Satellite channel estimation method and device and computer readable storage medium
Through the methods of compression perception and multi-scale sparse feature extraction, the problem of high computational complexity of satellite channel estimation is solved, and fast and accurate channel estimation is achieved at the computing-constrained reception end.
Patent Information
- Application Number
- CN202510657159.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-21
- Publication Date
- 2025-08-08
AI Technical Summary
In low-orbit satellite communication, due to the motion state of the transmitter and receiver, the calculation complexity of the satellite channel model is high and the calculation amount is large, and it is difficult for the prior art to effectively perform satellite channel estimation at the receiving end with limited computing.
Using compression perception theory, the compression perception measurement matrix is determined based on the output samples of the subcarrier and its pilot sequence, the initial input sample matrix is reconstructed, and the basic extension model of the satellite channel is constructed through multi-scale sparse feature extraction and sparse attention calculation, and the impulse response of the satellite channel is finally solved.
It reduces the computational complexity of satellite channel estimation, improves computing efficiency, and ensures the accuracy of channel estimation, and is suitable for receiving ends with limited computing volume.
Smart Images
Figure CN120455212A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to a channel estimation method, and more particularly to a satellite channel estimation method, device, and computer-readable storage medium. Background Art
[0002] In low-orbit satellite communications, satellite signals are transmitted through multiple paths, such as direct paths, reflection paths, and scattering paths. As a result, satellite signals on each path arrive at the receiver with different time delays and phases, resulting in a multipath effect. To reduce the impact of multipath effects on satellite signal transmission, OFDM (Orthogonal Frequency Division Multiplexing) technology is typically used to first modulate the satellite signal onto multiple different subcarriers, allowing each subcarrier to carry part of the satellite signal information. Each subcarrier is then transmitted to the receiver via multiple paths. Furthermore, because satellite signals have Doppler frequency shift, multiple subcarriers on the same path experience different fading and phase shifts. Therefore, when performing satellite channel estimation, it is necessary to correctly estimate the impulse responses of all subcarrier channels at any given moment.
[0003] However, in satellite communication systems, both the transmitter and receiver are usually in motion, causing the positions of buildings or trees in the signal transmission path to change accordingly. This results in extremely high computational complexity and a large amount of computational effort if a satellite channel model is directly constructed and used to estimate the impulse response of all subcarrier channels at any moment. Therefore, it is not suitable for the current receiver, which has limited computational effort. Summary of the Invention
[0004] The present invention provides a satellite channel estimation method, apparatus, and computer-readable storage medium to address the problems of related technologies. The technical solutions are as follows:
[0005] In a first aspect, an embodiment of the present application provides a satellite channel estimation method, comprising:
[0006] Determine a compressed sensing measurement matrix based on output samples of N subcarriers and pilot sequences carried by the output samples; N is a positive integer greater than or equal to 2;
[0007] The initial input sample matrix is reconstructed using the output samples of the N subcarriers and the compressed sensing measurement matrix; the i-th row vector of the initial input signal matrix is used to represent the initial input sample of the i-th subcarrier; 1≤i≤N, and i is an integer;
[0008] Perform multi-scale sparse feature extraction on the initial input sample matrix to obtain sparse feature matrices of the initial input sample matrix at multiple different scales;
[0009] Perform sparse attention calculation on the sparse feature matrix of the initial input sample matrix at different scales to obtain the global attention features at the corresponding scale;
[0010] The global attention features of the initial input sample matrix at different scales are fused to obtain the reconstructed target input sample;
[0011] A basis expansion model of a satellite channel is constructed based on a matrix of target output samples and target input samples; the target output samples are formed by superimposing output samples of N subcarriers;
[0012] The basis expansion model of the satellite channel is solved to obtain the impulse response of the satellite channel.
[0013] In one embodiment, determining a compressed sensing measurement matrix based on output samples of N subcarriers and pilot sequences carried by the output samples includes:
[0014] The output samples of the N subcarriers are sampled according to a preset sampling rate and sampling time to obtain M output subsamples of each subcarrier; M is a positive integer greater than or equal to 1;
[0015] The compressed sensing measurement matrix is obtained by iterative fitting using the M output subsamples of the N subcarriers and the pilot sequences they carry.
[0016] In one embodiment, the output samples of each subcarrier include M output subsamples, and the initial input sample matrix is reconstructed using the output samples of N subcarriers and the compressed sensing measurement matrix, including:
[0017] Determining M input subsamples corresponding to each subcarrier based on the M output subsamples of each subcarrier and a compressed sensing measurement matrix;
[0018] The M input subsamples of each subcarrier are fused to obtain an initial input sample matrix.
[0019] In one embodiment, a row vector of an initial input sample matrix includes M input subsamples; multi-scale sparse feature extraction is performed on the initial input sample matrix to obtain sparse feature matrices of the initial input sample matrix at multiple scales, including:
[0020] The initial input sample matrix is input into multiple convolution kernels of different sizes respectively, so that the multiple convolution kernels correspond one to one to output multiple sparse feature matrices of different scales; wherein the i-th row vector of each sparse feature matrix is used to represent the sparse features of the M input subsamples of the i-th subcarrier at the corresponding scale.
[0021] In one embodiment, the row vector of the initial input sample matrix includes M input subsamples, and the sparse feature matrix of the initial input sample matrix at each scale includes a query matrix, a key matrix, a value matrix, and a sparse feature dimension of a convolution kernel of a corresponding size. Sparse attention calculation is performed on the sparse feature matrices of the initial input sample matrix at different scales to obtain a global attention feature at the corresponding scale, including:
[0022] Determine a similarity score between the M input subsamples of each subcarrier at the corresponding scale based on the query matrix, the key matrix, and the sparse feature dimension at each scale;
[0023] The similarity scores between the M input subsamples of each subcarrier at each scale are normalized to obtain the corresponding probability distribution matrix; the i-th row vector of the probability distribution matrix is used to represent the importance of the M input subsamples of the i-th subcarrier at the corresponding scale as input subsamples;
[0024] The probability distribution matrix at each scale is weighted and summed with its corresponding value matrix to obtain the global attention features of the M input subsamples of each subcarrier at the corresponding scale.
[0025] In one embodiment, the global attention features of the initial input sample matrix at different scales include the global attention features of M input subsamples of each subcarrier, and the global attention features of the initial input sample matrix at different scales are fused to obtain a reconstructed target input sample matrix, including:
[0026] The global attention features of the M input subsamples of each subcarrier at multiple different scales are superimposed to obtain the global attention features of the M input subsamples of each subcarrier at all scales;
[0027] The global attention features of the M input subsamples of each subcarrier at all scales are concatenated to obtain the reconstructed target input sample matrix.
[0028] In one embodiment, constructing a basis extension model of a satellite channel based on a matrix of target output samples and target input samples includes:
[0029] Add the first timestamp information to the target input sample matrix to obtain the target input signal The first timestamp information includes the mth propagation moment and the lth propagation path of the target input sample matrix, and the first timestamp information is received synchronously when the output samples of N subcarriers are received; m and l are both positive integers greater than or equal to 1;
[0030] Add the second timestamp information to the target output sample to obtain the target output signal The second timestamp information includes a reception time m of the output samples of the N subcarriers;
[0031] Based on the target output signal The predicted value of the target input signal is predicted by the functional relationship between the impulse response h(ml) of the lth propagation path at the mth propagation time
[0032] Based on the target input signal Target output signal and the predicted value of the target input signal The constructed base expansion model of the satellite channel is expressed as:
[0033]
[0034] in, It is used to represent the square L2 norm of the deviation between the target input signal and its predicted value. min represents the minimization operation, and L is an integer greater than 2.
[0035] In one embodiment, the basis expansion model of the satellite channel is solved to obtain the impulse response h(ml) of the lth propagation path at the mth propagation moment of the satellite channel.
[0036] In a second aspect, embodiments of the present application provide a satellite channel estimation device, comprising: a memory and a processor. The memory and the processor communicate with each other via an internal connection path, the memory is configured to store instructions, and the processor is configured to execute the instructions stored in the memory. When the processor executes the instructions stored in the memory, the processor performs the method according to any of the aforementioned embodiments.
[0037] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a computer, the method in any one of the above-mentioned embodiments is executed.
[0038] The advantages or beneficial effects of the above technical solution include at least: first, determining a compressed sensing measurement matrix based on the output samples of N subcarriers and the pilot sequence carried by the output samples, and then reconstructing the initial input sample matrix using the output samples of the N subcarriers and the compressed sensing measurement matrix. The input samples of the N subcarriers can be compressed and reconstructed based on the compressed sensing theory to achieve a rough estimate of them, which can reduce the complexity of the calculation and quickly reconstruct the initial input sample matrix. Since the initial input sample matrix is a coarse signal, multi-scale sparse feature extraction is performed on the initial input sample matrix, and sparse attention calculation is performed on the sparse feature matrices of the initial input sample matrix at different scales to obtain global attention features at the corresponding scale. In this way, the initial input sample matrix can be used for detailed estimation at a finer scale. Subsequently, by fusing the global attention features of the sparse feature matrices of the initial input sample matrix at different scales, the sparse feature matrices of the initial input sample matrix at different scales can be promoted to complement and fuse each other, thereby reconstructing a more accurate target input sample matrix, making the target input sample matrix a more accurate signal. In this way, the basis expansion model of the satellite channel constructed using the received target output samples and the reconstructed target input sample matrix can be made more accurate, and then by simply solving the model, the impulse response of the satellite channel can be quickly and accurately estimated. In addition, this cascade method of first roughly estimating the initial input sample matrix and then using the initial input sample matrix to finely estimate the target input sample matrix can decompose the complex estimation problem into simpler sub-estimation problems, which is conducive to further reducing the complexity of the calculation and improving the implementation efficiency. The method of the present application can reduce the computational complexity of satellite channel estimation while ensuring the accuracy of satellite channel estimation, thereby reducing the computational cost and improving computational efficiency, which is conducive to application in the receiving end where the current computational cost is limited. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar components or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in this application and should not be construed as limiting the scope of this application.
[0040] Figure 1 The flowchart of the satellite channel estimation method according to an embodiment of the present application is shown.
[0041] Figure 2 FIG. 4 shows a structural block diagram of a satellite channel estimation device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0042] Hereinafter, only certain exemplary embodiments are briefly described. As will be appreciated by those skilled in the art, the described embodiments may be modified in various ways without departing from the spirit or scope of the present application. Therefore, the drawings and description are to be regarded as illustrative in nature and not restrictive.
[0043] Figure 1 FIG. 5 is a flow chart showing a satellite channel estimation method according to an embodiment of the present application. Figure 1 As shown, the satellite channel estimation method is applicable to a receiving end, and the method includes the following steps S110 to S170.
[0044] Step S110 : Determine a compressed sensing measurement matrix based on the output samples of N subcarriers and the pilot sequences carried by the output samples; N is a positive integer greater than or equal to 2. Each subcarrier is located in a different frequency band.
[0045] The pilot sequence is a known signal pre-embedded in the output samples of the subcarriers to assist the receiving end in channel estimation. The pilot sequence carried by the output samples of each subcarrier can be extracted from the output samples of each subcarrier.
[0046] Step S120: reconstruct an initial input sample matrix using the output samples of the N subcarriers and the compressed sensing measurement matrix; the i-th row vector of the initial input signal matrix is used to represent the initial input sample of the i-th subcarrier; 1≤i≤N, and i is an integer.
[0047] Step S130 : performing multi-scale sparse feature extraction on the initial input sample matrix to obtain sparse feature matrices of the initial input sample matrix at multiple different scales.
[0048] Step S140: Perform sparse attention calculation on the sparse feature matrices of the initial input sample matrix at different scales to obtain global attention features at corresponding scales.
[0049] Step S150: Fuse the global attention features of the initial input sample matrix at different scales to obtain a reconstructed target input sample matrix.
[0050] Step S160: construct a basis extension model of the satellite channel based on the target output sample and the target input sample matrix; the target output sample is formed by superimposing the output samples of N subcarriers.
[0051] Step S170: Solve the basis expansion model of the satellite channel to obtain the impulse response of the satellite channel.
[0052] The above scheme first determines the compressed sensing measurement matrix based on the output samples of N subcarriers and the pilot sequences carried by the output samples, and then reconstructs the initial input sample matrix using the output samples of the N subcarriers and the compressed sensing measurement matrix. Based on the compressed sensing theory, the input samples of the N subcarriers can be compressed and reconstructed to achieve a rough estimate of them. This can reduce the complexity of the calculation and quickly reconstruct the initial input sample matrix. Since the initial input sample matrix is a coarse signal, multi-scale sparse feature extraction is performed on the initial input sample matrix, and sparse attention calculation is performed on the sparse feature matrices of the initial input sample matrix at different scales to obtain global attention features at the corresponding scale. In this way, the initial input sample matrix can be used for detailed estimation at a finer scale. Subsequently, by fusing the global attention features of the sparse feature matrices of the initial input sample matrix at different scales, the sparse feature matrices of the initial input sample matrix at different scales can be promoted to complement and fuse each other, thereby reconstructing a more accurate target input sample matrix, making the target input sample matrix a more accurate signal. In this way, the basis expansion model of the satellite channel constructed using the received target output samples and the reconstructed target input sample matrix can be made more accurate, and then by simply solving the model, the impulse response of the satellite channel can be quickly and accurately estimated. In addition, this cascade method of first roughly estimating the initial input sample matrix and then using the initial input sample matrix to finely estimate the target input sample matrix can decompose the complex estimation problem into simpler sub-estimation problems, which is conducive to further reducing the complexity of the calculation and improving the implementation efficiency. The method of the present application can reduce the computational complexity of satellite channel estimation while ensuring the accuracy of satellite channel estimation, thereby reducing the computational cost and improving computational efficiency, which is conducive to application in the receiving end where the current computational cost is limited.
[0053] In one embodiment, determining a compressed sensing measurement matrix based on output samples of N subcarriers and pilot sequences carried by the output samples includes the following steps S111 to S112.
[0054] Step S111 : Sample the output samples of the N subcarriers respectively according to a preset sampling rate and sampling time to obtain M output subsamples of each subcarrier.
[0055] For example, if the sampling rate is 1 time / ms and the sampling time is 5ms, 5 output subsamples of each subcarrier can be obtained by sampling.
[0056] Step S112: Using the M output subsamples of the N subcarriers and the pilot sequences they carry, perform iterative fitting to obtain a compressed sensing measurement matrix.
[0057] For example, the output samples of N subcarriers are set to be y1, y2, ..., y N After executing step S111, the M output subsamples of the first subcarrier are y 11 、y 12 ,……,y 1M , the M output subsamples of the i-th subcarrier are y i1 、y i2 ,……,y iM , the M output subsamples of the Nth subcarrier are y N1 、y N2 ,……,y NM ; Correspondingly, the pilot sequence carried by the M output subsamples of the first subcarrier is xp 11 、xp 12 、……、xp 1M , the pilot sequence carried by the M output subsamples of the i-th subcarrier is xp i1 、xp i2 、……、xp iM , the pilot sequence carried by the M output subsamples of the Nth subcarrier is xp N1 、xp N2 、……、xp NM .
[0058] Based on the compressed sensing theory, the j-th output subsample y of the i-th subcarrier ij The pilot sequence xp it carries ij The mapping relationship between them can be expressed by the following formula (1):
[0059]
[0060] in, represents the initial compressed sensing measurement matrix, 1≤j≤M, and j is an integer.
[0061] Step S112 may be: first, the initial compressed sensing measurement matrix Set the initial value and set the pilot sequence xp carried by the first output subsample of the first subcarrier to 11 Input formula (1) to calculate the first output subsample y of the first subcarrier 11 The measured value y' 11 If y 11 and y' 11 If different, the initial compressed sensing measurement matrix is reduced or increased according to the preset step size The initial value of the first subcarrier; then the pilot sequence xp carried by the second output subsample of the first subcarrier 12 Enter formula (1) to calculate the measured value y' of the second output subsample of the first subcarrier12 , and so on, iterate until a certain output subsample of a certain subcarrier is the same as its measured value, and then the final compressed sensing measurement matrix is obtained
[0062] Based on this, the output samples of N subcarriers are first sampled according to a preset sampling rate and sampling time. This can make the data volume of the M output subsamples of each subcarrier less than the output samples of N subcarriers, which is conducive to compressing the output samples of N subcarriers, thereby reducing computing consumption. Then, the M output subsamples of N subcarriers and the pilot sequences carried by the output subsamples can be used to quickly construct a compressed sensing measurement matrix, which is conducive to improving the computing speed and thus improving the estimation efficiency of the satellite channel.
[0063] In one embodiment, the output samples of each subcarrier include M output subsamples, and the initial input sample matrix is reconstructed using the output samples of N subcarriers and the compressed sensing measurement matrix, including the following steps S121 to S122.
[0064] Step S121 : Determine M input subsamples corresponding to each subcarrier based on the M output subsamples of each subcarrier and the compressed sensing measurement matrix.
[0065] Step S122: Fuse the M input subsamples of each subcarrier to obtain an initial input sample matrix.
[0066] Exemplarily, steps S121 to S122 may be expressed using the following formula (2):
[0067]
[0068] in, represents the initial input sample matrix; r() represents the reshaping operation, which is used to characterize step S121, which is specifically to obtain the corresponding input subsamples by dividing the output subsamples of each subcarrier by the compressed sensing measurement matrix; k() represents the splicing operation, which is used to characterize step S122, which is specifically to obtain the initial input sample matrix by splicing the M input subsamples of N subcarriers
[0069] Based on this, the initial input sample matrix can be quickly reconstructed Make the initial input sample matrix A coarse signal with a small amount of data is beneficial to saving storage resources and improving the complexity of estimation, thereby reducing the computational consumption at the receiving end.
[0070] In one embodiment, a row vector of an initial input sample matrix includes M input subsamples; multi-scale sparse feature extraction is performed on the initial input sample matrix to obtain sparse feature matrices of the initial input sample matrix at multiple scales, including:
[0071] The initial input sample matrix is input into multiple convolution kernels of different sizes respectively, so that the multiple convolution kernels correspond one to one to output multiple sparse feature matrices of different scales; wherein the i-th row vector of each sparse feature matrix is used to represent the sparse features of the M input subsamples of the i-th subcarrier at the corresponding scale.
[0072] For example, multiple convolution kernels of different sizes include T, where T is an integer greater than or equal to 2, and the initial input sample matrix Input T convolution kernels of different sizes respectively, so that the T convolution kernels of different sizes can output the initial input sample matrix one by one. Sparse feature matrices at T different scales.
[0073] In one embodiment, the row vector of the initial input sample matrix includes M input subsamples. The sparse feature matrix at each scale includes the query matrix Q, the key matrix K, the value matrix V and the sparse feature dimension d of the convolution kernel of the corresponding size. t Among them, the query matrix Q, key matrix K and value matrix V at each scale can be generated by linear transformation of the sparse feature matrix at the corresponding scale.
[0074] Step S140: Initial input sample matrix Sparse attention calculation is performed on sparse feature matrices at different scales to obtain global attention features at corresponding scales, including the following steps S141 to S143.
[0075] Step S141 : Based on the query matrix, the key matrix, and the sparse feature dimension at each scale, determine the similarity scores between the M input subsamples of each subcarrier at the corresponding scale.
[0076] Exemplarily, step S141 can be expressed using the following formula (3):
[0077]
[0078] Among them, score t Represents the similarity score matrix between the M input subsamples of N subcarriers, score it represents the similarity score between the M input subsamples of the i-th subcarrier.
[0079] Step S142: Normalize the similarity scores between the M input subsamples of each subcarrier at each scale to obtain a corresponding probability distribution matrix; the i-th row vector of the probability distribution matrix is used to represent the importance of the M input subsamples of the i-th subcarrier at the corresponding scale as input subsamples.
[0080] For example, the similarity scores between the M input subsamples of each subcarrier at each scale are normalized to obtain the corresponding probability distribution matrix α t It can be expressed by the following formula (4):
[0081]
[0082] The softmax() function is used to convert the similarity scores between the M input subsamples of each subcarrier at each scale into a probability distribution. ik ) represents the conversion of the similarity scores between the M input subsamples of the i-th subcarrier into a probability distribution.
[0083] Step S143: Perform weighted summation on the probability distribution matrix at each scale and its corresponding value matrix V to obtain the global attention features of the M input subsamples of each subcarrier at the corresponding scale.
[0084] Specifically, in step S143, the global attention feature of the M input subsamples of the i-th subcarrier at the t-th scale is i (t) can be expressed by the following formula (5):
[0085]
[0086] Among them, a it =softmax(score it ), V j Represents the jth value in the value matrix V, feature i (t) is a 1×M matrix.
[0087] The above scheme first determines the similarity scores between the M input subsamples of each subcarrier at the corresponding scale through the query matrix, key matrix and sparse feature dimension at each scale, and then normalizes the similarity scores between the M input subsamples of each subcarrier at each scale to obtain the corresponding probability distribution matrix. Then, the probability distribution matrix at each scale is weighted and summed with its corresponding value matrix V to obtain the global attention features of the M input subsamples of each subcarrier at the corresponding scale. This can form a detailed estimation of the input subsamples of each subcarrier at a finer scale, which is conducive to improving the estimation accuracy.
[0088] In one embodiment, the global attention features of the initial input sample matrix at different scales include the global attention features of M input subsamples of each subcarrier at different scales. The global attention features of the initial input sample matrix at different scales are fused to obtain a reconstructed target input sample, including the following steps S151 to S152.
[0089] Step S151: superimpose the global attention features of the M input subsamples of each subcarrier at multiple different scales to obtain the global attention features of the M input subsamples of each subcarrier at all scales.
[0090] In step S151, the global attention features of the M input subsamples of the ith subcarrier at the 1st scale to the Tth scale are superimposed to obtain the global attention features of the M input subsamples of the ith subcarrier at all scales. i It can be expressed by the following formula (6):
[0091]
[0092] Step S152: Concatenate the global attention features of the M input subsamples of each subcarrier at all scales to obtain a reconstructed target input sample matrix Target input sample matrix It can be expressed by the following formula (7):
[0093]
[0094] Among them, feature ij represents the global attention feature of the j-th input subsample of the i-th subcarrier.
[0095] The above scheme first superimposes the global attention features of the M input subsamples of each subcarrier at multiple different scales to obtain the global attention features of the M input subsamples of each subcarrier at all scales, which can promote the complementarity and fusion of the global attention features of each subcarrier at multiple different scales. Then, the global attention features of the M input subsamples of each subcarrier at all scales are spliced to obtain the reconstructed target input sample matrix It can capture the global attention features of M input subsamples of N subcarriers at multiple different scales, which is beneficial to improve the reconstruction of the target input sample matrix This helps to improve the accuracy of channel estimation.
[0096] In one embodiment, constructing a basis extension model of a satellite channel based on a matrix of target output samples and target input samples includes the following steps S161 to S164.
[0097] Step S161: Add the first timestamp information to the target input sample matrix to obtain the target input signal The first timestamp information includes the mth propagation moment and the lth propagation path of the target input sample matrix, and the first timestamp information is received synchronously when the output samples of N subcarriers are received; m and l are both positive integers greater than or equal to 1;
[0098] Step S162: Add the second timestamp information to the target output sample to obtain the target output signal The second timestamp information includes a reception time m of the output samples of the N subcarriers;
[0099] Step S163: Output signal based on target The predicted value of the target input signal is predicted by the functional relationship between the impulse response h(ml) of the lth propagation path at the mth propagation time
[0100] Step S164: Based on the target input signal Target output signal and the predicted value of the target input signal The constructed base expansion model of the satellite channel is expressed as:
[0101]
[0102] in, It is used to represent the square L2 norm of the deviation between the target input signal and its predicted value. min represents the minimization operation, and L is an integer greater than 2.
[0103] In practical applications, the target input signal can be transmitted to the receiving end through at least one propagation path. When there are multiple propagation paths, after determining the target output sample and target input sample matrices, based on the similarity between the impulse responses of different propagation paths, the basis extension model of the satellite channel can be quickly constructed, and then the basis extension model of the satellite channel can be quickly solved to obtain the impulse response of the satellite channel.
[0104] More specifically, the basis expansion model of the satellite channel is solved to obtain the impulse response h(ml) of the lth propagation path of the satellite channel at the mth propagation moment, so that the satellite channel can be estimated.
[0105] Figure 2 FIG. 5 shows a structural block diagram of a satellite channel estimation device according to an embodiment of the present invention. Figure 2As shown, the satellite channel estimation apparatus includes a memory 10 and a processor 20. The memory 10 stores a computer program that can be executed on the processor 20. When the processor 20 executes the computer program, the satellite channel estimation method of the above embodiment is implemented. The number of the memory 10 and the processor 20 can be one or more.
[0106] The satellite channel estimation device also includes a communication interface 30 for communicating with external devices and performing data exchange. If the memory 10, processor 20, and communication interface 30 are implemented independently, the memory 10, processor 20, and communication interface 30 can be interconnected via a bus to enable communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc.
[0107] Optionally, in a specific implementation, if the memory 10, the processor 20 and the communication interface 30 are integrated on a chip, the memory 10, the processor 20 and the communication interface 30 can communicate with each other through an internal interface.
[0108] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method provided in the embodiment of the present application.
[0109] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. It is worth noting that the processor may be a processor that supports the advanced RISC machine (ARM) architecture.
[0110] Furthermore, optionally, the above-mentioned memory may include read-only memory and random access memory, and may also include non-volatile random access memory. The memory may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory may include random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus RAM (DR RAM).
[0111] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the present application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.
[0112] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.
Claims
1. A satellite channel estimation method, characterized in that: include: Determine a compressed sensing measurement matrix based on the output samples of the N subcarriers and the pilot sequences carried by the output samples; N is a positive integer greater than or equal to 2; Reconstructing an initial input sample matrix using the output samples of the N subcarriers and the compressed sensing measurement matrix; the i-th row vector of the initial input signal matrix is used to represent the initial input sample of the i-th subcarrier; 1≤i≤N, and i is an integer; Performing multi-scale sparse feature extraction on the initial input sample matrix to obtain sparse feature matrices of the initial input sample matrix at multiple different scales; Performing sparse attention calculation on the sparse feature matrices of the initial input sample matrix at different scales to obtain global attention features at corresponding scales; Fusing the global attention features of the initial input sample matrix at different scales to obtain a reconstructed target input sample; Constructing a basis expansion model of the satellite channel based on target output samples and the target input sample matrix; the target output samples are formed by superimposing output samples of N subcarriers; A basis expansion model of the satellite channel is solved to obtain an impulse response of the satellite channel.
2. The method according to claim 1, characterized in that Based on the output samples of the N subcarriers and the pilot sequences carried by the output samples, a compressed sensing measurement matrix is determined, including: Sampling the output samples of the N subcarriers respectively according to a preset sampling rate and sampling time to obtain M output subsamples of each subcarrier; M is a positive integer greater than or equal to 1; The compressed sensing measurement matrix is obtained by iterative fitting using the M output subsamples of the N subcarriers and the pilot sequences carried by them.
3. The method according to claim 1, characterized in that The output samples of each subcarrier include M output subsamples, and the initial input sample matrix is reconstructed using the output samples of the N subcarriers and the compressed sensing measurement matrix, including: Determining M input subsamples corresponding to each of the subcarriers based on the M output subsamples of each of the subcarriers and the compressed sensing measurement matrix; The M input subsamples of each of the subcarriers are fused to obtain the initial input sample matrix.
4. The method according to claim 1, wherein The row vector of the initial input sample matrix includes M input subsamples; performing multi-scale sparse feature extraction on the initial input sample matrix to obtain sparse feature matrices of the initial input sample matrix at multiple different scales, including: The initial input sample matrix is input into multiple convolution kernels of different sizes respectively, so that the multiple convolution kernels output the sparse feature matrices of multiple different scales in one-to-one correspondence; wherein the i-th row vector of each of the sparse feature matrices is used to characterize the sparse features of the M input subsamples of the i-th subcarrier at the corresponding scale.
5. The method according to claim 1, wherein The row vector of the initial input sample matrix includes M input subsamples, and the sparse feature matrix of the initial input sample matrix at each scale includes a query matrix, a key matrix, a value matrix, and a sparse feature dimension of a convolution kernel of a corresponding size. Sparse attention calculation is performed on the sparse feature matrices of the initial input sample matrix at different scales to obtain global attention features at the corresponding scale, including: Determining, based on the query matrix, the key matrix, and the sparse feature dimension at each scale, a similarity score between the M input subsamples of each subcarrier at the corresponding scale; Normalizing the similarity scores between the M input subsamples of each subcarrier at each scale to obtain a corresponding probability distribution matrix; the i-th row vector of the probability distribution matrix is used to represent the importance of the M input subsamples of the i-th subcarrier at the corresponding scale as input subsamples; The probability distribution matrix at each scale is weighted and summed with its corresponding value matrix to obtain the global attention features of the M input subsamples of each subcarrier at the corresponding scale.
6. The method according to claim 1, characterized in that The global attention features of the initial input sample matrix at different scales include the global attention features of the M input subsamples of each subcarrier, and the global attention features of the initial input sample matrix at different scales are fused to obtain a reconstructed target input sample matrix, including: Superimposing the global attention features of the M input subsamples of each subcarrier at multiple different scales to obtain the global attention features of the M input subsamples of each subcarrier at all scales; The global attention features of the M input subsamples of each of the subcarriers at all scales are concatenated to obtain the reconstructed target input sample matrix.
7. The method according to claim 1, characterized in that Constructing a basis extension model of the satellite channel based on target output samples and the target input sample matrix, comprising: Add the first timestamp information to the target input sample matrix to obtain the target input signal The first timestamp information includes the mth propagation moment and the lth propagation path of the target input sample matrix, and the first timestamp information is received synchronously when the output samples of the N subcarriers are received; m and l are both positive integers greater than or equal to 1; Add the second timestamp information to the target output sample to obtain the target output signal The second timestamp information includes a reception time m of the output samples of the N subcarriers; Based on the target output signal The predicted value of the target input signal obtained by predicting the functional relationship between the impulse response h(ml) of the lth propagation path at the mth propagation time Based on the target input signal The target output signal and the predicted value of the target input signal The base expansion model of the satellite channel is constructed as follows: in, Used to represent the square L2 norm of the deviation between the target input signal and its predicted value, min represents a minimization operation, and L is an integer greater than 2.
8. The method according to claim 7, characterized in that The basis expansion model of the satellite channel is solved to obtain the impulse response h(ml) of the lth propagation path at the mth propagation moment of the satellite channel.
9. A satellite channel estimation device, characterized in that: The system comprises a processor and a memory, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the method according to any one of claims 1 to 8. 10 . A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the method according to claim 1 is implemented.