CN119835124A

CN119835124A - DOA estimation method based on multi-time joint perception signal enhancement

Info

Publication number: CN119835124A
Application number: CN202510013639.1A
Authority: CN
Inventors: 幸锋; 李春国; 曹硕; 许铭诚; 徐澍
Original assignee: Southeast University
Current assignee: Southeast University
Priority date: 2025-01-06
Filing date: 2025-01-06
Publication date: 2025-04-15

Abstract

The invention discloses a DOA estimation method based on multi-time joint perception signal enhancement, which comprises the following steps of constructing a perception channel model for a moving target based on a water surface scene, optimizing a time-varying channel estimation result through multi-time joint denoising aiming at the problem of improving DOA estimation accuracy of a water surface object, denoising two dimensions of a space domain and an angle domain for the time-varying channel estimation by adopting a Double U-Net based deep learning algorithm, and performing DOA estimation by adopting a MUSIC algorithm by utilizing a denoised channel matrix. According to the invention, the Double U-Net algorithm based on the deep convolutional neural network is adopted to extract complex characteristics of the channel from the multi-time combined perception observation data, so that the denoising and enhancement of the channel are realized, and furthermore, the DOA estimation precision is improved, and thus, the accurate identification and tracking of the water surface object are effectively realized. Compared with the traditional method, the DOA estimation method provided by the invention has higher estimation precision.

Description

DOA estimation method based on multi-time joint perception signal enhancement

Technical Field

Ext> theext> inventionext> belongsext> toext> theext> technicalext> fieldext> ofext> DOAext> estimationext> inext> aext> 5ext> Gext> -ext> Aext> /ext> 6ext> Gext> senseext> -ext> ofext> -ext> generalext> intelligentext> computationext> integratedext> sceneext>,ext> andext> particularlyext> relatesext> toext> aext> DOAext> estimationext> methodext> basedext> onext> multiext> -ext> timeext> jointext> perceptionext> signalext> enhancementext>.ext>

Background

Ext> theext> fifthext> generationext> mobileext> communicationext> (ext> 5ext> Gext> -ext> aext>)ext> andext> theext> sixthext> generationext> mobileext> communicationext> (ext> 6ext> Gext>)ext> areext> conceivedext> asext> multipurposeext> systemsext> capableext> ofext> providingext> aext> communicationext> -ext> awareext> intelligentext> integratedext> serviceext> toext> aext> userext>,ext> oneext> ofext> theext> keyext> technologiesext> forext> providingext> thisext> serviceext> beingext> directionext> ofext> arrivalext> (ext> Directionext> ofext> Arrivalext>,ext> doaext>)ext> estimationext>.ext> The DOA estimation can help the system determine the direction of the signal source, thereby optimizing the reception and processing of the signal and improving the communication quality and the sensing performance. In this context, the importance of direction of arrival estimation is becoming increasingly prominent as one of the key technologies. Because DOA estimation can effectively extract the position information of a target, accurate identification and tracking of a water surface object can be realized. However, the complexity of the surface environment presents significant challenges for signal processing. The accuracy of DOA estimation is significantly affected by the specific electromagnetic properties of the water surface, the influence of fluctuations, reflections, scattering, multipath effects encountered during signal propagation, and the like.

The existing DOA estimation method, such as a conventional beam forming (Conventional Beamforming, CBF) method, is essentially a simple expansion of a time domain Fourier spectrum estimation method to a space domain, has a simple principle, but has limited resolving power and can be limited by the Rayleigh limit of an array. Multiple signal classification (Multiple Signal Classification, MUSIC) methods and rotation invariant subspace (Estimation ofSignal PARAMETERS VIA Rotational Invariance Technique, ESPRIT) methods, both of which utilize the properties of the signal subspace to achieve super-resolution direction finding. Both methods are covariance-based and require a sufficient number of data samples (snapshots) to accurately estimate the true covariance matrix. With the development of artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) in recent years, many DOA estimation methods based on deep learning have also emerged, such as using a denoising self-encoder (DenoisingAutoencoder, DAE) to learn the mapping from sample covariance to true covariance matrix, and then to make DOA estimation.

In order to improve the accuracy of DOA estimation and aiming at the identification and positioning of an object on the water surface, the DOA estimation is particularly performed under the general sense integrated technology framework. A channel model suitable for a water surface environment needs to be constructed, and special electromagnetic characteristics of the water surface, including the influence of fluctuation on signals and the influence of multipath effects on signal quality, are fully considered, so that DOA estimation performance is optimized based on the influence. The accurate channel modeling can provide reliable channel state information for DOA estimation, so that the accuracy and the signal-to-noise ratio of target identification are improved, and the cooperative work of communication and perception is optimized.

Disclosure of Invention

In the invention, from the perspective of improving DOA estimation precision, aiming at the movement characteristics of a water surface perception detection target, a channel estimation method based on multi-time joint perception signal enhancement is provided, and DOA estimation under a communication perception Integrated (Integrated SensingandCommunications, ISAC) system is disassembled into two processes of multi-time joint channel denoising and MUSIC spectrum estimation, wherein the performance of the two processes is superior to that of the traditional algorithm;

in order to achieve the purpose, the DOA estimation method based on multi-time joint perception signal enhancement is characterized by comprising the following steps of:

Step 1, determining path loss formulas under different scenes of a water surface and suburbs, and constructing a time-varying perception channel model for a moving target;

Step 2, based on the time-varying perception channel model in step 1, obtaining a noiseless channel H _gt (t), and preprocessing the noiseless channel H _gt (t) by least square to obtain an LS estimation result And sampling the values of the T moments;

step3 based on the noiseless channel in step2 And LS estimation resultsOffline training is carried out on the Double U-Net network;

Step 4, obtaining a new real channel to be estimated on line, carrying out least square processing on the real channel, sampling results at T moments, inputting the results into a Double U-Net network after offline training to carry out joint denoising, and obtaining a denoised channel matrix

Step 5, denoising the channel matrixAnd carrying out DOA estimation by adopting a joint MUSIC algorithm.

Further, the specific steps of the step 1 include:

Step 11, dividing the path loss into a line-of-sight type and a non-line-of-sight type, and introducing a roughness factor C for a water surface scene, wherein the calculation formula of the path loss of the water surface is as follows

Where n _OS denotes the path loss factor on the water surface, I ₀ denotes an indication function of the reflection path, Γ ₀ is the reflection coefficient,For the phase difference between the direct and reflected paths, I ₁ denotes the indication function of the scattering path, Γ ₁ is the scattering coefficient,Λ represents a wavelength, and j is an imaginary unit, which is a phase difference between a direct path and a scattering path;

step 12, based on the sparsity of the sensing channel, constructing a time-varying sensing channel model by using a classical SV channel model and combining the movement characteristics of the moving target

Where t is the time variable, γ is the scalar coefficient,The arrival angle and the departure angle of the first path at time t are respectively shown, alpha _l (t) is the attenuation of the first multipath,A steering vector indicating the angle of arrival of the first path at time t,A guide vector indicating the departure angle of the first path at time t.

Further, the expression of the roughness factor C in the step 11 is

Wherein σ _r represents roughness, ψ is a ground contact angle, which is the complementary angle of the arrival angle θ, and λ represents wavelength.

Further, the specific steps of the step2 include:

Step 21, generating an original uncontaminated data set by using a time-varying perception channel model, and marking the data set as a real channel matrix H _gt (t);

Step 22, preprocessing the real channel matrix H _gt (t) by LS method to obtain initial rough estimation value

Step 23 for the real channel matrix H _gt (t) and initial coarse estimateRespectively sampling T times to obtain a noiseless channelAnd LS estimation results

Further, the specific steps of the step 3 include:

Step 31, initializing training parameters, wherein the total training round number is E, the learning rate is eta, the side length dimension of a convolution kernel is k, the training batch size B, the weight coefficient alpha, the training layer number L and the parameters theta in a randomized initial vector Double U-Net network at T moments;

step 32, channel without noise LS estimation results as training tagsAs a data set to be trained;

step 33, LS estimation result Performing DFT conversion to obtain an angle domain result, and inputting the angle domain result into a Double U-Net network to obtain an angle domain noise estimated value and a space domain noise estimated value;

Step 34, combining LS estimation result, angle domain noise estimation value and space domain noise estimation value And (5) carrying out iteration for a plurality of times until the loss function value is minimum, and obtaining the trained Double U-Net network.

Further, the channel matrix after Double U-Net denoisingThe expression of (2) is

Wherein, Representing the spatial domain noise estimate value,The angle domain noise estimate is shown, and the estimate uses Θ= { W, b } as a parameter, including a weight W and an offset b.

Further, the loss function is

Wherein, |D _t || is the number of samples involved.

Further, the step 4 specifically includes:

step 41, obtaining a new real noise-free channel with estimation on line, carrying out LS estimation on the channel and sampling T moments;

Step 42, inputting the result and the DFT converted result into a trained Double U-Net network to obtain a denoised channel matrix

Further, the step 5 specifically includes:

step 51, according to the obtained channel matrix after denoising at T moments The covariance matrix was calculated by dividing into T groups at time, denoted H ^t (t=1, l, T), respectively:

R^t＝H^t(H^t)^H;

step 52, decomposing eigenvalue of covariance matrix into signal subspace and noise subspace:

wherein, Is the sub-space of the signal,Is a sub-space of the noise,AndRespectively corresponding characteristic value matrixes;

Step 53 noise subspace based For the reception direction θ _r (t), the MUSIC spectral function is defined as

And estimating an arrival angle θ _r (t) by searching for a peak of the spectral function P _MUSIC(θ_r (t);

And 54, detecting and estimating an arrival angle theta _r (t) by adopting the number of missed detection as a performance index.

Further, the step 54 specifically includes the steps of:

if L detection targets exist in the current t-th moment group, carrying out angle estimation on each target to obtain an estimated angle With true angleComparing to obtain an average angle error of

If it isThe set of data is considered to have a missed detection condition.

The beneficial effects are that:

1. Compared with DOA estimated scenes, the method uses the step of step 1, and considers a sensing channel wireless model based on a water surface movement detection target, which is different from the traditional communication sensing scene;

2. Compared with the traditional DOA estimation method, the method uses the steps of steps 2-5, considers the influence of low Signal-to-Noise Ratio (SNR) on DOA estimation in a water surface scene, and breaks down the DOA estimation problem in a time-varying perception channel into two stages of multi-time joint channel denoising and multi-time joint DOA estimation based on a Double U-Net algorithm;

3. Compared with the traditional method, the method has the advantages that the steps of steps 3-5 are adopted, the channel denoising ratio based on the Double U-Net algorithm is higher in accuracy than that of the traditional LS, denoising convolutional neural network (Denoising Convolutional neural network, dnCNN) algorithm and single-moment Double U-Net, and meanwhile DOA estimation accuracy based on the method is improved.

Drawings

FIG. 1 is a schematic flow chart of a DOA estimation method based on multi-time joint perceptual signal enhancement provided in an embodiment of the present invention;

FIG. 2 is a schematic diagram of a surface sensing system provided in an embodiment of the present invention;

FIG. 3 is a schematic diagram of a Double U-Net network architecture provided in an embodiment of the present invention;

Fig. 4 is a schematic diagram of NMSE results of different methods of channel estimation as a function of SNR according to an embodiment of the present invention;

fig. 5 is a schematic diagram of the number of missed samples of the DOA estimation according to SNR according to different methods when the number of paths is 5 in the embodiment of the present invention.

Detailed Description

The invention is further explained below with reference to the drawings.

As shown in fig. 1, the present invention provides a DOA estimation method based on multi-time joint perceptual signal enhancement, comprising:

And step 1, determining a path loss formula under a water surface scene, and constructing a time-varying perception channel model for a moving target. Step2, based on the time-varying perception channel model in step 1, obtaining a noiseless channel H _gt (t), and preprocessing the noiseless channel H _gt (t) by least square to obtain an LS estimation result And samples the values at T instants.

Step3 based on the noiseless channel in step2And LS estimation resultsOffline training is performed on the Double U-Net network.

The step1 specifically comprises the following steps:

step 11, dividing the path loss into two types of sight distance and non-sight distance, introducing a coarse factor C for a water surface scene to obtain a water surface path loss calculation formula of

Where n _OS denotes the path loss factor on the water surface, I ₀ denotes an indication function of the reflection path, Γ ₀ is the reflection coefficient,For the phase difference between the direct and reflected paths, I ₁ denotes the indication function of the scattering path, Γ ₁ is the scattering coefficient,In order to obtain a phase difference between the direct path and the scattered path, λ represents a wavelength, and j is an imaginary unit.

The expression of the roughness factor C in step 11 is

In step 12, based on the ULA model, the steering vector a _Ω (θ (t)) can be expressed as

Where N _Ω is the number of antenna arrays, d and λ represent the spacing between adjacent antennas and the signal wavelength, respectively, and thus can be calculated by the formula of the steering vector a _Ω (θ (t))And

As shown in fig. 2, the present embodiment is located in a water surface scene, which has 5 detection targets in total, the base station is disposed at the position (0,0,20), and there are 5 data points of the detection targets in a rectangular area formed by four points (-200,50,0), (-200,250,0), (200,50,0), and (200,250,0). The multi-target sensing of the object on the water surface is divided into active sensing and passive sensing according to whether one of the receiving signals knows the signal sent by the sender in advance, and when the active sensing and the receiving and sending sides are the same antenna, the parameters are set to be h=20m, the number of the antennas is N _t、N_r =64, the carrier frequency is f=2.7GHz, the distance between the water surface and the base station is R=10m, the length and the width of the water surface are W=500m, and the departure angle and the arrival angle of the antenna are theta. The object is perceived by the echo signals reflected by the object. For a moving object, the object has an initial velocity v (m/s), and accordingly, the channel transfer matrix H at this time also changes in real time, that is, becomes a function H (t) of time.

The specific steps of the step 2 include:

And 21, under the condition of considering the moving scene of the water surface detection target, generating an original uncontaminated data set by using a time-varying perception channel model, and marking the data set as a real channel matrix H _gt (t).

Step 22, considering the low complexity of LS estimation, LS estimation is performed on the real channel matrix H _gt (t) to obtain an initial rough estimation value

Step 23 joint denoising with time correlation due to time-variant of perceived channel, in particular for real channel matrix H _gt (t) and initial rough estimateRespectively sampling T times to obtain a noiseless channelAnd LS estimation results

The specific steps of the step 3 include:

Step 31, initializing training parameters, wherein the total training round number is E, the learning rate is eta, the side length dimension of a convolution kernel is k, the training batch size B, the weight coefficient alpha, the training layer number L and the parameters theta in the randomized initial vector Double U-Net network at T moments.

Step 32, channel without noiseLS estimation results as training tagsAs a data set to be trained, step 33, defining a loss function as:

Wherein, |D _t || is the number of samples involved.

Step 34, LS estimation resultDFT conversion is carried out to obtain an angle domain result, and the angle domain result is input into a Double U-Net network to obtain an angle domain noise estimated value and a space domain noise estimated value.

Step 35, combining LS estimation result, angle domain noise estimation value and space domain noise estimation valueAnd (5) carrying out iteration for a plurality of times until the loss function value is minimum, and obtaining the trained Double U-Net network.

In this embodiment, the total training round number is e=100, the learning rate is η=10 ^-2, the side length dimension of the convolution kernel is k=3, the training lot size b=128, the weight coefficient α=0.5, the training layer number l=9, and the objective of the double U-Net algorithm is to minimize the estimated denoised channel matrixAnd noiseless channelsNMSE values in between, thus defining the above-mentioned loss function.

As shown in fig. 3, the Double U-Net model of the present embodiment adopts a dual-path encoder-decoder architecture, and LS channel estimation values of T time sampling points are simultaneously introduced as multiple channels, so as to implement channel denoising through joint processing. One path processes the angular domain data passing through the DFT, the other path directly processes the original airspace data, the two paths are processed in parallel, the characteristics of the angular domain and the airspace are extracted respectively, and finally, the comprehensive suppression of noise is realized through the combination of the results, and then the specific output of the Double U-Net algorithm can be expressed as follows:

wherein, AndRespectively expressed in spatial domain and angular domain, LS estimation valueNoise estimation via U-Net network, i.eRepresenting the spatial domain noise estimate value,The angle domain noise estimate is shown, and the estimate uses Θ= { W, b } as a parameter, including a weight W and an offset b.

The specific steps of the step 4 include:

The SNR (dB) values are chosen to be-10, -5,0,5,10 respectively, and NMSE values are calculated for LS estimation results, dnCNN, double U-Net and multi-time joint estimation results, and the results are shown in figure 4.

The specific steps of the step 5 include:

R^t＝H^t(H^t)^H.

wherein, Is the sub-space of the signal,Is a sub-space of the noise,AndRespectively, their corresponding eigenvalue matrices.

Step 53 noise subspace basedFor the reception direction θ _r (t), the MUSIC spectral function is defined as

And estimates the arrival angle θ _r (t) by searching for the peak of the spectral function P _MUSIC(θ_r (t)).

The specific steps of adopting the missing detection number as the performance index in the step 54 include:

If it isThe set of data is considered to have a missed detection condition. The DOA estimation result can be detected.

The SNR (dB) value was chosen to be-10, -5,0,5,10. And carrying out DOA estimation on the LS estimation result, dnCNN and Double U-Net result under the single moment condition, and the LS estimation result and DoubleU-Net estimation result which are combined at multiple moments. A total of 2500 tests were performed and the missing test condition is shown in figure 5.

The invention compares the multi-time joint estimation based on DoubleU-Net algorithm with other traditional algorithms DnCNN and LS algorithm, and the estimation performance of the method without multi-time joint under different SNR. Compared with the Double U-Net algorithm and other algorithms, the DOA estimation method provided by the embodiment has better performance under the condition of the same SNR, which proves that the DOA estimation method provided by the embodiment effectively solves the problems of perceived positioning and tracking under the condition of a water surface moving target, and improves the DOA estimation accuracy.

The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.

Claims

1. A DOA estimation method based on multi-time joint perception signal enhancement, characterized by comprising:

Step 1: Determine the path loss formula for different scenarios on the water surface and in the suburbs, and build a time-varying perception channel model for the mobile target;

Step 2: Based on the time-varying perceptual channel model in step 1, obtain the noise-free channel H _gt (t), and preprocess it using least squares to obtain the LS estimation result And sample the values at T moments;

Step 3: Based on the noiseless channel in step 2 And the LS estimation results Offline training of the Double U-Net network;

Step 4: Obtain the new real channel to be estimated online, perform least squares processing on it and sample the results at T moments and input them into the Double U-Net network trained offline for joint denoising to obtain the denoised channel matrix

Step 5: Denoised channel matrix The joint MUSIC algorithm is used for DOA estimation.

2. According to claim 1, the DOA estimation method based on multi-time joint perception signal enhancement is characterized in that the specific steps of step 1 include:

Step 11: Path loss is divided into line-of-sight and non-line-of-sight. For water surface scenes, the roughness factor C is introduced, and the calculation formula for the path loss on the water surface is:

Where n _OS represents the path loss factor on the water surface, I ₀ represents the indicator function of the reflection path, Γ ₀ is the reflection coefficient, is the phase difference between the direct path and the reflected path, I ₁ represents the indicator function of the scattered path, Γ ₁ is the scattering coefficient, is the phase difference between the direct path and the scattered path, λ represents the wavelength, and j is the imaginary unit;

Step 12: Based on the sparsity of the perceptual channel, the classic SV channel model is used, and combined with the mobility characteristics of the mobile target, a time-varying perceptual channel model is constructed as

Where t is the time variable, γ is the scalar coefficient, denote the arrival angle and departure angle of the lth path at time t, α _l (t) is the attenuation of the lth multipath, The steering vector representing the arrival angle of the lth path at time t, The steering vector representing the departure angle of the lth path at time t.

3. The DOA estimation method based on multi-time joint perception signal enhancement according to claim 2 is characterized in that the expression of the roughness factor C in step 11 is

Among them, _σr represents the roughness, ψ is the ground contact angle, which is the complementary angle of the arrival angle θ, and λ represents the wavelength.

4. According to claim 1, the DOA estimation method based on multi-time joint perception signal enhancement is characterized in that step 2 specifically comprises:

Step 21: Generate the original uncontaminated data set using the time-varying perceptual channel model, denoted as the real channel matrix H _gt (t);

Step 22: Use the LS method to preprocess the real channel matrix H _gt (t) to obtain an initial rough estimate

Step 23: For the real channel matrix H _gt (t) and the initial rough estimate Sample T moments separately to obtain a noise-free channel And the LS estimation results

5. The DOA estimation method based on multi-time joint perception signal enhancement according to claim 1, characterized in that the specific steps of step 3 include:

Step 31: Initialize the training parameters, where the total number of training rounds is E, the learning rate is η, the side length of the convolution kernel is k, the training batch size is B, the weight coefficient is α, the number of training layers is L, and the random initial vector at T moments is the parameter Θ in the Double U-Net network;

Step 32: Noise-free channel As training labels, LS estimation results As the data set to be trained;

Step 33: LS estimation results Perform DFT transformation to obtain the angle domain result, input it into the Double U-Net network, and obtain the angle domain noise estimation value and the spatial domain noise estimation value;

Step 34: Combine the LS estimation results, the angle domain noise estimation value, and the spatial domain noise estimation value After multiple iterations until the loss function value is minimized, the trained Double U-Net network is obtained.

6. The DOA estimation method based on multi-time joint perception signal enhancement according to claim 5, characterized in that the channel matrix after Double U-Net denoising The expression is

in, represents the spatial noise estimate, Θ = {W, b} is used as a parameter, including the weight W and the offset b.

7. According to claim 5, the DOA estimation method based on multi-time joint perception signal enhancement is characterized in that the loss function is

Among them, ||D _t || is the number of samples included.

8. The DOA estimation method based on multi-time joint perception signal enhancement according to claim 1, wherein step 4 specifically comprises:

Step 41: obtain a new real noise-free channel with estimation online, perform LS estimation on it and sample T time instants;

Step 42: Input the above results and the results after DFT transformation into the trained Double U-Net network to obtain the denoised channel matrix

9. The DOA estimation method based on multi-time joint perception signal enhancement according to claim 1, wherein step 5 specifically comprises:

Step 51: Obtain the denoised channel matrix at T time points Divide into T groups according to time, denoted as H ^t (t＝1,L,T), and calculate the covariance matrix respectively:

R ^t =H ^t (H ^t ) ^H ;

Step 52: Perform eigenvalue decomposition on the covariance matrix to decompose it into a signal subspace and a noise subspace:

in, is the signal subspace, is the noise subspace, and are their corresponding eigenvalue matrices respectively;

Step 53: Noise Subspace Based For the receiving direction θ _r (t), the MUSIC spectrum function is defined as

And by searching for the peak value of the spectrum function P _MUSIC (θ _r (t)), the arrival angle θ _r (t) is estimated;

Step 54: Use the number of missed detections as a performance indicator to detect and estimate the arrival angle θ _r (t).

10. The DOA estimation method based on multi-time joint perception signal enhancement according to claim 9, characterized in that the specific steps of step 54 are:

If there are L detection targets in the group at the current t-th moment, estimate the angle of each target and get the estimated angle With real angle By comparison, the average angle error is

like It is considered that there is a missed detection in this group of data.