Authorization-free random access unified active device detection method in high-speed scene

Abstract

The invention relates to an unlicensed random access unified active device detection method in a high-speed scene, and belongs to the technical field of communication. According to the method, a GF-RA system model is established based on tensor CP decomposition, a Vandermonde structure of a factor matrix is explored and utilized through identical transformation, the ill-conditioned problem caused by large-scale equipment access is solved by adopting a subspace decomposition method, and the accuracy of factor matrix estimation is improved; and the active equipment detection is completed by using the reconstructed factor matrix design by means of an experience threshold and a known training precoder, so that the active equipment detection does not depend on complete channel state information any more and is simultaneously suitable for active and passive application scenarios.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically relating to a method for detecting unauthorized random access to a unified active device in high-speed scenarios. Background Technology

[0002] GF-RA (Grant-Free Random Access) is a novel multiple access technology that has garnered significant attention in recent years. Its primary goal is to enable efficient access for massive numbers of machine-type Internet of Things (IoT) devices, meeting the access requirements of future wireless communication networks for massive machine-type communication (mMTC). Research indicates that, to ensure communication reliability, traditional 4G networks employ frequent handshake processes in licensed random access mechanisms, leading to excessive signaling overhead and low energy and spectral efficiency for small data packet transmission. In contrast, the GF-RA mechanism can transmit information without waiting for base station scheduling, significantly reducing signaling overhead and access latency. It is well-suited to meet the demands of massive device access in post-5G and 6G applications, particularly those focused on the Internet of Things.

[0003] However, for the sake of GF-RA access and transmission efficiency, a large number of potential communication devices generally operate on non-orthogonal resources. This poses a significant technical challenge for base stations that need to accurately detect which devices are transmitting information (i.e., which devices are active). In recent years, scholars have proposed a series of solutions based on compressed sensing theory to address this challenge. The main idea is to fully utilize the inherent characteristics of sporadic communication in scenarios with massive device access. That is, the number of serving devices is large, but only a small number of devices simultaneously transmit short packets in a short period of time, which manifests as significant sparsity in the uplink transmission channel. Therefore, the receiving base station can use the received pilot signals to estimate the sparse channel, and then divide the elements in the estimated sparse channel matrix into blocks and apply empirical threshold decisions. If the value exceeds the threshold, the device corresponding to that block of the channel is considered to have transmitted information, i.e., the device is active. In summary, the GF-RA device detection scheme based on compressed sensing models the problem as a sparse matrix reconstruction problem. It uses traditional methods such as OMP (Orthogonal Match Pursuit) or AMP (Approximate Message Passing) to solve the sparse channel matrix, and then obtains the detection results of the active device set by calculating the support set of this sparse matrix. This scheme is generally referred to as the "joint active device detection and channel estimation" scheme. It can achieve satisfactory device detection errors based on obtaining complete CSI (Channel State Information). Applying non-orthogonal pilots to channel estimation and device detection can also significantly reduce the pilot overhead required to serve a large number of devices. However, most of these traditional schemes are currently geared towards traditional low-speed ground scenarios. Research shows that when facing emerging high-speed application scenarios such as high-speed rail, drones, and low-Earth orbit satellites, the channel estimation performance will be severely compromised by the Doppler effect, leading to a significant increase in active device detection errors.

[0004] In this situation, tensor decomposition, another classic channel estimation technique, has certain application potential. Its main idea is to fully utilize the angular domain sparsity and time domain sparsity of the system response in a multi-antenna MIMO (Multiple Input Multiple Output) system to construct a third-order tensor from the received pilot signal that can be decomposed using CP (CANDECOMP / PARAFAC). Based on rigorous mathematical derivation, this third-order tensor signal has been proven to be uniquely decomposed into a linear combination of a series of first-order tensors, with the channel parameters contained within these first-order tensors. Obtaining the specific values ​​of these first-order tensors generally involves reconstructing factor matrices. The most classic method is ALS (Alternating Least Squares) optimization, which iteratively calculates by fixing two factor matrices and updating another factor matrix. Finally, after reaching convergence, three factor matrices containing the channel parameters are obtained. Channel estimation is then completed by reconstructing the channel matrix using these parameters. The advantages of channel estimation based on tensor decomposition are mainly twofold: First, by fully utilizing the multidimensional low-rank structure of the channel, existing theoretical analysis and simulation results show that this method has lower computational complexity and better recovery performance compared to compressed sensing methods. Second, because this method decouples the parameters to estimate the channel, it can effectively guarantee the accuracy of channel parameter estimation even in high-speed scenarios affected by the Doppler effect, thus overcoming the performance loss caused by directly estimating CSI in such scenarios. Considering the current performance challenges faced by GF-RA and the active device detection scheme based on channel estimation, the above two advantages make the tensor decomposition method of great significance for development in GF-RA: active device detection, as a core problem of GF-RA, no longer needs to rely on the results of complete channel estimation, but only needs to use partial channel parameters, which significantly improves the efficiency of detection; at the same time, the detection module can be directly applied to both active and passive application deployments, which significantly improves the practicality of detection. However, tensor models also have their limitations. For example, ill-conditioned problems can occur when the factor matrix is ​​too large, leading to significant performance loss in traditional ALS optimization methods. This severely impacts their application in uplink multi-user systems, and GF-RA, which needs to serve a large number of potential devices, will face even greater challenges. Furthermore, how to detect device activity based on coupled partial channel parameters rather than complete channel estimation results under tensor models is a critical issue that urgently needs to be addressed.

[0005] It's worth noting that GF-RA can generally be divided into two types: active RA and passive RA. The main difference is that the former requires the specific detection of the unique ID of the active device, while the latter only cares about the specific information transmitted by the active device (such as weather forecasts). Therefore, the research approaches and design methods for active and passive RA are usually based on two independent frameworks. Although the aforementioned background on GF-RA focuses on active RA, research on passive RA applications has gradually emerged in recent years, especially in application scenarios such as UAV mapping. However, traditional passive RA research cannot be well compatible with active RA, making it difficult for a single GF-RA system to meet different practical needs. Therefore, there is an urgent need to design a unified detection method that can be applied to both active and passive RA.

[0006] In summary, designing a unified active device detection method for GF-RA based on tensor decomposition models in high-speed scenarios has significant research implications and practical value. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] The technical problem to be solved by this invention is how to provide a method for detecting unified active devices with unauthorized random access in high-speed scenarios, so as to solve the pathological problems caused by large-scale device access.

[0009] (II) Technical Solution

[0010] To address the aforementioned technical problems, this invention proposes a method for detecting unauthorized random access unified active devices in high-speed scenarios, characterized by the following steps:

[0011] Step S1: In the GF-RA system, a modeling method based on tensor CP decomposition is applied to establish a third-order tensor CP decomposition model, that is: the channel parameters are combined by utilizing the sparsity characteristics of the multi-antenna MIMO channel to obtain the low-rank structure of the tensor of the pilot signal received by the base station.

[0012] Step S2: Perform identity transformation on the aforementioned model, that is: based on the inherent properties of the tensor CP decomposition model, perform policy reorganization on the channel parameters, and obtain the Vandermonde structure of the factor matrix by performing identity transformation on the model.

[0013] Step S3: Reconstruct the factor matrix, that is: based on the received tensor signal and the Vandermonde structure of the factor matrix, perform subspace decomposition to obtain the reconstructed factor matrix;

[0014] Step S4, active device detection, namely: using the reconstructed factor matrix and the trained pre-encoder, a correlation method based on two-dimensional search is used to detect the estimated device activity factors, thereby obtaining a list of active devices.

[0015] (III) Beneficial Effects

[0016] This invention proposes a unified active device detection method for unlicensed random access in high-speed scenarios. Based on tensor CP decomposition, this invention establishes a GF-RA system model, discovers and utilizes the Vandermonde structure of the factor matrix through identity transformation, and employs subspace decomposition to solve the ill-conditioned problem caused by large-scale device access, thus improving the accuracy of factor matrix estimation. The obtained factor matrix is ​​designed to complete active device detection using empirical thresholds and a known trained precoder, thereby eliminating reliance on complete channel state information and making it applicable to both active and passive application scenarios.

[0017] Compared with the prior art, the beneficial effects of the present invention are: (1) it effectively solves the pathological problems caused by the access of a large number of potential devices; (2) it completes the detection of active devices without relying on the complete CSI, and maintains the effectiveness and practicality of the designed method while being applicable to both active RA and passive RA application scenarios. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the active device detection method provided by the present invention;

[0019] Figure 2 To demonstrate the device activity detection performance of the active device detection method according to the present invention under different signal-to-noise ratios;

[0020] Figure 3 The device activity detection performance of the active device detection method according to the present invention is evaluated under different training resource overheads. Detailed Implementation

[0021] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0022] This invention provides a method for detecting unlicensed random access unified active devices in high-speed scenarios based on a tensor decomposition model. It can solve the ill-conditioned problems caused by large-scale device access and further ensure the detection performance in fast time-varying channels in high-speed scenarios. At the same time, the detection of active devices no longer depends on the complete CSI and has the practicality to be applied to both active RA and passive RA application scenarios.

[0023] This invention provides a unified active device detection method for large-scale unlicensed random access in high-speed scenarios. The method establishes a GF-RA system model based on tensor CP decomposition, discovers and utilizes the Vandermonde structure of the factor matrix through identity transformation, and solves the ill-conditioned problem caused by large-scale device access by using the subspace decomposition method, thereby improving the accuracy of factor matrix estimation. The reconstructed factor matrix is ​​used to design active device detection with the help of empirical thresholds and known trained precoders, so that it no longer depends on complete channel state information and is applicable to both active and passive application scenarios.

[0024] The method for detecting unauthorized random access unified active devices in high-speed scenarios provided by this invention mainly includes the following steps:

[0025] Step S1: In the GF-RA system, a modeling method based on tensor CP decomposition is applied to establish a third-order tensor CP decomposition model. This involves combining channel parameters using the sparsity characteristics of multi-antenna MIMO channels to obtain the low-rank structure of the tensor of the base station received pilot signal. The specific method is as follows:

[0026] Consider selecting K subcarriers from a total of K0 subcarriers as pilot resources for each high-speed mobile IoT device, with a total of N frames, and introduce a device activity factor. (For active devices) , For inactive devices , Considering that the transmitting equipment and the receiving base station each have and Root antenna, and In a multi-radio chain MIMO system, the received pilot signal on the nth frame and kth subcarrier at the base station is determined. The vector representation of the total number of potentially connected devices. and the number of multipaths for each device Combined into the total number of transmission paths To facilitate quantitative modeling.

[0027] Furthermore, with the merging of transmission path numbers, the device activity factor will be... With path gain Combined into equivalent path gain ; the base station guidance vector With synthesizer matrix Merged into an equivalent receiver guidance vector Furthermore, based on the merging of transmission paths, it belongs to the first... Multipath, First The training precoder for each device is represented as follows: .

[0028] Next, based on the characteristics of the third-order tensor CP decomposition, the signals are merged according to the first-order frame number N to obtain the received pilot signal. The combined signal matrix on the k-th subcarrier is represented as follows: Then merge along the dimension of the number of training subcarriers K, and... Viewed as a slice of a third-order tensor along the subcarrier dimension (the total number of slices is K), the received pilot signal is obtained. Third-order tensor CP decomposition expression form At the same time, the composition of its corresponding three factor matrices is obtained.

[0029] Although the traditional ALS optimization method can reconstruct the above three factor matrices, The data is then used to further detect active devices. However, due to the large number of potential devices connected in GF-RA application scenarios, the three factor matrices are very large. Since the values ​​in the factor matrices originate from real-world problems, their errors are significantly amplified during the estimation process, ultimately leading to serious ill-conditioned problems. To address this issue, this invention utilizes a subspace-based decomposition method to reduce the dimensionality of the data. Research shows that although this method can effectively reconstruct large-sized factor matrices, it requires the factor matrices to possess a Vandermonde structure. Therefore, necessary identity transformations must be performed on the model established in step S1.

[0030] Step S2: Perform an identity transformation on the aforementioned model, namely: based on the inherent properties of the tensor CP decomposition model, perform policy reorganization on the channel parameters, and obtain the Vandermonde structure of the factor matrix by performing an identity transformation on the model; the specific method is as follows:

[0031] Using factor matrix The exhibited geometric series potential is extracted by taking the first term of each column vector and combining it with the equivalent path gain. Merged into a new equivalent path gain And reorganize it into the factor matrix that originally contained only one channel parameter. In this process, a factor matrix with a strict Vandermonde structure is obtained. and tensor signals New equivalent decomposition form This will result in a new factor matrix. It contains three coupled channel parameters, but the subsequent estimation method can solve this problem well.

[0032] Step S3: Reconstruct the three factor matrices That is, based on the received 3rd order tensor signal sum factor matrix The Vandermonde structure is subjected to subspace decomposition to obtain the reconstructed factor matrix; the specific method is as follows:

[0033] Select a set of intermediate parameters Used to process the received tensor pilot signal First dimension expansion Dimensional expansion yields And perform SVD (Singular Value Decomposition) on it, that is This yields three matrices for reconstruction. Furthermore, based on intermediate parameters To decompose a unitary matrix into two overlapping submatrices and ,Will Defined as Then it is the factor matrix. Generating matrix Similar matrices have the same eigenvalues; then, for Perform EVD (Eigenvalue Decomposition), that is This yields the unitary matrix used to reconstruct the factor matrix. Finally, using the obtained intermediate quantities, the three factor matrices are reconstructed based on the mixed product property of the Kronecker product, resulting in... And their relationship with the true factor matrix The correlation, including ranking differences. (Permutation matrix), scaling relation (Diagonal matrix) and error , and .

[0034] Step S4, Active Device Detection: This involves using the reconstructed factor matrix and the trained pre-encoder to detect the estimated device activity factors using a two-dimensional search-based correlation method, thereby obtaining a list of active devices. The specific method is as follows:

[0035] For the equivalent model obtained in step S2, the actual coupling is in the reconstructed factor matrix. The channel parameters are Doppler frequency shift. and leaving the corner It carries the activity information of the access device. Using a two-dimensional search-based correlation method, based on the different device ID detection requirements of active RA and passive RA application scenarios, active device detection is completed under the premise of preset experience threshold and known trained precoder.

[0036] Example 1:

[0037] This invention provides a method for detecting active devices in unlicensed random access in high-speed scenarios based on a tensor decomposition model. The main ideas are as follows: First, considering the Doppler effect of the channel and the device activity factor, a third-order tensor CP decomposition model is established. Second, the inherent properties of the CP decomposition model are used to perform an identity transformation on the system model to obtain the Vandermonde structure of the factor matrix. Then, using the geometric series property of the inherent elements of the Vandermonde matrix, tensor decomposition based on subspaces is performed to obtain the estimation results of the factor matrix. Finally, based on the obtained factor matrix, the active device factor is detected according to the coupling of channel parameters.

[0038] The present invention will be further described in detail below with reference to the accompanying drawings.

[0039] Appendix Figure 1 This demonstrates an application example of the GF-RA unified framework. Assume the number of antennas in the base station is N. R The device has N antennas. T The total potential number of devices served by the system is U, and the probability of each device being active is ε; the number of multipaths for each device is L. u ;

[0040] The joint active device detection and channel estimation method based on the tensor CP decomposition model GF-RA unified framework described in this invention mainly includes the following steps:

[0041] Step S1: Apply a modeling method based on tensor CP decomposition to the GF-RA system, that is: combine the channel parameters by utilizing the sparsity characteristics of the multi-antenna MIMO channel to obtain the low-rank structure of the tensor of the base station received pilot signal; the specific method is as follows:

[0042] Consider selecting K subcarriers from a total of K0 subcarriers as pilot resources for each device, with a total of N frames. Then, the received pilot signal on the nth frame and the kth subcarrier can be expressed as: (1)

[0043] in, It is the activity factor of device u; it is 1 if the device is active and 0 if the device is inactive. It is the synthesizer matrix at the base station; It is the channel of each device on the kth subcarrier in the nth frame; It is the training precoder for each device, which is obtained by multiplying the pilot signal and the precoding matrix; It is AWGN (Additive White Gaussian Noise). It is the system sampling rate; These are the guidance vectors for the receiver (base station) and transmitter (equipment), respectively; the channel parameters of the equipment are { , , , , }, respectively representing the first The first device The multipath channel gain, delay, angle of arrival, separation angle, and Doppler shift are considered. To facilitate further modeling, some of these parameters are combined, specifically including: the number of devices... and the number of multipaths for each device Combined into the total number of transmission paths ; to increase device activity factor With path gain Combined into equivalent path gain ; the base station guidance vector With synthesizer Merged into an equivalent receiver guidance vector Furthermore, based on the merging of transmission paths, it belongs to the first... Multipath, First The training precoder for each device is represented as follows: ;

[0044] By merging the first-order tensor (vector) signal represented by formula (1) along the dimension of frame number N, we can obtain the merged signal matrix of the received pilot signal on the k-th subcarrier, expressed as follows: ; further If we consider it as a slice of a third-order tensor along the subcarrier dimension (the total number of slices is K), then formula (1) can be modeled as a tensor CP decomposition model, expressed as: (2)

[0045] in, This is the equivalent transmitter guidance vector that couples the Doppler frequency shift and the separation angle, and it also includes the device's training pre-encoder. Information; This is the equivalent delay guide vector for the coupling path gain and delay. Defined as: (3) It is a third-order tensor quantized AWGN; These are the tensor CP decomposition models that can be used to reconstruct the receiving tensor. The three factor matrices are defined as follows:

[0046] (4)

[0047] Step S2: Perform an identity transformation on the model described in formula (2), that is: based on the inherent properties of the tensor CP decomposition model, perform policy recombination on the channel parameters to obtain the Vandermonde structure of the factor matrix; the specific method is as follows:

[0048] Utilizing the geometric series potential shown by formula (3), the first term is extracted and combined with the equivalent path gain. Merged into a new equivalent path gain Therefore, the time-delay guidance vector is rewritten as:

[0049] (5)

[0050] at the same time, Reassembled into an equivalent receiver guidance vector with only one independent channel parameter The equivalent transformation formula (2) can be expressed as:

[0051] (6)

[0052] in The new factor matrix combinations after parameter rearrangement are defined as follows:

[0053] (7)

[0054] The new factor matrix The Vandermonde matrix can be generated by a set of different generators. To represent; in addition, the new factor matrix It becomes a factor matrix that simultaneously couples three channel parameters: path gain, time delay, and angle of arrival. However, subsequent steps show that this does not affect the parameter estimation as long as the method is appropriate.

[0055] Step S3: Reconstruct the three factor matrices described in formula (7) That is, based on the received 3rd order tensor signal sum factor matrix The subspace decomposition method is used to perform subspace decomposition on the Vandermonde structure; the specific method is as follows:

[0056] Select a set of intermediate parameters Make it satisfy This forms a cyclic selection matrix:

[0057] (8)

[0058] Using the aforementioned circular matrix to receive tensor signals First dimension expansion Dimensional expansion is represented as:

[0059] (9)

[0060] in, It is an N-order identity matrix. Next, the dimension-expanded matrix... Perform SVD, i.e. This yields the unitary matrix used to reconstruct the factor matrix. and and diagonal array Furthermore, based on the aforementioned selected intermediate parameters... To decompose a unitary matrix into two overlapping submatrices and , respectively defined as

[0061]

[0062] (10)

[0063] in and They are respectively OK, Submatrices of columns.

[0064] Then define the aforementioned geometric series generator. The generated matrix ,in This represents a diagonal matrix whose diagonal elements are formed by the vectors enclosed in parentheses. Combined with the generating matrix... Based on the expression and the result of formula (10), the following key transformation relationship can be obtained under the premise of the Vandermonde structure:

[0065] (11)

[0066] in For something that must exist A non-singular matrix of order. A slight rearrangement of formula (11) yields:

[0067] (12)

[0068] Formula (12) is a typical expression of matrix similarity. Defined as Then it is obvious Similar to They have the same eigenvalues. Then... Perform EVD, i.e. This yields the unitary matrix used to reconstruct the factor matrix. ,in It is a rearranged eigenvalue matrix, which is related to the generating matrix. The only difference lies in the order, while unitary matrices... Non-singular matrices In addition to this, there is a scaling relationship, which can be expressed as follows:

[0069]

[0070] (13)

[0071] in This indicates retrieving the diagonal elements of the matrix. Let be the permutation matrix representing the ordering differences. This is a diagonal matrix representing the scaling relationship.

[0072] After completing the processing of the intermediate quantities mentioned above, from the factor matrix with the Vandermonde structure The reconstruction of the factor matrix begins. Specifically, the reconstructed factor matrix is ​​shown in formula (13). With the true factor matrix The only difference between them is the order, which is represented as:

[0073] (14)

[0074] In obtaining To continue reconstructing the remaining factor matrices, we first need to utilize the previously selected set of intermediate parameters again. right Take the front of each line and front Rows as submatrices are defined as follows: and Then, define their first... Listed as:

[0075]

[0076] (15)

[0077] The preceding 【1:Ω1】 indicates selecting rows 1 to Ω1; the following

:

[0078] At this point, we have completed the reconstruction of the factor matrix. and All preparations, following the method of column-by-column restoration, their first... The columns can be represented as follows:

[0079]

[0080] (16)

[0081] in The number of radio frequency chains provided for the base station for An identity matrix of order n. The resulting factor matrix. With the true factor matrix They are not completely equal; the relationship between them is expressed as:

[0082] (17)

[0083] in , and These represent the estimation errors of the three factor matrices, and their impact is negligible when the estimation results are accurate.

[0084] Step S4, active device detection, namely: using the reconstructed factor matrix and the trained preencoder, a correlation method based on two-dimensional search is used to detect active devices; the specific method is as follows:

[0085] The parameter estimation problem can be expressed as:

[0086] (18)

[0087] in for The first of the matrix diagonal elements, It is a diagonal matrix containing Doppler frequency shift information. For the equivalent device-side guide vector, This represents the matrix representation of the trained precoder after merging along the dimension of frame number N, where the above variables satisfy... .

[0088] Then, for the parameter to be estimated in formula (18) Propose Doppler frequency shift and leaving the corner Initial values ​​for these two parameters during iteration:

[0089]

[0090] (19)

[0091] Without considering full channel estimation, and focusing only on the active device detection problem, this approach processes only specific portions of the channel parameters. (Under consideration) It contains information about the device's activity, and and Since this information is known to the base station, it is broken down into its original product form for a two-dimensional search, as shown below:

[0092] (20)

[0093] in Defined as the first Is the path a transmission path of an active device (i.e., Functions related to active paths. Utilizing empirical thresholds. The settings for active device detection can be represented as follows:

[0094] (twenty one)

[0095] That is, when a set is found When the value of makes the maximum value of the correlation function greater than the empirical threshold, this is the th... A path is identified as an active path based on the pre-encoder known at the base station. It can be known that this transmission path belongs to the first... If there are multiple devices, then that device is considered an active device. Otherwise, it is judged as an inactive device.

[0096] The aforementioned operations mainly target active RA applications. However, for passive RA applications that do not require detecting the unique ID of active devices, the pre-assigned preambles are all shared. At this point, the relevant active device detection method, as shown in formulas (20)-(21), remains effective, and degenerates from a two-dimensional search to a one-dimensional search, specifically expressed as:

[0097] (twenty two)

[0098] Using the same experience threshold Under this premise, the detection of active devices can be similarly represented as:

[0099] (twenty three)

[0100] From the appendix Figure 2 Simulation results show that the active device detection scheme proposed in this invention has significant performance advantages over the traditional ALS scheme at different signal-to-noise ratios. On the one hand, and These represent device activity probabilities of 10% and 20%, respectively. A higher device activity probability means more active devices will connect simultaneously, leading to greater inter-device interference. On the other hand, it also results in higher training resource overhead. This will improve the performance of the proposed solution to some extent. (Appendix) Figure 2 The AER (Activity Error Rate) performance metrics demonstrate the accuracy of active device detection. More device access and increased training resource overhead respectively worsen and improve device detection performance. The results also show that when the signal-to-noise ratio is high, the proposed solution can reduce the device detection error to 10. -4 Magnitude.

[0101] From the appendix Figure 3 Simulation results show that the active device detection scheme proposed in this invention has significant performance advantages over the traditional ALS scheme under different training resource overheads. When the number of training subcarriers increases from... Increase to Within this range, the AER performance of active device detection is significantly improved, reaching 10% under high training overhead. -4 Quantity, and device activity factor The impact of signal-to-noise ratio (SNR) on performance is also similar to the aforementioned simulation results. Figure 1 At the same time, AER performance is... The significantly slower rate of improvement afterward also indicates that continuously increasing training resource overhead does not always yield significant performance gains. Instead, seeking a reasonable trade-off between overhead and performance is a more practical choice.

[0102] As can be seen from the foregoing implementation methods, this invention fully exploits and utilizes the Vandermonde structure in the factor matrix by performing necessary identity transformations on the GF-RA unified framework established based on the tensor CP decomposition model. It adopts a subspace-based decomposition method to replace the traditional ALS optimization and obtains more accurate large-size factor matrix estimation results. Furthermore, for the channel parameters and device activity information contained in the factor matrix, it designs a method to detect active devices by using empirical thresholds and known trained precoders. This allows active device detection to no longer rely on complete channel state information and remains effective in both active and passive RA application scenarios. Simulation results show that the proposed active device detection scheme can achieve good detection performance.

[0103] Industrial applicability

[0104] This invention presents a unified active device detection method for large-scale unlicensed random access in high-speed scenarios. It establishes a GF-RA system model based on tensor CP decomposition, and utilizes the Vandermonde structure of the factor matrix through identity transformation. A subspace decomposition method is employed to address the ill-conditioned problem caused by large-scale device access, improving the accuracy of factor matrix estimation. The reconstructed factor matrix is ​​used to design active device detection with the aid of empirical thresholds and a known trained precoder, thus eliminating reliance on complete channel state information and making it applicable to both active and passive application scenarios. This invention has good industrial applicability.

[0105] Example 2:

[0106] A method for detecting unauthorized random access unified active devices in high-speed scenarios includes the following steps:

[0107] Step S1: In the GF-RA system, a modeling method based on tensor CP decomposition is applied to establish a third-order tensor CP decomposition model, that is: the channel parameters are combined by utilizing the sparsity characteristics of the multi-antenna MIMO channel to obtain the low-rank structure of the tensor of the pilot signal received by the base station.

[0108] Step S2: Perform identity transformation on the aforementioned model, that is: based on the inherent properties of the tensor CP decomposition model, perform policy reorganization on the channel parameters, and obtain the Vandermonde structure of the factor matrix by performing identity transformation on the model.

[0109] Step S3: Reconstruct the factor matrix, that is: based on the received tensor signal and the Vandermonde structure of the factor matrix, perform subspace decomposition to obtain the reconstructed factor matrix;

[0110] Step S4, active device detection, namely: using the reconstructed factor matrix and the trained pre-encoder, a correlation method based on two-dimensional search is used to detect the estimated device activity factors, thereby obtaining a list of active devices.

[0111] In particular, within a unified framework that can simultaneously serve both active and passive application scenarios of GF-RA (Grant-Free Random Access) systems, the sparsity characteristics of multi-antenna MIMO channels are utilized to combine channel parameters, thereby obtaining the low-rank structure of the tensor of the base station received pilot signal, and modeling is performed based on tensor CP decomposition.

[0112] Based on the inherent properties of the tensor CP decomposition model, the channel parameters are reorganized according to the strategy. By performing an identity transformation on the modeling, the Vandermonde structure of the factor matrix is ​​obtained. Then, the subspace decomposition method is carried out to decompose the received tensor signal and reconstruct the corresponding three factor matrices.

[0113] For different combinations of channel parameters, the reconstructed factor matrix and the trained precoder are used to detect the device activity factors to be estimated using a two-dimensional search-based correlation method, thereby obtaining a list of active devices.

[0114] Specifically, the sparsity characteristics of multi-antenna MIMO channels are utilized to combine channel parameters, thereby obtaining the low-rank structure of the tensor of the base station received pilot signal. Modeling is then performed based on tensor CP decomposition, specifically including:

[0115] Consider selecting K subcarriers from a total of K0 subcarriers as pilot resources for each device, with a total of N frames, and introduce a device activity factor. (For active devices) , For inactive devices , Considering that the transmitting equipment and the receiving base station each have and Root antenna, and In a multi-radio chain MIMO system, the received pilot signal on the nth frame and kth subcarrier at the base station is determined. The vector representation of the total number of potentially connected devices. and the number of multipaths for each device Combined into the total number of transmission paths To facilitate quantitative modeling.

[0116] Furthermore, with the merging of transmission path numbers, the device activity factor will be... With path gain Combined into equivalent path gain ; the base station guidance vector With synthesizer Merged into an equivalent receiver guidance vector Furthermore, based on the merging of transmission paths, it belongs to the first... Multipath, First The training precoder for each device is represented as follows: .

[0117] Next, based on the characteristics of the third-order tensor CP decomposition, the signals are merged according to the first-order frame number N to obtain the received pilot signal. The second-order tensor (matrix) representation on the k-th subcarrier Then merge along the dimension of the number of training subcarriers K, and... Viewed as a slice of a third-order tensor along the subcarrier dimension (the total number of slices is K), the received pilot signal is obtained. Third-order tensor CP decomposition expression form At the same time, the composition of its corresponding three factor matrices is obtained.

[0118] Specifically, based on the inherent properties of the tensor CP decomposition model, channel parameters are restructured using a strategy. An identity transformation is performed on the modeling to obtain the Vandermonde structure of the factor matrix. Then, a subspace decomposition method is used to decompose the received tensor signal, reconstructing the corresponding three factor matrices. This includes:

[0119] Using factor matrix The exhibited geometric series potential is extracted by taking the first term of each column vector and combining it with the equivalent path gain. Merged into a new equivalent path gain And reorganize it into the factor matrix that originally contained only one channel parameter. In this process, a factor matrix with a strict Vandermonde structure is obtained. and tensor signals New equivalent decomposition form .

[0120] Select a set of intermediate parameters Used to process the received tensor pilot signal First dimension expansion Dimensional expansion yields And perform SVD (Singular Value Decomposition) on it, that is This yields three matrices for reconstruction, among which... To obtain the conjugate transpose of the matrix; further, based on the intermediate parameters To decompose a unitary matrix into two overlapping submatrices and ,Will Defined as Then it is the factor matrix. Generating matrix Similar matrices have the same eigenvalues; then, for Perform EVD (Eigenvalue Decomposition), that is This yields the unitary matrix used to reconstruct the factor matrix. ,in To The eigenvalues ​​are rearranged to form a diagonal matrix; finally, using the obtained intermediate quantities, the three factor matrices are reconstructed based on the mixed product property of the Kronecker product, resulting in... And their relationship with the true factor matrix The correlation, including ranking differences. (Permutation matrix), scaling relation (Diagonal matrix) and error , and .

[0121] Specifically, the reconstructed factor matrix and the trained preencoder are used to detect the device activity factors to be estimated using a two-dimensional search-based correlation method, thereby obtaining a list of active devices, including:

[0122] Using the reconstructed factor matrix By analyzing the Doppler frequency shift and leaving the corner These two parameters are used for correlation detection based on two-dimensional search. Within a unified framework that considers both active and passive application scenarios, this addresses the different needs of device ID detection and fully utilizes the parameters contained in the factor matrix. The device activity information carried in the system enables active device detection that can flexibly switch between active and passive modes, under the premise of a preset experience threshold and a known trained precoder.

[0123] In summary, this invention establishes a GF-RA system model based on tensor CP decomposition, discovers and utilizes the Vandermonde structure of the factor matrix through identity transformation, and solves the ill-conditioned problem caused by large-scale device access by using the subspace decomposition method, thereby improving the accuracy of factor matrix estimation. The obtained factor matrix is ​​designed to complete active device detection by means of empirical thresholds and known trained precoders, so that it no longer depends on complete channel state information and is applicable to both active and passive application scenarios.

[0124] Compared with the prior art, the beneficial effects of the present invention are: (1) it effectively solves the pathological problems caused by the access of a large number of potential devices; (2) it completes the detection of active devices without relying on the complete CSI, and maintains the effectiveness and practicality of the designed method while being applicable to both active RA and passive RA application scenarios.

[0125] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting unauthorized random access to a unified active device in a high-speed scenario, characterized in that, The method includes the following steps: Step S1: In the GF-RA system, a modeling method based on tensor CP decomposition is applied to establish a third-order tensor CP decomposition model, that is: the channel parameters are combined by utilizing the sparsity characteristics of the multi-antenna MIMO channel to obtain the low-rank structure of the tensor of the pilot signal received by the base station. Step S2: Perform identity transformation on the aforementioned model, that is: based on the inherent properties of the tensor CP decomposition model, perform policy reorganization on the channel parameters, and obtain the Vandermonde structure of the factor matrix by performing identity transformation on the model. Step S3: Reconstruct the factor matrix, that is: based on the received tensor signal and the Vandermonde structure of the factor matrix, perform subspace decomposition to obtain the reconstructed factor matrix; Step S4, active device detection, namely: using the reconstructed factor matrix and the trained pre-encoder, a correlation method based on two-dimensional search is used to detect the estimated device activity factors, thereby obtaining a list of active devices.

2. The method for detecting unauthorized random access to a unified active device in a high-speed scenario as described in claim 1, characterized in that, S1 includes: Consider selecting K subcarriers from a total of K0 subcarriers as pilot resources for each high-speed mobile IoT device, with a total of N frames, and introduce a device activity factor. For active devices , For inactive devices , Considering that the transmitting equipment and the receiving base station each have and Root antenna, and In a multi-radio chain MIMO system, the received pilot signal on the nth frame and kth subcarrier at the base station is determined. The vector representation of the total number of potentially connected devices. and the number of multipaths for each device Combined into the total number of transmission paths To facilitate quantitative modeling; As the number of transmission paths is merged, the device activity factor will be... With path gain Combined into equivalent path gain ; the base station guidance vector With synthesizer matrix Merged into an equivalent receiver guidance vector Furthermore, based on the merging of transmission paths, it belongs to the first... Multipath, First The training precoder for each device is represented as follows: ; Based on the characteristics of third-order tensor CP decomposition, the received pilot signal is obtained by merging along the dimension of the first-order frame number N. The combined signal matrix on the k-th subcarrier is represented as follows: Then merge along the dimension of the number of training subcarriers K, and... Consider it as a slice of a third-order tensor along the subcarrier dimension; the total number of slices is K, thus obtaining the received pilot signal. Third-order tensor CP decomposition expression form At the same time, the composition of its corresponding three factor matrices is obtained.

3. The method for detecting unauthorized random access to a unified active device in a high-speed scenario as described in claim 2, characterized in that, S2 includes: utilizing a factor matrix The exhibited geometric series potential is extracted by taking the first term of each column vector and combining it with the equivalent path gain. Merged into a new equivalent path gain And reorganize it into the factor matrix that originally contained only one channel parameter. In this process, a factor matrix with a strict Vandermonde structure is obtained. and tensor signals New equivalent decomposition form .

4. The method for detecting unauthorized random access to a unified active device in a high-speed scenario as described in claim 3, characterized in that, S3 includes: Select a set of intermediate parameters Used to process the received tensor pilot signal First dimension expansion Dimensional expansion yields And perform SVD on it, that is This yields three matrices for reconstruction. Based on intermediate parameters To decompose a unitary matrix into two overlapping submatrices and ,Will Defined as Then it is the factor matrix. Generating matrix Similar matrices have the same eigenvalues; then, for Perform EVD, i.e. This yields the unitary matrix used to reconstruct the factor matrix. Finally, using the obtained intermediate quantities, the three factor matrices are reconstructed based on the mixed product property of the Kronecker product, resulting in... And their relationship with the true factor matrix The correlation, including ranking differences. , scaling relationship and error , and .

5. The method for detecting unauthorized random access to a unified active device in a high-speed scenario as described in claim 4, characterized in that, S4 includes: For the equivalent model obtained in step S2, the actual coupling is in the reconstructed factor matrix. The channel parameters are Doppler frequency shift. and leaving the corner It carries the activity information of the access device. Using a two-dimensional search-based correlation method, based on the different device ID detection requirements of active RA and passive RA application scenarios, active device detection is completed under the premise of preset experience threshold and known trained precoder.

6. The method for detecting unauthorized random access to a unified active device in a high-speed scenario as described in claim 1, characterized in that, S1 specifically includes: Consider selecting K subcarriers from a total of K0 subcarriers as pilot resources for each device, with a total of N frames. Then, the received pilot signal on the k-th subcarrier in the n-th frame is represented as follows: （1） in, It is the activity factor of device u; it is 1 if the device is active and 0 if the device is inactive. It is the synthesizer matrix at the base station; It is the channel of each device on the kth subcarrier in the nth frame; It is the training precoder for each device, which is obtained by multiplying the pilot signal and the precoding matrix; It is additive white Gaussian noise (AWGN); It is the system sampling rate; These are the guidance vectors for the receiver (or base station) and the transmitter (or device), respectively; the channel parameters of the device are { , , , , }, respectively representing the first The first device Channel gain, delay, angle of arrival, separation angle, and Doppler shift of the multipath; The above parameters are partially combined, specifically including: the number of devices. and the number of multipaths for each device Combined into the total number of transmission paths ; to increase device activity factor With path gain Combined into equivalent path gain ; the base station guidance vector With synthesizer Merged into an equivalent receiver guidance vector Furthermore, based on the merging of transmission paths, it belongs to the first... Multipath, First The training precoder for each device is represented as follows: ; The first-order tensor (vector) signal represented by formula (1) is merged along the dimension of frame number N to obtain the merged signal matrix of the received pilot signal on the k-th subcarrier, which is expressed as follows: ; further If we consider it as a slice of a third-order tensor along the subcarrier dimension, then we can model equation (1) as a CP decomposition model of the tensor, which is expressed as: （2） in, This is the equivalent transmitter guidance vector that couples the Doppler frequency shift and the separation angle, and it also includes the device's training pre-encoder. Information; This is the equivalent delay guide vector for the coupling path gain and delay. Defined as: （3） It is a third-order tensor quantized AWGN; These are the tensors used to reconstruct the received tensor under the tensor CP decomposition model. The three factor matrices are defined as follows: （4）。 7. The method for detecting unauthorized random access unified active devices in high-speed scenarios as described in claim 6, characterized in that, S2 specifically includes: Utilizing the geometric series potential shown by formula (3), the first term is extracted and combined with the equivalent path gain. Merged into a new equivalent path gain Therefore, the time-delay guidance vector is rewritten as: （5） at the same time, Reassembled into an equivalent receiver guidance vector with only one independent channel parameter The equivalent transformation formula (2) is expressed as: （6） in The new factor matrix combinations after parameter rearrangement are defined as follows: （7） The new factor matrix The Vandermonde matrix can be generated by a set of different generators. To express.

8. The method for detecting unauthorized random access unified active devices in high-speed scenarios as described in claim 7, characterized in that, S3 specifically includes: Select a set of intermediate parameters Make it satisfy This forms a cyclic selection matrix: （8） Using the aforementioned circular matrix to receive tensor signals First dimension expansion Dimensional expansion is represented as: （9） in, It is an N-order identity matrix; then, after dimension expansion... Perform SVD, i.e. This yields the unitary matrix used to reconstruct the factor matrix. and and diagonal array Furthermore, based on the aforementioned selected intermediate parameters... To decompose a unitary matrix into two overlapping submatrices and , respectively defined as （10） in and They are respectively OK, Submatrices of columns; Then define the aforementioned geometric series generator. The generated matrix ,in This represents a diagonal matrix whose diagonal elements are formed by the vectors enclosed in parentheses; combined to generate a matrix. The expression and the result of formula (10), under the premise of the Vandermonde structure, yield the following key transformation relationship: （11） in For something that must exist The order is a non-singular matrix; by slightly rearranging formula (11), we can obtain: （12） Formula (12) is a typical expression of matrix similarity. Defined as Then it is obvious Similar to They have the same eigenvalues; then... Perform EVD, i.e. This yields the unitary matrix used to reconstruct the factor matrix. ,in It is a rearranged eigenvalue matrix, which is related to the generating matrix. The only difference lies in the order, while unitary matrices... Non-singular matrices In addition to this, there is a scaling relationship, which can be expressed as follows: （13） in This indicates retrieving the diagonal elements of the matrix. Let be the permutation matrix representing the ordering differences. This is a diagonal matrix representing the scaling relationship; After completing the processing of the intermediate quantities mentioned above, from the factor matrix with the Vandermonde structure Begin reconstructing the factor matrix; specifically, as shown in formula (13), the reconstructed factor matrix... With the true factor matrix The only difference between them is the order, which is represented as: （14） In obtaining To continue reconstructing the remaining factor matrices, we first need to utilize the previously selected set of intermediate parameters again. right Take the front of each line and front Rows as submatrices are defined as follows: and Then, define their first... Listed as: （15） At this point, the reconstruction of the factor matrix is ​​complete. and All preparations, following the method of column-by-column restoration, the first... The columns are represented as follows: （16） in The number of radio frequency chains provided for the base station for The first-order identity matrix; the resulting factor matrix With the true factor matrix They are not completely equal; the relationship is expressed as: （17） in , and These represent the estimation errors of the three factor matrices, respectively. When the estimation results are relatively accurate, the impact is almost negligible.

9. The method for detecting unauthorized random access unified active devices in high-speed scenarios as described in claim 8, characterized in that, In S4, for the application scenario of active RA, the parameter estimation problem is expressed as: （18） in for The first of the matrix diagonal elements, It is a diagonal matrix containing Doppler frequency shift information. For the equivalent device-side guide vector, This represents the matrix representation of the trained precoder after merging along the dimension of frame number N, where the above variables satisfy... ; Then, for the parameter to be estimated in formula (18) Propose Doppler frequency shift and leaving the corner Initial values ​​for these two parameters during iteration: （19） Without considering full channel estimation, and focusing only on the active device detection problem, only specific channel parameters are processed; under consideration... It contains information about the device's activity, and and Since this information is known to the base station, it is broken down into its original product form for a two-dimensional search, as shown below: （20） in Defined as the first Functions related to whether a path is a transmission path of an active device; using empirical thresholds. The settings for active device detection are as follows: （21） That is, when a set is found When the value of makes the maximum value of the correlation function greater than the empirical threshold, this is the th... A path is identified as an active path based on the pre-encoder known at the base station. It can be known that this transmission path belongs to the first... If there are multiple devices, then that device is considered an active device. Otherwise, it is judged as an inactive device.

10. The method for detecting unauthorized random access unified active devices in high-speed scenarios as described in claim 9, characterized in that, In S4, for passive RA application scenarios that do not require detection of the unique ID of active devices, the pre-assigned preambles of the devices are all shared, i.e. ; At this point, the relevant active device detection method, as shown in formulas (20)-(21), remains effective, and degenerates from a two-dimensional search to a one-dimensional search, specifically expressed as: （22） Using the same experience threshold Under the premise that active devices are detected, the detection method is as follows: （23）。

Images