Method for dynamically tracking and clustering multipath components in time-varying channel

Abstract

The invention belongs to the technical field of communication, and relates to a method for dynamically tracking and clustering multipath components in a time-varying channel, which comprises the following steps of: acquiring multipath data at a first moment and preprocessing the multipath data to obtain a characteristic parameter S1; setting a clustering number K1, clustering the multipath component at the first moment according to S1 and K1 to obtain a cluster # imgabs0 # at the first moment, and calculating a clustering index CI1 of the cluster # imgabs1 #; initializing a Q-learning model Q1; multipath data at the moment m are obtained and preprocessed, and a characteristic parameter Sm is obtained; matching the multipath components at the moments m-1 and m according to Sm-1 and Sm to obtain matched and unmatched multipath components at the moment m; obtaining an action am-1 according to the Km-1 and the CIm-1 at the moment m-1 and a model Qm-1, obtaining a clustering number Km according to the am-1, clustering multipath components which are matched and not matched at the moment m according to the Km, obtaining a cluster # imgabs2 #, and calculating a clustering index CIm of the cluster # imgabs3 #; updating the model Qm-1 according to the Km-1, the CIm-1, the am-1, the Km and the CIm to obtain a model Qm; according to the method, the Kalman filtering algorithm, the KM algorithm and reinforcement learning are combined to track and cluster the multipath components more accurately.

Description

Technical Field

[0001] The invention belongs to the technical field of communications and relates to a method for dynamic tracking and clustering of multipath components in a time-varying channel. Background Art

[0002] Drones have a broad market in both military and civil communications due to their low cost and high flexibility. Their applications in border surveillance, information collection in harsh environments, weather and environmental monitoring, traffic control, forest fire detection, and communication relay are developing rapidly. Wireless channel modeling is an abstract description of the wireless propagation environment and propagation characteristics. It is the primary problem to be solved in building wireless communication systems. Therefore, UAV channel modeling is crucial for UAV communication research.

[0003] Existing channel models are divided into two categories, one is a deterministic model and the other is a statistical model. The use of deterministic models requires obtaining the electromagnetic parameters of various scatterers in the environment through measurement methods, and calculating the channel propagation path of rays through path search algorithms. As the environment becomes more complex and the number of reflections increases, its complexity increases dramatically. The use of statistical channel models relies on channel measurement data. Millimeter wave channel measurement is time-consuming and labor-intensive, and channel data is difficult to obtain. Machine learning algorithms have powerful nonlinear fitting capabilities and can automatically learn data features from massive data. At present, many institutions and scholars at home and abroad have applied machine learning to the field of channel modeling and simulation.

[0004] The channel multipath generated when electromagnetic waves propagate in the private space exists in the form of clusters in the real physical channel. The cluster form can better represent the physical meaning of these multipaths, and the fitting effect of cluster-based channel modeling is also better, so most organizations write cluster-based channel modeling into the standard. For example, the KMeans algorithm, DBSCAN algorithm, KPM algorithm, GMM algorithm, etc. have been applied to channel modeling. Although these clustering algorithms have shown good results in channel modeling, most of them only focus on static channel models and ignore the time-varying nature of the channel. However, in actual application scenarios, the characteristics of the channel are not fixed, but will change over time. In particular, the drone channel has more obvious time-varying characteristics due to the high-speed mobility of the drone.

[0005] Ignoring these time-varying characteristics in the prior art will result in reduced applicability of channel estimation, which in turn affects the performance of the communication system. Summary of the invention

[0006] In order to solve the above-mentioned prior art problems, the present invention adopts a method for dynamic tracking and clustering of multipath components in a time-varying channel, comprising:

[0007] S1. Obtain the multipath data in the time-varying channel at the first moment and perform preprocessing to obtain the characteristic parameter S of the actual multipath component at the first moment. 1 ; Set the initial number of clusters K 1 , according to the characteristic parameter S of the multipath component 1 and the number of clusters K 1 The KPD algorithm is used to cluster the actual multipath components at the first moment to obtain multiple clusters at the first moment. Compute Cluster The clustering index CI 1 ; Initialize the Q-learning model; where Q-learning is Q-learning and KPD is the kernel power density algorithm;

[0008] S2. Obtain the characteristic parameter S of the actual multipath component at the previous moment m-1 m-1 , obtain the multipath data in the time-varying channel at the current time m and preprocess it to obtain the characteristic parameter S of the actual multipath component at the current time m m ; According to the characteristic parameter S m -1 and S m Matching the actual multipath component at the previous moment m-1 with the actual multipath component at the current moment m to obtain the matched actual multipath component and the unmatched actual multipath component at the current moment m;

[0009] S3. Get the number of clusters K at the previous moment m-1 m-1 and clustering index CI m-1 , according to the number of clusters K at the previous moment m-1 m-1 , clustering index CI m-1 And the Q-learning model gets action a m-1 , according to action a m-1 Get the number of clusters K at the current moment m , according to the number of clusters K at the current moment m The KPD algorithm is used to cluster the actual multipath components that are matched and the actual multipath components that are not matched at the current moment m to obtain the cluster at the current moment Compute Cluster The clustering index CI m ;

[0010] S4, according to the number of clusters K at the last moment m-1 m-1 , clustering index CI m-1 and action a m-1 and the number of clusters K at the current time m m and clustering index CI mUpdate the Q-learning model of the previous moment m-1 to obtain the Q-learning model of the current moment m.

[0011] The characteristic parameters of the actual multipath component m at the current time in, is the characteristic parameter of the nth multipath component, is power, is the arrival azimuth, is the departure azimuth, To reach the elevation angle, is the departure elevation angle, w is the angular frequency, represents the delay, j is the imaginary unit, and N is the number of m multipath components at the current moment.

[0012] Matching the actual multipath component at the previous time m-1 with the actual multipath component at the current time m includes:

[0013] S21, according to the characteristic parameter S of the multipath component m-1 Use the Kalman filter algorithm to predict the characteristic parameters of the multipath component at the current time m in, is the characteristic parameter of the n′th multipath component predicted at the current time m, and N′ is the number of actual multipath components at the previous time m-1;

[0014] S22, according to the characteristic parameter S m and characteristic parameters Use the Kuhn-Munkres algorithm to optimally match the actual multipath component of the current moment m with the predicted multipath component, and obtain the matched actual multipath component and the unmatched actual multipath component of the current moment m; wherein the Kuhn-Munkres algorithm is a Hungarian algorithm;

[0015] S23, the characteristic parameters of the multipath component predicted at the current moment m correspond one to one with the characteristic parameters of the actual multipath component at the previous moment, and the characteristic parameters of the matching multipath component pair are obtained. According to the multipath component matched with the actual multipath component matched at the current time m, the characteristic parameters of the matched multipath component pair are obtained. According to the characteristic parameters of the matched multipath component pairs and Get with Characteristic parameters of matched multipath components in, It is the characteristic parameter of the actual multipath component of the previous moment m-1n′.

[0016] Use the Kalman filter algorithm to predict the characteristic parameters of the multipath component at the current time m include:

[0017] S211, predicting the prior characteristic parameters at the current moment through the state transfer matrix according to the characteristic parameters at the previous moment;

[0018] S212, predicting the a priori error covariance matrix at the current moment according to the posterior error covariance matrix at the previous moment;

[0019] S213, using the a priori error covariance matrix at the current moment and the characteristic parameters of the actual multipath components at the current moment to correct the a priori characteristic parameters at the current moment, to obtain the a posteriori characteristic parameters at the current moment; updating the a posteriori error covariance matrix at the current moment according to the a priori error covariance matrix at the current moment;

[0020] S214, combining the posterior feature parameters at the current moment to obtain the feature parameters at the current moment

[0021] According to the characteristic parameter S m and characteristic parameters Using the Kuhn-Munkres algorithm to optimally match the actual multipath component at the current time m with the predicted multipath component includes:

[0022] Constructing the objective function in, is the minimum center distance, U is the set of multipath components; the Kuhn-Munkres algorithm is used to solve the objective function to obtain the actual multipath components that are matched and the actual multipath components that are not matched at the current moment m.

[0023] The clustering of the actual multipath components matched and the actual multipath components not matched at the current moment m includes:

[0024] For the actual multipath component matched at the current time m According to Characteristic parameters of matched multipath components Calculate trajectory change characteristics According to the trajectory change characteristics and characteristic parameters Calculate the relative density of multipath components For the actual multipath component that is not matched at the current time m According to the characteristic parameters Calculate the relative density of multipath components

[0025] The relative density The largest K mThe multipath components n are taken as the cluster center, and the distances between the remaining multipath components and the cluster center at the current time are calculated. The remaining multipath components are assigned to the cluster corresponding to the cluster center with the shortest distance to them, and the cluster is obtained.

[0026] According to Characteristic parameters of matched multipath components Calculate trajectory change characteristics include

[0027]

[0028] in, is the specific time of multipath component n at time m, is the specific time of multipath component n′ at time m-1

[0029] The Q-learning model includes:

[0030] State space s t =(K t ,CI t ), where s t is the state at time step t, K t represents the number of clusters at time step t, CI t represents the clustering index at time step t;

[0031] Action Space a t ={-1,0,1}; where a t is the action at time step t, -1 means reducing one cluster, 0 means keeping the current number of clusters unchanged, and 1 means adding one cluster;

[0032] Reward function R(s t ,a t )=-|CI t+1 -CI′|; where CI t+1 is the clustering index at time step t+1, CI′ is the clustering index threshold;

[0033] Updating the Q-learning model of the previous moment m-1 includes:

[0034] According to the number of clusters K at the last moment m-1 m-1 , clustering index CI m-1 and action a m-1 and the number of clusters K at the current time m m and clustering index CI m Construct training samples (s t ,a t ,R(s t ,a t ),st+1 ), where s t =(K m-1 ,CI m-1 ), a t =a m-1 ,R(s t ,a t )=|CI m -CI′|,s t+1 =(K m ,CI m ), according to the training samples (s t ,a t ,R(s t ,a t ),s t+1 ) Update the Q-learning model of the previous moment m-1 to obtain the Q-learning model of the current moment m.

[0035] Clustering index CI m The calculation process includes: calculating each cluster at the current moment Cluster of Heart and cluster center mean According to cluster center and cluster center mean Calculate the inter-cluster matrix tr(B) m and the intra-cluster matrix tr(W) m , calculate each cluster at the current moment Variance According to the cluster matrix tr(B) m , intra-cluster matrix tr(W) m And each cluster The variance of the clustering index CI is calculated m .

[0036] According to the cluster matrix tr(B) m , intra-cluster matrix tr(W) m And each cluster The variance of the clustering index CI is calculated m include: Wherein, N is the number of multipath components at time m.

[0037] Beneficial effects:

[0038] 1. The present invention is based on the characteristic parameter S of the multipath component m-1 Use the Kalman filter algorithm to predict the characteristic parameters of the multipath component at the current time m And according to the characteristic parameter S m and characteristic parameters The Kuhn-Munkres algorithm is used to optimally match the actual multipath component at the current moment m with the predicted multipath component, which reduces the burden of repeated calculations, can cope with the generation and disappearance of multipath components, and maintain the continuity of MPC tracking of multipath components, thereby greatly improving the MPC tracking accuracy in time-varying channels; 2. The trajectory change characteristics can identify drastic and mild rises or falls. The present invention achieves a detailed description of the dynamic changes of MPC through trajectory change characteristics, thereby greatly improving the MPC tracking accuracy in time-varying channels; 3. The present invention uses the Q-learning model to dynamically adjust the number of clusters K at each moment, and can still maintain the efficiency and accuracy of clustering in an environment with strong channel dynamics. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a method for dynamic tracking and clustering of multipath components in a time-varying channel proposed by an embodiment of the present invention;

[0040] Figure 2 The present invention is a flowchart of using the Kalman filter algorithm proposed in an embodiment of the present invention to predict time-varying channel data. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] like Figure 1 As shown, the present invention adopts a method for dynamic tracking and clustering of multipath components in a time-varying channel, comprising:

[0043] S1. Obtain the multipath data in the time-varying channel at the first moment and perform preprocessing to obtain the characteristic parameter S of the actual multipath component at the first moment. 1 ; Set the initial number of clusters K 1 , according to the characteristic parameter S of the multipath component 1 and the number of clusters K 1 The KPD algorithm (Kernel Power Density Based Algorithm) is used to cluster the actual multipath components at the first moment to obtain multiple clusters C at the first moment. k 1 , calculate cluster C k 1 The clustering index CI 1; Initialize the Q-learning model; where Q-learning is Q-learning; optional, K 1 The value range of is 2 to 30;

[0044] S2. Obtain the characteristic parameter S of the actual multipath component at the previous moment m-1 m-1 , obtain the multipath data in the time-varying channel at the current time m and preprocess it to obtain the characteristic parameters of the actual multipath component at the current time m Matching the actual multipath component at the previous moment m-1 with the actual multipath component at the current moment m to obtain the matched actual multipath component and the unmatched actual multipath component at the current moment m;

[0045] Acquiring multipath data in a time-varying channel and preprocessing it includes:

[0046] Extract the characteristic parameters of the actual multipath component at time m The description is as follows:

[0047]

[0048] in, is power, is the arrival azimuth, is the departure azimuth, To reach the elevation angle, is the departure elevation angle, m represents the measurement time, n represents the index of the multipath component, w is the angular frequency of the signal, represents the multipath delay of the time-varying channel, and j is an imaginary unit.

[0049] The matrix expression of the characteristic parameters is S m , the specific description is as follows:

[0050]

[0051] Wherein, N is the number of multipath components at time m.

[0052] Matching the actual multipath component at the previous time m-1 with the actual multipath component at the current time m includes:

[0053] S21, according to the characteristic parameter S of the multipath component m-1 Use the Kalman filter algorithm to predict the characteristic parameters of the multipath component at the current time m Where N′ is the number of actual multipath components at the previous moment m-1;

[0054] like Figure 2 As shown, the specific steps include:

[0055] Initialization state: the trajectory at the start of initialization and the posterior error covariance matrix at the start time

[0056] Update status:

[0057] S211, based on the characteristic parameters of the previous moment Predict the prior characteristic parameters at the current moment through the state transfer matrix Where F is the state transfer matrix, is the characteristic parameter of the actual multipath component n at the m-1th moment, is the prior characteristic parameter of the multipath component n predicted at the mth moment, is the process noise;

[0058] S212, according to the posterior error covariance matrix of the previous moment Predict the prior error covariance matrix for the current moment Where Q is the process noise covariance matrix;

[0059] S213, using the prior error covariance matrix at the current moment and the characteristic parameters of the actual multipath component m at the current time Prior feature parameters for the current moment Make corrections to obtain the posterior feature parameters at the current moment According to the prior error covariance matrix at the current moment Update the posterior error covariance matrix at the current moment

[0060]

[0061]

[0062]

[0063] Among them, H is the observation matrix used to map the state to the measurement space, K is the Kalman gain, and R is the measurement noise covariance matrix.

[0064] S214: The posterior feature parameters at the current moment Combine to get the posterior feature parameters at the current moment

[0065] S22, according to the characteristic parameter S m and characteristic parameters Use the Kuhn-Munkres algorithm to optimally match the actual multipath component of the current moment m with the predicted multipath component, and obtain the matched actual multipath component and the unmatched actual multipath component of the current moment m; wherein the Kuhn-Munkres algorithm is a Hungarian algorithm;

[0066] According to the characteristic parameter S m and characteristic parameters Using the Kuhn-Munkres algorithm to optimally match the actual multipath component at the current time m with the predicted multipath component includes:

[0067] Constructing the objective function in, is the minimum center distance, U is the set of multipath components; the Kuhn-Munkres algorithm is used to solve the objective function to obtain the actual multipath components that are matched and the actual multipath components that are not matched at the current moment m.

[0068] The predicted multipath component is taken as the left node, and the actual multipath component at the current moment m is taken as the right node. The key point matching is performed using the Kuhn-Munkres algorithm. If the multipath component at the current moment is more than that at the next moment, then the one that was not matched at the previous moment means that part of the multipath component at the previous moment has disappeared; if the multipath component at the next moment is more than that at the previous moment, then it means that a part of the multipath component is newly generated. This part of the newly generated multipath component has no matching multipath component and needs to be considered separately.

[0069] The calculation formula of the minimum center distance MCD is as follows:

[0070]

[0071] S23, the characteristic parameters of the multipath component predicted at the current moment m correspond one to one with the characteristic parameters of the actual multipath component at the previous moment, and the characteristic parameters of the matching multipath component pair are obtained. According to the multipath component matched with the actual multipath component matched at the current time m, the characteristic parameters of the matched multipath component pair are obtained. According to the characteristic parameters of the matched multipath component pairs and Get with Characteristic parameters of matched multipath components in, It is the characteristic parameter of the n′th multipath component predicted at the current time m.

[0072] S3. Get the number of clusters K at the previous moment m-1 m-1 and clustering index CI m-1 , according to the number of clusters K m-1 , clustering index CIm-1 And the Q-learning model of the previous moment m-1 obtains action a m-1 , according to action a m-1 Get the number of clusters K at the current moment m , according to the number of clusters K at the current moment m The KPD algorithm is used to cluster the actual multipath components that are matched and the actual multipath components that are not matched at the current moment m to obtain the cluster at the current moment Compute Cluster The clustering index CI m ;

[0073] When the KPD algorithm is used to cluster the actual multipath components at the first moment, the clustering density of the multipath components is and relative density All calculations were performed using the original formula.

[0074] Specifically, density and relative density The original calculation formula is as follows:

[0075]

[0076] in, is the set of Z multipath components that are closest to multipath component n at time m. The value range of Z is 1 to 20. The specific value depends on the actual situation of the multipath component. n' is the index of the multipath component. is the standard deviation of multipath component delay at time m, is the standard deviation of the multipath component transmission azimuth at time m, is the standard deviation of the multipath component receiving azimuth angle at time m, is the standard deviation of the multipath component transmission elevation angle at time m, is the standard deviation of the multipath component receiving elevation angle at time m.

[0077] Using the KPD algorithm to cluster the actual multipath components matched and the actual multipath components not matched at the current moment m includes:

[0078] For the actual multipath component matched at the current time m According to Characteristic parameters of matched multipath components Calculate trajectory change characteristics According to the trajectory change characteristics and characteristic parameters Calculate the relative density of multipath components For the actual multipath component that is not matched at the current time m According to the characteristic parameters Calculate the relative density of multipath components

[0079] The relative density The largest K m The multipath components n are taken as the cluster center, and the distances between the remaining multipath components and the cluster center at the current time are calculated. The remaining multipath components are assigned to the cluster corresponding to the cluster center with the shortest distance to them, and the cluster is obtained. Among them, distance is the minimum center distance.

[0080] Specifically, the trajectory change characteristics The calculation formula is:

[0081]

[0082] in, is the specific time of multipath component n at time m, is the specific time of multipath component n′ at time m-1

[0083] Preferably, the initial number of clusters K 1 The value range is 2 to 30.

[0084] density and relative density The improved calculation formula is as follows:

[0085]

[0086] S4, according to the number of clusters K at the last moment m-1 m-1 , clustering index CI m-1 and action a m-1 and the number of clusters K at the current time m m and clustering index CI m Update the Q-learning model of the previous moment m-1 to obtain the Q-learning model of the current moment m.

[0087] The Q-learning model includes:

[0088] State space s t =(K t ,CI t ), where s t is the state at time step t, K t represents the number of clusters at time step t, CI t represents the clustering index at time step t;

[0089] Action Space a t ={-1,0,1}; where a tis the action at time step t, -1 means reducing one cluster, 0 means keeping the current number of clusters unchanged, and 1 means adding one cluster;

[0090] Reward function R(s t ,a t )=-|CI t+1 -CI′|; where CI t+1 is the clustering index at time step t+1, CI′ is the clustering index threshold, which generally ranges from 2 to 15, indicating the clustering quality under ideal conditions; when the CI index is close to the set threshold, the absolute value of the reward is a negative value and is small, indicating that the system performs well; conversely, if the CI is greater than the set threshold, it means that the clustering is not ideal and the reward is low.

[0091] Updating the Q-learning model of the previous moment m-1 includes:

[0092] According to the number of clusters K at the last moment m-1 m-1 , clustering index CI m-1 and action a m-1 and the number of clusters K at the current time m m and clustering index CI m Construct training samples (s t ,a t ,R(s t ,a t ),s t+1 ), where s t =(K m-1 ,CI m-1 ), a t =a m-1 ,R(s t ,a t )=|CI m -CI|,s t+1 =(K m ,CI m );

[0093] According to the training samples (s t ,a t ,R(s t ,a t ),s t+1 ) Update the Q-learning model of the previous moment m-1 to obtain the Q-learning model of the current moment m.

[0094] The update formula of Q-learning is as follows:

[0095]

[0096] Among them, Q(st ,a t ) is the current state s t Take action a t The value of , α is the learning rate, γ is the discount factor, is the maximum Q value in the next state.

[0097] When defining the measurement indicators, we mainly follow two major principles: the closer the intervals between multipaths within a cluster, the better, and the larger the intervals between clusters, the better. Therefore, the CI indicator is used; the clustering indicator CI m The calculation process includes:

[0098]

[0099]

[0100]

[0101]

[0102]

[0103] Among them, tr(B) m represents the inter-cluster matrix at time m, tr(W) m represents the intra-cluster matrix at time m, K represents the number of clusters, L m is the number of multipath components at time m, that is, the number N mentioned above, Cluster The number of multipath components within Cluster The heart of cluster, is the cluster center mean of the cluster at time m Cluster The variance of Representation Cluster The jth multipath component of Representative cluster The power of the i-th multipath component, Cluster The characteristic parameters of the i-th multipath component are removed from the characteristic parameters of the power, represents the power of the lth multipath component at time m, It represents the characteristic parameter of the lth multipath component at time m, excluding the characteristic parameter of power, that is,

Claims

1. A method for dynamic tracking and clustering of multipath components in a time-varying channel, characterized in that: include: S1. Obtain the multipath data in the time-varying channel at the first moment and perform preprocessing to obtain the characteristic parameter S of the actual multipath component at the first moment. 1 ; Set the initial number of clusters K 1 , according to the characteristic parameter S of the multipath component 1 and the number of clusters K 1 The KPD algorithm is used to cluster the actual multipath components at the first moment to obtain multiple clusters at the first moment. Compute Cluster The clustering index CI 1 ; Initialize the Q-learning model; where Q-learning is Q-learning and KPD is the kernel power density algorithm; S2. Obtain the characteristic parameter S of the actual multipath component at the previous moment m-1 m-1 , obtain the multipath data in the time-varying channel at the current time m and preprocess it to obtain the characteristic parameter S of the actual multipath component at the current time m m ; According to the characteristic parameter S m-1 and S m Matching the actual multipath component at the previous moment m-1 with the actual multipath component at the current moment m to obtain the matched actual multipath component and the unmatched actual multipath component at the current moment m; S3. Get the number of clusters K at the previous moment m-1 m-1 and clustering index CI m-1 , according to the number of clusters K at the previous moment m-1 m -1 , clustering index CI m-1 And the Q-learning model gets action a m-1 , according to action a m-1 Get the number of clusters K at the current moment m , according to the number of clusters K at the current moment m The KPD algorithm is used to cluster the actual multipath components that are matched and the actual multipath components that are not matched at the current moment m to obtain the cluster at the current moment Compute Cluster The clustering index CI m ; S4, according to the number of clusters K at the last moment m-1 m-1 , clustering index CI m-1 And action a m-1 and the number of clusters K at the current time m m and clustering index CI m Update the Q-learning model of the previous moment m-1 to obtain the Q-learning model of the current moment m.

2. A method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 1, characterized in that: The characteristic parameters of the actual multipath component m at the current time in, is the characteristic parameter of the nth multipath component, is power, is the arrival azimuth, is the departure azimuth, To reach the elevation angle, is the departure elevation angle, w is the angular frequency, represents the delay, j is the imaginary unit, and N is the number of m multipath components at the current moment.

3. A method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 2, characterized in that: Matching the actual multipath component at the previous time m-1 with the actual multipath component at the current time m includes: S21, according to the characteristic parameter S of the multipath component m-1 Use the Kalman filter algorithm to predict the characteristic parameters of the multipath component at the current time m in, is the characteristic parameter of the n′th multipath component predicted at the current time m, and N′ is the number of actual multipath components at the previous time m-1; S22, according to the characteristic parameter S m and characteristic parameters Use the Kuhn-Munkres algorithm to optimally match the actual multipath component of the current moment m with the predicted multipath component, and obtain the matched actual multipath component and the unmatched actual multipath component of the current moment m; wherein the Kuhn-Munkres algorithm is a Hungarian algorithm; S23, the characteristic parameters of the multipath component predicted at the current moment m correspond one to one with the characteristic parameters of the actual multipath component at the previous moment, and the characteristic parameters of the matched multipath component pair are obtained. According to the multipath component matched with the actual multipath component matched at the current time m, the characteristic parameters of the matched multipath component pair are obtained. According to the characteristic parameters of the matched multipath component pairs and Get with Characteristic parameters of matched multipath components in, is the nth time at the previous moment m-1 ′ The characteristic parameters of the actual multipath components.

4. A method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 3, characterized in that: Use the Kalman filter algorithm to predict the characteristic parameters of the multipath component at the current time m include: S211, predicting the prior characteristic parameters at the current moment through the state transfer matrix according to the characteristic parameters at the previous moment; S212, predicting the a priori error covariance matrix at the current moment according to the posterior error covariance matrix at the previous moment; S213, using the a priori error covariance matrix at the current moment and the characteristic parameters of the actual multipath components at the current moment to correct the a priori characteristic parameters at the current moment, to obtain the a posteriori characteristic parameters at the current moment; updating the a posteriori error covariance matrix at the current moment according to the a priori error covariance matrix at the current moment; S214, combining the posterior feature parameters at the current moment to obtain the feature parameters at the current moment 5. The method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 3, characterized in that: According to the characteristic parameter S m and characteristic parameters Using the Kuhn-Munkres algorithm to optimally match the actual multipath component at the current time m with the predicted multipath component includes: Constructing the objective function in, is the minimum center distance, U is the set of multipath components; the Kuhn-Munkres algorithm is used to solve the objective function to obtain the actual multipath components that are matched and the actual multipath components that are not matched at the current moment m.

6. A method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 1, characterized in that: The clustering of the actual multipath components matched and the actual multipath components not matched at the current moment m includes: For the actual multipath component matched at the current time m According to Characteristic parameters of matched multipath components Calculate trajectory change characteristics According to the trajectory change characteristics and characteristic parameters Calculate the relative density of multipath components For the actual multipath component that is not matched at the current time m According to the characteristic parameters Calculate the relative density of multipath components The relative density The largest K m The multipath components n are taken as the cluster center, and the distances between the remaining multipath components and the cluster center at the current time are calculated. The remaining multipath components are assigned to the cluster corresponding to the cluster center with the shortest distance to them, and the cluster is obtained.

7. A method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 6, characterized in that: According to Characteristic parameters of matched multipath components Calculate trajectory change characteristics include in, is the specific time of multipath component n at time m, is the specific time of the multipath component n′ at time m-1.

8. The method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 1, characterized in that: The Q-learning model includes: State space s t =(K t ,CI t ), where s t is the state at time step t, K t represents the number of clusters at time step t, CI t represents the clustering index at time step t; Action Space a t ={-1,0,1}; where a t is the action at time step t, -1 means reducing one cluster, 0 means keeping the current number of clusters unchanged, and 1 means adding one cluster; Reward function R(s t ,a t )=-|CI t+1 -CI′|; where CI t+1 is the clustering index at time step t+1, CI′ is the clustering index threshold; Updating the Q-learning model of the previous moment m-1 includes: According to the number of clusters K at the last moment m-1 m-1 , clustering index CI m-1 and action a m-1 and the number of clusters K at the current time m m and clustering index CI m Construct training samples (s t ,a t ,R(s t ,a t ),s t+1 ), where s t =(K m-1 ,CI m-1 ), a t =a m-1 ,R(s t ,a t )=|CI m -CI′|,s t+1 =(K m ,CI m ), according to the training samples (s t ,a t ,R(s t ,a t ),s t+1 ) Update the Q-learning model of the previous moment m-1 to obtain the Q-learning model of the current moment m.

9. According to the method for dynamic tracking and clustering of multipath components in a time-varying channel in claim 1, the clustering index CI m The calculation process includes: Calculate each cluster at the current moment Cluster of Heart and cluster center mean According to cluster center and cluster center mean Calculate the inter-cluster matrix tr(B) m and the intra-cluster matrix tr(W) m , calculate each cluster at the current moment Variance According to the cluster matrix tr(B) m , intra-cluster matrix tr(W) m And each cluster The variance of the clustering index CI is calculated m .

10. A method for dynamic tracking and clustering of multipath components in a time-varying channel according to claim 9, characterized in that: According to the cluster matrix tr(B) m , intra-cluster matrix tr(W) m And each cluster The variance of the clustering index CI is calculated m include: Wherein, N is the number of multipath components at time m.

Images