Target sensing method and device, sensing equipment, program product and medium

Abstract

The invention provides a target sensing method and device, sensing equipment, a program product and a medium, and relates to the technical field of collaborative sensing. The method comprises the following steps: sending a sensing signal; receiving an echo signal of the sensing signal reflected by the target object; based on machine learning, a regularization parameter value of a random average RALS algorithm is determined; based on an RALS algorithm adopting regularization parameter values, the echo signals are separated, at least one factor array is obtained, and the factor array comprises at least one of an angle factor matrix, a time delay factor matrix and a Doppler factor matrix; and obtaining sensing result information corresponding to the target object according to the at least one factor array. According to the scheme of the invention, the problem of over-fitting or slow convergence speed in the existing target sensing method is solved.

Description

Technical Field

[0001] The present invention relates to the field of collaborative perception technology, and in particular to a target perception method, device, perception equipment, program product and medium. Background Art

[0002] Single-base station target perception mainly includes two aspects: target positioning and target true speed estimation. Among them, target positioning refers to the perception method in which a single sensing node transmits and receives echo signals reflected by the target, and then performs perception processing to achieve target positioning; true speed refers to the speed of a target in two-dimensional space, including speed magnitude and speed direction.

[0003] Based on multipath component fusion, single base station target positioning and real speed perception can be performed. Among them, multipath components refer to the echo signal components with different delays generated at the receiving end when the signal is reflected by buildings or other objects; multipath component fusion refers to a single sensing node using the direct path (Line of sight, LoS) from the node to the target and multiple multipath components generated by multiple buildings, also known as multiple non-direct paths (NLoS), so as to achieve target positioning and real speed perception by a single sensing node.

[0004] In single-base station target positioning and true speed perception based on multipath component fusion, it is necessary to separate the multipath components and then use them to solve the target positioning and true speed estimation. However, existing target perception methods have the problems of overfitting or slow convergence. Summary of the invention

[0005] The purpose of the present invention is to provide a target perception method, device, perception equipment, program product and medium, which solve the problem of overfitting or slow convergence in existing target perception methods.

[0006] To achieve the above object, an embodiment of the present invention provides a target perception method, which is applied to a perception node and includes:

[0007] Sending perception signals;

[0008] Receive an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, the direct path signal is a signal of the sensing signal directly reflected from the target object to the sensing node, and the non-direct path signal is a signal of the sensing signal reaching the target object through a reflective object and reflected from the target object to the sensing node;

[0009] Based on machine learning, determine the regularization parameter value of the random average RALS algorithm;

[0010] Separating the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix;

[0011] According to the at least one factor array, perception result information corresponding to the target object is obtained.

[0012] Optionally, the step of separating the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array includes:

[0013] Acquiring communication information carried by the perception signal;

[0014] According to the communication information, the communication information is stripped from the echo signal to obtain a third-order tensor corresponding to the echo signal;

[0015] Based on the RALS algorithm using the regularization parameter value, the third-order tensor is decomposed to obtain the at least one factor array.

[0016] Optionally, obtaining the perception result information corresponding to the target object according to the at least one factor array includes:

[0017] Perform parameter estimation on each of the factor arrays to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path;

[0018] Acquire environmental information around the sensing node, and construct at least one virtual anchor point based on the environmental information;

[0019] Matching the at least one virtual anchor point with an element in the parameter set to obtain parameter information corresponding to each virtual anchor point, wherein the parameter information includes at least one of angle information, delay information, and Doppler information;

[0020] The perception result information corresponding to the target object is determined according to the parameter information corresponding to the virtual anchor point.

[0021] Optionally, the perception result information includes position information, and determining the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point includes:

[0022] Determine angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0023] Based on the positioning principle of time difference of arrival TDOA or angle of arrival AOA, a first nonlinear equation group is constructed according to the angle information and delay information between the virtual anchor point and the target object;

[0024] Converting the first nonlinear equation group into a first expression corresponding to a maximum likelihood estimation problem;

[0025] The first expression is solved by convex optimization to obtain the position information of the target object.

[0026] Optionally, the perception result information includes speed information, and determining the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point includes:

[0027] Determine angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0028] Determine a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object;

[0029] Converting the Doppler expression into a second expression corresponding to the maximum likelihood estimation problem;

[0030] The second expression is solved by convex optimization to obtain the velocity information of the target object.

[0031] To achieve the above object, an embodiment of the present invention provides a target sensing device, which is applied to a sensing node, including:

[0032] A signal sending module, used for sending a sensing signal;

[0033] A signal receiving module, configured to receive an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, wherein the direct path signal is a signal of the sensing signal directly reflected from the target object to the sensing node, and the non-direct path signal is a signal of the sensing signal reaching the target object through a reflective object and reflected from the target object to the sensing node;

[0034] A first processing module is used to determine a regularization parameter value of a random average RALS algorithm based on machine learning;

[0035] A second processing module, configured to separate the echo signal based on a RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix;

[0036] The third processing module is used to obtain the perception result information corresponding to the target object according to the at least one factor array.

[0037] Optionally, the second processing module includes:

[0038] An information acquisition unit, used to acquire the communication information carried by the perception signal;

[0039] An information stripping unit, configured to strip the communication information from the echo signal according to the communication information, so as to obtain a third-order tensor corresponding to the echo signal;

[0040] A decomposition processing unit is used to decompose the third-order tensor based on the RALS algorithm using the regularization parameter value to obtain the at least one factor array.

[0041] Optionally, the third processing module includes:

[0042] A parameter estimation unit, configured to perform parameter estimation on each of the factor arrays to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path;

[0043] An anchor point construction unit, used to obtain environmental information around the sensing node and construct at least one virtual anchor point according to the environmental information;

[0044] a matching processing unit, configured to match the at least one virtual anchor point with an element in the parameter set to obtain parameter information corresponding to each virtual anchor point, wherein the parameter information includes at least one of angle information, delay information and Doppler information;

[0045] The first processing unit is used to determine the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point.

[0046] Optionally, the perception result information includes location information, and the first processing unit includes:

[0047] A first processing subunit, configured to determine angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0048] A second processing subunit is used to construct a first nonlinear equation group based on the positioning principle of the time difference of arrival TDOA or the angle of arrival AOA and according to the angle information and the delay information between the virtual anchor point and the target object;

[0049] A third processing subunit, configured to convert the first nonlinear equation group into a first expression corresponding to a maximum likelihood estimation problem;

[0050] The first solving subunit is used to perform convex optimization solution on the first expression to obtain the position information of the target object.

[0051] Optionally, the perception result information includes speed information, and the first processing unit includes:

[0052] a fourth processing subunit, configured to determine angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0053] a fifth processing subunit, configured to determine a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object;

[0054] a sixth processing subunit, configured to convert the Doppler expression into a second expression corresponding to a maximum likelihood estimation problem;

[0055] The second solving subunit is used to perform convex optimization solution on the second expression to obtain the speed information of the target object.

[0056] To achieve the above object, an embodiment of the present invention provides a sensing device, which is a sensing node, including a processor and a transceiver, wherein the transceiver is used to send a sensing signal;

[0057] The transceiver is further used to receive an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, the direct path signal is a signal of the sensing signal directly reflected from the target object to the sensing node, and the non-direct path signal is a signal of the sensing signal reaching the target object through a reflective object and reflected from the target object to the sensing node;

[0058] The processor is used to determine a regularization parameter value of a random average RALS algorithm based on machine learning;

[0059] The processor is further configured to separate the echo signals based on a RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix;

[0060] The processor is further configured to obtain perception result information corresponding to the target object according to the at least one factor array.

[0061] Optionally, when the processor separates the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, the processor is specifically configured to:

[0062] Acquiring communication information carried by the perception signal;

[0063] According to the communication information, the communication information is stripped from the echo signal to obtain a third-order tensor corresponding to the echo signal;

[0064] Based on the RALS algorithm using the regularization parameter value, the third-order tensor is decomposed to obtain the at least one factor array.

[0065] Optionally, when the processor obtains the perception result information corresponding to the target object according to the at least one factor array, it is specifically configured to:

[0066] Perform parameter estimation on each of the factor arrays to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path;

[0067] Acquire environmental information around the sensing node, and construct at least one virtual anchor point based on the environmental information;

[0068] Matching the at least one virtual anchor point with an element in the parameter set to obtain parameter information corresponding to each virtual anchor point, wherein the parameter information includes at least one of angle information, delay information, and Doppler information;

[0069] The perception result information corresponding to the target object is determined according to the parameter information corresponding to the virtual anchor point.

[0070] Optionally, the perception result information includes position information, and when the processor determines the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point, it is specifically configured to:

[0071] Determine angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0072] Based on the positioning principle of time difference of arrival TDOA or angle of arrival AOA, a first nonlinear equation group is constructed according to the angle information and delay information between the virtual anchor point and the target object;

[0073] Converting the first nonlinear equation group into a first expression corresponding to a maximum likelihood estimation problem;

[0074] The first expression is solved by convex optimization to obtain the position information of the target object.

[0075] Optionally, the perception result information includes speed information, and when the processor determines the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point, it is specifically configured to:

[0076] Determine angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0077] Determine a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object;

[0078] Converting the Doppler expression into a second expression corresponding to the maximum likelihood estimation problem;

[0079] The second expression is solved by convex optimization to obtain the velocity information of the target object.

[0080] To achieve the above-mentioned purpose, an embodiment of the present invention provides a perception device, including a transceiver, a processor, a memory, and a program or instruction stored in the memory and executable on the processor; when the processor executes the program or instruction, the target perception method as applied to the perception node as above is implemented.

[0081] In order to achieve the above-mentioned object, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the target perception method as applied to a perception node as described above.

[0082] To achieve the above objectives, an embodiment of the present invention provides a readable storage medium having a program or instruction stored thereon, which, when executed by a processor, implements the steps in the target perception method applied to a perception node as described above.

[0083] The beneficial effects of the above technical solution of the present invention are as follows:

[0084] According to the method of the embodiment of the present invention, after the sensing node sends the sensing signal, it can receive the echo signal reflected by the target object, and then determine the regularization parameter value of the RALS algorithm based on machine learning, and then separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, and then obtain the sensing result information corresponding to the target object based on the at least one factor array. In this way, based on machine learning, the problem of difficulty in selecting regularization parameters in the RALS algorithm can be solved, and the problems of overfitting or slow convergence in target perception caused by improper selection of regularization parameters of the RALS algorithm can be avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 is a flow chart of a target perception method according to an embodiment of the present invention;

[0086] Figure 2 A schematic diagram of a target perception method according to an embodiment of the present invention;

[0087] Figure 3 A schematic diagram of a perception scene according to an embodiment of the present invention;

[0088] Figure 4 A schematic diagram of the geometric relationship between a target object and a sensing node according to an embodiment of the present invention;

[0089] Figure 5 is a structural diagram of a target sensing device according to an embodiment of the present invention;

[0090] Figure 6 is a structural diagram of a sensing device according to an embodiment of the present invention;

[0091] Figure 7 This is a structural diagram of a sensing device according to another embodiment of the present invention. DETAILED DESCRIPTION

[0092] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0093] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present invention. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0094] In various embodiments of the present invention, it should be understood that the size of the serial numbers of the following processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0095] Additionally, the terms "system" and "network" are often used interchangeably herein.

[0096] In the embodiments provided in the present application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined according to A. However, it should also be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0097] Here, the relevant technologies are introduced as follows:

[0098] In the research on single-node target positioning and true speed perception based on multipath components, there are two main challenges. The first challenge is how to separate the multipath of the mixed signal of multipath aliasing, and the second challenge is how to use the multipath parameter information to achieve target perception.

[0099] At present, the solutions for hybrid multipath signal separation mainly include: a solution based on tensor decomposition, a solution based on independent component analysis (ICA), and a solution based on deep learning.

[0100] Among them, the scheme based on tensor decomposition mainly makes full use of the multidimensional structural information of the data and the high-order statistical characteristics of the tensor, so as to improve the performance and robustness of signal separation, but the premise of this scheme is that the rank of the tensor needs to be known. Common tensor decomposition methods include CP decomposition, Tucker decomposition and non-negative tensor decomposition. Among them, CP decomposition is the most commonly used for the separation of perceptual multipath mixed signals. In CP decomposition, the most classic processing methods include alternating least squares (ALS), regularized alternating least squares (RALS) and semidefinite relaxation (SDR). However, these methods all have some disadvantages, such as: ALS has the disadvantages of slow convergence speed, poor robustness and easy to fall into local optimum, RALS has the disadvantages of difficulty in selecting regularization coefficients and easy overfitting, and SDR has the problem of easy falling into local optimum and high computational complexity.

[0101] In the ICA-based solution, the core idea is to find a set of linear transformations by maximizing the mutual information of the signals to separate the observed multiple mixed signals into independent source signals. The advantage of this solution is that it does not require prior knowledge (such as the rank or dimension of the data) and can handle nonlinear mixed and non-Gaussian signals well. Common ICA solution methods include: fast ICA, information maximization ICA, and ICA based on minimum mutual information. However, these ICA solution methods also have disadvantages, such as being prone to falling into local optimal solutions, poor effects on mixed signals with Gaussian distributions, and inaccurate estimation of the mixing matrix.

[0102] In deep learning-based solutions, such as deep learning-based blind source separation methods, neural networks are mainly used for separation, especially autoencoders and variational autoencoders (VAE), which have attracted widespread attention in the fields of signal processing and machine learning in recent years. The core idea of ​​these methods is to learn how to decouple independent source signals from complex mixed signals without knowing the specific structure of the mixed signals or the statistical characteristics of the source signals in advance. The advantage of this method is that it has strong generalization ability. Once fully trained, the model can effectively separate the source signals in unseen mixed situations. Since it does not rely on prior signal distribution assumptions, it is more suitable for diversity and complexity in practice. In addition, the flexibility of deep learning models allows them to adapt to various complex situations that may occur in the real world through appropriate data enhancement and model design. However, deep learning methods also face some challenges and limitations. For example, they usually require a large amount of labeled data for training, the training time is too long, and especially when processing large-scale data, it requires large computational and resource costs. In addition, the performance and effect of the model largely depend on the design of the network architecture, the selection of hyperparameters, and the tuning strategy during training, all of which require in-depth experiments and analysis to optimize and are difficult to adapt to dynamic environments.

[0103] The existing target perception scheme based on multipath information usually uses multipath information and mixed time difference of arrival (TDOA) and angle of arrival (AOA) to construct a solution problem and use weighted least squares (WLS) to solve it. Among them, the hybrid TDOA and AOA algorithm uses the different arrival time differences and received signal directions of virtual anchor points to jointly solve the overdetermined equations, and WLS estimates the unknown quantity by minimizing the residual equation and. However, these methods are difficult to achieve the true speed estimation of the target, and WLS is not suitable for solving nonlinear problems.

[0104] like Figure 1 As shown, a target perception method according to an embodiment of the present invention is applied to a perception node, including:

[0105] Step 101: Send a perception signal.

[0106] When the sensing node starts sensing, it will emit uniform signal radiation into the space (that is, send a sensing signal), the receiving antenna of the sensing node will start working, and can obtain the communication information carried in the transmitted sensing signal through the feedback mechanism.

[0107] Step 102, receiving an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one indirect path signal, the direct path signal is a signal that the sensing signal is directly reflected from the target object to the sensing node, and the indirect path signal is a signal that the sensing signal passes through a reflective object to reach the target object and is reflected from the target object to the sensing node.

[0108] It should be noted that the echo signal here includes the direct path signal LoS and / or the multipath component (i.e., the non-direct path signal NLoS), that is, the echo signal not received by the sensing node here is a mixed multipath signal. In the embodiment of the present invention, the reflection phenomenon of the electromagnetic wave (such as the above-mentioned sensing signal) on the surface of the target object is mainly used to locate and measure the speed of the target object, and finally obtain the sensing result of the sensing node on the target object (including the position information obtained by the target positioning and the speed information obtained by the speed measurement).

[0109] Specifically, Figure 2 As shown, the target perception method of the present invention can be specifically performed by a target perception system deployed on a perception node, and the system can include the following modules: a mixed signal receiving and separation module, a regularization parameter estimation module based on machine learning, a parameter estimation module, a priori environmental information acquisition module, and a perception module based on multipath fusion. Among them, the above step 102 can be performed by the mixed signal receiving and separation module, which is responsible for receiving the echo signal, including the perception signal reflected by the target object and the multiple two-way NLoS path mixed signals reflected by the reflective surface such as the building to reach the receiving end of the perception node.

[0110] Step 103: Determine a regularization parameter value of the random average RALS algorithm based on machine learning.

[0111] like Figure 2As shown, this step can be specifically performed by a regularization parameter estimation module based on machine learning, which is mainly responsible for finding the optimal regularization parameter value of the RALS algorithm in the current scenario. Usually, the learning of this process is independently run and pre-learned. Specifically, the module can generate or collect the received data in the test environment as a training data set, and use the true value results as a verification data set to find the optimal regularization parameter value through the cross-validation algorithm in machine learning. The score is composed of 20% decomposition error and 80% convergence speed. In a specific embodiment, the module obtains the optimal regularization parameter value by learning 1,000 diverse data sets.

[0112] It should be noted that in step 103, based on machine learning, the optimal regularization parameter can be obtained in a variety of scenarios, thereby overcoming the problem that the traditional RALS method is difficult to select the optimal regularization parameter, and can obtain higher separation accuracy and faster convergence speed, avoiding problems such as overfitting or slow convergence speed caused by improper selection of RALS regularization parameters.

[0113] Step 104: Separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix.

[0114] In this step, the RALS method based on machine learning is used to achieve the optimal selection of regularization parameters in diverse scenarios, and further achieve high-precision signal separation of mixed multipath signals (i.e., echo signals).

[0115] Step 105: Obtain perception result information corresponding to the target object according to the at least one factor array.

[0116] It should be noted that the perception result information corresponding to the target object here mainly includes the position information and speed information of the target object.

[0117] In this embodiment, after the sensing node sends the sensing signal, it can receive the echo signal reflected by the target object, and then determine the regularization parameter value of the RALS algorithm based on machine learning, and then separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, and then obtain the sensing result information corresponding to the target object based on the at least one factor array. In this way, based on machine learning, the problem of difficulty in selecting regularization parameters in the RALS algorithm can be solved, and then the problems of overfitting or slow convergence in target perception caused by improper selection of regularization parameters of the RALS algorithm can be avoided.

[0118] It should be noted that the target sensing method of the embodiment of the present invention mainly considers the sensing scenario of a single sensing node. In the following, the sensing node is a static sensing node (such as a base station) as an example for description.

[0119] like Figure 3 As shown in the figure, the sensing node (such as base station G) performs target perception in a complex multipath environment. The main process includes: the sensing node transmits a sensing signal, the sensing signal is reflected by L buildings (such as building 1, building 2, building 3...building L) and reaches the target object (i.e., target O), then returns through the original path, and is finally received by the sensing node. After that, the sensing node obtains the location information and true speed information of target O through sensing signal processing.

[0120] In such Figure 3 In the sensing scenario shown, the OFDM signal transmitted by the sensing node occupies N c subcarriers and M sym The sensing node is configured with a uniform linear antenna array (ULA), which includes: N T Transmitting antennas and N R Root receiving antenna. It should be noted that in this sensing scenario, the reflection generated by the building is considered as specular reflection, and because the energy of multiple reflections is extremely small, only one reflection path is considered (that is, the two-way multipath from the sensing node to the building and then to the target object). Here, assuming that each building generates a reflection, the number of multipaths is equal to the number of buildings (that is, there are L multipaths).

[0121] Here, assuming that the position of the sensing node (that is, the stationary base station G) is (0,0), the up-converted OFDM signal transmitted by the sensing node (that is, the sensing signal model) can be expressed as:

[0122]

[0123] Where m∈{0,1,…,M sym -1} represents the index value of OFDM symbol; n∈{0,1,…,N c -1} represents the subcarrier index value; d(m,n) represents the Quadrature Amplitude Modulation (QAM) or Quadrature Phase Shift Keying (QPSK) modulation symbol sent by the sensing node; j represents the imaginary unit, t represents the time; f c represents the total duration of the carrier frequency; Δf represents the subcarrier spacing; T represents the total duration of the OFDM symbol; rect(·) represents the rectangular window function.

[0124] In the sensing time slot, the OFDM signal transmitted by the sensing node is reflected by L buildings and arrives at a location (x g ,y g ), the speed is The velocity angle is θ v The single target O is then returned to the sensing node through the same reciprocal path and received by the sensing node. The down-conversion of the echo received on the nth subcarrier during the mth OFDM symbol time at the sensing node can be expressed as:

[0125]

[0126] in, l represents the index value of the multipath, l∈{0,1,…,L}; α l represents the attenuation of the lth multipath component (including the reflection coefficient of the target O and the building and the path loss). When l = 0, it represents the LoS path between the sensing node and the target O; represents the Doppler shift, where v l represents the relative speed of target O relative to the lth reflector (here refers to the building), f c Indicates the total duration of the carrier frequency, c 0 represents the speed of light of the target relative to the lth reflector; T represents the total duration of the OFDM symbol; Δf represents the subcarrier spacing; τ l Indicates delay, Where R l represents the sum of the distance from the target to the lth reflector and the distance from the lth reflector to the sensing node; θ l represents the arrival angle AoA of the lth multipath component; a r (·) indicates receiving the steering vector; a t (·) represents the launch steering vector; (·) T Represents the transpose operation; represents a modulation symbol vector; Represents Additive White Gaussian Noise (AWGN).

[0127] Among them, a r (·) can be expressed by the following formula:

[0128]

[0129] a t (·) can be expressed by the following formula:

[0130]

[0131] Among them, dr represents the interval between two antennas; k represents the index value of the receiving antenna, k = {0, 1, ..., N R -1}; λ represents the wavelength; p represents the index value of the transmitting antenna, p={0,1,…,N T -1}.

[0132] In some embodiments, step 104, separating the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array includes:

[0133] Step 1041: Acquire communication information carried by the perception signal.

[0134] Step 1042: Strip the communication information from the echo signal according to the communication information to obtain a third-order tensor corresponding to the echo signal.

[0135] Here, for the stripping of communication signals (i.e., communication information), the "modulation symbol domain" signal processing method can usually be used to achieve the stripping of known communication symbols (i.e., communication information), and represent the mixed multipath signal at the receiving end (i.e., echo signal) as a third-order tensor.

[0136] It should be noted that when the sensing node receives a sensing service request, it uses an omnidirectional antenna mode (i.e., provides uniform radiation coverage on the horizontal plane) without knowing the location of the target object in advance. At this time, the transmitted signal will be reflected by reflective objects (such as buildings), generating multipath components, and the receiving antenna of the sensing node receives the mixed multipath signal (i.e., the echo signal, including the multipath component and the LoS path signal of the target object) and performs subsequent processing. Among them, the mixed multipath signal mainly undergoes down-conversion, digital-to-analog conversion, coarse synchronization, de-cyclic prefix, discrete Fourier transform, channel equalization, multipath merging and other operations, and then according to the previously known communication information (i.e., the communication information carried by the sensing signal), the corresponding received modulation symbol is stripped of communication information; after the communication information is stripped, the obtained space-time-frequency three-dimensional complex matrix can be expressed in the form of a third-order tensor, which is convenient for subsequent processing and analysis.

[0137] In a specific embodiment of the present invention, in N R There are N receiving antennas, c subcarriers and M sym On OFDM symbols, the third-order tensor of the echo signal received by the sensing node can be expressed as:

[0138] in, Z represents an AWGN tensor; · represents the inner product of two matrices or vectors; θ l represents the arrival angle AoA of the lth multipath component; A represents the angle factor matrix: B represents the delay factor matrix: C represents the Doppler factor matrix:

[0139]

[0140] Among them, a an (θ l ) can be expressed as:

[0141] a de (τ l ) can be expressed as:

[0142] f(f D,l ) can be expressed as:

[0143] in, represents a unit vector.

[0144] Step 1043: Decompose the third-order tensor based on the RALS algorithm using the regularization parameter value to obtain the at least one factor array.

[0145] It should be noted that the column vectors (i.e., elements) in each factor matrix (i.e., angle factor matrix, delay factor matrix, and Doppler factor matrix) are corresponding, that is, the column vectors (such as angle information, delay information, and Doppler information) corresponding to the same index value of different factor matrices come from the same multipath.

[0146] In this step, the RALS method based on the regularization parameter selected by machine learning can be used to achieve separation of multipath mixed signals to obtain at least one factor array.

[0147] Specifically, when the number of multipaths is greater than 1, the separation condition is met. Under the premise of knowing the number of multipaths, the CP decomposition problem (the basic idea is to express a three-dimensional tensor as the sum of several rank-1 tensors) can be expressed as a low-rank approximation problem, which can be expressed as: Among them, A represents the angle factor matrix, B represents the delay factor matrix, C represents the Doppler factor matrix, and Y represents the third-order tensor. represents the Frobenius norm.

[0148] It should be noted that the low-rank approximation problem can be solved by the RALS algorithm. Specifically, the RALS algorithm can reshape the problem into three coupled sub-problems, which are expressed as follows:

[0149]

[0150] Among them, λ represents the regularization parameter.

[0151] In some embodiments, step 105, obtaining the perception result information corresponding to the target object according to the at least one factor array, includes:

[0152] Step 1051, perform parameter estimation on each of the factor arrays respectively to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path.

[0153] It should be noted that the above-mentioned parameter estimation module is mainly responsible for the angle, delay and Doppler parameter estimation of the separated multipath signal (that is, the angle factor matrix, delay factor matrix and Doppler factor matrix output by the mixed signal receiving and separation module). Here, the parameter estimation method can adopt the MUSIC method, the FFT method based on compressed sensing and the FFT method based on the grid. Through the parameter estimation method, the angle information (that is, the angle set), the delay information (that is, the delay set) and the Doppler information (that is, the Doppler set) can be obtained. It should be noted that the three parameters are one-to-one corresponding, and the elements in the parameter set can be renumbered, for example, angle 1, delay 1 and Doppler 1 correspond to the first multipath. It should be noted that this does not affect the subsequent perception, and the multipath and virtual anchor point matching will be performed later.

[0154] Here, the angle factor matrix is ​​used as an example. The angle factor matrix can be expressed as:

[0155]

[0156] in, represents the angle factor matrix, and Z represents the AWGN matrix.

[0157] By observing the angle factor matrix, we can find that each column vector in the matrix corresponds to a multipath, and the phase offset generated by the angle is along the receiving antenna dimension of each column vector. Therefore, we can estimate the parameters of each column and obtain L+1 angle information.

[0158] It should be noted that any parameter estimation method is applicable in the embodiments of the present invention. Here, the 2D-FFT method is used to estimate the l-th angle information as an example for explanation:

[0159] First, take out the angle factor matrix The lth column vector of

[0160] Then, the column vector is subjected to a density functional theory (DFT) operation, which can be expressed as:

[0161]

[0162] in, represents the angle factor matrix, k′=0,1,…,N R -1.

[0163] After that, a peak search is performed on g(k′) to obtain the corresponding peak index value ξ l , then the estimated angle information can be expressed as:

[0164]

[0165] In this way, through the 2D-FFT method, L+1 angle information can be obtained, and finally these angle information can be represented as an angle set Similarly, the delay set and Doppler set can be obtained, where the delay set is expressed as The Doppler set is expressed as

[0166] Step 1052: Acquire environmental information around the sensing node, and construct at least one virtual anchor point based on the environmental information.

[0167] like Figure 2 As shown, the above-mentioned prior environment information acquisition module is mainly responsible for acquiring the environment information around the sensing node, such as the position and angle of reflective objects such as buildings around the sensing node.

[0168] It should be noted that, for the acquisition of the physical space environment information around the sensing node, it can be assumed that the physical space is a building. For static sensing nodes (such as base stations), it is generally believed that the reflective objects around the sensing nodes are stationary and can be pre-calculated (that is, information can be acquired and stored in advance); for dynamic sensing nodes (that is, non-static sensing nodes, such as user equipment (UE), and for example, vehicle-mounted synaesthesia integrated equipment, drones, etc.), since the surrounding environment of the sensing node is updated in real time, it is usually necessary to use part of the time slots of the sensing frame (such as the first few time slots of the sensing frame) to reconstruct the environment before the sensing node performs perception, and then obtain the physical space environment information, that is, the environmental information acquisition can be performed in the first few time slots of the sensing node to perform the perception task. At this time, it is believed that the surrounding environment is unchanged during the total time of acquiring environmental information and performing perception. For example, for the current fifth generation mobile communication technology (5th Generation Mobile Communication Technology, 5G) new wireless (New Radio, NR) frame structure, a frame structure containing 160 time slots per frame can be selected, and 20 time slots are reserved for perception. These 20 time slots are divided into: 10 time slots for environment reconstruction and another 10 time slots for perception.

[0169] It should also be noted that the virtual anchor point refers to a virtual node constructed by using the geometric relationship between the sensing node and the reflective object in the physical space. In the embodiment of the present invention, the role of the virtual anchor point is: assume that the multipath component received by the sensing node is received by the virtual anchor point. In other words, the Ath multipath component signal is converted into a signal sensed at the virtual anchor point, and each reflective object can generate a virtual anchor point. Figure 4 As shown, for the l′∈{1,2,…,L}th reflective object, assume that its coordinates in two-dimensional space are (x l′ ,y l′ ), the direction is φ l′ , we can obtain the horizontal coordinate (x l′,VA ) and the ordinate (y l′,VA ).

[0170] Among them, the horizontal axis (x l′,VA ) can be expressed as:

[0171]

[0172] The vertical axis (y l′,VA ) can be expressed as:

[0173]

[0174] Step 1053: Match the at least one virtual anchor point with the elements in the parameter set to obtain parameter information corresponding to each virtual anchor point, where the parameter information includes at least one of angle information, delay information, and Doppler information.

[0175] In this step, in order to know which virtual anchor point the L+1 angle information, delay information and Doppler information in the parameter estimation module correspond to, matching can be performed through the following process:

[0176] First, the angle information, delay information and Doppler information are grouped and named as multipath groups, where the elements contained in the lth multipath group are Then, the angle between the center point of each reflective object and the sensing node is calculated to obtain the angles between L reflective objects and the sensing node, which is recorded as Finally, Take as an example, traverse the angle information in L+1 multipath groups and The multipath group corresponding to the closest angle information is determined as and The corresponding virtual anchor point l′ is matched. Through the above operations, the corresponding multipath group can be found for each virtual anchor point, and the remaining multipath group is considered to match the sensing node. So far, each multipath component can be regarded as being received separately, for example: multipath component A is regarded as being received by virtual anchor point A, and LoS path component is regarded as being received by sensing node. In this way, the target perception problem based on multipath component fusion can be transformed into a joint perception problem of multiple sensing nodes. Here, for the convenience of subsequent representation, it is assumed that the received information of virtual anchor point l′ is expressed as The received information of the sensing node is expressed as

[0177] Step 1054: Determine the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point.

[0178] In some embodiments, the perception result information includes position information, and determining the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point includes:

[0179] (i) determining angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node.

[0180] See also Figure 4 , for the virtual anchor point l′, the delay (and Doppler) of the sensing signal reaching the sensing node after reflection is the same as the delay (and Doppler) of the sensing signal reaching the virtual anchor point, but the angle is different. Therefore, it is necessary to change the angle θ l Convert to It can be considered that the multipath component is received by the virtual anchor point. According to the geometric relationship, we can get:

[0181]

[0182] In this way, all the angle information received at the virtual anchor point is updated. For the virtual anchor point l′, the received information can be expressed as For the sensing node, the received information is expressed as

[0183]

[0184] (ii) Based on the positioning principle of the time difference of arrival TDOA or the angle of arrival AOA, a first nonlinear equation group is constructed according to the angle information and the time delay information between the virtual anchor point and the target object.

[0185] Here, the angle information and delay information in the received information can be used to estimate the location of the target. Combined with the TDOA or AOA method, the target positioning problem can be expressed as a nonlinear equation group for solution. Here, the first nonlinear equation group can be expressed as:

[0186]

[0187] Among them, r l′,0 =R l′ -R 0 represents the difference between the distance from the target object to the l′th virtual anchor point and the distance from the target to the sensing node, and and the position of the target is expressed as (x tr ,y tr ).

[0188] (3) Converting the first nonlinear equation group into a first expression corresponding to the maximum likelihood estimation problem.

[0189] Here, the first nonlinear equation set can be transformed into a maximum likelihood estimation problem, thereby improving accuracy and robustness. The estimated error vector can be expressed as:

[0190]

[0191] The first nonlinear equation system is transformed into a maximum likelihood estimation problem, which can be expressed by the first expression:

[0192]

[0193] (iv) performing convex optimization on the first expression to obtain the position information of the target object.

[0194] It should be noted that the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm can be used to solve the target positioning result (that is, the position information of the target object) as follows: Thereby achieving target positioning.

[0195] In some embodiments, the perception result information includes speed information, and determining the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point includes:

[0196] (i) determining angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node.

[0197] (ii) determining a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object.

[0198] In some specific embodiments, the Doppler at the virtual anchor point can be expressed by the following Doppler expression:

[0199]

[0200] (iii) Converting the Doppler expression into a second expression corresponding to the maximum likelihood estimation problem.

[0201] Among them, the estimated error vector can be expressed as:

[0202]

[0203] Here, the Doppler expression is transformed into a maximum likelihood estimation problem, which can be expressed by the second expression, which is:

[0204]

[0205] (iv) performing convex optimization on the second expression to obtain velocity information of the target object.

[0206] Here, the second expression can be solved by convex optimization using the BFGS algorithm to obtain the true speed of the target object, that is, the speed information, which includes: angle and size This enables the estimation of the true speed of the target object.

[0207] In the above embodiment, the positioning problem of the target object and the true speed estimation problem of the target object are converted into a maximum likelihood estimation problem, and then the BFGS method is used to solve the convex optimization problem, which can obtain more flexibility and accuracy in problem solving.

[0208] In a specific embodiment of the present invention, for a single-node target perception scenario in a complex multipath environment, based on multipath component fusion, target perception (including target positioning and true speed estimation) under a single node of synaesthesia integration can be assisted. Specifically, the above-mentioned perception module based on multipath fusion can use a priori physical space environment information to construct a virtual anchor point. Then, the three parameter sets obtained in the parameter estimation module are matched with the virtual anchor point. For example, for virtual anchor point A, the angle between the reflector A corresponding to the virtual anchor point A and the perception node can be calculated. When the angle value B of multipath B is very close to the angle value of this reflector A and the perception node, it can be considered that multipath B is the multipath reflected by the reflector A, that is, the receiving end of multipath B is the virtual anchor point A. At this time, each virtual anchor point and perception node has a set of perception information from the target object, and the single-node perception problem based on multipath components is converted into a problem of multi-point collaborative perception under perception nodes and multiple virtual anchor points. Finally, for target positioning, the angle information and delay information of the multipath corresponding to each virtual anchor point can be converted into the angle information and delay information between the virtual anchor point and the target object, and then the TDOA or AOA positioning principle is used to construct a nonlinear equation group and convert it into a maximum likelihood estimation problem. The BFGS algorithm is further used to solve the convex optimization problem to obtain the two-dimensional spatial position information of the target object. For the target true velocity estimation, the Doppler expression between the target object and the virtual anchor point can be expressed first, and then converted into a maximum likelihood estimation problem. The BFGS algorithm is further used to solve the convex optimization problem to obtain the two-dimensional spatial true velocity estimation of the target object.

[0209] The target perception method of this embodiment can achieve higher-precision target positioning than existing solutions and can also achieve high-precision target true speed estimation by converting nonlinear equations into maximum likelihood estimation problems and using the BFGS method to solve convex optimization problems.

[0210] like Figure 5 As shown, a target sensing device according to an embodiment of the present invention is applied to a sensing node, including:

[0211] A signal sending module 510, configured to send a sensing signal;

[0212] The signal receiving module 520 is used to receive an echo signal of the sensing signal reflected by the target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, the direct path signal is a signal that the sensing signal is directly reflected from the target object to the sensing node, and the non-direct path signal is a signal that the sensing signal reaches the target object through a reflective object and is reflected from the target object to the sensing node;

[0213] A first processing module 530 is used to determine a regularization parameter value of a random average RALS algorithm based on machine learning;

[0214] A second processing module 540 is used to separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix and a Doppler factor matrix;

[0215] The third processing module 550 is used to obtain the perception result information corresponding to the target object according to the at least one factor array.

[0216] In this embodiment, after the sensing node sends the sensing signal, it can receive the echo signal reflected by the target object, and then determine the regularization parameter value of the RALS algorithm based on machine learning, and then separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, and then obtain the sensing result information corresponding to the target object based on the at least one factor array. In this way, based on machine learning, the problem of difficulty in selecting regularization parameters in the RALS algorithm can be solved, and then the problems of overfitting or slow convergence in target perception caused by improper selection of regularization parameters of the RALS algorithm can be avoided.

[0217] Optionally, the second processing module 540 includes:

[0218] An information acquisition unit, used to acquire the communication information carried by the perception signal;

[0219] An information stripping unit, configured to strip the communication information from the echo signal according to the communication information, so as to obtain a third-order tensor corresponding to the echo signal;

[0220] A decomposition processing unit is used to decompose the third-order tensor based on the RALS algorithm using the regularization parameter value to obtain the at least one factor array.

[0221] Optionally, the third processing module 550 includes:

[0222] A parameter estimation unit, configured to perform parameter estimation on each of the factor arrays to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path;

[0223] An anchor point construction unit, used to obtain environmental information around the sensing node and construct at least one virtual anchor point according to the environmental information;

[0224] a matching processing unit, configured to match the at least one virtual anchor point with an element in the parameter set to obtain parameter information corresponding to each virtual anchor point, wherein the parameter information includes at least one of angle information, delay information and Doppler information;

[0225] The first processing unit is used to determine the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point.

[0226] Optionally, the perception result information includes location information, and the first processing unit includes:

[0227] A first processing subunit, configured to determine angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0228] A second processing subunit is used to construct a first nonlinear equation group based on the positioning principle of the time difference of arrival TDOA or the angle of arrival AOA and according to the angle information and the delay information between the virtual anchor point and the target object;

[0229] A third processing subunit, configured to convert the first nonlinear equation group into a first expression corresponding to a maximum likelihood estimation problem;

[0230] The first solving subunit is used to perform convex optimization solution on the first expression to obtain the position information of the target object.

[0231] Optionally, the perception result information includes speed information, and the first processing unit includes:

[0232] a fourth processing subunit, configured to determine angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0233] a fifth processing subunit, configured to determine a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object;

[0234] a sixth processing subunit, configured to convert the Doppler expression into a second expression corresponding to a maximum likelihood estimation problem;

[0235] The second solving subunit is used to perform convex optimization solution on the second expression to obtain the speed information of the target object.

[0236] It should be noted here that the above-mentioned target perception device provided by the embodiment of the present invention can implement all the method steps implemented by the above-mentioned target perception method embodiment applied to the perception node, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0237] like Figure 6 As shown, a sensing device 600 according to an embodiment of the present invention is a sensing node, including a processor 610 and a transceiver 620, wherein the transceiver 620 is used to send a sensing signal;

[0238] The transceiver 620 is further used to receive an echo signal of the sensing signal reflected by the target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, the direct path signal is a signal of the sensing signal directly reflected from the target object to the sensing node, and the non-direct path signal is a signal of the sensing signal reaching the target object through a reflective object and reflected from the target object to the sensing node;

[0239] The processor 610 is used to determine a regularization parameter value of a random average RALS algorithm based on machine learning;

[0240] The processor 610 is further configured to separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix;

[0241] The processor 610 is further configured to obtain perception result information corresponding to the target object according to the at least one factor array.

[0242] In this embodiment, after the sensing node sends the sensing signal, it can receive the echo signal reflected by the target object, and then determine the regularization parameter value of the RALS algorithm based on machine learning, and then separate the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, and then obtain the sensing result information corresponding to the target object based on the at least one factor array. In this way, based on machine learning, the problem of difficulty in selecting regularization parameters in the RALS algorithm can be solved, and then the problems of overfitting or slow convergence in target perception caused by improper selection of regularization parameters of the RALS algorithm can be avoided.

[0243] Optionally, when the processor 610 separates the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, it is specifically configured to:

[0244] Acquiring communication information carried by the perception signal;

[0245] According to the communication information, the communication information is stripped from the echo signal to obtain a third-order tensor corresponding to the echo signal;

[0246] Based on the RALS algorithm using the regularization parameter value, the third-order tensor is decomposed to obtain the at least one factor array.

[0247] Optionally, when obtaining the perception result information corresponding to the target object according to the at least one factor array, the processor 610 is specifically configured to:

[0248] Perform parameter estimation on each of the factor arrays to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path;

[0249] Acquire environmental information around the sensing node, and construct at least one virtual anchor point based on the environmental information;

[0250] Matching the at least one virtual anchor point with an element in the parameter set to obtain parameter information corresponding to each virtual anchor point, wherein the parameter information includes at least one of angle information, delay information, and Doppler information;

[0251] The perception result information corresponding to the target object is determined according to the parameter information corresponding to the virtual anchor point.

[0252] Optionally, the perception result information includes position information, and when the processor 610 determines the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point, it is specifically configured to:

[0253] Determine angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0254] Based on the positioning principle of time difference of arrival TDOA or angle of arrival AOA, a first nonlinear equation group is constructed according to the angle information and delay information between the virtual anchor point and the target object;

[0255] Converting the first nonlinear equation group into a first expression corresponding to a maximum likelihood estimation problem;

[0256] The first expression is solved by convex optimization to obtain the position information of the target object.

[0257] Optionally, the perception result information includes speed information, and when the processor 610 determines the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point, it is specifically configured to:

[0258] Determine angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node;

[0259] Determine a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object;

[0260] Converting the Doppler expression into a second expression corresponding to the maximum likelihood estimation problem;

[0261] The second expression is solved by convex optimization to obtain the velocity information of the target object.

[0262] It should be noted here that the above-mentioned perception node provided by the embodiment of the present invention can implement all the method steps implemented by the above-mentioned target perception method embodiment applied to the perception node, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0263] A sensing node according to another embodiment of the present invention, such as Figure 7As shown, it includes a transceiver 710, a processor 700, a memory 720, and a program or instruction stored in the memory 720 and executable on the processor 700; when the processor 700 executes the program or instruction, the target perception method applied to the perception node is implemented.

[0264] The transceiver 710 is used to receive and send data under the control of the processor 700.

[0265] Among them, Figure 7 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically linking together various circuits of one or more processors represented by processor 700 and memory represented by memory 720. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 710 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium. The processor 700 is responsible for managing the bus architecture and general processing, and the memory 720 may store data used by the processor 700 when performing operations.

[0266] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0267] A computer-readable storage medium according to an embodiment of the present invention stores a program or instruction thereon, and when the program or instruction is executed by a processor, the steps in the target perception method described above are implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. The computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0268] It should be further explained that the terminals described in this specification include but are not limited to smart phones, tablet computers, etc., and many of the functional components described are called modules in order to more particularly emphasize the independence of their implementation methods.

[0269] In the embodiment of the present invention, module can be implemented with software so that it can be executed by various types of processors. For example, an executable code module of an identification can include one or more physical or logical blocks of computer instructions, for example, it can be constructed as an object, process or function. Nevertheless, the executable code of the identified module does not need to be physically located together, but can include different instructions stored in different positions, and when these instructions are logically combined together, it constitutes a module and realizes the specified purpose of the module.

[0270] In fact, executable code module can be a single instruction or many instructions, and can even be distributed on a plurality of different code segments, distributed among different programs, and distributed across a plurality of memory devices. Similarly, operating data can be identified in the module, and can be implemented and organized in the data structure of any appropriate type according to any appropriate form. The operating data can be collected as a single data set, or can be distributed in different locations (including on different storage devices), and can only be present on a system or network as an electronic signal at least in part.

[0271] When a module can be implemented by software, considering the level of existing hardware technology, a person skilled in the art can build a corresponding hardware circuit to implement the corresponding function of the module that can be implemented by software without considering the cost. The hardware circuit includes a conventional very large scale integration (VLSI) circuit or gate array and existing semiconductors such as logic chips, transistors, or other discrete components. The module can also be implemented by a programmable hardware device, such as a field programmable gate array, a programmable array logic, a programmable logic device, etc.

[0272] The above exemplary embodiments are described with reference to the accompanying drawings, and many different forms and embodiments are feasible without departing from the spirit and teachings of the present invention. Therefore, the present invention should not be constructed as a limitation of the exemplary embodiments proposed herein. More specifically, these exemplary embodiments are provided so that the present invention will be perfect and complete, and the scope of the present invention will be conveyed to those who are familiar with the technology. In these figures, the component sizes and relative sizes may be exaggerated for clarity. The terms used here are only based on the purpose of describing specific exemplary embodiments and are not intended to be limiting. As used herein, unless the text clearly indicates otherwise, the singular forms "one", "an" and "the" are intended to include these multiple forms. It will be further understood that the terms "including" and / or "comprising" when used in this specification indicate the presence of the features, integers, steps, operations, components and / or components, but do not exclude the presence or increase of one or more other features, integers, steps, operations, components, components and / or their groups. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of that range and any subranges therebetween.

[0273] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A target perception method, characterized in that: Applied to perception nodes, including: Sending perception signals; Receive an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one indirect path signal, the direct path signal is a signal that the sensing signal directly reflects from the target object to the sensing node, and the indirect path signal is a signal that the sensing signal reaches the target object through a reflective object and is reflected from the target object to the sensing node; Based on machine learning, determine the regularization parameter value of the random average RALS algorithm; Separating the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix; According to the at least one factor array, perception result information corresponding to the target object is obtained.

2. The method according to claim 1, characterized in that The step of separating the echo signal based on the RALS algorithm using the regularization parameter value to obtain at least one factor array includes: Acquiring communication information carried by the perception signal; According to the communication information, the communication information is stripped from the echo signal to obtain a third-order tensor corresponding to the echo signal; Based on the RALS algorithm using the regularization parameter value, the third-order tensor is decomposed to obtain the at least one factor array.

3. The method according to claim 1, characterized in that The obtaining, according to the at least one factor array, the perception result information corresponding to the target object includes: Perform parameter estimation on each of the factor arrays to obtain a parameter set corresponding to each of the factor arrays, wherein the parameter set includes at least one item of an angle set, a delay set, and a Doppler set, and elements with the same index value in the angle set, the delay set, and the Doppler set correspond to the same signal propagation path; Acquire environmental information around the sensing node, and construct at least one virtual anchor point based on the environmental information; Matching the at least one virtual anchor point with an element in the parameter set to obtain parameter information corresponding to each virtual anchor point, wherein the parameter information includes at least one of angle information, delay information, and Doppler information; The perception result information corresponding to the target object is determined according to the parameter information corresponding to the virtual anchor point.

4. The method according to claim 3, characterized in that The perception result information includes position information, and determining the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point includes: Determine angle information and delay information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node; Based on the positioning principle of time difference of arrival TDOA or angle of arrival AOA, a first nonlinear equation group is constructed according to the angle information and delay information between the virtual anchor point and the target object; Converting the first nonlinear equation group into a first expression corresponding to a maximum likelihood estimation problem; The first expression is solved by convex optimization to obtain the position information of the target object.

5. The method according to claim 3, characterized in that: The perception result information includes speed information, and determining the perception result information corresponding to the target object according to the parameter information corresponding to the virtual anchor point includes: Determine angle information and Doppler information between the virtual anchor point and the target object according to parameter information corresponding to the virtual anchor point and a geometric relationship between the virtual anchor point, the target object and the sensing node; Determine a Doppler expression between the virtual anchor point and the target object according to the angle information and Doppler information between the virtual anchor point and the target object; Converting the Doppler expression into a second expression corresponding to the maximum likelihood estimation problem; The second expression is solved by convex optimization to obtain the velocity information of the target object.

6. A target sensing device, characterized in that: Applied to perception nodes, including: A signal sending module, used for sending a sensing signal; A signal receiving module, configured to receive an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, wherein the direct path signal is a signal of the sensing signal directly reflected from the target object to the sensing node, and the non-direct path signal is a signal of the sensing signal reaching the target object through a reflective object and reflected from the target object to the sensing node; A first processing module is used to determine a regularization parameter value of a random average RALS algorithm based on machine learning; A second processing module, configured to separate the echo signal based on a RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix; The third processing module is used to obtain the perception result information corresponding to the target object according to the at least one factor array.

7. A sensing device, characterized in that: Applied to a sensing node, the sensing device comprises: a transceiver and a processor; The transceiver is used to send a sensing signal; The transceiver is further used to receive an echo signal of the sensing signal reflected by a target object, wherein the echo signal includes: a direct path signal and / or at least one non-direct path signal, the direct path signal is a signal of the sensing signal directly reflected from the target object to the sensing node, and the non-direct path signal is a signal of the sensing signal reaching the target object through a reflective object and reflected from the target object to the sensing node; The processor is used to determine a regularization parameter value of a random average RALS algorithm based on machine learning; The processor is further configured to separate the echo signals based on a RALS algorithm using the regularization parameter value to obtain at least one factor array, wherein the factor array includes at least one item of an angle factor matrix, a delay factor matrix, and a Doppler factor matrix; The processor is further configured to obtain perception result information corresponding to the target object according to the at least one factor array.

8. A sensing device, comprising: A transceiver, a processor, a memory, and a program or instruction stored in the memory and executable on the processor; characterized in that when the processor executes the program or instruction, the target perception method as described in any one of claims 1 to 5 is implemented.

9. A computer program product, characterized in that It comprises computer instructions, which, when executed by a processor, implement the steps of the target perception method as described in any one of claims 1 to 5.

10. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by the processor, the steps in the target perception method as described in any one of claims 1-5 are implemented.

Images