A signal delay estimation method based on block processing

Abstract

The application discloses a signal delay estimation method based on block processing, comprising the following steps: acquiring CSI data and preprocessing (validity test, noise filtering, selecting valid index); obtaining overlapping data blocks by block processing according to a preset block quantity and overlapping ratio; building Hankel matrix and enhancement matrix for each block, and obtaining delay estimation value through singular value decomposition, subspace separation and characteristic polynomial solving; assigning weight to each estimation value in combination with signal quality evaluation; and determining the final result based on the delay estimation value and weight information. In the application, a large matrix is disassembled through block overlapping, the singular value decomposition complexity and storage occupation are reduced, and lightweight hardware is adapted; the processing time consumption is shortened, the real-time demand in multiple scenes is met, the application can be run on ordinary equipment, and positioning deviation caused by delay is avoided.

Description

Technical Field

[0001] This invention belongs to the field of wireless signal processing and positioning technology, specifically a signal delay estimation method based on block processing. Background Technology

[0002] In modern wireless communication and positioning systems, signal delay estimation technology based on CSI (Content Component Analysis) has become a key support for achieving high-precision positioning. Its core lies in extracting time delay information during signal propagation from CSI data using signal processing algorithms, thereby determining the distance between transmitting and receiving devices. The traditional MP (Multi-Level Video) algorithm, a classic estimation method in this field, uses a path matrix to extract signal features, separates the signal subspace and noise subspace through matrix factorization, and finally solves for the signal delay parameter based on the subspace characteristics. However, as wireless positioning scenarios evolve towards higher real-time performance, higher robustness, and lower resource consumption, traditional algorithms have gradually revealed a series of technical shortcomings in practical engineering applications.

[0003] First, the problem of excessive computational complexity is particularly prominent. The computational load of traditional algorithms is mainly concentrated on the accuracy of matrix factorization estimation for large-dimensional matrices. CSI data typically needs to be of considerable length, which leads to a significant increase in the dimensionality of the matrices being processed. The computational complexity of matrix factorization increases with increasing matrix dimension, which not only makes the algorithm processing time far exceed the requirements of real-time applications, but also prevents the algorithm from being deployed on mobile devices or IoT terminals with limited storage resources and computing power, greatly limiting its application scope.

[0004] Secondly, the traditional block-based processing strategy introduced to reduce computational complexity has led to severe block boundary effects. Traditional block-based methods divide continuous CSI signals into unrelated data segments. This approach artificially truncates signal features across block boundaries. In multipath propagation environments, wireless signals reach the receiver via multiple paths. The signal features of different paths are often distributed across a continuous time domain. If key features fall within block boundaries, they will be dispersed into adjacent data blocks, preventing a single data block from fully reflecting the signal characteristics and resulting in a significant decrease in delay estimation accuracy.

[0005] Furthermore, the result selection strategy of traditional algorithms is too simplistic, further exacerbating the instability of the estimation results. Traditional methods generally adopt a simple selection strategy to determine the final delay estimate. This strategy does not fully consider the differences in signal quality between different data blocks. In scenarios with strong noise interference, some data blocks may produce estimates with large deviations due to noise contamination. The simple selection strategy is highly susceptible to outliers, resulting in drastic fluctuations in the estimation results and insufficient robustness.

[0006] Finally, traditional block-based methods have shortcomings in their handling of data blocks at different locations. In a block sequence, data blocks at different locations exhibit varying signal integrity and quality due to differing levels of contextual support. However, traditional methods apply the same processing strategy to all data blocks, failing to address the specific characteristics of data blocks at different locations. This results in estimations of varying quality being equally included in the final fusion, impacting the overall estimation accuracy and stability.

[0007] In summary, the shortcomings of traditional algorithms in terms of computational efficiency, estimation accuracy, and robustness can no longer meet the technical requirements of current wireless positioning systems for a balance between high accuracy and low complexity, as well as strong anti-interference capabilities. Therefore, there is an urgent need to propose an optimized solution that can systematically address the above-mentioned problems. Summary of the Invention

[0008] The purpose of this invention is to provide a signal delay estimation method based on block processing, so as to solve the problems of low computational efficiency, poor estimation accuracy and insufficient robustness of traditional algorithms proposed in the background art.

[0009] A signal delay estimation method based on block processing includes the following steps:

[0010] Step S1: Obtain Channel State Information (CSI) data, perform validity checks and noise filtering on the CSI data, and select a valid index range to obtain preprocessed CSI data.

[0011] Step S2: Based on the preset number of blocks and overlap ratio, the preprocessed CSI data is divided into blocks to obtain several overlapping data blocks.

[0012] Step S3: Delay estimation is performed on each overlapping data block to obtain the delay estimate value of each data block; specifically: a Hankel matrix is ​​constructed for each data block, and the Hankel matrix is ​​vertically concatenated with its conjugate transpose matrix to construct an enhancement matrix. Singular value decomposition is performed on the enhancement matrix, and the signal subspace and noise subspace are separated based on the decomposition result. Then, the delay estimate value is obtained by constructing a characteristic polynomial and solving its roots.

[0013] Step S4: Calculate the final weight for each delay estimate. Specifically, based on the singular values ​​obtained from singular value decomposition, calculate the quality weight reflecting the signal quality of the data block by analyzing the ratio relationship between the singular values, allocate the position weight according to the spatial position of the data block in the block sequence, and combine the quality weight and the position weight to obtain the final weight.

[0014] Step S5: Determine the final delay estimation result based on the delay estimate and the corresponding final weight.

[0015] According to the above technical solution, in step S2, the preprocessed CSI data is divided into blocks to obtain several overlapping data blocks. Specifically, the number of effective blocks required to cover the total length of the CSI data without overlap is calculated first, and then the block size of each data block is determined in combination with the number of effective blocks. Next, the overlap size of adjacent data blocks is calculated, and finally, an index range is allocated to each data block and the corresponding data block is extracted.

[0016] According to the above technical solution, in step S2, the calculation of the effective number of blocks is based on the total length of the preprocessed CSI data, combined with the preset number of blocks and overlap ratio, and is derived by deriving the theoretical number of blocks required to cover the total length in a non-overlapping scenario.

[0017] According to the above technical solution, in step S2, the overlap size of adjacent data blocks is calculated by multiplying the block size of each data block by a preset overlap ratio, and the overlap size must satisfy the requirement that signal features crossing block boundaries can be repeatedly captured in adjacent data blocks.

[0018] According to the above technical solution, in step S3, when constructing the Hankel matrix, the value range of the row number parameter L of the Hankel matrix is ​​1 / 4 to 1 / 2 of the block size.

[0019] According to the above technical solution, in step S3, the enhancement matrix is ​​constructed by vertical splicing, specifically: the Hankel matrix constructed for the data block is used as the upper part, and the conjugate transpose of the Hankel matrix is ​​used as the lower part. After splicing, an enhancement matrix with a dimension of 2 × the number of rows of the Hankel matrix is ​​formed.

[0020] According to the above technical solution, the calculation of quality weights in step S4 includes the following sub-steps: extracting the top s values ​​from the singular value decomposition results. n There are 1 maximum singular values, s n Given a preset number of singular values ​​to participate in the calculation, the ratio of each singular value to the largest singular value is calculated, and the average of all the ratios is taken as the quality weight of the corresponding data block.

[0021] According to the above technical solution, in step S4, the weight allocation takes into account the positional characteristics of the data block in the processing sequence, and adopts a corresponding weight strategy for data blocks at different positions.

[0022] According to the above technical solution, in step S5, a threshold parameter is set based on a preset screening strategy.

[0023] According to the above technical solution, in step S5, the final delay estimation result is determined based on the delay estimate and the corresponding weight. Specifically, the delay estimate is sorted and the final result is determined by a filtering mechanism in combination with the weight information.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] In this invention, an innovative block processing mechanism is used to decompose the large-dimensional matrix that traditional algorithms need to process into multiple small-dimensional sub-matrices, which greatly reduces the computational complexity of matrix decomposition and significantly reduces the algorithm's occupation of storage resources. This effectively alleviates the pressure on resource-constrained devices in terms of computing power and storage, and lays the foundation for the deployment of the algorithm on various lightweight hardware.

[0026] In terms of real-time performance, thanks to the reduction in computational complexity, the overall processing time of the algorithm is significantly shortened, which can meet the real-time response requirements of scenarios such as wireless positioning, indoor navigation, and IoT device positioning. It can run smoothly on common hardware such as ordinary mobile terminals and IoT sensing devices, avoiding positioning deviations or application experience degradation caused by processing delays.

[0027] In terms of numerical stability, traditional algorithms are prone to numerical oscillations and accuracy loss when dealing with large-dimensional matrices. However, this invention controls the matrix size within a reasonable range through block processing, fundamentally avoiding such numerical problems, ensuring the consistency and reliability of delay estimation results in different scenarios, and improving the stability of algorithm output. Attached Figure Description

[0028] Figure 1 This is a flowchart of a signal delay estimation method based on block processing according to the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1

[0031] like Figure 1 As shown, this invention proposes a complete technical solution covering data preprocessing, block processing, single-block estimation, weight allocation, and result filtering. Through the coordinated optimization of each step, it achieves an optimal balance between computational efficiency, estimation accuracy, and robustness. Specific technical details are as follows:

[0032] Step 1: CSI data preprocessing.

[0033] In this step, CSI data serves as the input basis for the algorithm, and its quality directly determines the upper limit of the accuracy of subsequent processing. Therefore, it is necessary to perform systematic preprocessing operations to eliminate invalid information and noise interference in the original data, so as to provide high-quality data support for block processing and delay estimation.

[0034] First, data validity verification is performed: During the acquisition process, raw CSI data may experience data loss or abnormal jumps due to hardware sampling errors, packet loss, or other issues. Invalid data segments need to be identified through continuity detection. For consecutive invalid data segments, interpolation is used for repair. This method constructs a repair model based on valid data points before and after the invalid segment, ensuring repair accuracy while avoiding the introduction of additional noise.

[0035] Secondly, noise filtering is performed: CSI data usually contains quantization noise and environmental interference noise. A low-pass filter is used to filter the data. The cutoff frequency of the filter is set according to the sampling frequency characteristics, which can effectively suppress high-frequency noise and avoid signal characteristic distortion caused by over-filtering.

[0036] Finally, perform valid index selection: Since the frequency domain edges of CSI data are easily affected by the characteristics of RF front-end filters and have poor signal stability, it is necessary to remove unstable data segments at the frequency domain edges and retain only the valid data in the middle area to ensure that the data processed subsequently has high stability and reliability.

[0037] Step 2: Block overlap processing mechanism.

[0038] To address the boundary effects and computational complexity issues of traditional block partitioning, this invention designs a block overlap mechanism. By rationally partitioning and overlapping coverage, the integrity of signal features is guaranteed while reducing computational complexity.

[0039] First, initialize the parameters, pre-setting the number of blocks N according to application requirements. block N block The overlap ratio r is used to determine the total length Ndata of the preprocessed CSI data. The choice of the number of blocks needs to balance the amount of data per block with the total number of blocks processed. If the number of blocks is too small, the amount of data per block will be too large, which will not effectively reduce the computational complexity; if the number of blocks is too large, the amount of data per block will be insufficient, which will lead to insufficient signal feature extraction.

[0040] Then calculate the number of valid blocks N. effective This parameter reflects the theoretical number of blocks required to cover all CSI data in a non-overlapping scenario, and its calculation formula is as follows:

[0041]

[0042] In the formula, Ndata represents the length of the input CSI data sequence, r represents the overlap ratio, and N effective Indicates the number of valid blocks.

[0043] The core purpose of this calculation method is to ensure that the subsequently determined block size can completely cover Ndata with all overlapping data blocks, avoiding data omissions or redundancy. Based on the effective number of blocks, the block size of each data block is further determined as block_size:

[0044]

[0045] In the formula, N block This indicates the preset number of blocks, and Ndata represents the length of the input CSI data sequence. effective This indicates the number of valid blocks, and block_size indicates the size of each data block.

[0046] Consistency in block size is an important prerequisite for subsequent processing. If there are differences in block size, the accuracy of signal feature extraction in different blocks will vary, which will affect the reliability of the final estimation result.

[0047] The overlap size of adjacent data blocks, overlap_size, is calculated using the following formula:

[0048]

[0049] In the formula, block_size represents the block size of each data block, and r represents the overlap ratio.

[0050] The overlap size design must ensure that signal features crossing block boundaries can be repeatedly captured in at least two adjacent data blocks, so that signal features crossing boundaries can be fully extracted in both data blocks, thereby eliminating the boundary effect of traditional block partitioning.

[0051] Finally, data block index allocation is performed: for the i-th (i=1,2,…,N) block ) data block, its starting index start i The calculation logic is as follows:

[0052]

[0053] End of index i Then it is:

[0054]

[0055] This index allocation method ensures that there is exactly an overlap of overlap_size sampling points between adjacent data blocks, and that all data blocks can completely cover the preprocessed CSI data without any data omissions or duplications.

[0056] The choice of overlap ratio r needs to balance boundary effect suppression and computational efficiency: when r is too small, the overlapping area is insufficient to cover the signal features across the boundary, and boundary effects will still exist; when r is too large, there is too much overlap between adjacent data blocks, which leads to increased computational redundancy and prolonged algorithm processing time. Experiments have verified that a suitable overlap ratio can achieve a balance between boundary suppression and efficiency improvement.

[0057] Step 3: Single-block signal delay estimation.

[0058] For each data block obtained after block overlap processing, a signal processing algorithm is used to estimate the delay, obtaining a preliminary delay estimate for that data block. The core of this process lies in accurately extracting the signal delay information from the data block through matrix construction and decomposition. The specific operations are as follows:

[0059] First, construct a Hankel matrix H for the i-th data block xi (of length Ni). i The Hankel matrix effectively reflects the time-domain correlation characteristics of a signal. Its matrix elements H... i The values ​​of (m,n) are:

[0060]

[0061] In the formula, m and n are the row and column indices of the matrix. The dimension of the Hankel matrix is ​​determined by a key parameter L (number of rows), the selection of which must strike a balance between the sufficiency of signal feature extraction and computational complexity. To ensure both sufficiency of information extraction and computational efficiency, parameter L can be adaptively selected based on the data block length Ni.

[0062] To improve the numerical stability and estimation accuracy of subsequent processing, the Hankel matrix H can be... i Its conjugate transpose matrix H i Combine them to construct an enhancement matrix Y i :

[0063]

[0064] This combination method can make more comprehensive use of the complex characteristics of the signal, thereby improving the accuracy of subsequent signal and noise separation.

[0065] Next, for the enhancement matrix Y i Perform Singular Value Decomposition (SVD):

[0066]

[0067] Among them, U i and V i For unitary matrices, Σi The matrix is ​​a diagonal matrix, and the singular values ​​on its diagonal are arranged in descending order. SVD can effectively separate the information contained in a matrix into a signal subspace and a noise subspace. Typically, larger singular values ​​correspond to the signal subspace, while smaller singular values ​​correspond to the noise subspace.

[0068] Finally, based on the separated signal subspace information, the delay estimate is solved using the core logic of a super-resolution estimation algorithm (such as the Matrix Pencil algorithm). This process involves solving the delay estimate from the left singular vector matrix U. i The vectors corresponding to the preset number of signal paths are extracted to form a signal subspace matrix. Based on this subspace matrix, a characteristic polynomial is constructed. The delay estimate τ of the i-th data block is calculated by solving its roots. i This method can achieve high-precision delay estimation.

[0069] Step 4: Weighting of the delayed estimates.

[0070] To differentiate the reliability of latency estimates obtained from different data blocks, this invention designs a comprehensive weighting mechanism, which assigns a weight to each latency estimate τ. i Allocate appropriate weights. This mechanism aims to provide a scientific basis for subsequent screening steps.

[0071] This weighting mechanism can comprehensively consider the intrinsic signal quality of a data block and its positional characteristics in the sequence.

[0072] On the one hand, signal quality is evaluated based on the inherent characteristics of the signal. This evaluation can utilize the singular value characteristics obtained from the SVD decomposition in step three. The degree of separation between the signal subspace and the noise subspace directly reflects the signal quality; therefore, a mapping relationship can be designed to quantify the distribution characteristics of singular values ​​(such as the relative magnitudes of singular values ​​in the signal subspace) into a quality weight. Generally, the better the signal quality, the higher this weight value.

[0073] On the other hand, consider the positional characteristics of data blocks within the block sequence. Due to the block overlap mechanism, data blocks located in the middle of the sequence have more complete contextual information than those at the edges, and their signal integrity is generally higher. Therefore, higher positional weights can be assigned to middle data blocks, and relatively lower positional weights to edge data blocks.

[0074] Finally, the evaluation results obtained from the two dimensions of signal quality and location characteristics are combined to generate the final weight w for each data block. i This combination approach enables a comprehensive assessment of the reliability of each estimation result.

[0075] Step 5: Final delay estimation based on weighted filtering.

[0076] To address the issue of traditional selection strategies being susceptible to outliers, this invention proposes a weighted filtering strategy to determine the most reliable final result from delay estimates of multiple data blocks. This strategy effectively combines the weight information of each estimation result to eliminate outliers.

[0077] First, the latency estimate τ for all data blocks. i Sort in ascending order to obtain the sorted delay sequence τ. sorted :

[0078]

[0079] Simultaneously, based on the sorting of delay values, their corresponding final weights wi are also sorted accordingly, resulting in a sorted weight sequence wsorted. Ascending sorting is based on the physical laws of wireless signal propagation, namely, the delay along a direct path is usually the minimum.

[0080] Next, the cumulative weights after sorting are calculated. Starting from the first weight after sorting, the cumulative sum of the first k weights is calculated sequentially to obtain the cumulative weight sequence cum_weights.

[0081] Next, a filtering threshold is set. This threshold can be determined based on the sum of the final weights of all data blocks and a preset threshold ratio α:

[0082]

[0083] The choice of threshold ratio α aims to balance the robustness of the screening with the preservation of effective information, so as to effectively eliminate outlier estimates while ensuring that the screening results tend to select smaller delay values ​​that are more likely to be direct paths.

[0084] Finally, the final delay estimation result is determined. This is done by iterating through the cumulative weight sequence cum_weights, finding the smallest index that satisfies cum_weights[k] ≥ threshold. :

[0085]

[0086] The first in the sorted delay sequence The delay value corresponding to each position As the final delay estimate τ final :

[0087]

[0088] The advantage of this filtering logic is that it prioritizes delay values ​​with higher weights. If a small delay value at the beginning of the sequence has a high weight, the accumulated weight will quickly exceed the threshold, thus selecting that reliable small delay value. Conversely, if there are outliers with low weights at the beginning, more items need to be accumulated to reach the threshold, thus automatically skipping these outliers and selecting a more reliable estimate.

[0089] Step six, parameter configuration principles.

[0090] To enable the method in this invention to adapt to different operating environments and requirements, the core parameters in the algorithm can be flexibly adjusted to ensure that excellent performance is maintained in different scenarios.

[0091] For parameters in the block overlap mechanism, such as the overlap ratio r and the number of blocks N block Its configuration requires a trade-off between computational efficiency and estimation accuracy. For example, in scenarios with extremely high estimation accuracy requirements or strong signal interference, the overlap ratio r can be appropriately increased to enhance the protection of signal characteristics; while in scenarios with higher processing efficiency requirements, the overlap ratio can be appropriately decreased. Similarly, the number of blocks N block The adjustments also follow similar principles and can be configured according to the device's computing and storage resources and the required level of estimation precision.

[0092] For parameters in single-block delay estimation, such as the row number parameter L of the Hankel matrix, their selection directly affects the algorithm's resolution and computational complexity. This invention employs an adaptive parameter selection strategy, correlating the value of L with the length of the data block, thereby optimizing computational efficiency while ensuring the algorithm's theoretical performance.

[0093] The flexible configuration strategy of these parameters enables the method in this invention to be adjusted according to the accuracy requirements, efficiency requirements, interference intensity and equipment resources of the actual application scenario, avoiding the problem of insufficient scenario adaptability caused by fixed parameters, and ensuring that the algorithm can perform optimally in various scenarios.

[0094] Example 2

[0095] This embodiment is a further refinement of Embodiment 1.

[0096] To make the technical solution of this invention easier to understand and implement, this embodiment verifies the complete implementation process of the algorithm in a typical indoor environment, including experimental environment configuration, operational details of each step, and performance analysis. This experimental scenario includes both direct path signals and multipath interference caused by walls, furniture, etc., effectively simulating complex channel environments in real-world applications. In the experiment, the transceiver devices were deployed at a known distance apart. The theoretical propagation delay calculated based on this distance and the signal propagation speed was used as the benchmark for subsequent accuracy evaluation.

[0097] CSI Data Acquisition and Configuration:

[0098] Wireless channel data was captured using a wireless network card equipped with CSI acquisition capabilities and corresponding acquisition tools. Acquisition parameters (such as sampling frequency and data update rate) were set to general values. During data acquisition, to ensure data quality, interference from other radio frequency sources in the environment was avoided, and the transceiver equipment was kept in a fixed position to prevent path changes. Hundreds of CSI frames were acquired in the experiment, providing ample data support for subsequent algorithm verification.

[0099] Algorithm core parameter settings:

[0100] Based on the optimization strategy described in this invention and combined with the channel characteristics of the experimental scenario, the core parameters of the algorithm were configured. Number of blocks N block The settings are designed to effectively reduce computational complexity while ensuring that the amount of data in a single block is sufficient for feature extraction. The overlap ratio r adopts a general value that achieves a good balance between suppressing boundary effects and computational efficiency. The number of singular values ​​participating in the quality weight calculation, as well as parameters such as the threshold ratio α in the intelligent screening mechanism, are all set to empirically proven optimal values ​​that ensure the robustness and accuracy of the algorithm.

[0101] Complete algorithm implementation steps:

[0102] For each frame of CSI data acquired, the following preprocessing and core algorithm operations are performed sequentially:

[0103] Data preprocessing:

[0104] First, data validity is verified by detecting anomalous amplitude jumps between adjacent sampling points to identify and locate outliers. For the few detected outlier data points, linear interpolation based on the valid data points before and after them is used to repair them, ensuring data continuity and integrity. Second, to suppress high-frequency noise interference, a low-order FIR low-pass filter is used to filter the data, with the filter's cutoff frequency set appropriately according to the signal's sampling frequency. The processed data shows that high-frequency noise components are effectively suppressed. Finally, valid index selection is performed. Since the frequency domain edges of CSI data are susceptible to the roll-off characteristics of hardware RF front-end filters, resulting in poor signal stability, this portion of data is discarded, retaining only the middle data segments with stable and reliable signal quality, yielding valid CSI data of length Ndata for subsequent processing.

[0105] Block overlap processing:

[0106] Based on the preprocessed valid CSI data of length Ndata, this embodiment strictly follows the block overlap mechanism described in Embodiment 1 for data segmentation. This process transforms a continuous data stream into a set of overlapping data blocks of the same size through a series of precise calculations, laying the foundation for subsequent parallel processing and high-precision estimation.

[0107] First, the key parameters are calculated. The first step is to calculate the number of effective blocks N. effective This parameter is not the actual number of blocks, but a theoretical intermediate value that reflects the equivalent number of blocks required to cover the same data length without overlap. Its calculation formula is as follows:

[0108]

[0109] In the formula, N data N represents the total length of the preprocessed valid CSI data. block This is the pre-defined actual number of blocks, typically an integer chosen based on a trade-off between computational resources and feature extraction requirements (e.g., between 3 and 8); r is the overlap ratio, which is generally selected based on the application scenario's tolerance for boundary effect suppression and computational redundancy, usually set within the range of 15% to 30% (i.e., ...). The core of this calculation lies in quantifying the overlapping configurations, providing a basis for subsequently determining a uniform block size that can guarantee complete coverage of all data.

[0110] Subsequently, based on the calculated number of effective blocks N effective The actual size of each data block, `block_size`, is determined. To ensure consistency in subsequent block processing and comparability of results, all data blocks must be of uniform size. The calculation logic for the block size is as follows:

[0111]

[0112] This formula is passed through N data With N effective The ratio of N to N is used to deduce the theoretical length of each block after considering overlap, and then compared with N. block Related. Among them, This indicates a floor operation, ensuring that the final block_size is an integer, because the size of the data block must be an integer number of sampling points.

[0113] Once a uniform block size (block_size) is determined, the size of the overlapping area between adjacent data blocks (overlap_size) can be accurately calculated. This value is directly determined by the block size and the overlap ratio (r), and the calculation formula is as follows:

[0114]

[0115] Similarly, rounding down ensures that the number of sampling points in the overlapping region is also an integer. This overlap_size is key to eliminating boundary effects in this invention; it ensures that any signal feature crossing block boundaries can be fully captured and analyzed in at least two adjacent data blocks.

[0116] Finally, based on the calculated block_size and overlap_size, for all N block Precise indexing is performed on each data block. For the i-th data block in the sequence (where i = 1, 2, ..., N), its starting index is start. i and end index end i end i The calculation logic is as follows:

[0117]

[0118] This index allocation method ensures that the starting position of the i-th data block is "stepped" based on the non-overlapping portion of the previous data block, with a step length of (block_size − overlap_size). This systematic allocation method guarantees that all data blocks can seamlessly and completely cover the entire data sequence of length Ndata, and that there is an overlap region of exactly overlap_size sampling points between any two adjacent data blocks. Through the above series of calculations and allocations, the original CSI data is successfully segmented into a set of clearly structured overlapping data blocks that meet the design requirements.

[0119] Single-block signal delay estimation:

[0120] Taking one data block as an example, the implementation process of single-block latency estimation is explained in detail. First, based on the length of the data block and the adaptive parameter selection strategy, the number of rows L of the Hankel matrix is ​​determined, and a Hankel matrix H with appropriate dimensions is constructed. i Subsequently, H i By vertically concatenating its conjugate transpose, an enhanced matrix Y is constructed. i This is to fully utilize the complex characteristics of CSI data. Next, the enhancement matrix Y... i Performing SVD decomposition yields a singular value distribution exhibiting a clear step-like pattern. A clear boundary exists between the first few larger singular values ​​and the subsequent smaller singular values, indicating effective separation of the signal subspace and noise subspace. Finally, based on the SVD decomposition results, the left singular vectors corresponding to the first few largest singular values ​​are extracted to construct the signal subspace matrix. Following the core logic of the MP algorithm, the delay estimate τ of this data block is calculated by solving the eigenvalue problem. i Using the same method, latency estimation is performed on all data blocks to obtain a preliminary set of latency estimates.

[0121] Weighting:

[0122] To differentiate the reliability of delay estimates obtained from different data blocks and to provide a scientific and quantitative basis for subsequent intelligent selection, this invention designs a comprehensive weighting mechanism that integrates the intrinsic signal quality and spatial location characteristics. This mechanism evaluates each delay estimate τ through multi-dimensional assessment. i Allocate appropriate weights w final Its core lies in transforming the invisible "reliability" into a calculable value.

[0123] The mechanism mainly consists of two mutually orthogonal components: quality weights and position weights.

[0124] First, calculate the quality weight w. qualityThe calculation of this weight is based on an objective assessment of the inherent signal quality of each data block, and its theoretical basis stems from the singular value characteristics of SVD decomposition in step three. SVD can decompose signal energy into a series of singular values ​​arranged in descending order of size, where larger singular values ​​correspond to the signal subspace (containing effective information such as direct paths and major multipaths), and smaller singular values ​​correspond to the noise subspace. Therefore, the degree of separation between the signal subspace and the noise subspace directly reflects the signal-to-noise ratio and quality of the signal. Specifically, by analyzing the distribution characteristics of the first sn largest singular values, the signal quality can be quantitatively evaluated. If the first few singular values ​​are all large and close in size, it indicates that the signal energy is distributed on multiple stable paths, the signal structure is clear, and the quality is high; conversely, if all singular values ​​except the largest singular value decrease sharply, it may mean that the signal is overwhelmed by strong noise or has a simple structure, resulting in low reliability. Based on this principle, the quality weight is designed to be positively correlated with the uniformity of energy distribution in this signal subspace, with blocks having higher signal quality being assigned larger quality weights.

[0125] Secondly, assign position weights w position,i The weighting design is based on the spatial location characteristics of data blocks in the block sequence. The core consideration is that data blocks at different positions receive varying degrees of contextual support from the block overlap mechanism. Data blocks in the middle of the sequence are covered by overlapping areas of adjacent data blocks on both sides. This means that the beginning and end of their signals are effectively supported by contextual information, resulting in the highest signal integrity and stability. In contrast, edge blocks at the ends of the sequence (i.e., the first and last data blocks) only receive support from adjacent data blocks on one side, with no data coverage on their outer edges. This makes the beginning or end of their signals more susceptible to boundary truncation effects, and their signal quality is generally lower than that of middle blocks. Therefore, a higher base weight (e.g., normalized to 1.0) is assigned to data blocks in the middle, while a moderately reduced weight (e.g., a coefficient value less than 1.0) is assigned to edge blocks at the ends of the sequence. This reduction is carefully considered, reflecting the objective fact that their reliability is lower than that of middle blocks, while avoiding excessive penalty that could completely discard the potentially valid information contained in edge blocks.

[0126] Finally, the final weight w for each data block is obtained by multiplying the quality weight by the location weight. final，i :

[0127]

[0128] The advantage of this multiplicative combination method is that it requires an estimate to perform well in both signal quality and spatial location dimensions to obtain the highest weight. Significant deficiencies in either dimension (such as extremely low signal quality or being located at an edge) will significantly lower its final weight. The calculated final weight sequence intuitively reflects that data blocks with better signal quality and better locations receive higher weights, providing a scientific and comprehensive reliable basis for subsequent intelligent selection.

[0129] Smart Filtering:

[0130] To overcome the shortcomings of traditional minimum or maximum weight selection strategies, which are susceptible to interference from single outliers, this invention proposes an intelligent screening strategy based on a weight accumulation threshold. This strategy uses a logic of sorting, accumulation, and threshold judgment to select the most reliable final result from multiple delay estimations. Its core advantage lies in its ability to combine the weight information of each estimation result to effectively eliminate outliers, while also deeply aligning with the physical laws of wireless signal propagation.

[0131] The implementation steps of this strategy are as follows:

[0132] First, perform sorting. Then, calculate the latency estimate τ for all data blocks. i Sort in ascending order to obtain the sorted delay sequence τ sorted At the same time, the final weights w are adjusted synchronously according to the sorting order of the delay estimates. final The order of i is used to obtain the sorted weight sequence wsorted. This step is significant beyond simple numerical sorting; it utilizes the fundamental physical laws of wireless signal propagation: among all possible propagation paths, the direct path has the shortest propagation distance, and its corresponding delay value is usually the smallest. Therefore, sorting the delay values ​​in ascending order essentially performs a physical sorting of the estimation results from "most likely a direct path" to "more likely a long-range multipath path," providing crucial prior information for subsequent selection.

[0133] Subsequently, the cumulative value of the sorted weights is calculated. Starting from the first sorted weight wsorted[1], the cumulative sum of the first k weights is calculated sequentially to obtain a monotonically increasing cumulative weight sequence cum_weights. The physical meaning of this cumulative sequence can be understood as "cumulative confidence", which represents the sum of the reliability we have accumulated starting from the estimate that is most likely to be the direct path. The growth rate of the cumulative weights intuitively reflects the reliability difference of the estimates at the front end of the sequence.

[0134] Next, a dynamic filtering threshold is set. This threshold is not a fixed value, but is obtained by multiplying the sum of the final weights of all data blocks in the current frame by a preset threshold ratio α:

[0135]

[0136] The threshold ratio α is a key adjustment parameter, and its selection must simultaneously satisfy robust statistical theory and the laws of wireless propagation. From a robust statistical perspective, α sets a reasonable cutoff point to exclude a certain proportion of outliers; from the perspective of wireless propagation laws, a moderate α value ensures that the screening results tend to select estimates with smaller delay values ​​at the beginning of the sequence.

[0137] Finally, a threshold check is performed to determine the final result. The cumulative weight sequence cum_weights[k] is traversed to find the first minimum index that satisfies cum_weights[k] ≥ threshold. The delay value corresponding to this position. This result is selected as the final delay estimation result in this embodiment. The ingenuity of this filtering logic lies in its adaptability: if the delay value at the beginning of the sequence (small delay, possibly a direct path) has a high weight, its own weight may directly exceed the threshold, thus being quickly selected; if there is an outlier at the beginning of the sequence with extremely low weight, its own weight is insufficient to exceed the threshold, and the algorithm will naturally add the weights of subsequent estimates until the accumulated "confidence" is high enough, thus cleverly "skipping" the outlier at the beginning and selecting a more reliable subsequent estimate. This method avoids the sensitivity of traditional minimum value selection strategies to outliers and makes full use of weight information and physical laws, ensuring the reliability and accuracy of the final result.

[0138] Performance verification and analysis:

[0139] Comparing the final delay estimate obtained in this embodiment with the theoretical propagation delay, the results show that the absolute error between the two is at the extremely low nanosecond (ns) level, which translates to sub-meter level accuracy in distance error. This level of accuracy is significantly better than traditional processing methods, fully demonstrating the significant advantage of the algorithm of this invention in estimation accuracy.

[0140] To further verify the stability of the algorithm, the complete processing flow described above was executed on hundreds of frames of CSI data collected. Statistical analysis of the error distribution of all final delay estimation results showed that the errors of the vast majority of estimation results were concentrated within a very small range, and no outliers with extremely large deviations appeared. These statistical results demonstrate that the algorithm of this invention is not only highly accurate but also highly stable, maintaining consistent high performance in multiple repeated experiments.

[0141] Performance comparison with existing algorithms:

[0142] To highlight the comprehensive advantages of the algorithm of this invention, its performance was compared with that of a current mainstream delay estimation algorithm (MUSIC algorithm). The comparative experiments were based on the same CSI dataset, with mean absolute error (MAE), single-processing time, and memory usage as key evaluation metrics.

[0143] In terms of estimation accuracy: the mean absolute error of the algorithm of this invention is significantly better than the two comparative algorithms, with an accuracy improvement of several times, and the advantages are extremely significant.

[0144] In terms of computational efficiency: the single processing time of the algorithm of this invention is much lower than that of traditional super-resolution algorithms, and the efficiency is significantly improved, which can meet the real-time requirements of most positioning systems.

[0145] Regarding memory usage: Due to the use of block processing, the maximum memory usage required by the algorithm of this invention during processing is only a small fraction of that of traditional super-resolution algorithms, greatly reducing the requirements for hardware resources and making it more suitable for deployment on resource-constrained devices. A comprehensive comparison shows that the algorithm of this invention achieves a complete improvement in the balance of accuracy, efficiency, and resource usage. It solves the problem of insufficient accuracy in traditional algorithms and overcomes the computational complexity of super-resolution algorithms, demonstrating significant performance advantages and broad application prospects.

[0146] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0147] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A signal delay estimation method based on block processing, characterized in that: Includes the following steps: Step S1: Obtain Channel State Information (CSI) data, perform validity checks and noise filtering on the CSI data, and select a valid index range to obtain preprocessed CSI data. Step S2: Based on the preset number of blocks and overlap ratio, the preprocessed CSI data is divided into blocks to obtain several overlapping data blocks. Step S3: Perform delay estimation for each overlapping data block to obtain the delay estimate value for each data block; Specifically, a Hankel matrix is ​​constructed for each data block. The Hankel matrix is ​​then vertically concatenated with its conjugate transpose to construct an enhancement matrix. Singular value decomposition is performed on the enhancement matrix. Based on the decomposition results, the signal subspace and noise subspace are separated. Finally, the delay estimate is obtained by constructing a characteristic polynomial and solving for its roots. Step S4: Calculate the final weight for each delay estimate. Specifically, based on the singular values ​​obtained from singular value decomposition, calculate the quality weight reflecting the signal quality of the data block by analyzing the ratio relationship between the singular values, allocate the position weight according to the spatial position of the data block in the block sequence, and combine the quality weight and the position weight to obtain the final weight. Step S5: Determine the final delay estimation result based on the delay estimate and the corresponding final weight.

2. The signal delay estimation method based on block processing according to claim 1, characterized in that: In step S2, the preprocessed CSI data is divided into several overlapping data blocks. Specifically, the number of effective blocks required to cover the total length of the CSI data without overlap is calculated first. Then, the block size of each data block is determined based on the number of effective blocks. Next, the overlap size of adjacent data blocks is calculated. Finally, an index range is allocated to each data block and the corresponding data block is extracted.

3. The signal delay estimation method based on block processing according to claim 2, characterized in that: In step S2, the number of effective blocks is calculated based on the total length of the preprocessed CSI data, combined with the preset number of blocks and overlap ratio, and is derived by deriving the theoretical number of blocks required to cover the total length in a non-overlapping scenario.

4. The signal delay estimation method based on block processing according to claim 3, characterized in that: In step S2, the overlap size of adjacent data blocks is calculated by multiplying the block size of each data block by a preset overlap ratio, and the overlap size must satisfy the requirement that signal features crossing block boundaries can be repeatedly captured in adjacent data blocks.

5. The signal delay estimation method based on block processing according to claim 1, characterized in that: In step S3, when constructing the Hankel matrix, the value of the row number parameter L of the Hankel matrix ranges from 1 / 4 to 1 / 2 of the block size.

6. The signal delay estimation method based on block processing according to claim 4, characterized in that: In step S3, the enhancement matrix is ​​constructed by vertical concatenation. Specifically, the Hankel matrix constructed for the data block is used as the upper part, and the conjugate transpose of the Hankel matrix is ​​used as the lower part. After concatenation, an enhancement matrix with a dimension of 2 × the number of rows of the Hankel matrix is ​​formed.

7. The signal delay estimation method based on block processing according to claim 1, characterized in that: In step S4, the calculation of the quality weights includes the following sub-steps: extracting the top s values ​​from the singular value decomposition results. n There are 1 maximum singular values, s n Given a preset number of singular values ​​to participate in the calculation, the ratio of each singular value to the largest singular value is calculated, and the average of all the ratios is taken as the quality weight of the corresponding data block.

8. The signal delay estimation method based on block processing according to claim 1, characterized in that: In step S4, the weight allocation takes into account the positional characteristics of the data block in the processing sequence and adopts a corresponding weighting strategy for data blocks at different positions.

9. The signal delay estimation method based on block processing according to claim 1, characterized in that: In step S5, threshold parameters are set based on a preset filtering strategy.

10. The signal delay estimation method based on block processing according to claim 1, characterized in that: In step S5, the final delay estimation result is determined based on the delay estimate and the corresponding weight. Specifically, the delay estimate is sorted and the final result is determined by a filtering mechanism in combination with the weight information.

Images