Method for generating voxelized environment model for signal processing in wireless communications, method for processing wireless signals on receiver side or on transmitter side using voxelized environment model, and receiver or transmitter implementing these methods

Abstract

A method of generating a voxelized environment model for signal processing in wireless communications includes receiving information about spatial locations of a UE and an AP, about frequencies or frequency bands used in current communications, about spatial sizes of voxels within the voxelized environment, and receiving a puncturing threshold; modeling all theoretically possible geometric paths and associated attributes between each UE and each AP in consideration of all possible voxels; determining attributes of communication channels corresponding to the respective modeled geometric paths; identifying unavailable paths; identifying an unrealistic path; and generating a voxelization environment model, and removing an unavailable path and an unrealistic path from the voxelization environment model according to the received puncturing threshold. Further, random blocking may be introduced into the environmental model. A voxelized environment or data derived therefrom is used in a method of joint communication and environment awareness and / or for pre-processing signals prior to transmission.

Description

Technical Field

[0001] This invention relates to the field of wireless communication, and more particularly to wireless communication that uses a spatial representation of the environment to optimize operating parameters in transmitters and / or receivers.

[0002] symbol

[0003] Scalar values ​​are represented by the italic lowercase letter x in this paper, while complex vectors and matrices are represented by the bold lowercase and uppercase letters x and X, respectively. (·) T and (·)* denote the transpose operator and the complex conjugate operator, respectively, and diag(·) denotes the diagonalization operator. |·| denotes the absolute value operator, while || || l Represents the l-norm. (IE) x (x) and Var x (x) respectively represent x relative to the given information. The expected value operator and variance operator for the x-distribution are given. and Let N(μ,ν) and CN(μ,ν) represent the real number field and the complex number field, respectively, and let N(μ,ν) and CN(μ,ν) represent the real Gaussian distribution and the complex Gaussian distribution with mean μ and variance ν, respectively.

[0004] Throughout the specification, the term "environmental sensing" can be used in a wide range of expressions for capturing information about the environment in order to create its three-dimensional representation. Background Technology

[0005] Spatial representation of the environment is widely used in robot vision, localization and mapping, as well as computer graphics and medical imaging. One known technique involves discretizing the real environment by voxelizing an occupied mesh. Typically, voxels are regular cubes arranged in a 3D space covering the environment, and these cubes are assigned information about the space they occupy. The discretization of the resulting voxelized space is well-suited for 3D modeling, where objects in the region of interest can be represented by coarse estimates or more accurate shapes, depending on the size of the unit voxels.

[0006] Figure 1 a) An exemplary environment is shown, in Figure 1 b) and Figure 1 In c), the environment is represented using voxels of different sizes with voxelization at different resolutions. Figure 1 b) and Figure 1 c) illustrates the 3D voxelized occupied mesh, where the total region of interest (ROI) is defined as having a size of L. x ×L y ×L zThe ROI is a cubic space, where each dimension represents the length of the x-axis, y-axis, and z-axis, respectively, in meters. A grid composed of individual elements, in which, and These represent the number of voxels or partitions along the x-axis, y-axis, and z-axis, respectively, and L V This is the side length of the voxel cube, in meters, corresponding to the image resolution. If we use a three-dimensional tensor (N... x ×N y ×N z If the voxelization occupies the grid, it directly represents the discretized model of the ROI (e.g., Figure 1 b) and Figure 1 (as shown in c)). The elements of the three-dimensional tensor indicate the occupancy of voxels, thus indicating whether that part of the space is empty or filled with a given material.

[0007] Besides the flexibility in estimation accuracy offered by voxelized meshes, one of the biggest advantages of mesh-based models is the simplicity of data representation (especially compared to 3D vertex-based or point cloud-based methods), which makes voxelized meshes more advantageous from a machine learning perspective.

[0008] Recent developments in communication technology have demonstrated the benefits of modeling the space between the transmitter and receiver using voxelized environments for Joint Communication and Sensing (JCAS) (sometimes also called Integrated Sensing and Communication or ISAC). Throughout this specification, the terms JCAS and ISAC are used interchangeably. JCAS is a technique in wireless communication that aims to obtain information about the environment from signal scattering present in effective channel state information (CSI) (e.g., caused by objects, obstructions, user activity, etc.) while simultaneously enabling data communication.

[0009] To this end, the electromagnetic scattering behavior of objects in the real environment can be added to the 3D geometric information provided by the classical voxelized occupancy mesh to customize the communication channel modeling method used in wireless communication scenarios. To account for scattering behavior, a voxel occupancy coefficient v is assigned to each voxel. k ∈{0,1} (where k∈{1,…,N) V}), where v k =0 indicates that the k-th voxel is empty, meaning the corresponding environment is free space, and v k =1 indicates that the k-th voxel is occupied by a scattering object (e.g., an object on a table, chair, or wall), such as Figure 1 As shown. Note that the binary voxel occupancy coefficient can also be extended to the complex voxel scattering coefficient, i.e. To capture the effect of occupied voxels on reflected electromagnetic waves. The constant β k and ω k It depends not only on the material itself, but also on the frequency and angle of incidence of the propagating signal, and the effect of the material occupying a given voxel on the electromagnetic wave it reflects or refracts. In other words, the value of the voxel scattering coefficient is expected to be highly dependent on the electromagnetic properties of the scattering object and the impact wave, which can be empirically measured and modeled as functions of properties such as frequency and material properties.

[0010] Figure 2 a) and Figure 2 (b) illustrates the general concept of line-of-sight (LOS) and non-line-of-sight (NLOS) paths in voxelized space, where all paths are available and the reflection angles are within the range of the actual reflected impact electromagnetic waves. Figure 2 In a), the LOS path is the direct path between the User Equipment (UE) and the Access Point (AP), indicated by the solid line, while the two occupied voxels in the ROI reflect signals transmitted by the UE back to the AP. The dashed line represents the NLOS path from the UE to the voxel, and the dotted line represents the NLOS path from the voxel to the AP. Figure 2 In b), one of the previously filled voxels is now empty, which will obviously affect the scattering coefficient of that voxel. The dashed line from the empty voxel to AP represents... Figure 2 The path that exists in a) is not considered an existing path.

[0011] X. Tong, Z. Zhang, J. Wang, C. Huang, and M. Debbah, in “Joint Multi-User Communication and Sensing Exploiting Both Signal and Environment Sparsity,” IEEE Journal of Selected Topics in Signal Processing, Vol. 15, No. 6, pp. 1409–1422, November 2021, discuss an exemplary regular voxelized 3D space with some scattering objects accommodating a single access point (AP), a single reconfigurable smart surface (RIS), and multiple single-antenna user equipments (UEs). Multiple UEs communicate with the AP via sparse code multiple access (SCMA) through multiple frequency subcarriers and multiple transmit instances, via LOS and NLOS paths from UE to scattering object to RIS and finally to AP.

[0012] Figure 3A schematic representation of the 3D space (including RIS) considered in known systems and methods is shown. Here, signals reflected from RIS toward AP are shown with dotted lines to highlight their specific sources.

[0013] Based on the known system model and assumptions, the signal received at the AP will not only carry the UE's payload or data, but also contain scattering path information that can be used to acquire or sense the environment. To aid in estimation, it is assumed that the UE transmits a pilot symbol of known length at the beginning of the transmission interval, i.e., a known preamble. The goal of the known method is to return an estimate of the UE's transmit SCMA code (i.e., user detection or identification), and a binary voxel occupancy grid corresponding to the environment estimate; for example, 0 represents voxels representing free space, and 1 represents voxels representing occupied space.

[0014] The known system can be mathematically represented by the following model.

[0015] Y = (H + PRQVB)[X] p X]+W

[0016] Where Y is the received signal matrix of all receiving antennas and AP on all symbol instances, H is the LOS channel matrix from UE to AP, P is the NLOS channel matrix from RIS to AP, R is the RIS reflection coefficient matrix, Q is the NLOS channel matrix from voxel to RIS, B is the NLOS channel matrix from UE to voxel, V is the voxel occupancy matrix, and X... p Let P be the pilot symbol matrix, X be the data symbol matrix, and W be the AWGN noise matrix. Since matrices P, R, Q, and B are known, the model can be simplified to...

[0017] Y = (H - AVB)[X] p X]+W

[0018] Where A = PRQ is the effective NLOS channel matrix from voxels to AP.

[0019] Thus, the main joint communication and sensing problem is formulated, namely, using the known channel matrices H, A, B and the known pilot matrix X. p To estimate the environment matrix V and the data symbol matrix X.

[0020] Therefore, existing technical methods use three modules:

[0021] • The environment estimation module uses the Linear Generalized Approximate Message Passing (GAMP) algorithm, which estimates the environment V using a known channel and given data symbols X (pilot or estimated symbols from the previous iteration). Its model is as follows:

[0022] Y = (H + AVB)[X] + W

[0023] • The effective channel reconstruction module integrates known channel and estimated environment information and uses V estimated by the GAMP algorithm to combine the information with known H, A, and B to calculate the effective total channel.

[0024]

[0025] • The data signal estimation module uses the linear SCMA message passing algorithm (MPA) and the estimated effective channel to estimate the unknown data symbol X.

[0026]

[0027] These three modules iterate in the time domain using a sliding window approach (e.g.) Figure 4 As shown, this allows for initial environmental estimation using known pilot symbols, but for data signal estimation, the window slides to estimate unknown data symbols, and so on.

[0028] However, like other known methods of JCAS in voxelized environments, the prior art methods discussed further above rely on the following idealized assumptions: all paths between all transmit and receive antennas are fully available, i.e., no path is blocked; the channel gain of the LOS path from UE to AP is known; the channel gain of the NLOS paths from UE to voxel, voxel to RIS, and RIS to AP is known; the RIS reflection coefficient is known; the voxelized environment model is binary, i.e., only discrete occupancy values ​​of 0 or 1 are possible; all scattering paths have real reflection angles; an SCMA communication scheme is used; and only a single AP exists. It should be noted that throughout this specification, the terms voxelized environment, voxelized space, 3D space, 3D mesh, and similar expressions are used interchangeably unless the specific use is obvious from the context, and these terms may also be represented by the abbreviation VE.

[0029] Besides simplifying known models, it also fails to consider energy losses, frequency-dependent scattering behavior, and the relative distance between voxels and devices, which depend on the material and surface structure of the scattering object and their impact on channel gain during reflection scattering.

[0030] These assumptions severely limit the application of known methods to real-world 3D environments, leading to a need for a method for generating improved voxelized environment models for signal processing in wireless communications, a method for transmitting or receiving using said improved voxelized environment models, and a transmitter and receiver for implementing these methods. Summary of the Invention

[0031] This need is met by the methods as described in claims 1, 7, and 8, the computer program product as described in claim 10, the wireless communication device as described in claim 12, and the communication system as described in claim 14. A corresponding computer-readable storage medium is presented in claim 11.

[0032] Before describing the various aspects of the present invention, the concepts of unavailable and impractical paths in a voxelized environment will be clarified.

[0033] Figure 5 a) and Figure 5 b) illustrates exemplary voxelized spaces where some of these paths may become infeasible or unavailable due to various physical phenomena. Figure 5 a) and Figure 5 (b) In both cases, a solid line from the User Equipment (UE) to the Access Point (AP) represents the available LOS path. Clearly, if the occupied voxel lies in a straight line with the path between the UE and the AP (not shown in the figure), the corresponding path will be unavailable. If the scattering angle between the incident and reflected paths is too large and exceeds a critical angle (e.g., close to 180°), the corresponding NLOS path will be impractical and practically unavailable. Figure 5 As shown in a), the dashed line from the voxel to the AP represents the NLOS path of the reflected voxel to the AP, which has an unrealistic reflection angle, i.e., unusable. The determination of the critical angle depends on the electromagnetic properties of the RF wave and the environment (e.g., operating frequency), therefore this invention proposes a simplified model to approximately incorporate this phenomenon into the channel matrix of the voxelized mesh environment model. Path unusability can also be caused by tilted surfaces, such as... Figure 5 As shown in b), the signal is reflected away from the AP. Figure 5 Same as in a), Figure 2 In b), the theoretically possible NLOS path for reflection from an empty voxel will be unusable because the object in the voxel will reflect the incident wave away.

[0034] Furthermore, before describing the various aspects of the present invention, a general underlying channel model will be described. It is assumed that the ROI includes N. U Each single-antenna UE and each equipped with N R N receiving antennas A A multi-antenna AP. For example... Figure 5The effective channel between the UE and AP, as shown, comprises two different types of components: LOS components and NLOS components. The LOS component is the direct path between the UE and AP, while the NLOS component covers the path reflected from occupied voxels corresponding to portions of objects in the environment—that is, the path scattered by voxels representing the scattering objects, as further described above. For simplicity, we further assume that the bandwidth is high enough that the power of paths with more than one reflection is negligible. Therefore, the NLOS component can be decomposed into two sub-paths: a sub-path from the UE to the voxel and a sub-path from the voxel to the AP. These two sub-paths, together with the voxel scattering coefficients, constitute the aggregated NLOS channel. It should be noted that NLOS paths with more than one reflection can also be considered.

[0035] In view of the above, N U A single-antenna UE and all APs N A N R A set of receiving antennas (i.e., N) A All N of each of the APs R The effective channel between (a number of antennas) is given by the following formula.

[0036]

[0037] in, It is the effective channel matrix. and These are the channel matrices formed by the LOS path from UE to AP, the NLOS sub-path from voxel to AP, and the NLOS sub-path from UE to voxel, respectively. It is a vector containing all scattering coefficients of the voxelized grid. Assume the elements of the channel matrices H, A, and B follow a variance of... and It follows a zero-mean complex normal distribution.

[0038] It should be noted that the channel model presented above can be directly extended to multi-carrier systems, thereby generating...

[0039]

[0040] Where f represents the set of all subcarrier frequencies. Specific frequencies within.

[0041] It is important to note that the channel model in the equations shown above assumes that all paths between the UE, AP, and voxel are fully available, as in the known systems discussed in the background section. However, in reality, many paths may become unavailable due to various physical phenomena, such as obstruction by airborne particles, absorption, path loss, or the limited resolution of the voxelization model itself. Furthermore, as mentioned above, if the angle between the incident and reflected paths is too large and exceeds a critical angle, the corresponding NLOS path will be unavailable, such as... Figure 5 As shown in a). Similarly, if the occupied voxel is in a straight line with the LOS path, the corresponding path will be unavailable. The determination of the critical angle depends on the electromagnetic properties of the RF wave and the environment (e.g., operating frequency), therefore a simplified model is proposed to approximately incorporate this phenomenon into the channel matrix of the voxelized mesh environment model.

[0042] Realistically, modeling all paths may be computationally unfeasible. Therefore, this invention proposes to eliminate unrealistic and unusable paths and further suggests considering the statistical feasibility of paths to improve the voxelized mesh model.

[0043] Since it is assumed that the geometric positions of all voxels, as well as the UE and AP, are known relative to each other within a uniform voxelized mesh, the triangular distances of all paths can be directly calculated. Given these distances and operating frequencies, the channel loss of the paths is obtained using, for example, the Friis transmission formula, which gives the path loss coefficients as a function of distance and frequency. Furthermore, the reflection angles of all voxels in the path between the UE and AP can be calculated. This information may have already been used to eliminate impractical or unusable paths.

[0044] Therefore, consider the physical phenomena that occur when a propagating wave is reflected, especially for any given frequency, and consider the following facts:

[0045] i) A critical angle θ exists * , making such Figure 5 a) As shown, if the incident angle θ > θ * The wave is absorbed instead of reflected, therefore the corresponding voxel-to-AP NLOS subpath is unavailable, and

[0046] ii) The curvature of the surface exposed to the impact wave may cause no signal to be reflected toward the AP, such as Figure 5 b) As shown.

[0047] As previously mentioned, modeling this phenomenon at each voxel can be too complex to perform, especially when the resolution of the voxelized ROI is high (i.e., the individual voxels are small). Therefore, a statistical method is proposed to consider the angle between the impact wave and the reflected wave at each voxel (hereinafter referred to as the scattering angle) and thus the availability of the NLOS subpath from each voxel to the AP, which significantly reduces computational complexity.

[0048] In view of the above, the present invention proposes the following stochastic geometric model to integrate the above phenomena into the channel matrix of the voxelized grid environment model.

[0049] Note that it is assumed that the multiple antennas of the corresponding AP are entirely located within a single voxel, therefore it is assumed that they have the same angle of arrival (AoA) and different channel path coefficients. Given the known 3D coordinates of the UE, AP, and the voxel under consideration, respectively...

[0050] and c V =[x V ,y V ,z V ] T The scattering angle θ of the path at the corresponding voxel can then be calculated as follows:

[0051]

[0052] The arccos(·) operator represents the inverse cosine trigonometric function.

[0053] The empirical probability distribution function (PDF) of the scattering angle θ can be obtained by applying c to the scattering angle θ respectively. U c A and c V The above equation is used to evaluate all possible combinations of the permissible locations of the UE, AP, and voxels within the given ROI, and... Figure 6 The document shows exemplary PDFs of environments voxelized at various resolutions. Permissible locations refer to all possible locations within the grid area, excluding locations outside the grid, locations outside the ROI, and also excluding situations where two or more UEs or APs are located in the same voxel.

[0054] exist Figure 6 As can be seen from this, for a sufficiently large N V The distribution of the scattering angle θ can be well modeled by a mixture of two β distributions, i.e., f x(x) = γ·β(a1,b1) + (1-γ)·(a2,b2), where x ∈ [0; 180°], where γ is a weighting factor, and quantities a1, a2, b1, b2 are shape parameters optimized to match empirical data obtained by evaluating the aforementioned equation, where c U c A and c V Randomly selected within the voxelized mesh.

[0055] This empirical stochastic geometry approach improves the voxelized environment model obtained to date by introducing stochastic blocking of NLOS subpaths based on a chosen critical angle, a complementary cumulative distribution proportional to the approximate β-mixed PDF, and the scattering angle at each voxel. The choice of critical angle can take into account the frequency or band used in the corresponding communication and / or a fixed minimum and / or maximum value.

[0056] As mentioned earlier, the scattering angle θ at all voxels is calculated for all UE and AP pairs. If θ > θ 临界 (θ 临界 If the scattering path is ∈[0°,180]), then the scattering path is determined to be unavailable, and the corresponding path is removed from the NLOS channel matrices A and B respectively.

[0057] exist Figure 7 The critical angle θ is shown in the figure. 临界 The impact on the severity of channel matrix puncture was assessed through numerical evaluation using random numbers and locations of UEs and APs at various voxelized grid resolutions. The average channel puncture rate was determined by the total number of path vertices (N). A +N R Normalization. As expected, the perforation rate exhibits normalization at θ. 临界 =180° (without punch) and θ 临界 A smooth increase between θ = 0° (full punching). This essentially includes the case of blocked LOS, since blocking can be considered as the case where θ = 180°. As mentioned above, the true critical angle depends on the complex electromagnetic properties of the environment, but its determination is beyond the scope of this invention. However, a very interesting behavior was observed where the severity of punching is actually unaffected by the resolution of the pixelated environment model (i.e., voxel size), converging to the same relationship at sufficiently high resolutions.

[0058] like Figure 7 The convergence curve shown can be approximated by scaling the Gaussian curve height, where the scaling factor obtained by the heuristic search is... N(-9.8,54) 2The optimal parameterization of ) is then achieved. In summary, an efficient model for drilling behavior is proposed by introducing a feasibility coefficient ξ∈[0,1], which follows a probability p. ξ The Bernoulli distribution, the probability is obtained by evaluating θ. 临界 Scaling Gaussian distribution at time To obtain this, the independent and identically distributed (iid) feasibility coefficients are then multiplied by each element of the channel matrix, capturing the behavior of unavailable and punctured infeasible paths.

[0059] The benefits of the improved environment model will be illustrated below for two exemplary ISAC methods. Here, a set of N under the above model is considered. U UEs and a total of N A Uplink communication scenario between N APs, where N A Each AP is connected to the central processing unit (CPU) via a throughput-unrestricted, error-free fronthaul link, enabling all N APs to access the central processing unit (CPU) via a throughput-unrestricted, error-free fronthaul link. A N R Signals received at each receiving antenna are aggregated without loss of information or delay.

[0060] N T The aggregated received signal matrix Y over a discrete transmit instance (symbol slot) is given by the following formula.

[0061]

[0062] in, It is the effective channel matrix as described above.

[0063] It is a collection of data from all N. U The transmitted signal matrix of each UE (each symbol is from a base N). χ (Extracted from the constellation χ), and It is the receiver-side additive white Gaussian noise (AWGN) matrix with independent and identically distributed (iid) elements drawn from CN(0,N0), where N0 is the noise variance.

[0064] Transmitted signal X includes pilot block and data blocks Make

[0065]

[0066] Where, N P and N D N represents the number of symbol time slots allocated to the pilot sequence and the data sequence, respectively. T =N P +N DAnd in which, assuming the pilot symbol matrix X P It is fully known at the CPU level.

[0067] Given the equations provided above, the objective of the exemplary ISAC scheme discussed below can be briefly stated. The CPU's communication objective is to estimate the channel matrix G, given only that X is known. P Estimating the unknown data symbol matrix X in the case of intermediate pilot symbols D Furthermore, the sensing target extracts a voxelized model of the environment as an occupancy coefficient vector v from the channel matrix G.

[0068] By combining the previously proposed channel decomposition model of G, the received signal model of Y, and the transmitted signal model of X, the entire system model becomes

[0069]

[0070] The unknown variables of interest are the environment, namely the voxel coefficient vector v and the data symbol matrix X. D .

[0071] In the following description, it is assumed that the LOS channel H and the sub-paths from the UE to the voxel and from the voxel to the AP in the aforementioned equation of Y are known, which is still in the case of the two variables diag(v) and X. D An atypical relationship is preserved between them. In particular, the latter unknowns are related to the asymmetric bilinear system under the aforementioned equation of Y, requiring decoupling or joint estimation of complex computational schemes.

[0072] The exemplary ISAC methods discussed below for the joint estimation problem of an asymmetric bilinear system represented by the aforementioned equation for Y utilize the well-known Gaussian belief propagation (GaBP) message passing (MP) framework. Both of these exemplary ISAC methods benefit from the improved modeling of VE proposed in this paper. Note that other ISAC methods can also benefit from the improved modeling of VE.

[0073] The first proposed method is called Alternating Linear ISAC (AL-ISAC), which takes unknown variables v and X as inputs. D Each variable in the equation is combined with two separate linear estimation processes, which estimate the unknown variables in an alternating manner via a feedback chain between the two processes. Both estimation processes are based on equation (6) and can be described as follows:

[0074] - An iterative first linear GaBP MP process for estimating the environment vector v given the transmitted signal matrix X, and conversely,

[0075] - An iterative second linear GaBP MP process for estimating the transmitted signal matrix X given an environment vector v.

[0076] The following section derives the two linear GaBP processes and the MP rules, and then describes the composition of the complete AL-ISAC process covering the two derived processes.

[0077] The iterative first linear GaBP MP process for the environment vector v operates on only one unknown variable. Therefore, to estimate v, in addition to the known channel matrices H, A, and B, it must be assumed that the entire transmitted signal matrix X is known. Assuming X is known, the system described by the aforementioned equation for Y can be reformulated as follows:

[0078]

[0079] Among them, because of the channel matrix Matrix product

[0080] and It is known that the system described is linear on v, from which we can obtain... Figure 8 The corresponding factor graph is shown.

[0081] Each element of the signal received by Y is y m,t (where m∈{1,…,N) A N R} and t∈{1,…,N T}) corresponds to the factor nodes represented by squares in the graph, and each element v of the unknown environment variable v k (where k∈{1,…,N) V}) corresponds to the variable nodes represented by circles in the graph. Conversely, each (m,t)th factor node in the factor graph has each variable node element v k The corresponding soft copy is represented as The corresponding mean square error (MSE) is given by the following formula.

[0082]

[0083] Using soft copies and their MSEs, factor nodes for each variable v k Received signal y m,t Perform soft interference cancellation (IC) to generate IC symbols.

[0084]

[0085] Where, x n,t and w m,tLet X and W be the (n,t)th and (m,t)th elements, respectively, where n∈{1,…,N}. U}, and among them, auxiliary variables This represents the aggregated incident signal from all UEs at the k-th voxel.

[0086] Next, by utilizing the Central Limit Theorem (CLT), the sum of the difference and noise terms is approximated as a complex Gaussian scalar, allowing the sign after interference cancellation to be approximated. PDF modeling for

[0087]

[0088] Wherein, the corresponding variance Given by the following formula

[0089]

[0090] in, It is the noise variance.

[0091] All v k The conditional variance is calculated through all factor nodes, and the message is sent to the corresponding variable node. Therefore, the k-th variable node obtains N from all factor nodes. A N R N T Each conditional variance is used to calculate the external confidence level. In GaBP, self-interference is eliminated at the k-th variable node. Conditional PDFs are used to suppress the generation of PDFs using the following methods.

[0092]

[0093] Among them, external mean and variance The following formulas are given respectively.

[0094]

[0095] Finally, according to Bayes' rule, the updated posterior can be obtained by comparing the PDF of the external confidence with v. k The prior distribution is combined to obtain the updated soft copy, which is then obtained through the following formula.

[0096]

[0097] The normalization factor in the denominator is the updated posterior of the integral over the complex field.

[0098] Similarly, the error variance of the soft copy update is obtained by evaluating the following formula.

[0099]

[0100] Given information about the voxel coefficient distribution, i.e., binary coefficients with discrete priors given by the Bernoulli distribution (with occupancy probabilities) The soft copy and its MSE can be efficiently obtained in a closed-form manner, as given by the following formulas.

[0101]

[0102] Then, the updated soft copy and MSE of each variable node are transmitted back to all factor nodes for the next iteration of the GaBPMP computation process. After a given number of GaBP iterations to improve the soft estimates, a confidence consensus is reached on the soft copies at each variable node to obtain a single estimate via the following formula.

[0103]

[0104] Among them, consensus mean and variance Represented as

[0105]

[0106] Therefore, the final estimate is obtained using the following formula.

[0107]

[0108] Equations (8) to (20) fully describe the first linear GaBPMP process used to estimate the voxel environment v (i.e., voxel coefficients), referencing Figure 9 The steps of method 500 shown are summarized below.

[0109] The inputs to the first process are the received signal matrix Y, the channel matrix H, A and B, the transmitted signal matrix X, the noise variance N0, and the prior distribution of the environment voxels.

[0110] The first iteration begins with initialization, because the data comes from earlier iterations. soft copy and corresponding MSE Currently unavailable. In step 502, according to The first signal matrix C is calculated, which describes the effective signal modified by the channel between the UE and the voxel. Then, in step 504, soft copies of all variables are initialized. And in step 506, according to equation (8) Initialize all variable nodes Perform initialization.

[0111] The following steps are iterated until the termination criterion is met. In step 510, equation (9) is used to evaluate the received signal y. m,t Execute soft IC, generate In step 512, the corresponding conditional variance is determined according to equation (11). In step 514, the external mean is determined according to equation (13). and variance In step 516, the new soft copy is calculated according to equations (14) and (15), respectively. and And in step 518, the soft copy is updated via damping. and For example, as discussed in P. Som, T. Datta, A. Chokalingam, and B.S. Rajan, “Improved large-MIMO detection based on damped belief propagation,” IEEE Inf. Theory Workshop on Inf. Theory, 2010, pp. 1-5. The damping factor can be selected from the interval η∈[0,1] and is used to prevent premature convergence to a local optimum. It should be noted that... Perform each of steps 510 to 518. In step 520, perform a check to determine if the termination criterion is met. If negative, proceed to the "No" branch of step 520, repeating steps 510 to 518 using the latest values ​​of the soft copy and MSE as input. If positive, proceed to the "Yes" branch of step 520, in step 522... Calculate the consensus mean according to equation (19). and variance And in step 524, the final soft estimate is calculated according to equation (20). The termination criteria for the iterative loop may include a predetermined maximum number of iterations, or a predetermined convergence threshold that can be evaluated in terms of the soft-copy MSE. Note that... Perform subsequent steps that meet the termination criteria.

[0112] It is important to note that the signal matrix X is assumed to be given; that is, X is not estimated by the process and therefore remains constant throughout the iterations, as with the effective signal c. k,t As seen in the pre-calculation.

[0113] The complementary second linear GaBP MP process, which is specifically used to estimate the transmitted signal matrix X given v, will now be described.

[0114] To derive the second linear GaBP MP process for estimating the signal matrix X given v, the overall system model, further derived above and represented by equation (6), is first simplified to the aggregated received signal matrix Y initially presented in equation (4), where the effective channel... The corresponding factor plot is in Figure 10 The diagram shows that each element x of the unknown signal matrix X is... n,t (where n∈{1,…,N) U}) are variable nodes, represented by circles.

[0115] Note that in this case, since the variable X is two-dimensional, the linear system produces a factor graph that is split into "pages" such that different time indices t∈{1,…,N} T The variable nodes and factor nodes are independent, and messages are exchanged only by nodes with the same time index t. Apart from this separation of the factor graph, the derivation of the MP rule is similar to the derivation of the linear GaBP MP process for the environment vector v presented further above.

[0116] Transmitted signal matrix element x n,t The soft replica to the (m,t)th factor node is generated by This indicates that the corresponding MSE is given by the following formula.

[0117]

[0118] Soft copies and MSE are used for x in accordance with the following formula n,t The received signal in the soft IC

[0119]

[0120] The conditional PDF is described by the following formula.

[0121]

[0122] And conditional variance Given by the following formula

[0123]

[0124] Conditional PDF is combined with self-interference elimination at variable nodes to generate external confidence following the formula below.

[0125]

[0126] Among them, external mean and variance The following formulas are given respectively.

[0127]

[0128] Furthermore, since the symbols have uniformly discrete priors from the symbol constellation χ, where the symbol probability is... Therefore, its soft copy and MSE are obtained through the following formula.

[0129]

[0130] For the specific case of M-QAM (where M=4), the soft copy and MSE calculations simplify to an efficient closed-form expression given by the following equation.

[0131]

[0132] in, Let denot χ represent the average sign power of the constellation, and tanh(·) denote the trigonometric hyperbolic tangent function.

[0133] Finally, the consensus PDF obtained after iteration is given by the following formula.

[0134]

[0135] Among them, consensus mean and variance Represented as

[0136]

[0137] This results in a soft estimate.

[0138]

[0139] Equations (21) to (33) fully describe the second linear GaBP MP process for estimating the transmitted signal matrix X given the environment vector v, referencing Figure 11 The steps of method 600 shown are summarized below.

[0140] The inputs to the second process are the received signal matrix Y, the channel matrix H, A and B, the environment vector v, the noise variance N0, and the prior distribution of the transmitted symbols.

[0141] Similar to the first calculation process, the second calculation process begins with initialization, wherein the effective channel matrix is ​​calculated in step 602. Then, in step 604, the soft copies at all variable nodes are initialized. Furthermore, in step 606, the MSE at all variable nodes is initialized according to equation (21). Perform initialization.

[0142] The following steps are iterated until the termination criterion is met. In step 610, the received signal after the soft IC is calculated according to equation (22). Furthermore, in step 612, the corresponding conditional variance is calculated using equation (24). Next, in step 614, the external mean is calculated according to equation (26). and variance This allows the new soft copy to be calculated in step 616 according to equations (27) and (28), respectively. and corresponding Now, in step 618, the soft copy can be updated via damping. and Similar to the update process in Method 500 discussed further above. Here, damping is also used to prevent premature convergence to a local optimum. It should be noted that... Perform each of steps 610 to 618. In step 620, perform a check to determine if the termination criterion is met. If negative, proceed to the "No" branch of step 620, repeating steps 610 to 618 using the latest values ​​of the soft copy and MSE as input. If positive, proceed to the "Yes" branch of step 620, in step 622... Calculate the consensus mean according to equation (32). and variance And in step 624, the final soft estimate is calculated according to equation (33). Then, in step 626, the final soft estimate is... Projected onto the symbol constellation χ, and in step 628, the projected... The output is the final hard estimate. The termination criteria for the iterative loop may include a predetermined maximum number of iterations or a predetermined convergence threshold that can be evaluated in terms of the soft replica MSE.

[0143] The two procedures described above need to be combined to complete the proposed AL-ISAC procedure for estimating the two unknown variables.

[0144] Each of the two estimation processes presented above is suitable for estimating only one of the two variables v or X (assuming that complete information about the corresponding other variable is available in each case). However, a very inherent problem with ISAC in the overall system model further derived above is that the variables are not fully known, making it possible that linear processes cannot be directly applied to estimation.

[0145] To address this issue, the proposed AL-ISAC process sequentially calls two linear GaBP MP processes to estimate two sets of variables. This requires separating the received signal into blocks corresponding to the pilot and data phases, as follows:

[0146]

[0147] in, and For use Defined.

[0148] First, using only the pilot phase of the system as represented by equation (34a), the first linear GaBP MP process for estimating the voxel environment v is invoked to utilize the pilot block X. P The initial environment vector is estimated using the known input signal matrix. Next, using only the data phase of the system as represented by equation (34b), the second linear GaBP MP process for estimating the transmitted signal matrix X is invoked to use the initial environment estimates. Using known input environment vectors to estimate unknown data blocks Finally, by calling the first linear GaBP MP process for estimating the voxel environment v again, but using the initial environment estimate... As the initial value for soft copies at all factor nodes and used The environment vector is obtained by using the known input signal matrix. Figure 12 The diagram illustrates the alternating properties of an exemplary AL-ISAC process.

[0149] While offering potential advantages in terms of complexity, particularly for scenarios with a large number of UEs, the alternation pattern of the AL-ISAC process, as will be further illustrated below, has the following drawbacks: it heavily relies on the pilot sequence X. P The length of the signal, as will be further shown below, not only involves a clear trade-off in total communication throughput, but also affects the performance of both environmental and data signal estimation.

[0150] The purpose of the second of two alternative exemplary estimation processes, known as Bi-ISAC, is to circumvent this shortcoming. This alternative process uses a single bilinear estimation module, more specifically a bilinear MP, to estimate the two unknowns v and X in parallel. D This bilinear MP incorporates the uncertainty of the two estimates at each iteration. The bilinear MP requires only a single estimation module to obtain the two variables, such as... Figure 13 The exemplary block diagram is shown.

[0151] Based on the system equations reformulated using the AL-ISAC method discussed further above, v and X D The unique asymmetric bilinear relationship hinders the application of known bilinear estimators (such as Bilinear Generalized Approximate Message Passing (BiGAMP) or Parametric BiGAMP). BiGAMP works only for symmetric systems described by equations of the form Y = VX + W, and is used to jointly estimate the unknowns V and X, while Parametric BiGAMP works only for systems with structure Y = ∑ k v k A k The X+W system works, used to address situations where A is known. k Joint estimation of v under the circumstances k And X.

[0152] Compared to these two examples, the unique asymmetric system discussed in this paper is neither described in the above form nor can it be transformed to fit a general bilinear form. This requires deriving a new, specially constructed bilinear Gaussian belief propagation (BiGaBP) MP rule for its solution.

[0153] In the second exemplary method discussed in this paper, BiGaBP messaging is performed as follows: Figure 14 This is performed on the shown three-factor plot. Here, the factor nodes, represented as square nodes, are the received symbols, and the two sets of variable nodes, represented as circular nodes, correspond to the environment vector v and the signal matrix X, respectively. There is an important difference between these two types of variable nodes: data variable nodes only originate from N corresponding to the same time instance t. A N R Each factor node receives messages, while the environment variable node receives messages from all N. A N R N T Each factor node receives a message.

[0154] Figure 14 The factor graph shown is Figure 8 and Figure 10 The factor graph of the linear GaBP process shown has higher complexity compared to that of the linear system structure. This higher complexity stems from the fact that, unlike the lower-complexity equations of the initial system model further presented above, the equations of the initial system model have an asymmetric and embedded system structure with respect to two variables together. Furthermore, the messages transmitted through the graph edges carry the same information items as in the linear GaBP process further presented above: soft copies of the variables of interest, the MSE, and the conditional PDF. However, since the latter two, except for the soft copies, are not known, the corresponding computation of the messages must incorporate the uncertainty of the two variables in the form of the corresponding MSE.

[0155] Given the above, similar to the derivation in the section handling the AL-ISAC process, the exchanged messages are constructed based on soft copies of variables. Since X contains variables with t∈{1,…N} P The corresponding entry is the pilot symbol X. P Therefore, the corresponding soft copy is set to its known pilot value, i.e., And the corresponding MSE value is set to 0. For t∈{N} p +1,…N T The remaining soft copies and MSE are given as in the equations for the AL-ISAC process.

[0156] Under the conditions of soft replicas and their MSE, the factor node is determined by the following formula for each variable v. k and x n,t Execution Soft IC

[0157]

[0158] Among them, the soft IC of the data variables given in equation (35b) only applies to those with index t∈{N} p +1,…N T The unknown variable node is executed.

[0159] Following the soft IC, the corresponding condition PDF is

[0160]

[0161] Among them, the corresponding conditional variance and Given by the following formula

[0162]

[0163] In order to facilitate symbol labeling, an expectation has been introduced.

[0164] Furthermore, the external confidence score PDF after interference cancellation is calculated for all variable nodes, as given by the following formula.

[0165]

[0166] The corresponding external mean and variance are given by the following formula.

[0167]

[0168] Using equations (41a) to (42b), we can obtain the same equations as those in the linear GaBP process (i.e., for v) k Equations (14) and (15) and for x n,tEquations (27) and (28) are used to obtain the updated soft copy and MSE, where the converged consensus is respectively related to the target Equations (19) and (20) and for Equations (32) and (33) are the same.

[0169] refer to Figure 15 The steps of method 700 shown are summarized below for estimating and transmitted data signal matrix A complete description of an exemplary Bi-ISAC process.

[0170] The inputs to this iterative process are the received signal matrix Y, the channel matrices H, A and B, and the pilot matrix X. P Noise variance N0 and prior distribution of the environment and the prior distribution of the emitted symbols The output of this process is an estimated environment vector. and the estimated transmitted signal matrix

[0171] Again, the iteration process begins with the initialization in step 702. Here, for the data variable nodes corresponding to the pilot block, i.e., for t∈{1,…,N} P}, initialize the soft copy as a pilot signal: The corresponding Initialize to 0. For the data variable node corresponding to the data block, that is, for t∈{N} P +1,…,N T}, initialize the soft copy to

[0172] and The corresponding initialization is achieved via equation (21) Finally, for the environment variable node, that is, for t∈{1,…,N} T The environment soft copy is initialized to The corresponding initialization is performed via equation (8).

[0173] The core of this method involves repeating the following steps for all m, n, k, and t until a termination criterion is met:

[0174] 710: Calculate the soft IC signal using equation (35) and

[0175] 712: Calculate the conditional variance using equation (38) and

[0176] 714: Calculate the external mean using equation (41) and variance

[0177] 716: Calculate the external mean using equation (42) and variance

[0178] 718: Calculate the new soft copy using equations (14) and (27) and

[0179] 720: Calculate the new equation using equations (15) and (28) and

[0180] 722: Update all soft copies and MSE via damping.

[0181] 724: Does it meet the termination criteria?

[0182] The method further includes, after the termination criteria are met, for all n, k, t:

[0183] 726: Calculate the consensus PDF, whereby equation (32) is used to calculate... And calculate using equation (19) and

[0184] 728: Calculate the final soft estimate using equations (20) and (33) and

[0185] 730: Output As the final estimate.

[0186] 732: The final soft estimate Project onto the symbolic constellation χ.

[0187] 734: Output the projected As a hard estimate.

[0188] It is worth noting that in the MP-type process, removing signal paths from the communication environment is similar to removing edges from the factor graph representation of the system. This means fewer messages are passed between fixed and variable nodes. While fewer messages may lead to lower estimation accuracy, it can significantly reduce the computational load. By appropriately selecting the puncturing threshold, the number of paths removed, i.e., the number of edges removed from the factor graph representation of the system, can be controlled according to the corresponding requirements of the use case. This can also be referred to as the controllable “aggressiveness” of puncturing the channel matrix.

[0189] Therefore, according to a first aspect of the invention, a method is proposed for generating a voxelized environment model for wireless communication signal processing in a region of interest (ROI). The ROI includes N A ≥1 access point (AP) and N U ≥1 UE (UE). N A Each of the APs (APs) has N R ≥1 antenna, and all APs (APs) serving the ROI are communicatively connected to the central processing unit (CPU). VE consists of voxels arranged in a three-dimensional mesh.

[0190] The method includes receiving, at the CPU, information about the spatial dimensions of voxels within a voxelized environment (VE), information about the spatial locations of the UE and AP within the voxels of the VE, and receiving a punching threshold. The spatial locations of the UE and AP can be obtained, for example, from previously executed JCAS methods, geographic location information provided by the UE and / or AP, dead reckoning considering the movement of the device executing the method, or a combination thereof. The spatial locations are then mapped onto a 3D voxel mesh of the VE. The punching threshold may include a threshold for attenuation along the total path length, a critical angle for scattering at the voxel surface, information about frequency-dependent attenuation or frequency-dependent critical scattering angle along the path length, a value for the maximum number of reflections along the path, a maximum or range of allowed phase shifts during reflection, a maximum or range of polarization changes during reflection, etc., provided that the path is still considered usable or realistic as long as it does not exceed the threshold. The punching threshold may further include a probability threshold related to the probability that a voxel is occupied or empty. The punching threshold can be considered a selective input for determining the realism of the environment being modeled.

[0191] The method further includes modeling all theoretically possible geometric paths between each UE and each AP, considering all voxels in the region of interest (ROI). Modeling all theoretically possible geometric paths only requires knowing the locations of all APs and UEs within the VE, and all paths will be generated regardless of whether the voxels in the VE are filled or empty. At this stage, the actual or assumed properties of the voxels are unknown and unimportant; theoretically possible geometric paths can be determined simply by considering the reflections from one or more surfaces of the corresponding parallelepiped of the voxel.

[0192] In the next step, the properties of the communication channel corresponding to the possible geometric paths modeled are determined. The properties associated with the communication channel include at least the total path length corresponding to the length of the geometric path, and may further include information about: the length of the path segments, the reflection angle (also referred to herein as the scattering angle) between the path segments reflected from the surface of the voxel's parallelepiped, and reflection-related parameters (such as frequency-dependent attenuation, phase shift, polarization changes, etc., that may occur during reflection). These properties may, in particular, include frequency-dependent attenuation based on the total channel length.

[0193] Once the properties of the communication channel are determined, unusable and / or unrealistic paths are identified and removed from the entire set of theoretically possible paths. A corresponding VE model is generated containing only the remaining usable and realistic signal paths. The removal of unusable and / or unrealistic paths can be controlled based on the received punching threshold.

[0194] Then, the VE model, the prior distribution of environmental voxels and their corresponding occupancy parameters obtained from the VE model, the feasibility coefficient matrix of the corresponding channel matrices for all available and / or realistic signal paths obtained from the VE model, and / or the channel matrix punched according to the feasibility coefficient matrix are provided for the process of receiving and processing wirelessly transmitted signals in the communication environment represented by VE, or for the process of preprocessing signals to be wirelessly transmitted into the communication environment represented by VE. The feasibility coefficient matrix can hold binary values ​​of 0 or 1, which results in the deletion or punching of channel coefficients associated with feasibility coefficient 0. However, non-integer feasibility coefficients can also be used, and punching can be performed using a threshold.

[0195] Identifying unavailable and / or impractical paths may include identifying all geometric paths between the UE and AP that have more than a predetermined number of reflections, identifying all geometric paths whose total path length or path attenuation exceeds a predetermined value, and / or identifying at least one reflection whose scattering angle exceeds a predetermined critical scattering angle θ. ＊All geometric paths. When determining the scattering angle, for simplicity, reflections on the surface of the voxel's parallelepiped can be assumed. The information provided in the punch threshold can be used in particular to control the reflection angle, which is assumed to still provide a useful signal at the corresponding receiving node.

[0196] When the critical scattering angle θ * When the frequency is related, embodiments of the method may further include receiving information about the frequency and the corresponding critical scattering angle θ. * The system receives information about the relationship between the UE and the AP, and information about the frequency or frequency band used in the current communication between the UE and the AP. It then considers the received information to determine unavailable and / or impractical paths.

[0197] In one or more embodiments of the method, determining unavailable and / or unrealistic paths may further include: calculating, for all antennas of all APs, and for each voxel of the VE, or for each voxel of the VE that contributes to an available or realistic path, the probability of a signal from the UE scattering from that voxel toward the corresponding AP. A punching threshold may include a predetermined probability threshold, and probabilities below the threshold indicate that the path is unavailable or unrealistic and will be removed. The probability may be represented by a probability density function (PDF) of the scattering angle, for example, by perturbation of the path by c. U c A and c V Equation (3) is evaluated by considering all possible combinations of the permissible locations of the UE, AP, and voxels within the given ROI.

[0198] This probability can be further improved based on previously or currently received signals and / or previous executions of the method, thereby generating the most probable voxel occupancy state, which can be transformed into a more accurate voxelized environment model.

[0199] Identifying unavailable and / or impractical paths can further include: based on the selected critical scattering angle θ * A random blocking of the NLOS subpath is introduced, proportional to the complementary cumulative distribution of the probability of the signal from the UE scattering from that voxel toward the corresponding AP, and based on the scattering angle at each voxel of the VE. When the critical scattering angle θ... * When it is frequency-dependent, introducing random blocking can also be considered for the corresponding frequency or band transmitted from the UE.

[0200] In message-passing algorithms used for estimating channels, determining environmental representations, or recovering received signals, eliminating unavailable and impractical paths from the proposed VE reduces the amount of computation or messages to be transmitted. However, this reduction also decreases the accuracy of the channel model and environmental representation, and may impact the speed or quality of signal recovery.

[0201] As further indicated above, the method according to the first aspect of the invention can be used in a receiver of a communication system configured to perform JCAS. Therefore, according to the second aspect of the invention, a method is proposed for processing wireless communication signals for joint communication and environmental awareness in a region of interest (ROI). The ROI includes N A ≥1 access point (AP) and N U ≥1 UE (UE). N A Each of the APs (APs) has N R ≥1 antenna, and all APs (APs) serving the ROI are communicatively connected to the central processing unit (CPU). The ROI is represented by a voxelized environment (VE), which comprises voxels arranged in a three-dimensional mesh. The method includes: receiving at the CPU wireless signals received at one or more antennas of one or more APs serving the ROI; and generating, using the method according to the first aspect of the invention presented above, a priori distribution of the environment voxels and their corresponding occupancy parameters based on the VE model, a feasibility coefficient matrix of the corresponding channel matrices of all available and / or realistic signal paths based on the VE model, and / or a channel matrix punched according to the feasibility coefficient matrix. Alternatively, such environmental information can be received. The generated or received environmental information is then provided as input to a JCAS process in a receiver associated with the ROI. It should be noted that, for example, when the communication system is configured for centralized signal reception for the ROI, the receiver may be co-located with the CPU. Exemplary JCAS processes to which environmental information may be provided include the AL-ISAC process and the Bi-ISAC process, but other JCAS processes may also benefit from environmental information.

[0202] The method according to the first aspect of the invention can also be used in a transmitter of a communication system. Therefore, according to a third aspect of the invention, a method for processing wireless communication signals on the transmitter side is proposed. The transmitter is communicatively connected to a central processing unit (CPU), which is communicatively connected to an N serving a region of interest (ROI). A≥1 access point (AP). The ROI is represented by a voxelized environment (VE), which comprises voxels arranged in a three-dimensional mesh. The method includes receiving from the CPU a VE model of the ROI determined by the method according to the first aspect of the invention. Alternatively, a priori distribution of environmental voxels and their corresponding occupancy parameters based on such a VE model, or at least one feasibility coefficient matrix based on such a VE model for a corresponding channel matrix of a signal path between a transmitter () and a user equipment (UE) located within the ROI. In a further step, at least one channel matrix of a signal path between the transmitter and the UE punched according to the received environmental information is determined, or at least one such punched channel matrix is ​​received from the CPU. The at least one punched channel matrix is ​​then provided as input to an equalizer process and / or a precoder process in the transmitter to pre-equalize and / or precode the signal to be transmitted, respectively. The pre-equalized and / or precoded signal can then be transmitted into the ROI after any optional further processing.

[0203] In the following sections, the robustness of the exemplary methods discussed further above under various degrees of punching and elimination of unrealistic paths is analyzed, and the advantages of each method relative to the others are also revealed.

[0204] Specifically, regarding sensing capabilities, it should be noted that due to the unique nature of the voxelized occupancy grid-based method utilized in this paper, the metrics used for radar-based ISAC cannot be directly applied. Therefore, it is useful to alternatively introduce a new metric called Voxel Occupancy Error Rate (VOER), which measures the ratio of false positive (FP) and false negative (FN) elements. False positive and false negative elements are defined, respectively, as the cases where a voxel is incorrectly estimated as occupied in the presence of empty true data and the cases where it is incorrectly estimated as empty voxel in the presence of occupied true data. Mathematically, VOER is defined as... Wherein, v is the actual data and It is the voxel coefficient estimation vector, and ||·||0 represents the l0 norm of the vector.

[0205] Note that for a trivial, fully empty (or "blind") estimator, its return value is... VOER decreases to This is the average sparsity of the environment. Therefore, this figure can be used as an absolute reference for performance; in a sense, VOER << VOER 空 This indicates that the considered ISAC method has good sensing performance.

[0206] Finally, instead of SER, which is frequently used in related literature, we evaluate the communication performance of the exemplary ISAC method here in terms of the more descriptive BER, which is defined as Among them, B e X represents D The number of data bits used for error detection in X, and B is X. D The total number of bits transmitted.

[0207] In the first evaluation, the effect of changing the number of pilot symbols (e.g., captured by parameter ρ) on the performance of the proposed method was investigated. Figure 16 The sensing performance results shown in (a) demonstrate that, for the same ρ value, the Bi-ISAC method outperforms the AL-ISAC method across the entire SNR range. In other words, for a given SNR, the Bi-ISAC method requires significantly fewer pilots to achieve the same VOER performance as the AL-ISAC method.

[0208] exist Figure 16 (b) evaluates the impact of pilot ratio on communication performance in terms of BER. It can be seen that for smaller pilot ratios, i.e., ρ < 0.2, the two proposed methods achieve similar performance. However, for larger pilot ratios, i.e., ρ > 0.3, the AL-ISAC scheme outperforms the Bi-ISAC scheme at moderate SNR (i.e., 5 dB). This result may seem counterintuitive, but it is actually expected and can be explained by the fact that the linear MP module of the AL-ISAC scheme is built on the assumption of perfect symbol knowledge (i.e., zero uncertainty in symbol estimation), and this assumption is increasingly satisfied at larger pilot ratios.

[0209] However, ultimately, when the SNR is significantly higher (e.g., SNR ≥ 15 dB), the Bi-ISAC scheme once again outperforms the AL-ISAC scheme, which is a direct result of the erroneous leveling behavior exhibited by the AL-ISAC scheme.

[0210] Figure 17 The effects of random channel blocking are elucidated. Specifically, the figure compares the two proposed ISAC methods with respect to the critical angle θ. ＊ The VOER and BER performance follow the stochastic geometric empirical model further derived above, where the critical angle determines the channel blocking rate (see [link to model]). Figure 6 ).

[0211] Figure 17 The environmental sensing performance shown in (a) exhibits similar behavior to the effect of ρ, where the Bi-ISAC scheme achieves superior performance in all cases. Furthermore, the curves for the Bi-ISAC scheme show a slower gradient increase compared to the AL-ISAC scheme, indicating that the Bi-ISAC scheme is more robust to path hindrance.

[0212] The superior robustness of the Bi-ISAC scheme under short pilot lengths and against random channel blocking can be attributed to the increased number of edges in the full-factor graph caused by the bilinear representation of the system. This can be achieved by transforming the linear case... Figure 8 and Figure 10 With the bilinear case Figure 14 A comparison reveals that, compared to the factor nodes of a linear factor graph, each factor node in a bilinear factor graph connects to a significantly larger number of variable nodes. This means that even when a large number of edges are removed, as proposed in this invention, more edges remain for stable message passing. Message passing via a pruned graph is only feasible when there is still sufficient pilot data connected to the main graph, thus making the Bi-ISAC scheme more dependent on the pilot ratio for stability.

[0213] Regarding Figure 17 The communication performance comparison in (b) reveals that the performance of the two schemes differs with the pilot ratio. For low pilot ratios, the AL-ISAC scheme performs slightly better than the Bi-ISAC scheme and exhibits similar robustness to path obstruction, while for high pilot ratios, the Bi-ISAC scheme demonstrates significantly better performance than the AL-ISAC scheme. This prompts adjustments to the pilot ratio based on the puncturing process, which itself depends on the environment.

[0214] Therefore, in an embodiment of the method according to the third aspect of the invention, the puncturing factor or threshold determined for the communication connection between the AP and the UE in the ROI is provided as input to the transmitter of one or more APs undertaking the corresponding communication connection. As already shown above, the puncturing factor or threshold carries information about the critical angle, which, together with the pilot ratio, affects the VOER and BER performance of the receiver. The transmitter adapts the pilot ratio of the corresponding communication connection based on the received puncturing factor or threshold.

[0215] The methods presented above can be represented by computer program instructions of a computer program product. Therefore, according to a third aspect of the invention, a computer program product includes computer program instructions that, when executed by a processor of a receiver or a processor functionally coupled to the receiver, cause the processor or controls the receiver to perform the methods according to the first and / or second aspects of the invention.

[0216] These computer program instructions may be retrievably stored on or retrievably transmitted on a computer-readable medium or data carrier. The medium or data carrier may be physically embodied, for example, in the form of a hard disk, solid-state drive, flash memory device, etc. However, the medium or data carrier may also include modulated electromagnetic, electrical, or optical signals, which are received by the computer through a corresponding receiver and transmitted to and stored in the computer's memory.

[0217] According to a fourth aspect of the invention, a wireless communication device configured to process wireless communication signals for transmission and / or processing wireless communication signals upon reception includes at least one antenna, circuitry for processing radio frequency signals, a microprocessor, volatile memory, and non-volatile memory, these elements or components being connected via one or more data lines and / or signal lines or buses. The non-volatile memory stores computer program instructions that, when executed by the microprocessor, configure the elements or components of the wireless communication device to implement or perform the methods according to the first, second, and / or third aspects of the invention.

[0218] In one or more embodiments, the wireless communication device includes a receiver co-located with the transmitter.

[0219] In one or more embodiments, the circuitry of the wireless communication device for processing radio frequency signals includes a low-noise amplifier and / or a mixer configured to provide a representation of the received signal at an intermediate frequency. The mixer preferably uses the same oscillator signal as the transmitter located co-located with the receiver. The latter enables environmental perception using signals reflected from objects emitted by an entity including the receiver.

[0220] Two or more communication devices according to the fourth aspect of the invention can form a system that allows communication and estimation of NLOS channels according to the method presented above. The signals used in the system preferably have at least one pilot signal.

[0221] This invention advantageously allows for more realistic and accurate modeling of voxelized environment models that take into account path blocking, scattering, behavior based on non-ideal geometry, unrealistic reflection angles, energy loss during scattering, other frequency-dependent behaviors, and geometry-only channel path loss for use in communication systems. This not only enables significantly improved and more accurate channel estimation but also reduces computational effort because unavailable and / or unrealistic paths are removed from the process, i.e., not considered. Statistical methods (where an attempt is made to account for the fact that scattering is statistically possible at any voxel) can further refine the model, yielding the most probable voxel states. Improved channel estimation contributes to improved and faster symbol detection, and in particular opens the path to lighter forward error correction, which can translate into higher spectral efficiency.

[0222] Unlike previous VE models, the method according to the invention further considers the behavior during scattering. Typically, it can be assumed that scattering of the incident signal wave exists at the locations of at least some voxels. Prior art assumes that scattering occurs without energy loss and occurs equally in all directions. In contrast, the method according to the invention captures the actual physical behavior of electromagnetic waves during scattering. This allows for modeling of scattering paths based on geometry and frequency, and further allows for modeling of energy loss during scattering. With this information, unrealistic scattering paths can be punched based on physical electromagnetic wave theory (i.e., Mie or Rayleigh scattering formulas, Maxwell's equations), and the scattering coefficients of voxels can be extended from the binary 0 or 1 found in prior art methods to more realistic continuous values ​​between 0 and 1, or even complex values.

[0223] Finally, the method according to the invention allows for adjustment of the intensity of unrealistic (i.e., blocked) paths and scattering by adjusting the "drilling threshold" input to the method. This adds a margin range to the calculation of blocked paths and scattering angles to provide more lenient drilling.

[0224] The methods presented herein are applicable to any scenario where the environment is modeled using a voxelized mesh model, i.e., scenarios that may be considered in a joint communication and sensing setup. In particular, the invention can be used in indoor scenarios with fixed access points, for communication and environmental detection via: mobile vehicles communicating with roadside units (RSUs) to simultaneously achieve vehicle / pedestrian detection, for multiple vehicles collaboratively sensing environmental and road conditions in the absence of RSUs, for multiple connected UEs (Bluetooth, Wi-Fi, IoT, etc.) to passively sense the environment (i.e., without using specific sensing signals), and so on. The use cases and benefits of the communication and environmental sensing methods presented herein relate to the same wireless interface and can be implemented within that wireless interface. Attached Figure Description

[0225] The figures in the accompanying drawings are used to illustrate various aspects of the invention. In the drawings:

[0226] Figure 1 An exemplary environment with objects in the region of interest and its voxelized representation are shown.

[0227] Figure 2 A representation of the general concept of LOS and NLOS paths in voxel space is shown.

[0228] Figure 3 A schematic representation of the 3D space considered in known systems and methods is shown.

[0229] Figure 4 The process modules of the prior art JCAS method, which iterates using a sliding window approach, are shown.

[0230] Figure 5 A schematic representation of an unavailable and impractical communication path in voxelized space is shown.

[0231] Figure 6 An empirical β-mixing model of the scattering angle distribution is shown.

[0232] Figure 7 The critical angle θ is shown. 临界 The impact of channel matrix perforation severity at different voxel grid resolutions

[0233] Figure 8 A factor plot of a linear system, designed for estimating the voxel coefficients v, is shown.

[0234] Figure 9 An exemplary flowchart is shown, illustrating the method steps invoked to estimate the voxel coefficients v.

[0235] Figure 10 The factor plot of a linear system, designed for estimating the signal matrix X, is shown.

[0236] Figure 11 An exemplary flowchart of the method steps invoked to estimate the signal matrix X is shown.

[0237] Figure 12 A schematic flowchart of an exemplary AL-ISAC method is shown.

[0238] Figure 13 A schematic diagram of an exemplary Bi-ISAC method is shown.

[0239] Figure 14 The ternary factor plot of the bilinear system is shown.

[0240] Figure 15An exemplary flowchart of an exemplary Bi-ISAC method is shown.

[0241] Figure 16 The VOER and BER performance of exemplary AL-ISAC and Bi-ISAC methods at different SNR values ​​are shown as a function of pilot length ratio ρ.

[0242] Figure 17 The VOER and BER performance of exemplary AL-ISAC and Bi-ISAC methods with different pilot length ratios ρ as a function of critical angle θ are shown. * Changes,

[0243] Figure 18 A schematic overview of the inputs and outputs of the method according to the first aspect of the invention is shown.

[0244] Figure 19 An exemplary block diagram of a wireless communication device according to a fourth aspect of the present invention is shown.

[0245] Figure 20 An exemplary flowchart of a method according to a first aspect of the present invention is shown.

[0246] Figure 21 An exemplary flowchart of a method according to a second aspect of the present invention is shown, and

[0247] Figure 22 An exemplary flowchart of a method according to a third aspect of the present invention is shown. Detailed Implementation

[0248] Figures 1 to 17 This has already been described in more detail above and will not be discussed further.

[0249] Figure 18 A schematic overview of the inputs and outputs of a method according to a first aspect of the invention is shown. The inputs to the method are physical parameters (e.g., spatial locations of the UE and AP) and configurable parameters (e.g., voxel cell size and punching threshold). The output is a voxelized environment model that includes scattering voxels and actually available channel paths.

[0250] Figure 19An exemplary block diagram of a wireless communication device 400 according to a fourth aspect of the present invention is shown. The wireless communication device 400 includes at least one antenna 402, circuitry 404 for processing radio frequency signals, a microprocessor 412, volatile memory 414, and non-volatile memory 416. The aforementioned components are communicatively connected via at least one signal or data connection or bus 418. The non-volatile memory 416 stores computer program instructions that, when executed by the microprocessor 412, cause the wireless communication device 400 to implement or perform embodiments of the methods according to the first or second aspect of the present invention.

[0251] Figure 20 An exemplary flowchart of a method 100 according to a first aspect of the present invention is shown. In step 110, information about the spatial dimensions of voxels within the VE, information about the spatial locations of UEs and APs within the voxels of the VE, and a punching threshold are received. In step 120, considering all voxels in the VE of the ROI, all theoretically possible geometric paths between each UE and each AP are modeled. In step 130, the properties of the communication channels corresponding to the corresponding modeled geometric paths are determined. In step 140, unavailable and / or unrealistic paths are identified. In step 160, a VE model is generated, and unavailable and / or unrealistic paths are removed from the VE model according to the received punching threshold. In step 170, the VE model, the prior distribution of environmental voxels and their corresponding occupancy parameters obtained from the VE model, the feasibility coefficient matrix of the corresponding channel matrices for all available and / or realistic signal paths obtained from the VE model, and / or the channel matrix punched according to the feasibility coefficient matrix are output.

[0252] Figure 21 An exemplary flowchart of a method 200 according to a second aspect of the invention is shown. In step 210, wireless signals received at the CPU at one or more antennas of one or more APs serving the ROI are received. In step 220, the method 100 according to a first aspect of the invention generates an VE model, a priori distribution of environmental voxels and their corresponding occupancy parameters based on such VE model, a feasibility coefficient matrix of corresponding channel matrices for all available and / or realistic signal paths based on such VE model, and / or a channel matrix punched according to the feasibility coefficient matrix, or receives such environmental information. In step 230, the environmental information and optionally the received wireless signals are provided as input to a JCAS process in a receiver associated with the ROI.

[0253] Figure 22An exemplary flowchart of a method 300 according to a third aspect of the present invention is shown. In step 310, the CPU receives from the VE model of the ROI determined by the method 100 according to the first aspect of the present invention, or a priori distribution of environmental voxels and their corresponding occupancy parameters based on such VE model, or at least one feasibility coefficient matrix based on such VE model for a corresponding channel matrix of a signal path between a transmitter and a user equipment (UE) located within the ROI. In step 320, at least one channel matrix of the signal path between the transmitter and the UE, punctured according to the received environmental information, is determined, or at least one such punctured channel matrix is ​​received from the CPU. In step 330, the at least one punctured channel matrix is ​​provided as input to an equalizer process and / or a precoder process in the transmitter to pre-equalize and / or pre-code the signal to be transmitted, respectively. Finally, in step 340, the signal is transmitted into the ROI.

Claims

1. A method (100) for generating a voxelized environment model for wireless communication signal processing in a region of interest (ROI), the ROI comprising N A ≥1 access point (AP) and N U ≥1 UE (UE), the N A Each of the APs has N R The method comprises ≥1 antenna, and all APs serving the ROI are communicatively connected to a central processing unit (CPU). The VE includes voxels arranged in a three-dimensional mesh. The CPU receives (110) information about the spatial dimensions of voxels within the voxelized environment (VE), information about the spatial locations of user equipment (UE) and access points (AP) within the voxels of the VE, and a punching threshold. Considering all voxels in the VE of the region of interest (ROI), model all theoretically possible geometric paths between each UE and each AP (120). Determine (130) the properties of the communication channel corresponding to the corresponding modeling geometric path. Identify (140) unavailable and / or impractical paths. Generate a (160) VE model, and remove unusable and / or unrealistic paths from the VE model based on the received punching threshold, and Provide (170) the VE model, the prior distribution of environmental voxels and their corresponding occupancy parameters obtained from the VE model, the feasibility coefficient matrix of the corresponding channel matrices of all available and / or realistic signal paths obtained from the VE model, and / or the channel matrix punched according to these feasibility coefficient matrices, for the process of receiving and processing wirelessly transmitted signals in the communication environment represented by the VE, or for the process of preprocessing signals to be wirelessly transmitted into the communication environment represented by the VE.

2. The method as described in claim 1, wherein, Identifying (140) unavailable and / or impractical paths includes: identifying all geometric paths between the UE and AP with more than a predetermined number of reflections, identifying all geometric paths with a total path length or path attenuation exceeding a predetermined value, and / or identifying at least one reflection with a scattering angle exceeding a predetermined critical scattering angle (θ). * All geometric paths of ).

3. The method of claim 2, further comprising when the critical angle (θ) * When it is frequency-dependent: The CPU receives information about the frequency and the corresponding critical scattering angle (θ). * Information about the relationship between ) The CPU receives information about the frequency or frequency band used in the current communication between the UE and the AP, and Consider the corresponding critical scattering angle (θ) * To determine (140) unavailable and / or impractical paths.

4. The method as described in any of the preceding claims, wherein, Determining (140) unavailable and / or unrealistic paths includes: for all antennas of all APs, and for each voxel of the VE or for each voxel of the VE that contributes to the availability or realism of the path, calculating the probability that a signal from the UE will scatter from that voxel toward the corresponding AP, and removing the unavailable and / or unrealistic paths accordingly based on the received punch threshold.

5. The method of claim 4, wherein, Calculating this probability involves evaluating the scattering angle at each voxel for all theoretically possible geometric paths between the AP, UE, and voxel's permissible locations within the ROI.

6. The method as described in claim 4 or 5, wherein, Determining that a path (140) is unavailable further includes: based on a predetermined critical scattering angle (θ) * The random blocking of NLOS subpaths is introduced in proportion to the complementary cumulative distribution of the probability of the signal from the UE scattering from the voxel toward the corresponding AP, and based on the scattering angle at each voxel of the VE or the scattering angle at each voxel of the VE that contributes to the available or realistic path.

7. A method (200) for processing wireless communication signals for joint communication and environmental awareness (JCAS) in a region of interest (ROI), the ROI comprising N A ≥1 access point (AP) and N U ≥1 UE (UE), the N A Each of the APs has N R ≥1 antenna, and all APs serving the ROI are communicatively connected to a central processing unit (CPU), the ROI being represented by a voxelized environment (VE) comprising voxels arranged in a three-dimensional mesh, the method comprising: The CPU receives (210) wireless signals received at one or more antennas of one or more APs serving the ROI. Generate (220) an VE model, a priori distribution of environmental voxels and their corresponding occupancy parameters based on the VE model, a feasibility coefficient matrix of the corresponding channel matrices of all available and / or realistic signal paths based on the VE model, and / or a channel matrix punched according to the feasibility coefficient matrix, according to one or more of claims 1 to 6; or receive such environmental information. The environmental information, along with optionally received wireless signals, is provided as input (230) to the JCAS process in the receiver associated with the ROI.

8. A method (300) for processing wireless communication signals on the transmitter side, the method comprising performing the following operations at a transmitter communicatively connected to a central processing unit (CPU), the CPU being communicatively connected to an N serving a region of interest (ROI). A ≥1 access point (AP), the ROI is represented by a voxelized environment (VE) which consists of voxels arranged in a 3D mesh: The CPU receives (310) a VE model of the ROI determined by one or more of the methods (100) according to claims 1 to 6, or receives a prior distribution of environmental voxels and their corresponding occupancy parameters based on the VE model, or receives at least one feasibility coefficient matrix based on the VE model for the corresponding channel matrix of the signal path between the transmitter and the user equipment (UE) located within the ROI. (320) Determine at least one channel matrix punctured based on received environmental information for the signal path between the transmitter and the UE, or receive at least one punctured channel matrix from the CPU. The at least one punctured channel matrix is ​​provided as input (330) to the equalizer process and / or pre-encoder process in the transmitter to pre-equalize and / or pre-encode the signal to be transmitted, respectively. Transmit the signal (340) into the ROI.

9. The method of claim 8, further comprising providing a puncturing coefficient determined for the communication connection between the AP and the UE in the ROI as input to the transmitter of one or more APs undertaking the corresponding communication connection, and adapting the pilot ratio of the corresponding communication connection according to the puncturing coefficient.

10. A computer program product comprising computer program instructions that, when executed by a processor of a central processing unit (CPU), a receiver, or a transmitter, or a processor functionally coupled to the CPU, receiver, or transmitter, cause the processor or controls the CPU, the receiver, or the transmitter to perform the method as described in one or more of claims 1 to 6, 7, or 8 to 9.

11. A computer-readable medium or data carrier retrievably transmits or stores a computer program product as described in claim 10.

12. A wireless communication device (400) configured to process wireless communication signals during transmission and / or reception, the device comprising at least one antenna (402) connected via one or more data and / or signal lines or buses (418), circuitry (404) for processing radio frequency signals, a microprocessor (412), volatile memory (414), and non-volatile memory (416), wherein, The non-volatile memory stores computer program instructions that, when executed by the microprocessor, configure components of the wireless communication device to implement or perform the method as described in any one of claims 1 to 9.

13. The wireless communication device (400) as claimed in claim 12, wherein, The circuitry for processing radio frequency signals includes a low-noise amplifier and / or a mixer configured to provide a representation of the received signal at an intermediate frequency, preferably using the same oscillator signal as the co-located transmitter.

14. A communication system comprising two or more communication devices according to any one of claims 12 or 13.

Images