t, and SNR. ij (t) represents the signal-to-noise ratio, which is approximately the ratio of RSS to the interference noise I in the network;

[0042] (2) Network latency: The relationship between latency and network load is set as an exponential function; if the latency of network j is d′ j Then the time delay d for terminal i to access network j at time t is... ij (t) can be expressed as:

[0043]

[0044] (3) Packet loss rate: The packet loss rate is the ratio of the number of lost data packets to the total number of sent data packets within a certain period of time. Assume that the number of data packets sent by network j in the first t time steps is ψ. total The number of confirmed data packets received is ψ ack Then, at time t, the packet loss rate γ of terminal i accessing network j is... ij (t) can be expressed as:

[0045]

[0046] (4) Skip Factor: Define a skip factor δ to mark the networks that need to be skipped in the candidate network set. After the handover is triggered at time t, the skip factor δ of terminal i in the j-th candidate network is... ij (t) can be expressed as:

[0047]

[0048] In the formula, cl represents the coverage area of ​​candidate network j. j Let τ represent the chord length of the trajectory of candidate network j. ij This indicates the dwell time of terminal i in network j.

[0049] Furthermore, the specific steps of the interval number multi-attribute decision algorithm for correcting Jaccard similarity in step 103 are as follows:

[0050] (1) Constructing the interval number decision matrix: Assume there are N networks in the network set to be evaluated, and M network attributes are involved in the evaluation. Before making a decision, the interval numbers of the M attributes of the N networks are collected. The interval number of each network attribute is determined by the maximum and minimum values ​​obtained from multiple data samplings. The maximum and minimum values ​​obtained by sampling the k-th attribute of network j are respectively... and Then the interval number of the k-th attribute of network j can be expressed as: Therefore, the interval number decision matrix to be decided can be represented as:

[0051]

[0052] (2) Normalized attribute interval number: For the interval number matrix Normalization is performed to obtain the normalized matrix, denoted as . in Formulas (15) and (16) represent the normalization processes for benefit-based and cost-based network parameters, respectively:

[0053]

[0054]

[0055] (3) Determine the interval-type ideal scheme: In order to better measure the differences between networks, assume that the interval-type ideal scheme of each network attribute is Θ=[Θ1,Θ2,...,Θ M ], where Θ k It can be represented as:

[0056]

[0057] (4) Calculate the corrected Jaccard similarity: Jaccard similarity is used to describe the similarity and difference between sets. The larger the Jaccard similarity, the higher the similarity between the sets. Since interval numbers can also be regarded as sets of numbers, the Jaccard similarity of the normalized attribute values ​​with respect to the ideal solution Θ can be expressed as:

[0058]

[0059] Since the similarity between two interval numbers cannot be compared using the Jaccard similarity score when their midpoints are the same, the right endpoints of the interval numbers can be added to the Jaccard calculation to correct them. The corrected Jaccard similarity score can be expressed as:

[0060]

[0061] Therefore, the corrected Jaccard similarity of each network attribute value in the normalized decision matrix corresponding to the ideal solution can be represented by the matrix ζ = (ζ jk ) NM ;

[0062] (5) Determine the optimal weights for each network attribute: Determine the weights of each network attribute in the decision-making process based on minimizing the sum of the biases. The corresponding optimization model is as follows:

[0063]

[0064] (6) Calculate the overall similarity: After obtaining the weights of each network attribute, the overall similarity θ of network j in the set of networks to be evaluated can be obtained by weighted summation. j ;

[0065]

[0066] Furthermore, the generation of the optimal switching strategy specifically includes:

[0067] After the vehicle terminal triggers the handover, the IKF location prediction model can generate the candidate network set CNS_1 and the cooperative network set CNS_2 in advance. The MJS-INMADM algorithm can then be used to calculate the comprehensive similarity score of all networks in the two network sets, denoted as θ. 1 and θ 2 By combining the hop factor δ of each network in the candidate network set, an optimal handover strategy can be generated for the terminal. The generation process is as follows:

[0068] The network O1 with the highest overall similarity was selected from CNS_1. If the jump factor of network O1 The optimal strategy is to directly switch to network O1; otherwise, obtain network O2 from CNS_2, which has the highest overall similarity and a jump factor of 1. The optimal strategy is to switch directly to network O2.

[0069] The advantages and beneficial effects of this invention are as follows:

[0070] 1. This invention addresses the scenario of frequent handovers caused by highly dynamic vehicular terminals continuously traversing networks such as 5G microcells and WiFi. It proposes a cross-segment cooperative adaptive handover decision method in ultra-dense heterogeneous wireless networks.

[0071] 2. In traditional Kalman filter models, the correction of the next prediction result by new observations is suppressed by old data. When the vehicle speed changes significantly, the prediction result may have a large error compared with the actual position. Therefore, in step 102, an improved Kalman filter model is introduced to predict the position after the vehicle terminal triggers the handover. By strengthening the weight of new observations, the dependence on old values ​​is reduced, the prediction error is lowered, and it is easier to generate candidate network sets in advance.

[0072] 3. Due to the influence of network topology and terminal motion state, accessing a new target network may cause frequent terminal switching. Therefore, in step 103, a jump factor is defined by cell area, trajectory chord length and dwell time to mark the network in the candidate network set that is likely to cause frequent switching.

[0073] 4. In real-world network environments, network parameters fluctuate within a certain range. Traditional multi-attribute decision-making algorithms typically use fixed network attribute values ​​to evaluate candidate networks, introducing errors into the decision results. Furthermore, this invention predicts the terminal's location information at future moments to obtain relevant network attributes at that time. In this case, treating network parameters as fixed values ​​for handover decisions will inevitably lead to even greater errors. Therefore, in step 103, a multi-attribute decision-making algorithm with interval number of modified Jaccard similarities is used to calculate the scores of each network in the candidate network set. Combined with a jump factor, an adaptive cross-regional cooperative handover decision-making scheme is proposed. This scheme can adaptively generate the optimal handover strategy for the handover terminal, effectively reducing the number of handovers and lowering the handover failure rate. Attached Figure Description

[0074] Figure 1 This is a simulation scenario diagram of a heterogeneous wireless network in a core urban area, provided by a preferred embodiment of the present invention.

[0075] Figure 2 Flowchart of an adaptive cross-regional cooperative handover decision scheme to alleviate frequent handovers;

[0076] Figure 3 This is a diagram illustrating the motion state of a vehicle in a Kalman filter.

[0077] Figure 4 A diagram of the improved Kalman filter's position prediction model;

[0078] Figure 5 Comparison of location data predicted by different prediction models;

[0079] Figure 6 A comparison of the proportion of prediction results from different prediction models at different error distances;

[0080] Figure 7 A topology diagram of the alternative network set for vehicle-mounted terminals;

[0081] Figure 8 Here is a flowchart of the interval number multi-attribute decision algorithm based on modified Jaccard similarity;

[0082] Figure 9 Comparison of cumulative switching counts for different methods;

[0083] Figure 10 Comparison of the number of switching times for different table tennis methods;

[0084] Figure 11 Comparison of failure rates for switching between different methods;

[0085] Figure 12 A comparison of the total network throughput using different methods;

[0086] Figure 13 A comparison of the time costs of different algorithmic methods; Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0088] The technical solution of the present invention to solve the above-mentioned technical problems is:

[0089] This method comprehensively considers the network congestion problem caused by the short-term clustering and movement of a large number of vehicle-mounted terminals during peak urban traffic hours in ultra-dense heterogeneous wireless networks that incorporate vehicle self-organizing networks. It can effectively alleviate network congestion, balance the load between networks, and improve user experience.

[0090] The adaptive cross-regional cooperative handover decision method proposed in this invention includes the following steps:

[0091] A cross-regional cooperative adaptive handover decision method in ultra-dense heterogeneous wireless networks is designed for urban core areas, addressing the frequent handover problem caused by highly dynamic vehicle-mounted terminals continuously traversing 5G microcells and WiFi networks. The method comprises the following steps:

[0092] 101. Handover Trigger: Periodically collect the signal strength and network bandwidth of the vehicle terminal in the current network, calculate the handover trigger factor α, and if α ij If (t) = 1, a switch is triggered; otherwise, no switch is triggered.

[0093] 102. Mobility prediction: The vehicle's motion data is collected by the positioning device on the vehicle and recorded in the historical trajectory database. When the vehicle terminal triggers a handover, the historical trajectory information of the vehicle is used to predict the position of the terminal at the next two time points through the IKF model.

[0094] 103. Handover Decision: First, candidate network set CNS_1 for handover and cooperative network set CNS_2 for hopping are generated in advance based on the predicted location. Then, the hopping factor is defined using network topology and terminal motion state, and the score of each network in CNS_1 and CNS_2 is calculated using the interval number multi-attribute decision algorithm with modified Jaccard similarity. Finally, the optimal handover strategy is generated for the terminal based on the hopping factor and network score results.

[0095] Furthermore, based on the handover triggering described in step 101, this invention proposes periodically collecting the received signal strength and network bandwidth of the vehicle terminal in the current network to calculate the handover triggering factor α. The handover triggering decision is made using the value of the handover triggering factor, and the relevant definitions and specific steps are as follows:

[0096] Received signal strength: Received signal strength is a fundamental indicator for terminals to evaluate a network, reflecting the network's channel quality. Due to path loss during signal transmission, the received signal strength of terminal i accessing network j at time t can be expressed as:

[0097] RSS ij (t)=P j -ηlgdis ij (t)+μ (0,σ) (1)

[0098] Among them, P j Let η represent the wireless signal transmit power of network j, and η represent the path loss factor. ij (t) represents the distance from terminal i to network j at time t, μ (0,σ) Let be a Gaussian random variable with a mean of 0 and a variance of σ.

[0099] Network bandwidth: Assume that the bandwidth resources of each network are divided into several physical resource blocks, and the bandwidth of each resource block is . When the number of vehicle terminals connected to network j is below the rated number, each terminal can obtain a fixed number of resource blocks. When the rated number is exceeded, all resource blocks are shared equally among all terminals. Therefore, the network bandwidth obtained by terminal i when accessing network j at time t can be expressed as:

[0100]

[0101] Where u represents the number of vehicle terminals accessing network j at time t, and prb ij (t) represents the number of resource blocks allocated to vehicle terminal i in network j, PRB j Let represent the maximum number of resource blocks that network j can provide. Therefore, the handover trigger factor α of vehicle terminal i in its original access network j at time t can be expressed as:

[0102]

[0103] Furthermore, in the mobility prediction described in step 102, the characteristic is that the vehicle's motion data is collected by the positioning device on the vehicle and recorded in the historical trajectory database. When the on-board terminal triggers a handover, the historical trajectory information of the vehicle is used to predict the terminal's position at the next two time points using the IKF model, specifically including:

[0104] Theoretical prediction: Assume the vehicle's state at time t is S(t) = [la(t), lo(t)] T Where la(t) represents the dimension data at time t, and lo(t) represents the longitude data. If the best estimated state at time t-1 is S′(t-1), then the predicted state at time t, estimated according to the theoretical model, is:

[0105]

[0106] Here, F represents the state transition matrix, which is mainly used to describe how the state at the previous time step transitions to the next state. To represent prediction noise, during the vehicle's motion, there are many uncertainties that cause a discrepancy between the predicted state obtained by the theoretical model and the best estimated state. The error caused by this uncertainty is represented by the covariance matrix:

[0107]

[0108] Furthermore, since the prediction model itself may also have errors, if Q is used... p To represent the noise in the prediction model, substituting equation (4) into equation (5) yields the following process for the propagation of the error covariance matrix between adjacent time steps:

[0109] P′(t)=FP(t-1)F T +Q p (6)

[0110] GPS Measurement: Since each vehicle is equipped with a GPS device, it can collect and record motion information during the vehicle's movement. If the state observed by the GPS positioning device at time t is denoted as , the observation matrix is ​​denoted as , and the observation noise is denoted as , then the transformation process from the predicted state to the observed state of the vehicle at time t can be expressed as:

[0111] Z(t)=HS(t)+Q g (7)

[0112] State Update: The predicted state and observed state at time t are obtained in formulas (4) and (7), respectively. Since there are errors in the prediction process, the predicted value needs to be corrected by the observed value to obtain the best estimated state after correction:

[0113]

[0114] in Let K(t) be the residual between the actual and expected observations, and K(t) be the Kalman gain at time t. The calculation process is as follows:

[0115] K(t)=P′(t)H T (HP′(t) T H T +Q g ) -1 (9)

[0116] The role of Kalman gain is to balance the predicted state covariance P and the observed state covariance Q. g The magnitude of the Kalman gain is used to determine the proportion of the roles played by the prediction model and the observation model in the prediction process. After obtaining the Kalman gain, the noise covariance matrix P(t) of the best estimated state needs to be updated for the next prediction, where E is the identity matrix.

[0117] P(t)=(Ε-K(t)H)P′(t) (10)

[0118] Improved Prediction Model: In the Kalman filter model, the correction of the next prediction result by the new observation is suppressed by the old data. When the vehicle speed changes significantly, this may lead to a large error between the prediction result and the actual position. Therefore, researchers have proposed methods such as particle filtering and attenuated memory filtering to reduce the error in the prediction process. Attenuated memory filtering is widely used in the improvement of Kalman filters due to its simple calculation and excellent performance. Specifically, it multiplies the noise covariance matrix P(t-1) in formula (6) by an attenuation factor η0 with a value greater than 1 to strengthen the weight of the new observation and thus reduce the dependence on the old value. Therefore, this invention introduces the attenuated memory filtering method to improve the Kalman filter, resulting in an improved Kalman filter position prediction model.

[0119] Furthermore, according to the handover decision described in step 103, the key feature is that, firstly, a candidate network set CNS_1 for handover and a cooperative network set CNS_2 for hopping are generated in advance based on the predicted location. Then, a hopping factor is defined using network topology and terminal motion state, and the scores of each network in CNS_1 and CNS_2 are calculated using an interval number multi-attribute decision algorithm that modifies Jaccard similarity. Finally, based on the hopping factor and network score results, an optimal handover strategy is generated for the terminal. The relevant parameter definitions and algorithm process in the adaptive cross-regional cooperative handover decision are as follows:

[0120] Generating a candidate network set: Assuming that vehicle terminal i triggers a handover at time t, its position at time t+1 can be predicted using the IKF model based on its historical trajectory over the previous t time steps. When the user's handover request arrives, the backend identifies all networks within the predicted location's connectivity range and uses the network at that location as the candidate network set after triggering the handover, denoted as CNS_1. Similarly, based on the historical trajectory over the previous t+1 time steps, the position at time t+2 can be predicted, and the network at this location is used as the cooperative network set for skip handover after the handover is triggered, denoted as CNS_2. In this invention, CNS_1 and CNS_2 are collectively referred to as the candidate network set.

[0121] Switching Decision Parameters: When the vehicle terminal triggers a switch, a new network needs to be selected from the candidate network sets CNS_1 and CNS_2 for the terminal to access. Since data transmission rate, network latency, and packet loss rate are key indicators for evaluating network performance during vehicle terminal movement, this invention uses these three parameters to assess network performance. Due to the influence of network topology and terminal movement status, accessing a new target network may cause frequent switching of the terminal. Therefore, this invention defines a skip factor, marking networks in the candidate network set that are prone to frequent switching as networks that need to be skipped. The definitions of each network parameter and the skip factor are given below:

[0122] (1) Data transmission rate: According to Shannon's formula, the data transmission rate of a terminal accessing the network is related to parameters such as bandwidth and signal-to-noise ratio. Therefore, the data transmission rate e obtained by terminal i accessing network j at time t is... ij (t) can be expressed as:

[0123] e ij (t)=B ij (t)×log2(1+SNR ij (t)) (11)

[0124] Among them, B ij (t) represents the bandwidth resources allocated by network j to terminal i at time t, and SNR. ij (t) represents the signal-to-noise ratio, which is approximately the ratio of RSS to the interference noise I in the network;

[0125] (2) Network Latency: Network latency is generally related to load. Increasing the same load on the network will result in more severe latency. Therefore, the relationship between latency and network load can be set as an exponential function. If the latency of network j is d′... j Then the time delay d for terminal i to access network j at time t is... ij (t) can be expressed as:

[0126]

[0127] (3) Packet loss rate: The packet loss rate is the ratio of the number of lost data packets to the total number of sent data packets within a certain period of time. Assume that the number of data packets sent by network j in the first t time steps is ψ. total The number of confirmed data packets received is ψ ack Then, at time t, the packet loss rate γ of terminal i accessing network j is... ij (t) can be expressed as:

[0128]

[0129] (4) Skip Factor: Due to the diversity of network topologies and the high mobility of terminals, the coverage of each network, the chord length of the vehicle's trajectory and the network's path, and the vehicle's dwell time are all significantly different. These factors can easily lead to frequent terminal handovers. To avoid terminals accessing networks that are prone to frequent handovers, this invention considers the above factors and defines a skip factor δ to mark the networks that need to be skipped in the candidate network set. After a handover is triggered at time t, the skip factor δ of terminal i in the j-th candidate network is... ij (t) can be expressed as:

[0130]

[0131] In the formula, cl represents the coverage area of ​​candidate network j. j Let τ represent the chord length of the trajectory of candidate network j. ij This indicates the dwell time of terminal i in network j.

[0132] A Modified Jaccard Similarity Interval-Number Multi-Attribute Decision Algorithm: In real-world network environments, network parameters fluctuate within a certain range. Traditional multi-attribute decision algorithms typically use fixed network attribute values ​​to evaluate candidate networks, introducing errors into the decision results. Furthermore, this invention predicts the terminal's future location information to obtain relevant network attributes at that time. In this case, treating network parameters as fixed values ​​for handover decisions will inevitably lead to even greater errors. Therefore, this invention employs a modified Jaccard similarity interval-number multi-attribute decision algorithm (MJS-INMADM) to calculate the comprehensive score of each network in the candidate network set. This algorithm can evaluate network performance using the range of network attributes even when attribute weights are unknown, improving the accuracy of handover decisions. The specific process is as follows:

[0133] (1) Constructing the Interval Number Decision Matrix: Assuming there are N networks in the network set to be evaluated, and M network attributes involved in the evaluation, the interval numbers of the M attributes of the N networks are collected before making a decision. This chapter determines the interval number of each network attribute by using the maximum and minimum values ​​from multiple data samplings. For example, the maximum and minimum values ​​obtained by sampling the k-th attribute of network j multiple times are as follows: and Then the interval number of the k-th attribute of network j can be expressed as: Therefore, the interval number decision matrix to be decided can be represented as:

[0134]

[0135] (2) Normalized attribute interval number: For the interval number matrix Normalization is performed to obtain the normalized matrix, denoted as . in Formulas (15) and (16) represent the normalization processes for benefit-based and cost-based network parameters, respectively:

[0136]

[0137]

[0138] (3) Determine the interval-type ideal scheme: In order to better measure the differences between networks, assume that the interval-type ideal scheme of each network attribute is Θ=[Θ1,Θ2,...,Θ M ], where Θk It can be represented as:

[0139]

[0140] (4) Calculate the corrected Jaccard similarity: Jaccard similarity is used to describe the similarity and difference between sets. The larger the Jaccard similarity, the higher the similarity between the sets. Since interval numbers can also be regarded as sets of numbers, the Jaccard similarity of the normalized attribute values ​​with respect to the ideal solution Θ can be expressed as:

[0141]

[0142] Since the similarity between two interval numbers cannot be directly compared using the Jaccard similarity score when their midpoints are the same, the right endpoints of the interval numbers can be added to the Jaccard calculation to correct their similarity. The corrected Jaccard similarity score can be expressed as:

[0143]

[0144] Therefore, the corrected Jaccard similarity of each network attribute value in the normalized decision matrix corresponding to the ideal solution can be represented by the matrix ζ = (ζ jk ) NM .

[0145] (5) Determine the optimal weights for each network attribute: Since a network attribute is better the closer it is to the ideal solution, meaning the deviation between the network and the ideal solution is smaller, the weights of each network attribute in the decision-making process can be determined by minimizing the sum of deviations. The corresponding optimization model is:

[0146]

[0147] (6) Calculate the overall similarity: After obtaining the weights of each network attribute, the overall similarity θ of network j in the set of networks to be evaluated can be obtained by weighted summation. j .

[0148]

[0149] Generating the optimal handover strategy: After the vehicle terminal triggers the handover, the IKF location prediction model can generate the candidate network set CNS_1 and the cooperative network set CNS_2 in advance. The MJS-INMADM algorithm can calculate the comprehensive similarity score of all networks in the two network sets, denoted as θ. 1 and θ 2By combining the hop factor δ of each network in the candidate network set, an optimal handover strategy can be generated for the terminal. The process of generating the optimal handover strategy is as follows: select the network O1 with the highest comprehensive similarity from CNS_1, where... If the jump factor of network O1 The optimal strategy is to directly switch to network O1; otherwise, obtain network O2 from CNS_2, which has the highest overall similarity and a jump factor of 1. The optimal strategy is to switch directly to network O2.

[0150] Based on the above analysis, the present invention designs Figure 2 The algorithm flowchart is shown.

[0151] To verify this invention, we conducted simulation experiments on the MATLAB platform, setting up the following simulation scenario: a network composed of three access technologies—5G, WLAN, and ad hoc networks—as a super-dense ad hoc heterogeneous network model. The simulation scenario was built on the MATLAB platform for simulation analysis. It is assumed that the scenario deploys 50 5G macro base stations, 200 5G micro base stations, 150 wireless LANs, and several vehicle ad hoc networks. The simulation scenario of the heterogeneous wireless network in the core urban area is as follows: Figure 1 As shown.

[0152] In the simulation, it is assumed that the arrival of vehicles within the entire network coverage area follows a Poisson distribution with an arrival rate of 1 ≤ λ ≤ 10. To further highlight the superiority of this invention, the proposed method (Location Prediction and Cross Region Cooperation, LPCRC) is compared and analyzed with the multi-attribute decision-based skip handover method (MD-SHO) in the literature [Costa A, Pacheco L, D Rosário, et al. Skipping-based handover algorithm for video distribution over ultra-denseVANET[J]. Computer Networks, 2020, 176: 1-12.] and the fuzzy logic-based vertical handover method (FL-VHO) in the literature [Kim J., Cho J., Jeong J., et al. Fuzzy logic based handoff scheme for heterogeneous vehicular mobile networks[C] / / International Conference on HighPerformance Computing & Simulation. Bologna, Italy: IEEE, 2014: 863-870.].

[0153] Figure 5 The figures show the actual vehicle trajectory, the trajectory obtained from the KF prediction model, and the IKF prediction model. As can be seen from the figures, both prediction models match the actual trajectory well. However, further analysis of the prediction results reveals... Figure 6 The figures show the proportions of prediction results from the two prediction models at different error distances. In 1700 sets of data, the proportions of data points predicted by the KF model with distance errors greater than 5m, 10m, 20m, and 30m from the actual trajectory points were 3.17%, 2.05%, 1.23%, and 1%, respectively, while the proportions for the IKF position prediction model were 2.29%, 1.52%, 1%, and 0.82%. Therefore, the IKF prediction model can predict the vehicle's position more accurately.

[0154] Figure 9The graph shows the cumulative number of handovers for the three algorithms as the vehicle terminal travels further. As can be seen, the cumulative number of handovers for all algorithms gradually increases with the vehicle terminal's travel distance. When the travel distance is less than 3000 meters, the cumulative number of handovers for all three algorithms is only a few. However, when the distance exceeds 3000 meters, the cumulative number of handovers for the FL-VHO algorithm is consistently much higher than that for the MD-SHO and LPCRC-VHO algorithms. This is because the latter two algorithms reduce the number of handovers by introducing a skip handover mechanism. Furthermore, since the LPCRC-VHO algorithm incorporates an Ad Hoc network, when the vehicle terminal triggers a handover, it may connect to an Ad Hoc network that can maintain a persistent connection. Therefore, as the vehicle terminal continues to move, the cumulative number of handovers for the LPCRC-VHO algorithm is also slightly lower than that for the MD-SHO algorithm.

[0155] Figure 10 The figure shows the number of ping-pong handovers for the three algorithms at different speeds. The number of ping-pong handovers directly reflects the impact of frequent handovers on the vehicle terminal during movement. As the vehicle speed increases, the number of ping-pong handovers for each algorithm increases, but the LPCRC-VHO and MD-SHO algorithms have significantly fewer ping-pong handovers than the FL-VHO algorithm. This is because the change in speed triggers more handovers for the vehicle terminal, while the MD-SHO and LPCRC-VHO algorithms, by introducing a skip handover mechanism, can to some extent avoid frequent handovers initiated by the vehicle terminal, thus reducing the number of handovers. Furthermore, the LPCRC-VHO algorithm consistently has the lowest number of ping-pong handovers because this chapter defines a skip factor to avoid ping-pong handovers caused by the vehicle terminal accessing networks with short dwell times.

[0156] Figure 11 The graph shows the handover failure rate of the three algorithms as the number of vehicle terminals increases. As can be seen, the handover failure rate of all three algorithms gradually increases with the increase in the number of vehicle terminals. The FL-VHO algorithm consistently has a higher handover failure rate than the MD-SHO and LPCRC-VHO algorithms. This is mainly because the MD-SHO and LPCRC-VHO algorithms employ a skip mechanism, reducing the number of handovers for the vehicle terminal in the network. They also consider the terminal's dwell time in the network, avoiding connection to networks prone to frequent handovers. Furthermore, the LPCRC-VHO algorithm can predict the terminal's future location after a handover is triggered using an IKF location model, allowing the target network to reserve network resources for the handover terminal in advance, thus improving the terminal's access success rate.

[0157] Figure 12The graph shows the changes in total network throughput for the three algorithms as the number of in-vehicle terminals increases. As can be seen, the total network throughput increases with the number of terminals. When the number of terminals is below 500, the total throughput of all three algorithms increases rapidly. When the number of terminals exceeds 500, the upward trend in total throughput gradually stabilizes due to limited network resources. However, when the number of terminals is the same, the total throughput of the LPCRC-VHO algorithm is consistently higher than the other two algorithms. This is because the LPCRC-VHO algorithm introduces an Ad Hoc network, increasing network capacity. Furthermore, during handover decisions, the LPCRC-VHO algorithm increases the network selection range for in-vehicle terminals by generating candidate network sets and cooperative network sets; and the introduction of the skip handover mechanism also helps to avoid in-vehicle access to overloaded networks, thus improving network throughput.

[0158] Figure 13 The figure shows the time cost of the three algorithms as the number of experiments increases. As can be seen from the figure, the time cost of the FL-VHO algorithm is significantly higher than that of the MD-SHO and LPCRC-VHO algorithms. This is because when considering a large number of decision parameters, the FL-VHO algorithm has a larger rule base, and the fuzzy logic reasoning to determine the optimal network will take longer. Additionally, since the algorithm in this chapter needs to evaluate networks in both the candidate network set and the cooperative network set when generating the optimal switching strategy, its time cost is also slightly higher than that of the MD-SHO algorithm.

[0159] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0160] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0161] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0162] The above embodiments should be understood as illustrative only and not as limiting the scope of protection of the present invention. After reading the description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A cross-regional cooperative adaptive handover decision method in an ultra-dense heterogeneous wireless network, characterized in that, Includes the following steps:

101. Handover Triggering Steps: Periodically collect the signal strength and network bandwidth of the vehicle terminal in the current network, and calculate the handover triggering factor. If the handover trigger factor of terminal i in network j If the switch is triggered, it will be switched; otherwise, it will not be triggered.

102. Mobility prediction steps: The vehicle's motion data is collected by the positioning device on the vehicle and recorded in the historical trajectory database. When the on-board terminal triggers a switch, the terminal's position at the next two time moments is predicted using the vehicle's historical trajectory information through an improved Kalman filter model.

103. Handover decision steps: First, based on the predicted location, a candidate network set CNS_1 for handover and a cooperative network set CNS_2 for hopping are generated in advance. Then, the hopping factor is defined using network topology and terminal motion state, and the score of each network in CNS_1 and CNS_2 is calculated using the interval number multi-attribute decision algorithm with modified Jaccard similarity. Finally, based on the hopping factor and network score results, the optimal handover strategy is generated for the terminal. Step 102 involves collecting vehicle motion data using the vehicle's positioning device and recording it in the historical trajectory database. When the on-board terminal triggers a switch, the historical trajectory information of the vehicle is used to predict the terminal's position at the next two time points using the IKF model. Specifically, this includes:

301. Theoretical prediction: Assume the vehicle's state at time t is... ,in Here are the dimensions of the data at time t. The data is longitude; if the best estimated state at time t-1 is... Then, based on the theoretical model, the predicted state at time t is estimated as follows: (4) in, This represents the state transition matrix, used to describe how the state transitions from the previous time step to the next state. To represent the prediction noise, the error is expressed using the covariance matrix: (5) use To represent the noise in the prediction model, substituting equation (4) into equation (5) yields the following process for the propagation of the error covariance matrix between adjacent time steps: (6) 302. GPS Measurement: The state observed by the GPS positioning device at time t is recorded as... The observation matrix is ​​denoted as The observation noise is expressed as Then, the transformation process from the predicted state to the observed state of the vehicle at time t can be expressed as: (7) 303. State Update: The predicted state and observed state at time t are obtained in formulas (4) and (7), respectively. The predicted value is corrected by the observed value, so that the corrected best estimated state is obtained as follows: (8) in The residual between the actual and expected observations. Let be the Kalman gain at time t, and its calculation process is as follows: (9) The role of Kalman gain is to balance the predicted state covariance. and observation state covariance The magnitude of the Kalman gain is used to determine the proportion of the roles of the prediction model and the observation model in the prediction process; after obtaining the Kalman gain, the noise covariance matrix of the best estimated state also needs to be updated for the next prediction. ,in It is the identity matrix; (10) 304. Improved Prediction Model: An improved Kalman filter position prediction model is obtained by introducing an attenuation memory filtering method. Specifically, the noise covariance matrix in formula (6) is given an improved Kalman filter position prediction model. Multiply by a decay factor greater than 1 ; The specific steps of the interval number multi-attribute decision algorithm for correcting Jaccard similarity in step 103 are as follows: (1) Constructing the interval number decision matrix: Assume there are N networks in the network set to be evaluated, and M network attributes are involved in the evaluation. Before making a decision, the interval numbers of the M attributes of the N networks are collected. The interval number of each network attribute is determined by the maximum and minimum values ​​in multiple data samplings. The maximum and minimum values ​​obtained by sampling the k-th attribute of network j are respectively and Then the number of intervals for the k-th attribute of network j can be expressed as: Therefore, the decision matrix for the interval number to be decided can be represented as: (14) (2) Normalized attribute interval number: For the interval number matrix Normalization is performed to obtain the normalized matrix, denoted as . ,in Formulas (15) and (16) represent the normalization processes for benefit-based and cost-based network parameters, respectively: (15) (16) (3) Determine the interval-type ideal solution: In order to better measure the differences between networks, assume that the interval-type ideal solution for each network attribute is as follows: ,in It can be represented as: (17) (4) Calculate the corrected Jaccard similarity: Jaccard similarity is used to describe the similarity and difference between sets. The larger the Jaccard similarity, the higher the similarity of the sets. Since the interval number can also be regarded as a set of numbers, the normalized attribute value is related to the ideal solution. The Jaccard similarity can be expressed as: (18) Since the similarity between two interval numbers cannot be compared using the Jaccard similarity score when their midpoints are the same, the right endpoints of the interval numbers can be added to the Jaccard calculation to correct them. The corrected Jaccard similarity score can be expressed as: (19) Therefore, the corrected Jaccard similarity of each network attribute value in the normalized decision matrix corresponding to the ideal solution can be represented by a matrix as follows: ; (5) Determine the optimal weights for each network attribute: Determine the weights of each network attribute in the decision-making process based on minimizing the sum of the biases. The corresponding optimization model is as follows: (20) (6) Calculate the overall similarity: After obtaining the weights of each network attribute, the overall similarity of network j in the network set to be evaluated can be obtained by weighted summation. ; (21)。 2. The cross-regional cooperative adaptive handover decision method in an ultra-dense heterogeneous wireless network according to claim 1, characterized in that, Step 101 periodically collects the signal strength and network bandwidth of the vehicle terminal in the current network, and calculates the switching trigger factor. Specifically, it includes:

201. Received signal strength: At time t, the received signal strength of terminal i accessing network j is expressed as: (1) in, This represents the wireless signal transmission power of network j. Indicates the path loss factor. This represents the distance from terminal i to network j at time t. To conform to a mean of 0 and a variance of , Gaussian random variables; 202. Network bandwidth: The network bandwidth obtained by terminal i when accessing network j at time t can be expressed as: (2) in, This represents the number of vehicle terminals accessing network j at time t. This represents the bandwidth of each resource block. This represents the number of resource blocks allocated to vehicle terminal i in network j. This represents the maximum number of resource blocks that network j can provide. Therefore, the handover trigger factor for vehicle terminal i in its original access network j at time t is... It can be represented as: (3)； Among them, when RSS is lower than the set threshold The sum of the hysteresis margin (HM) and the bandwidth is lower than the minimum bandwidth requirement for terminal i to run its services. hour, A switch needs to be triggered; otherwise, It will not trigger.

3. The cross-regional cooperative adaptive handover decision method in an ultra-dense heterogeneous wireless network according to claim 1, characterized in that, Step 103 generates a candidate network set CNS_1 for switching and a cooperative network set CNS_2 for hopping, specifically including: Assuming that vehicle terminal i triggers a handover at time t, its position at time t+1 can be predicted using the historical motion trajectory of the vehicle terminal at the previous t times according to the IKF model. When the user's handover request arrives, the backend discovers all networks within the connection range based on the predicted position and uses the network obtained at this position as the candidate network set after triggering the handover, denoted as CNS_1. Similarly, based on the historical motion trajectory at the previous t+1 times, the position at time t+2 can be predicted, and the network obtained at this position is used as the cooperative network set for skip handover after the handover is triggered, denoted as CNS_2. CNS_1 and CNS_2 are collectively referred to as the candidate network set.

4. The cross-regional cooperative adaptive handover decision method in an ultra-dense heterogeneous wireless network according to claim 3, characterized in that, The handover decision parameters specifically include: when the vehicle terminal triggers a handover, a new network needs to be selected for the terminal to access from the candidate network sets CNS_1 and CNS_2; since data transmission rate, network latency, and packet loss rate are key indicators for evaluating the performance of the access network during the vehicle terminal's movement, these three parameters are used to evaluate the network performance; due to the influence of network topology and terminal movement state, accessing a new target network may cause frequent handovers of the terminal, a skip factor is defined, and networks in the candidate network set that are prone to causing frequent handovers are marked as networks that need to be skipped.

5. The cross-regional cooperative adaptive handover decision method in an ultra-dense heterogeneous wireless network according to claim 1, characterized in that, The specific steps for generating the optimal switching strategy include: After the vehicle-mounted terminal triggers the handover, the IKF location prediction model can generate the candidate network set CNS_1 and the cooperative network set CNS_2 in advance. The MJS-INMADM algorithm can then be used to calculate the comprehensive similarity score of all networks in the two sets, denoted as... and Combining the hop factors of each network in the candidate network set This allows the generation of an optimal handover strategy for the terminal. The generation process is as follows: Select the network with the highest overall similarity from CNS_1. ,in If the network Jump factor The optimal strategy is to switch directly to the network. Otherwise, obtain the network with the highest overall similarity and a jump factor of 1 from CNS_2. ,in The optimal strategy is to switch directly to the network. .

Images