Computational imaging method based on cell-free large-scale multi-input multi-output system

Abstract

The present invention discloses a computational imaging method based on a non-cellular large-scale multi-input multi-output system. In the non-cellular large-scale multi-input multi-output system, multiple access points are connected to a central processing unit by wireless or wired means, and target imaging is achieved by utilizing wireless signals from multiple devices. In order to restore the target of interest in the environment, multiple user devices simultaneously transmit perception signals, which reach each access point after being scattered by the target. The access point calculates the signals received by each access point to obtain a local preliminary image of the object. The access point then transmits the preliminary image data to the central processing unit for fusion processing, thereby achieving accurate imaging of the target. The present invention provides a simple and effective computational imaging method for accurate imaging of targets of interest in complex environments.

Description

Technical Field

[0001] The present invention relates to the field of wireless network perception, and in particular to a computational imaging method based on a non-cellular large-scale multi-input multi-output system. Background Art

[0002] The future 6G system will have higher frequency bands, larger bandwidths, and denser distribution of large-scale antenna arrays. Therefore, a single system will integrate more functions, such as wireless communication, perception, and positioning, which will greatly reduce the cost of additional perception equipment. In addition, the communication collaboration between widely deployed access points and user devices can be used to improve perception performance. Based on the software and hardware research on the coexistence and cooperation issues of current communication systems and radar systems, it is feasible to use radio signals for perception. The entire communication system can be used as a sensor, using the transmission, reflection, and scattering of radio waves to better understand the physical world and provide more services. And because of the unique advantages of radio perception - the ability to overcome factors such as weather and light, all-weather perception can be achieved.

[0003] Cell-free massive MIMO technology is one of the core technologies of future wireless networks, with the characteristics of ultra-high reliability, data throughput, energy efficiency and uniform coverage. Compared with current cellular systems, there are no cells or cell boundaries in cell-free networks. Therefore, this approach can alleviate a major bottleneck and inherent limitation of wireless networks, namely strong inter-cell interference. As a typical distributed system, a large number of distributed access points in a cell-free massive MIMO system are connected to a central processing unit to serve a smaller number of users on the same time-frequency resources. The performance of the system depends on efficient distributed processing algorithms.

[0004] At present, radar imaging and optical imaging have achieved long-term and mature development, but there are few studies on imaging in the field of communications. In the vast field of radar sensing or computational imaging, the key to effective detection is to use the intrinsic sparsity of objects or scatterers in the environment. This type of problem is often modeled as a sparse signal recovery problem based on pixel partitioning according to the theory of compressed sensing, and solved by widely used methods such as sparse Bayesian learning, orthogonal matching pursuit, and approximate message passing. However, most of them only consider single access point computational imaging. A single access point can restore the target object of interest in the environment, but due to factors such as channel noise pollution and incomplete received signals, the imaging effect is often not ideal. Through the non-cellular large-scale multi-input multi-output system architecture, it is possible to integrate the estimation information of multiple access points, fuse them to obtain accurate estimates, and the restored image effect is better than that of a single access point. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of difficult computational imaging of communication networks, poor imaging effect of single access points, low computational efficiency, etc. in the above-mentioned scheme, and proposes a computational imaging method for a non-cellular large-scale multi-input multi-output system.

[0006] The specific technical solutions adopted by the present invention are as follows:

[0007] A computational imaging method based on a non-cellular large-scale multi-input multi-output system comprises the following steps:

[0008] S1. The entire environment space is treated as a three-dimensional space and discretized to obtain N = P*P*P small cubes, each of which represents a pixel point. The target object of interest exists in the environment space. The target object consists of K pixels and is sparse, with a sparseness rate of λ = K / N. The scattering coefficient x is used n represents the scattering coefficient of the nth pixel point; any pixel point in the entire environment space has two possibilities: belonging to a non-target object or a target object; the scattering coefficient of the entire environment space is expressed as x = [x1, x2, ..., x N ] T , each element in x is independent of each other and obeys the Bernoulli-Gaussian distribution, and its probability function is expressed as Where λ is the sparsity rate, δ(x) is the Dirac function at x = 0, It means that any pixel x obeys the Gaussian distribution probability density function with mean θ and variance φ, and q = (λ, θ, φ) is the set of prior parameters;

[0009] S2. In a non-cellular large-scale MIMO system, a central processing unit is deployed, B access points are configured with M antennas, and D single-antenna user devices are deployed; the perception signal sent by the dth user device is s d , the transmission power is d=1,2,...,D,where ||s d || means to find s d The model, Expressing hope;

[0010] S3, the bth access point receives the signal y b , using the single access point computational imaging method, we can get the local preliminary estimated image and the variance matrix Σ b , b=1,2,...,B, and then send the estimation information of each of the B access points to the central processing unit;

[0011] S4, the central processing unit estimates the signal based on B access points and the variance matrix {Σ1,Σ2,…,Σ B}, using the multi-signal fusion method, the imaging signal of the target object is obtained

[0012] Based on the above technical solution, each step in the present invention can be further implemented in the following preferred manner.

[0013] Preferably, the specific implementation steps of the single access point computational imaging method described in step S3 are as follows:

[0014] a) The channel matrix from the dth user equipment to the target object is The channel matrix of the dth user equipment signal reflected from the target to the bth access point is: where α d and β b,d are large-scale fading coefficients based on path loss and shadow fading, H d and G b,d are all small-scale fading coefficients and obey the standard Gaussian distribution; then the signal model received by the b-th access point is expressed as The signal model expression is further simplified into the standard linear model y b =A b x b +w b =z b +w b , where the matrix x b is the target object estimated by the bth access point, w b is additive Gaussian white noise with a noise variance of ψ b ;

[0015] b) For any access point b, the received signal y in a) b The vector product of is converted to the product of all scalar elements, where x b =[x b,1 ,...,x b,n ,...,x b,N ] T , z b =[z b,1 ,...z b,m ,...,z b,M ] T ,y b =[y b,1 ,...y b,m ,...,y b,M ] T , then the posterior probability of the ambient scattering coefficient is:

[0016]

[0017] Where p(yb |x b ) is y b The likelihood function, p(x b |q) is x b The prior probability, p(y b,m |z b,m ) is y b,m The scalar likelihood function, p(x b,n |q) is x b,n The scalar prior probability of ;

[0018] c) For m=1,2,...,M, find z b,m About (y b ; The marginal posterior probability of q) Where p(y b,m |z b,m ) is y b,m The scalar likelihood function of Indicates z b,m The mean is The variance is Gaussian distribution probability density function; find the p(z b,m |y b ; q) and then calculate the expectation to get z b ; Then perform matrix inversion operation A b -1 *z b =r b , and get the target object r polluted by noise b , that is, r b =x b +v b , where v b is additive Gaussian white noise; finally, for n = 1, 2, ..., N, we can obtain x b,n About (y b ; The marginal posterior probability of q) Where p(x b,n |q) is x b,n The scalar prior probability of , For x b,n The mean is The variance is Gaussian distribution probability density function;

[0019] d) The prior probability p X (x b,n |q) Substitute into c) x b,n About (y b ; The marginal posterior probability formula of q) is obtained b,n The marginal posterior probability of is:

[0020]

[0021] Among them, δ(x b,n ) represents x b,n = 0, the Dirac function, ζ b,n is the normalization factor:

[0022]

[0023] parameter

[0024] e) Through the linear minimum mean square error estimation, the expectation of the marginal posterior probability in d) is obtained and variance μ b,n :

[0025]

[0026]

[0027] Thus, the target image signal estimated by the b-th access point is obtained and the diagonal variance matrix Σ b =diag[μ b,1 ,...,μ b,n ,...,μ b,N ];

[0028] f) Repeat steps c) to e) for iterative calculation until convergence, and output the final and Σ b .

[0029] Furthermore, in step f), after each iteration, the target image signal is determined. The calculation results of all N elements in the current iteration round The calculation result in the previous iteration is Whether the convergence condition is met ε is the set error accuracy. If it is not satisfied, the iteration continues; if it is satisfied, the iteration stops.

[0030] Furthermore, the error precision ε is preferably 10 -5 .

[0031] Preferably, the multi-signal fusion method in step S4 is specifically implemented as follows:

[0032] The central processing unit is based on the local target image signal uploaded by B access points and the variance matrix {Σ1,Σ2,…,Σ B}, weighted sum of the elements at the nth position to get the estimated value of that position n=1,2,...,N, where weight And to meet μ b,n for The corresponding variance; all N The final accurate estimation of the target imaging signal is formed

[0033] The invention has the following beneficial effects: the distributed computational imaging method based on a non-cellular large-scale multi-input multi-output system proposed in the invention utilizes the scattering of radio signals in the communication system in the environment, obtains the image of the target object through perception and calculation, and can solve the problems of difficult computational imaging in the communication field, poor imaging effect of a single access point, and low computational efficiency. The single access point computational imaging method and the central processing unit multi-signal fusion method proposed in the invention have low implementation complexity, reduce the overhead of additional sensing equipment, and better observe and understand the physical world as the cornerstone of 6G, and improve the performance of computational imaging in a non-cellular large-scale multi-input multi-output system. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a block diagram of a computational imaging method based on a cell-free large-scale multiple-input multiple-output system;

[0035] Figure 2 is the original target imaging image;

[0036] Figure 3 It is a single access point computational imaging map;

[0037] Figure 4 It is the central processing unit multi-signal fusion imaging image. DETAILED DESCRIPTION

[0038] The present invention will be further described and illustrated below in conjunction with the accompanying drawings and specific implementation methods.

[0039] In a preferred embodiment of the present invention, a block diagram of a computational imaging method based on a non-cellular large-scale multi-input multi-output system is shown in FIG. Figure 1 As shown in the figure, only the uplink mechanism is considered, that is, D single-antenna user devices simultaneously send the same sensing signal to B access points through a shared channel. The signal reaches the access point through direct radiation and scattering of the target of interest. Since the direct radiation part of the signal does not contain target information, only the scattered received signal is retained. The access point first needs to use a linear mean square error estimator to perform a rough estimate at the local end, which can restore the target object of interest in the environment. Subsequently, the posterior probability mean and variance calculated locally are synchronously transmitted to the central processing unit, and then fused on the central processing unit through weighted summation, and finally the environmental target vector is obtained.

[0040] The present embodiment adopts a computational imaging method based on a non-cellular large-scale multi-input multi-output system, which specifically includes the following steps:

[0041] S1. The entire environment space is treated as a three-dimensional space and discretized to obtain N = P*P*P small cubes, each of which represents a pixel point. The target object of interest exists in the environment space. The target object is composed of K (K << N) pixels and is sparse, with a sparse rate of λ = K / N. The scattering coefficient x is used n represents the scattering coefficient of the nth pixel point; any pixel point in the entire environment space has two possibilities: belonging to a non-target object or a target object; the scattering coefficient of the entire environment space is expressed as x = [x1, x2, ..., x N ] T , each element in x is independent of each other and obeys the Bernoulli-Gaussian distribution, and its probability function is expressed as Where λ is the sparsity rate, δ(x) is the Dirac function at x = 0, It means that any pixel x obeys the Gaussian distribution probability density function with mean θ and variance φ, and q = (λ, θ, φ) is the set of prior parameters.

[0042] S2. In a non-cellular large-scale MIMO system, a central processing unit is deployed, B access points are configured with M antennas, and D single-antenna user devices are deployed; the perception signal sent by the dth user device is s d , the transmission power is d=1,2,...,D,where ||s d || means to find s d The model, Expressing expectation.

[0043] S3, the bth access point receives the signal y b , using the single access point computational imaging method, we can get the local preliminary estimated image and the variance matrix Σ b , b=1,2,...,B, and then send the estimation information of each of the B access points to the central processing unit.

[0044] In this embodiment, the specific implementation steps of the single access point computational imaging method described in step S3 are as follows:

[0045] a) The channel matrix from the dth user equipment to the target object is The channel matrix of the dth user equipment signal reflected from the target to the bth access point is: where α d and β b,dare large-scale fading coefficients based on path loss and shadow fading, H d and G b,d are all small-scale fading coefficients and obey the standard Gaussian distribution; then the signal model received by the b-th access point is expressed as The signal model expression is further simplified into the standard linear model y b =A b x b +w b =z b +w b , where the matrix x b is the target object estimated by the bth access point, w b is additive Gaussian white noise with a noise variance of ψ b ;

[0046] b) For any access point b, the received signal y in a) b The vector product of is converted to the product of all scalar elements, where x b =[x b,1 ,...,x b,n ,...,x b,N ] T , z b =[z b,1 ,...z b,m ,...,z b,M ] T ,y b =[y b,1 ,...y b,m ,...,y b,M ] T , then the posterior probability of the ambient scattering coefficient is:

[0047]

[0048] Where p(y b |x b ) is y b The likelihood function, p(x b |q) is x b The prior probability, p(y b,m |z b,m ) is y b,m The scalar likelihood function, p(x b,n |q) is x b,n The scalar prior probability of , with the symbol ∝ meaning “proportional to ”.

[0049] c) For m=1,2,...,M, find z b,m About (y b ; The marginal posterior probability of q) The symbol It means "defined as", p(y b,m |z b,m ) is y b,m The scalar likelihood function of Indicates z b,m The mean is The variance is Gaussian distribution probability density function; find the p(z b,m |y b ; q) and then calculate the expectation to get z b ; Then perform matrix inversion operation A b -1 *z b =r b , and get the target object r polluted by noise b , that is, r b =x b +v b , where v b is additive Gaussian white noise; finally, for n = 1, 2, ..., N, we can obtain x b,n About (y b ; The marginal posterior probability of q) Where p(x b,n |q) is x b,n The scalar prior probability of , For x b,n The mean is The variance is Gaussian distribution probability density function;

[0050] d) The prior probability p X (x b,n |q) Substitute into c) x b,n About (y b ; The marginal posterior probability formula of q) is obtained b,n The marginal posterior probability of is:

[0051]

[0052] Among them, δ(x b,n ) represents x b,n = 0, the Dirac function, ζ b,n is the normalization factor:

[0053]

[0054] parameter

[0055] e) Through the linear minimum mean square error estimation, the expectation of the marginal posterior probability in d) is obtained and variance μ b,n :

[0056]

[0057]

[0058] Thus, the target image signal estimated by the b-th access point is obtained and the diagonal variance matrix Σ b =diag[μ b,1 ,...,μ b,n ,...,μ b,N ];

[0059] f) Repeat steps c) to e) for iterative calculation until convergence, and output the final and Σ b .

[0060] In the above step f), whether the iterative calculation converges can be achieved by setting a convergence judgment condition. In this embodiment, the target image signal can be judged after each iterative calculation is completed. The calculation results of all N elements in the current iteration round (recorded as round t+1) and the calculation result in the previous iteration round (recorded as round t) Whether the convergence condition is met ε is the set error precision. If it is not satisfied, the iteration continues. If it is satisfied, the iteration stops. The error precision ε can be set according to actual conditions. In this embodiment, it is set to 10 -5 .

[0061] All access points repeat steps a) to f) to obtain their respective estimated a posteriori expectations and variances, and then the estimation information of each of the B access points can be sent to the central processing unit.

[0062] S4, the central processing unit estimates the signal based on B access points and the variance matrix {Σ1,Σ2,…,Σ B}, using the multi-signal fusion method, the imaging signal of the target object is obtained

[0063] In this embodiment, the specific implementation steps of the multi-signal fusion method in step S4 are as follows:

[0064] The central processing unit is based on the local target image signal uploaded by B access points and the variance matrix {Σ1,Σ2,…,Σ B}, weighted sum of the elements at the nth position to get the estimated value of that position n=1,2,...,N, where weight And to meet μ b,n for Corresponding variance; do the same for other elements to get the estimated value composition of all elements That is, all N The final accurate estimation of the target imaging signal is formed

[0065] Next, the computational imaging method based on a non-cellular large-scale multi-input multi-output system shown in S1 to S4 is applied to a specific simulation example, in which the image of the real target is as follows: Figure 2 As shown. Computer simulation shows that the single access point computational imaging method proposed in the present invention is Figure 3 As shown in FIG. 1 , the target of interest in the environment can be better restored according to the system parameters and channel conditions. However, due to factors such as channel noise pollution and incomplete received signals, the imaging effect is often not ideal. The multi-signal fusion method proposed in the present invention is as follows: Figure 4 As shown, the imaging effect is significantly better than single access point imaging, and the target object is basically accurately restored. It can be seen that the present invention provides an effective calculation method for realizing imaging in a non-cellular large-scale multi-input multi-output system.

Claims

1. A computational imaging method based on a non-cellular large-scale multi-input multi-output system, characterized in that: The steps include: S1. The entire environment space is treated as a three-dimensional space and discretized to obtain N = P*P*P small cubes, each of which represents a pixel point. The target object of interest exists in the environment space. The target object consists of K pixels and is sparse, with a sparseness rate of λ = K / N. The scattering coefficient x is used n represents the scattering coefficient of the nth pixel point; any pixel point in the entire environment space has two possibilities: belonging to a non-target object or a target object; the scattering coefficient of the entire environment space is expressed as x = [x1, x2, ..., x N ] T , each element in x is independent of each other and obeys the Bernoulli-Gaussian distribution, and its probability function is expressed as Where λ is the sparsity rate, δ(x) is the Dirac function at x = 0, It means that any pixel x obeys the Gaussian distribution probability density function with mean θ and variance φ, and q = (λ, θ, φ) is the set of prior parameters; S2. In a non-cellular large-scale MIMO system, a central processing unit is deployed, B access points are configured with M antennas, and D single-antenna user devices are deployed; the perception signal sent by the dth user device is s d , the transmission power is where ||s d || means to find s d The model, Expressing hope; S3, the bth access point receives the signal y b , using the single access point computational imaging method, we can get the local preliminary estimated image and the variance matrix Σ b , b=1,2,...,B, and then send the estimation information of each of the B access points to the central processing unit; S4, the central processing unit estimates the signal based on B access points and the variance matrix {Σ1,Σ2,…,Σ B }, using the multi-signal fusion method, the imaging signal of the target object is obtained 2. The method according to claim 1, characterized in that The specific implementation steps of the single access point computational imaging method described in step S3 are as follows: a) The channel matrix from the dth user equipment to the target object is The channel matrix of the dth user equipment signal reflected from the target to the bth access point is: where α d and β b,d are large-scale fading coefficients based on path loss and shadow fading, H d and G b,d are all small-scale fading coefficients and obey the standard Gaussian distribution; then the signal model received by the b-th access point is expressed as The signal model expression is further simplified into the standard linear model y b =A b x b +w b =z b +w b , where the matrix x b is the target object estimated by the bth access point, w b is additive Gaussian white noise with a noise variance of ψ b ; b) For any access point b, the received signal y in a) b The vector product of is converted to the product of all scalar elements, where x b =[x b,1 ,...,x b,n ,...,x b,N ] T , z b =[z b,1 ,...z b,m ,...,z b,M ] T ,y b =[y b,1 ,...y b,m ,...,y b,M ] T , then the posterior probability of the ambient scattering coefficient is: Where p(y b |x b ) is y b The likelihood function, p(x b |q) is x b The prior probability, p(y b,m |z b,m ) is y b,m The scalar likelihood function, p(x b,n |q) is x b,n The scalar prior probability of ; c) For m=1,2,...,M, find z b,m About (y b ; The marginal posterior probability of q) Where p(y b,m |z b,m ) is y b,m The scalar likelihood function of , Indicates z b,m The mean is The variance is Gaussian distribution probability density function; find the p(z b,m |y b ; q) and then calculate the expectation to get z b ; Then perform matrix inversion operation A b -1 *z b =r b , and get the target object r polluted by noise b , that is, r b =x b +v b , where v b is additive Gaussian white noise; finally, for n = 1, 2, ..., N, we can obtain x b,n About (y b ; The marginal posterior probability of q) Where p(x b,n |q) is x b,n The scalar prior probability of For x b,n The mean is The variance is Gaussian distribution probability density function; d) The prior probability p X (x b,n |q) Substitute into x in c) b,n About (y b ; The marginal posterior probability formula of q) is obtained, and x b,n The marginal posterior probability of is: Among them, δ(x b,n ) represents x b,n = 0, the Dirac function, ζ b,n is the normalization factor: parameter e) Through the linear minimum mean square error estimation, the expectation of the marginal posterior probability in d) is obtained and variance μ b,n : Thus, the target image signal estimated by the b-th access point is obtained and the diagonal variance matrix Σ b =diag[μ b,1 ,...,μ b,n ,...,μ b,N ]; f) Repeat steps c) to e) for iterative calculation until convergence, and output the final and Σ b .

3. The method according to claim 2, characterized in that In step f), after each iteration, the target image signal is determined. The calculation results of all N elements in the current iteration round The calculation result in the previous iteration is Whether the convergence condition is met ε is the set error accuracy. If it is not satisfied, the iteration continues; if it is satisfied, the iteration stops.

4. The method according to claim 3, characterized in that The error precision ε is preferably 10 -5 .

5. The method according to claim 1, characterized in that The specific implementation steps of the multi-signal fusion method in step S4 are as follows: The central processing unit is based on the local target image signal uploaded by B access points and the variance matrix {Σ1,Σ2,…,Σ B }, weighted sum of the elements at the nth position to get the estimated value of that position The weight And to meet μ b,n for The corresponding variance; all N The final accurate estimation of the target imaging signal is formed

Images