Non-cooperative multi-user positioning method in near-field scattering environment

Abstract

The invention discloses a non-cooperative multi-user positioning method in a near-field scattering environment, which comprises two stages of signal source acquisition and user identification, and realizes multi-user positioning in a complex scattering environment by means of mining of own characteristics of a system and a machine learning algorithm when the number of users is unknown and pilot signals are sent, and in the first stage, the number of users is not known, and the number of users is not known. The spatial orthogonality between signals from different positions in a near field is utilized, a synchronous orthogonal matching pursuit algorithm based on a deep convolutional neural network CNN is used, iteration stopping indication is provided, and the positions of all users and scattering points are obtained; and in the second stage, a k-means clustering algorithm assisted by signal energy is used, and the distance between data points and signal energy information are simultaneously incorporated into an updating criterion of a clustering center so as to accurately identify the user. According to the invention, the problem of non-cooperative multi-user positioning in a near-field scattering environment is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of near-field user positioning, and specifically to a non-cooperative multi-user positioning method in a near-field scattering environment. Background Technique

[0002] User positioning is an important research direction in the sixth-generation mobile communication network. Although the Global Positioning System (GPS) is widely used for navigation and positioning, in densely populated urban areas or indoor environments, the GPS signal will attenuate severely. In addition, the positioning accuracy of GPS in a scattering environment also needs to be improved. In recent years, the positioning technology based on Wireless Local Area Network (WLAN) has received extensive attention.

[0003] In the far field, wireless positioning usually relies on three types of measurements: Received Signal Strength (RSS), Time of Arrival (ToA), and Angle of Arrival (AoA). The RSS-based method usually has worse performance than the other two methods. ToA measurement implicitly requires transceiver synchronization, while AoA measurement becomes invalid in a scattering environment. With the increase in the number of antennas and the improvement of the carrier frequency, in the sixth-generation mobile communication (6G) system, most users may be located in the near-field region. In the near-field region, the electromagnetic wave propagation wavefront is modeled as a spherical wave. The properties of the spherical wavefront bring new opportunities for user positioning, that is, the position of the user can be inferred according to the phase difference of the spherical wave received on different antennas without transceiver time synchronization. Therefore, in the near field, the Curvature of Arrival (CoA) of the spherical wave can be utilized to infer the position of the user.

[0004] Near-field positioning technology has recently become a research hotspot. Many studies have given the theoretical Cramer-Rao bound of near-field positioning accuracy, and there are also many studies on specific positioning algorithms, including the maximum likelihood method and subspace method using second-order statistics (SOS). Different from the plane wavefront that only contains angle information, the spherical wavefront contains both user angle and distance parameters. Therefore, for subspace positioning methods, such as Multiple Signal Classification (MUSIC) and Estimation of Signal Parameters via Rotational Invariance Techniques (ESPRIT), their extended two-dimensional version algorithms have been proposed and used to obtain the near-field user position. However, the positioning accuracy of the statistic-based method depends on the estimation accuracy of the autocorrelation matrix of the received signal, and the eigenvalue decomposition of the autocorrelation matrix also results in a high computational complexity. There are also studies that have proposed subspace positioning methods that do not require eigenvalue decomposition. However, most of these subspace-based algorithms require a symmetric uniform linear array to utilize the Toeplitz property of the autocorrelation matrix, and the positioning performance will deteriorate severely in the presence of unknown noise or multipath fading environment.

[0005] On the other hand, some studies have proposed near-field localization and trajectory prediction methods based on deep learning. For example, a convolutional neural network (CNN) is used to process the classification and localization of near-field and far-field hybrid sources. This method treats the position parameter estimation as a regression problem, and its computational complexity is much lower than that of the method that treats the position parameter estimation as a classification problem. There are also studies on the near-field user localization problem in the presence of broadband beam splitting effect and spatial non-stationarity characteristics. A CNN-based localization scheme is designed, which uses the input and output of a controllable beam splitting scheme to generate estimates of position parameters such as angles and distances. When the direct path of the user signal reaching the receiving end is blocked, the intelligent reflecting surface (RIS) technology provides the possibility for user localization. RIS can provide a controllable refraction path between the user and the base station. Therefore, there are also many studies on RIS-based near-field localization technology, including the theoretical analysis of RIS localization in the near-far hybrid field scenario, and the joint optimization of RIS parameters and the design of localization algorithms. However, there is a synchronization problem among the user, RIS, and base station in RIS-based localization. Summary of the Invention

[0006] In order to overcome the non-cooperative near-field multi-user localization problem in the case of unknown number of users and transmitted pilot signals, the present invention provides a non-cooperative multi-user localization method in a near-field scattering environment, which makes full use of the exploration of the system's own characteristics and machine learning algorithms to achieve multi-user localization in a complex scattering environment.

[0007] To achieve the above object, the present invention provides the following technical solution: A non-cooperative multi-user localization method in a near-field scattering environment, including a new algorithm framework, which includes two stages: signal source acquisition and user identification. When the number of users and transmitted pilot signals are unknown, by exploring the system's own characteristics and machine learning algorithms, multi-user localization in a complex scattering environment is achieved;

[0008] The method includes the following steps:

[0009] 1) In the first stage, use the simultaneous orthogonal matching pursuit (SOMP) algorithm based on the deep convolutional neural network CNN to obtain the positions of all users and scattering points;

[0010] 2) In the second stage, use the signal energy-assisted k-means clustering algorithm to incorporate both the distance between data points and signal energy information into the update criterion of the clustering center, and then accurately identify users.

[0011] Preferably, in step 1), the simultaneous orthogonal matching pursuit (SOMP) algorithm makes full use of the sparsity of the channel in the near-field multi-user localization problem and the orthogonality between signals from different sources to achieve position acquisition under unknown user pilot signals. This algorithm is based on the following sparse representation model:

[0012] Y = DU + W

[0013] where Y = [y 1 ,..., y T represents the received signal matrix at T time instants, and the received signal vector y t at each time instant contains the received signals of M antennas; similarly, the matrix U = [u 1 ,..., u T is composed of sparse vectors at T time instants, and each sparse vector u t is the projection of the received signal t t in the dictionary space, and the matrix W = [w 1 ,..., w T represents noise; due to the sparsity of the channel, the rows of the matrix U are sparse; therefore, the localization problem can be solved by the SOMP algorithm; in addition, in the above formula, D represents the near-field dictionary matrix, which consists of the steering vectors b(θ, r) to form a complete angle-distance space, that is, this dictionary covers the steering vectors at all possible angles and distances; specifically, the angle space is divided into θ i , i = 1,..., M, where cosθ i ∈ [-1, 1]; the distance space is divided into r j = [r min , r max , j = 1,..., N; then the dictionary D is expressed as:

[0014] D = [b(θ 1 , r 1 ),..., b(θ 1 , r N ),..., b(θ M , r 1 ),..., b(θ M , r N )]

[0015] where

[0016]

[0017] The above formulas are the expressions of the steering vectors and the distances from the data source (θ, r) to the m-th antenna respectively, where fc represents the carrier frequency, c is the speed of light, and d represents the antenna spacing, which is set to be half a wavelength;

[0018] After the received signal is projected onto the above dictionary D, the obtained sparse vector u t is regarded as an indicator vector with MN elements, and the value of each element is either zero or non-zero; when it is non-zero, it indicates that there is a user or scatterer in the corresponding grid of the dictionary column space.

[0019] Preferably, in step 1), the deep convolutional neural network CNN provides a stop iteration indication for the SOMP algorithm. Specifically, CNN determines whether the residual signal after each iteration is close to the ambient noise. If not, the iteration of SOMP will continue, and another path will be extracted from the received signal. If so, the algorithm terminates.

[0020] Preferably, in step 1), the constructed deep convolutional neural network CNN structure includes two CBR fusion blocks, a pooling layer, several fully connected (FC) layers, and a Softmax layer. The CBR fusion block contains a convolutional layer, a batch normalization layer, and a rectified linear unit (ReLU) layer.

[0021] Preferably, in step 2), the signal energy-assisted k-means clustering algorithm finds all users and the surrounding scatterers in an iterative manner. Each iteration includes three steps: distance calculation, data point assignment, and cluster center update. In the first step, the distance from all data points to each cluster center is calculated using the Euclidean distance, that is, the distance from the data point s i to the cluster center s' j is expressed as:

[0022]

[0023] where s i represents the i-th data point among many data points, and the vector v i = [x i , y i T represents the Cartesian coordinates corresponding to the polar coordinates (θ i , r i , r i ) of the point s.

[0024] Preferably, in step 2), after the data point assignment, the maximum signal energy in each cluster is detected. When the maximum value is lower than the average energy of all input data points, it is considered that there is a high probability that there is no user in this cluster. Therefore, the total number of set clusters is reduced by 1, and the data points in this cluster will be incorporated into other clusters in the next iteration.

[0025] Preferably, in step 2), in the update of the cluster center, two parameters, signal energy and distance, are introduced as the measurement criteria, that is

[0026]

[0027] where (x k' , y k' ) are the polar coordinates (θ k' , r k') are the corresponding Cartesian coordinates, and k’ is an index variable used to traverse each data point in cluster C k in;

[0028] ρ k' is defined as:

[0029]

[0030] where ρ k' represents the amplitude |u k | of each data point in cluster c k' | weighted coefficient.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1. Solve the positioning problem in complex scattering environments

[0033] The method proposed by the present invention can effectively solve the multi-user positioning problem in the near-field scattering environment when the number of users and the transmitted pilot signals are unknown, breaking through the limitations of traditional methods in complex environments; and this method is applicable to a variety of complex scattering scenarios, such as dense urban environments, indoor scenarios, etc., and has wide applicability.

[0034] 2. Innovation in the signal source acquisition stage

[0035] The present invention proposes a Synchronous Orthogonal Matching Pursuit (SOMP) algorithm based on a Deep Convolutional Neural Network (CNN). When the number of users and the number of scatterers are unknown, by introducing CNN to provide a stop iteration indication for the SOMP algorithm, the problem that the traditional SOMP algorithm is difficult to determine the iteration stop condition is solved; and the introduction of CNN enhances the accuracy of signal source acquisition. Especially in complex scattering environments, it can more accurately locate users and scatter points.

[0036] 3. Innovation in the user identification stage

[0037] The present invention proposes a signal energy-assisted k-means clustering algorithm, which incorporates both the distance between data points and signal energy information into the update criterion of the clustering center, overcoming the limitation of the traditional k-means algorithm that only clusters based on distance; and by combining signal energy information, the accuracy of user identification is significantly improved. Especially in the near-field scattering environment, it can more accurately identify users.

[0038] 4. Overall advantages of the algorithm framework

[0039] The algorithm framework proposed by the present invention includes signal source acquisition and user identification. While ensuring the positioning accuracy, it improves the calculation efficiency, is applicable to real-time positioning requirements, and in complex scattering environments, this method shows strong robustness and can stably complete the multi-user positioning task.

[0040] 5. Practical application value

[0041] The present invention can be applied to wireless communication systems, especially in dense urban environments or indoor scenarios, where it can effectively locate multiple users and improve the resource allocation efficiency of the communication system; in Internet of Things (IoT) applications, this method can help locate a large number of unknown IoT devices and enhance the capabilities of network management and device monitoring; meanwhile, in emergency rescue and security monitoring scenarios, this method can quickly locate multiple targets and improve the rescue efficiency and monitoring accuracy. Description of the drawings

[0042] The drawings are used to provide a further understanding of the present invention and form a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation to the present invention.

[0043] In the drawings:

[0044] Figure 1 is a schematic diagram of a multi-user system in a near-field scattering environment;

[0045] Figure 2 is a schematic diagram of the multi-user positioning algorithm framework in the present invention;

[0046] Figure 3 is a schematic diagram of the CNN network structure in the present invention;

[0047] Figure 4 is a training performance graph of the CNN network in the present invention;

[0048] Figure 5 is a test performance graph of the CNN network in the present invention;

[0049] Figure 6 is a clustering effect display graph of the signal energy-assisted k-means algorithm in the present invention;

[0050] Figure 7 is a clustering performance graph of the signal energy-assisted k-means algorithm in the present invention;

[0051] Figure 8 is a positioning accuracy graph of the multi-user positioning algorithm in the present invention. Detailed implementation manners

[0052] The following describes the preferred embodiments of the present invention with reference to the drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0053] Embodiment: The present invention provides a non-cooperative multi-user positioning method in a near-field scattering environment, including the following steps:

[0054] As shown Figure 1 in the figure, consider a single-cell massive MIMO system, where the base station (BS) is equipped with M antennas in the form of a uniform linear array (ULA). Usually in a 6G network, it is assumed that M ranges from hundreds to thousands. Deploying such a large antenna array, combined with the use of high-frequency carriers, means that the near-field region can extend to dozens to hundreds of meters. As shown Figure 1 in the figure, the near-field region considered in this embodiment is located between the Fresnel boundary and the Fraunhofer boundary. The former is the outer boundary of the inductive near-field, and the latter is the dividing line between the radiative near-field region and the far-field region, which are respectively defined as:

[0055] and

[0056] where D is the antenna diameter and λ is the carrier wavelength.

[0057] Suppose there are K single-antenna users and some scatterers randomly distributed in the near-field region of the BS, and the electromagnetic wave front is a spherical wave. Take the first antenna of the uniform linear array ULA as the reference antenna, and the antenna spacing is Suppose the ULA is parallel to the y-axis, and the coordinates of its m-th antenna are (0, (m - 1)d), where m = 1,..., M. In this system, the BS array steering vector corresponding to a signal with a distance of r and an incident angle of θ is:

[0058]

[0059] where the calculation formula for the distance from the signal source to the m-th antenna is:

[0060]

[0061] For the k-th user, assume that in addition to the direct path, there are L k scatterers around it, then its channel vector is expressed as:

[0062]

[0063] where the first term on the right side of the equation represents the LoS path, and the second term is the sum of L k scattering paths. The symbols g l,k , θ l,k and r l,k represent the coefficient, arrival angle, and distance of the l-th path respectively. And ψ l,k is the phase shift caused by the l-th scatterer, and α l,k is the distance between the user and the l-th scatterer. Assume there are a total of K users, and denote the signal transmitted by the k-th user as s k , then the signal received at the BS is:

[0064]

[0065] Among them, the vector w represents the noise at the receiving end.

[0066] As Figure 2 shown, in order to obtain multi-user location information from the received signal, first, the above-mentioned received signal is re-expressed as:

[0067]

[0068] where L = K + L 1 +L 2 +...+L k is the total number of paths of signal superposition at the receiving end, including the direct path (LoS) of each user and the refracted paths (NLoS) of all scatterers. represents the equivalent coefficient of each path, and s l represents the signal transmitted through path l. It can be seen from the above formula that the signal received by the base station is the superposition of the signals of all paths. Therefore, the information of all paths, that is, the angle of arrival θ l of the signal source on each path and the distance r l to the BS and other parameters can be extracted from y.

[0069] To obtain these parameters, considering the sparsity of the near-field channel, that is, the number of paths is limited, this embodiment proposes to use a method based on compressive sensing (CS) to obtain the location information of the signal source by utilizing the approximate orthogonality of the steering vectors of different paths. Different from the subspace-based method, the CS-based method does not need to calculate the autocorrelation matrix of the received signal, which reduces the pilot overhead and the computational complexity of the matrix eigenvalue decomposition. To use the CS-based method, we re-express the above-mentioned received signal y as:

[0070] y = Du + w

[0071] where D represents the near-field dictionary matrix, which consists of the steering vectors b(θ, r) to form a complete angle-distance space, that is, this dictionary covers the steering vectors at all possible angles and distances. Specifically, the angle space is divided into θ i , i = 1,..., M, where, cosθ i ∈[-1, 1]; the distance space is divided into r j = [r min , r max , j = 1,..., N. Then the dictionary D is expressed as:

[0072] D = [b(θ 1 , r 1 ),..., b(θ 1 , rN ),...,b(θ M ,r 1 ),...,b(θ M ,r N )]

[0073] In the above formula, the vector u is a sparse vector, representing the projection of the received signal in the dictionary space. It can actually be regarded as an indicator vector with MN elements, each element being either zero or non-zero. The non-zero elements indicate that there is a signal source on the corresponding grid in the column space of the dictionary.

[0074] Now the problem is to obtain the positions of the non-zero elements in u, and then the path parameters can be obtained by indexing the relevant columns in the dictionary D. This is solved by the Orthogonal Matching Pursuit (OMP) algorithm in this practical example. This practical example assumes that there are received signals at T moments, and then the Multiple Measurement Vector (MMV) version of the received signal is obtained:

[0075] Y = DU + W

[0076] where Y = [y 1 ,...,y T , U = [u 1 ,...,u T , and W = [w 1 ,...,w T represent the received signal matrix, the sparse vector matrix, and the noise matrix at T moments respectively. Due to the sparsity of the channel, the matrix U is row-sparse. Therefore, the localization problem can be solved by the Simultaneous Orthogonal Matching Pursuit (SOMP) version of OMP.

[0077] As Figure 3 shown, since the number information of users and scatterers is unknown, in the present invention, a deep convolutional network CNN is considered to provide an iterative stopping indication for the SOMP algorithm.

[0078] Specifically, this network consists of two CBR fusion blocks and several fully connected (FC) layers. The CBR fusion block includes a convolutional layer, a batch normalization (BN) layer, and a rectified linear unit (ReLU) layer. Since the received signal is complex, the number of input neurons is set to twice the number of receiving antennas M to process the real and imaginary parts of the received signal. In this practical example, by dividing the signal received on each antenna into real and imaginary parts, the neural network input data at each moment is represented as:

[0079]

[0080] In the formula, z t,m represents the remaining signal on the m-th antenna, where and respectively represent its real part and imaginary part. Starting from the sequence input layer, the input signal is passed to the CBR fusion layer: First, using one-dimensional convolution, by sliding the kernel function or filter over the input data and calculating the dot product of the input data and the kernel, the spatial correlation of the input features is extracted. The mathematical expression is:

[0081]

[0082] where, w conv represents the weights of the convolutional layer. Through padding operations, z t,conv and have the same size. And when using F filters, the output signal includes a total of F channels, denoted as: After convolution, through the BN operation, we get:

[0083]

[0084] where and σ 2 (z t,conv ) are the mean and variance of z t,conv respectively. Adding ∈ in the formula is for the numerical calculation stability, and:

[0085]

[0086]

[0087] This layer is used to normalize the input so that it has zero mean and unit variance. This operation reduces the problem of gradient vanishing and improves the robustness of the network to hyperparameters.

[0088] Subsequently, apply the ReLU layer to the output of the BN layer, that is, the activation function f(x) = max(0, x), to obtain z outTo facilitate network learning of non - linear relationships, the ReLU operation maps negative input values to zero and keeps positive input values unchanged. After two CBR fusion blocks, a max - pooling layer is used for downsampling. Its output is reconstructed and passed to the fully - connected layer (FC). In this network, in this embodiment, a 2M×100 FC layer, a 100×200 FC layer, and a 200×2 FC layer are used, with ReLU activation functions interspersed in the FC layers. From the perspective of representation learning, the convolutional and pooling layers in a convolutional neural network can extract features from input data, while the role of the FC layer is to combine the extracted features in a non - linear way to obtain the output. Among them, the FC layer itself does not have the ability to extract features, but tries to use existing high - order features to achieve the learning goal. Finally, the output of the last FC layer is further processed by the softmax layer to calculate the probability of each class, that is, whether the remaining signal still contains a valid path or is already close to environmental noise. There are a total of T time - moment signals in this embodiment, and each signal is individually input into the network to obtain an output. The final result is obtained by voting, that is, when more than T / 2 outputs indicate that the remaining signal contains a valid path, this embodiment will continue the iteration of the SOMP algorithm.

[0089] Figure 4 and Figure 5 respectively show the training and test performance graphs of the CNN network in the present invention. Figure 4 shows the changes in accuracy and loss with iterations. The network is trained using 5000 training samples, minimizing the cross - entropy loss through the Adam optimizer, with a learning rate of 0.03, using 100 epochs, and a mini - batch size of 20 observations. The training samples consist of two classes, one is the useful signal from users or scatterers, and the other is white noise. These two signals are randomly generated, labeled 1 and 0 respectively, and are normalized before being input into the network. Due to the small number of training samples and the simple architecture of the CNN, it only takes a few minutes to train on a standard CPU. Figure 4 shows the superior performance of the training results, that is, high accuracy and low loss. The results on the test samples further verify its generalization ability, with a classification error rate of 0.0027 on 3000 samples, as Figure 5 shown. The left figure is the label distribution of the test data, and the right figure is the prediction result.

[0090] Figure 6Shows the clustering effect of the signal energy-assisted k-means clustering algorithm in the present invention. This algorithm attempts to find all users and the surrounding scatterers in an iterative clustering manner. Each iteration includes three steps: calculating the distance from all data points to the cluster centers, point assignment, and updating the cluster centers. More specifically, in the distance calculation, in this embodiment, the distances between L input data points and each cluster center are calculated. In most applications, the distance metric is the Euclidean distance, that is, the distance between the data point s i to the cluster center s' j is expressed as:

[0091]

[0092] where s i represents the i-th data point among many data points, and the vector v i =[x i ,y i T represents the Cartesian coordinates corresponding to the polar coordinates (θ i ,r i ,r i ) of the point s

[0093] In the point assignment, according to the distances calculated in the previous step, each data point in the set S is assigned to the nearest cluster, resulting in k clusters c 1 ,...,c k .

[0094] Here, the cluster center update step has two functions: adjusting the number of clusters and updating the center of each cluster. For each cluster, if the maximum signal energy value of the data points in it is still lower than the given average signal energy value, it is considered that there is no user in this cluster, and the data points of this cluster will be incorporated into other clusters in the next iteration. This is reasonable in the scenario considered in this embodiment because the signals from users are usually much stronger than those from scatterers. Therefore, when the maximum value in a cluster is still relatively small, it is very likely that there is no user in this cluster, or the signal of this user is very weak and can be ignored. Then, for other clusters, this embodiment proposes an amplitude-weighted cluster center update criterion, that is:

[0095]

[0096] where (x k' ,y k' ) are the Cartesian coordinates corresponding to the polar coordinates (θ k' ,r k' ), ρ k' represents the weighting coefficient of the amplitude |u k | of each data point in the cluster c k' , and is defined as:

[0097] ​

[0098] Compared with the traditional k-means clustering algorithm that only uses location to calculate the clustering center, the metric proposed in the present invention involves both the amplitude information of each data in the clustering and the location information of each data in the clustering. In this way, this embodiment can make full use of the prior knowledge of the considered scenario, that is, each user and its associated scatterers form a relatively small cluster, and the signal transmitted through the LoS path is usually stronger than the signal from the scatterers, thereby driving the cluster center point to gradually approach the users in each cluster. Therefore, as the algorithm iterates, a user-centered clustering partition will be formed.

[0099] The above process continues until there are no obvious changes in the number of clusters and the location of each cluster center. Then, the clustering center is replaced by the point with the largest amplitude in the clustering, that is, the finally identified user.

[0100] Figure 7 This is the clustering performance graph of the energy-assisted k-means clustering algorithm in the present invention. In Figure 7 (a), the accuracy is defined to evaluate the clustering accuracy of the proposed method, where n(c k ) is the number of data points obtained after clustering of user k, and n is the total number of data points. And in Figure 7 (b), is defined to evaluate the effect of the proposed method in identifying users, where is the number of identified users, and K is the total number of users. The larger these two metrics are, the better the clustering performance. In Figure 7 (a), it can be seen that the proposed unsupervised clustering algorithm shows excellent performance, and its accuracy reaches more than 96% in different scenarios with different numbers of users. However, as the number of users increases, the clustering accuracy will also decrease slightly. This is because as the number of users increases, the scatterers generated in the simulation settings may overlap or be very close in space, resulting in clustering errors. But the loss can be ignored. In Figure 7 (b), it can be seen that the user identification accuracy of this algorithm is basically close to 100%. Since user identification is the main task of multi-user positioning, it can be concluded that this method has achieved excellent performance in non-cooperative multi-user positioning.

[0101] Figure 8 The simulation curve of the positioning accuracy of the algorithm varying with the number of users is given. The positioning accuracy is represented by the mean square error of the position estimation result. In the simulation, the base station uses 256 antennas, the carrier frequency is 28 GHz, and the received signals at 10 moments are used in total. The users are randomly distributed in the near field of 10 - 50 meters. From Figure 8It can be seen that under different numbers of users, the positioning accuracy can reach an error of about 0.8 meters, indicating that the proposed algorithm effectively solves the non-cooperative multi-user positioning problem in the near-field scattering environment.

[0102] Finally, it should be noted that the above are only preferred examples of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A non-cooperative multi-user positioning method in a near-field scattering environment, characterized by: It includes two stages: signal source acquisition and user identification. When the number of users is unknown and pilot signals are sent, multi-user positioning in complex scattering environments is achieved by mining the system's own characteristics and using machine learning algorithms. The method comprises the following steps: 1) In the first stage, the locations of all users and scattering points are obtained using a synchronous orthogonal matching pursuit algorithm based on a deep convolutional neural network (CNN); 2) In the second stage, the signal energy-assisted k-means clustering algorithm is used to incorporate the distance between data points and signal energy information into the update criteria of the cluster center, thereby accurately identifying users.

2. The non-cooperative multi-user positioning method in a near-field scattering environment according to claim 1, characterized in that: In step 1), the synchronous orthogonal matching pursuit algorithm is based on the following sparse representation model: Y=DU+W Where Y = [y1,...,y T ] represents the received signal matrix at T moments, and the received signal vector y at each moment t Contains the received signals of M antennas; similarly, the matrix U=[u1,...,u T ] is composed of sparse vectors at T moments, each sparse vector u t The received signal y t Projection in dictionary space, matrix W = [w1, ..., w T ] represents noise; In addition, in the above formula, D represents the near-field dictionary matrix, which is a complete angle-distance space composed of the steering vector b(θ, r), that is, the dictionary covers the steering vectors at all possible angles and distances; By dividing the angle space into θ i , i = 1, ..., M, where cosθ i ∈[-1,1]; divide the distance space into r j =[r min ,r max ], j = 1, ..., N; then the dictionary D is expressed as: D=[b(θ1,r1),...,b(θ1,r N ),...,b(θ M ,r1),...,b(θ M ,r N )] in, The above formulas are the expressions of the steering vector and the distance from the data source (θ, r) to the mth antenna, respectively, where fc represents the carrier frequency, c is the speed of light, and d represents the antenna spacing, which is set to half a wavelength; After the received signal is projected onto the above dictionary D, the resulting sparse vector u t is considered as an indicator vector with MN elements.

3. The non-cooperative multi-user positioning method in a near-field scattering environment according to claim 2, characterized in that: In step 1), the deep convolutional neural network CNN provides a stop iteration instruction for the SOMP algorithm. Specifically, CNN determines whether the residual signal after each iteration is close to the ambient noise. If not, the iteration of SOMP will continue and another path will be extracted from the received signal. If yes, the algorithm terminates.

4. The non-cooperative multi-user positioning method in a near-field scattering environment according to claim 3, characterized in that: In step 1), the constructed deep convolutional neural network (CNN) structure includes two CBR fusion blocks, a pooling layer, several fully connected layers and a Softmax layer, wherein the CBR fusion block contains a convolution layer, a batch normalization layer and a rectified linear unit layer.

5. The non-cooperative multi-user positioning method in a near-field scattering environment according to claim 1, characterized in that: In step 2), the signal energy-assisted k-means clustering algorithm finds all users and their surrounding scatter points in an iterative manner; each iteration includes three steps: distance calculation, data point allocation, and cluster center update; in the first step, the distance from all data points to each cluster center is calculated using the Euclidean distance, that is, the data point s i To cluster center s' j The distance is expressed as: Among them, s i Represents the i-th data point among many data points, vector v i =[x i ,y i ] T Indicates point s i The polar coordinates (θ i ,r i ) corresponding to the Cartesian coordinates.

6. The non-cooperative multi-user positioning method in a near-field scattering environment according to claim 1, characterized in that: In step 2), after the data points are allocated, the maximum signal energy in each cluster is detected.

7. The non-cooperative multi-user positioning method in a near-field scattering environment according to claim 1, characterized in that: In step 2), in the update of cluster centers, two parameters, signal energy and distance, are introduced as measurement criteria, namely Among them, (x k' ,y k' ) is the polar coordinate (θ k' ,r k' ) is the corresponding Cartesian coordinate, k' is an index variable used to traverse the cluster C k For each data point in ρ k' Defined as: Among them, ρ k' Represents cluster c k The amplitude of each data point |u k' |Weighting coefficient.

Images