Low-orbit satellite communication system low-overhead channel estimation method based on OTSM
By constructing a low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM, and utilizing a three-layer structured prior probability model and a variational Bayesian off-network estimation framework, the contradiction between Doppler resolution and frame duration in OTFS systems in low-Earth orbit satellite communication is resolved. This achieves high-precision channel estimation while reducing computational complexity, supporting efficient deployment of satellite-to-ground communication.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-15
- Publication Date
- 2026-03-13
AI Technical Summary
The OTFS system suffers from an inherent contradiction between Doppler resolution and frame duration in low-Earth orbit satellite communications, which leads to the destruction of channel sparsity. At the same time, its hardware implementation is highly complex and difficult to deploy efficiently on platforms with limited computing resources.
A low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM is constructed. The time-varying channel is modeled by a three-layer structured prior probability model, and the channel is estimated using a variational Bayesian off-network estimation framework and OTSM unitary transform. The channel estimation module, parameter update module, and data detection module are alternately and iteratively updated to reduce computational complexity.
While achieving high-precision channel estimation, it reduces computational complexity, supports future OTSM modulation satellite-to-ground communication scenarios, and improves the accuracy and efficiency of channel estimation.
Smart Images
Figure CN121664338A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, and in particular to a low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM. Background Technology
[0002] As a core waveform for 5G, traditional Orthogonal Frequency Division Multiplexing (OFDM) waveforms have been widely studied and applied due to their high spectral efficiency and resistance to multipath fading. For scenarios such as monitoring power facilities in remote areas and emergency disaster relief, where traditional terrestrial communication systems have insufficient coverage, reliable communication can be achieved through low-Earth orbit (LEO) satellite communication. However, the relatively high-speed motion between the transceiver ends of LEO satellite communication causes Doppler shift, which disrupts the orthogonality between OFDM subcarriers and severely affects system transmission performance. To address this issue, a communication system based on Orthogonal Time-Frequency-Space (OTFS) modulation, which offers superior resistance to Doppler shift compared to OFDM systems, can be considered.
[0003] However, the OTFS system suffers from the following drawbacks when applied to low-Earth orbit satellite communications: First, there is an inherent contradiction between Doppler resolution and frame duration. To meet the requirements of low-latency communication, the orthogonal time-frequency modulation frame length is limited, resulting in fractional Doppler frequency shift in the channel's Doppler domain, which disrupts the sparsity of the time-delay Doppler channel. Second, hardware implementation is challenging. OTFS modulation and demodulation are based on symplectic finite Fourier transform and inverse symplectic finite Fourier transform, resulting in high computational complexity and making efficient deployment difficult on platforms with limited computing resources, such as satellites. Summary of the Invention
[0004] The main objective of this invention is to propose a low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM, which can reduce computational complexity while achieving high-precision channel estimation.
[0005] This invention is achieved through the following technical solution:
[0006] A low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM includes the following steps:
[0007] Step S1: Construct a system model based on OTSM. This system model includes a transmitter, a time-varying channel, and a receiver. The time-varying channel is modeled based on a three-layer structured prior probability model. The transmitter performs OTSM transformation on the transmitted data and transmits it through the time-varying channel.
[0008] Step S2: The receiver constructs a variational Bayesian off-network estimation framework, which includes a channel estimation module, a parameter update module, and a data detection module. The data detection module uses variational Bayesian inference based on two-dimensional OTSM unitary transform to quickly iteratively estimate the transmitted data. The channel estimation module reconstructs the measurement matrix using the data detected by the data detection module and iteratively estimates the time-varying channel using a subspace-constrained variational Bayesian inference method. The parameter update module updates the time delay, Doppler hyperparameters, and measurement matrix related to the time-varying channel by maximizing the log-likelihood function.
[0009] Step S3 involves alternating and iterative updates of the channel estimation module, parameter update module, and data detection module to obtain the final channel estimation result.
[0010] Furthermore, in step S1, the time-varying channel Modeled as a Gaussian distribution ,in, For time-varying channels The precision vector follows a Bernoulli-Gamma distribution. , To represent time-varying channels The support vectors of the activation states of the elements follow a Bernoulli prior distribution. , Indicates the number of grid points for time delay measurement. Indicates the number of grid points in the Doppler measurement. and They are vectors and The There are two elements, a and b, which represent the shape and scale parameters of the time-varying channel element in its active state. , The shape and scale parameters of the time-varying channel element in its inactive state are represented by s. i for The i-th element in Indicates the channel sparsity. Indicates a Gaussian distribution. This represents the gamma distribution.
[0011] Furthermore, in step S1, the system model transmits data in each channel coherence period consisting of MN symbols, including pilot symbols and data symbols. The patterns of the pilot symbols and data symbols are represented as follows: ,in, Indicates pilot symbol, This represents the data symbol at position (l, p) in the time-delay sequence domain, where l0 and p0 represent pre-defined positions. max This represents the grid point value of the system's maximum time delay.
[0012] Furthermore, in step S2, the channel estimation module aims to solve for the set of latent variables. Regarding the received observation vector Maximum posterior distribution The problem is transformed into solving for a probability distribution using variational Bayesian inference. and The problem of minimizing KL divergence: According to mean-field theory, Factorized The minimization problem is solved using an alternating optimization method, where... , All are variational distributions. This refers to the noise accuracy in a three-layer structured prior probability model. They are variational distributions. The mean and variance, This indicates the operation of diagonalizing a vector into a matrix.
[0013] Furthermore, in step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered... The minimization problem simplifies to a quadratic optimization problem involving the variational mean and variance. Solving this quadratic optimization problem yields the following results: Closed-form solution To avoid inverting high-dimensional matrices, the gradient descent method is used to iteratively solve for the mean. The (t+1)th iteration Represented as Set the initial value of the iteration to The iteration eventually yielded ,in, , , yes The i-th diagonal element, They are respectively and The i-th diagonal element, To take off-network latency into account and Doppler components Measurement matrix, subspace for The main energy index set, matrix sum matrix Each is a matrix sum matrix submatrix, subspace Set all external elements to 0. For matrix The inverse matrix, In order to seek the truth, Represents the space of complex numbers. Represents the gradient. This indicates the gradient descent step size.
[0014] Furthermore, in step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered... , Approximately gamma distribution , The mean is expressed as ,in, For shape parameters, For scale parameters, , and They represent The mean and variance of.
[0015] Furthermore, in step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered... , Approximates Bernoulli distribution , The mean is expressed as ,in, for The normalized posterior probability.
[0016] Furthermore, in step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered... , Approximately gamma distribution , The mean is expressed as ,in, For shape parameters, Let c be the scale parameter, and c represent the prior probability of noise accuracy. In the shape parameter, d represents the prior probability of noise accuracy. The scale parameter in This represents the square of the second norm of a vector.
[0017] Furthermore, in step S2, for the parameter update module, the optimization problem of maximizing the log-likelihood function is expressed as: ,based on The first-order Taylor expansion decouples the optimization problem into two parts, each with respect to a different order of operations. and The problem is a quadratic optimization problem, where C represents a constant.
[0018] Furthermore, in step S2, the data detection module utilizes OTSM unitary transform. The received signal and the signal to be detected are converted to the time-frequency domain to obtain the following results: and By analyzing dual-observation variables Perform variational Bayesian inference to solve for latent variables. The corresponding maximum posterior distribution , This indicates the data to be tested. Represents the Walsh-Hadamard matrix. Represents the Fourier matrix. It represents the Kronecker product.
[0019] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:
[0020] The system model based on OTSM constructed in this invention includes a transmitter, a time-varying channel, and a receiver. The time-varying channel is modeled based on a three-layer structured prior probability model. The transmitter performs OTSM transformation on the transmitted data and transmits it through the time-varying channel. The receiver constructs a variational Bayesian off-network estimation framework, which includes a channel estimation module, a parameter update module, and a data detection module. The data detection module uses variational Bayesian inference based on two-dimensional OTSM unitary transform to quickly iteratively estimate the transmitted data. The channel estimation module reconstructs the measurement matrix using the data detected by the data detection module and iteratively estimates the time-varying channel using a subspace-constrained variational Bayesian inference method. The parameter update module updates the time delay, Doppler hyperparameters, and measurement matrix related to the time-varying channel by maximizing the log-likelihood function. The channel estimation module, parameter update module, and data detection module iteratively update the data to obtain the final channel estimation result. Thus, while meeting the required posterior estimation accuracy of the time-varying channel, the computational complexity is minimized by avoiding large-scale matrix inversion, enabling support for future OTSM-modulated satellite-to-ground communication scenarios. Attached Figure Description
[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0022] Figure 1 This is a flowchart of the present invention.
[0023] Figure 2 This is a schematic diagram of the system model of the present invention.
[0024] Figure 3 The observation vector of this invention The curve of the normalized mean square error changing with the number of iterations is compared with the comparison diagram of the comparison scheme.
[0025] Figure 4This is a comparison chart of the signal-to-noise ratio curve obtained by this invention as a function of the normalized mean square error of the equivalent time delay sequence domain channel and the comparison scheme. Detailed Implementation
[0026] The present invention will be further described below through specific embodiments.
[0027] like Figure 1 As shown, the low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM includes the following steps:
[0028] Step S1: Construct a system model based on OTSM. This system model includes a transmitter, a time-varying channel, and a receiver. The time-varying channel is modeled based on a three-layer structured prior probability model. The transmitter performs OTSM transformation on the transmitted data and transmits it through the time-varying channel.
[0029] System model such as Figure 2 As shown. The bandwidth of the OTSM system constructed in this embodiment is... The frame period is , and These represent the number of subcarriers and the number of time slots, respectively. and These represent the subcarrier spacing and symbol duration, respectively. The system model transmits MN symbols (pilot and data symbols) within each channel coherence period. The time-delay sequence domain data is transformed by performing Walsh-Hadamard transform and Discrete Fourier transform on the two-dimensional data matrix composed of pilot and data symbols along the sequence domain and time-delay domain, respectively. Convert to time-frequency domain data : ,in, The Walsh function represents the sorting of discrete sequences. Then, through Heisenberg transform and pulse shaping, the time-frequency domain symbols are converted into continuous-time signals: .definition and These are the maximum time delay and the maximum Doppler frequency shift, respectively, and satisfy the following conditions: After the transmitted signal passes through the satellite-to-ground time-varying channel, the received signal in the time domain is represented as: ,in, For the time-delay Doppler representation of the space-to-ground time-varying channel, ,in and Representing the complex gain of the satellite-to-ground channel, respectively, the... The time delay and Doppler frequency shift of each channel path, This indicates the number of channel paths. Next, the received signal undergoes a Wigner transform and an inverse OTSM transform to obtain the received data symbols in the time-delay sequence domain: .
[0030] The input and output of the system model can be represented as a two-dimensional quasi-convolution relationship. ,in, Represents the equivalent time delay sequence domain channel. Represents decimal numbers and The decimal number obtained by XORing the binary expansion, parameter and In fact, latency and Doppler shift It can be an integer or a fraction. The sampling function is defined by multiple coupled variables, including the symbol position in the time delay sequence field and the off-grid delay. and Doppler shift When the transceiver meets the biorthogonal pulse shaping condition, Decoupling , and This is the subsampling function.
[0031] Under integer time-delay Doppler resolution conditions, the time-delay sequence domain equivalent channel The system exhibits sparse multipath propagation in the time-delay domain and paired multipath propagation in the sequence domain. This is due to the mismatch between the Walsh-Hadamard transform of the OTSM system and the symplectic finite Fourier transform of the OTFS system. Performing the aforementioned two transforms sequentially on the equivalent channel in the time-delay sequence domain yields a sparse time-delay Doppler equivalent tapped channel. However, due to the presence of fractional time delay and fractional Doppler frequency shift, both the time-delay Doppler channel and the time-delay sequence domain channel exhibit bi-domain propagation. Therefore, it is necessary to directly estimate the taps of the original physical time-delay sequence domain channel, rather than the equivalent time-delay sequence domain channel. Once a more accurate fractional Doppler frequency shift and fractional delay are obtained, the equivalent time-delay sequence domain channel and the corresponding measurement matrix can be reconstructed, leading to more accurate results in subsequent iterations.
[0032] Based on the equivalent time-delay sequence domain channel and the decoupled sampling function, the orthogonal time-multiplexed signal input and output are reconstructed into matrix form, i.e. ,in Represents the measurement matrix. This represents the set of non-zero channel gains. Due to the subsampling function... and Corresponding to the respective Fourier transform and Walsh-Hadamard transform, the matrix and It can be diagonalized to and Among them, the diagonal array and The Element and the The elements are represented as follows: and .
[0033] The graphical representation of pilot symbols and data symbols is as follows: ,in, Indicates pilot symbol, This represents the data symbol at position (l, p) in the time-delay sequence domain, where l0 and p0 represent pre-defined positions. max This represents the grid point value of the system's maximum time delay.
[0034] According to the measurement matrix virtual resolution with latency and Doppler virtual resolution Building a virtual mesh Meanwhile, offline delay and Doppler components are also considered. The measurement matrix is obtained by using a first-order Taylor expansion. Expressed as ,matrix sum matrix They are matrices Regarding time delay grid points and Doppler grid points The first derivative, Let be the constant term in the Taylor expansion. The channel estimation model can be expressed as: The data transmission model is represented as: In the channel model, the measurement matrix... sparse channel vector ,noise In the data transmission model, the channel matrix Data vector .
[0035] Consider the channel estimation and data detection problem as a compressed sensing problem. In the channel estimation model, based on the received signal and pilot data, the sparse channel vector is simultaneously realized using compressed sensing methods. Accurate estimation and measurement matrix Off-network correction; in the data detection model, the data vector is recovered using compressed sensing based on the received signal and the channel vector estimated in the previous step. However, due to the frame length of the OTSM system... Due to the relatively long channel length, traditional compressed sensing methods cannot balance the issues of channel estimation accuracy and computational complexity.
[0036] Traditional sparse Bayesian methods typically model the signal to be estimated as a two-layer sparse prior, namely the sparse signal vector and its accuracy. However, the performance of two-layer sparse prior Bayesian methods is limited when dealing with incomplete prior information and ill-conditioned measurement matrices. For the sparse channel estimation problem, to more fully capture the prior structure of the sparse signal, a three-layer structured prior probability model is used for sparse time-varying channels. Modeling is performed.
[0037] First, assume noise. Prior probability Noise accuracy Prior probability ,parameter and The shape parameter and scale parameter of the noise accuracy are represented respectively, hence the conditional probability distribution. Assuming support vectors Indicates channel The activation states of elements follow a Bernoulli prior distribution. ,in This represents the channel sparsity. Based on support vectors. Assuming the channel precision vector Follows Bernoulli-Gamma distribution, i.e. The sparse channel is then modeled as a circular cyclic complex Gaussian distribution. ,in, Indicates the number of grid points for time delay measurement. Indicates the number of grid points in the Doppler measurement. and They are vectors and The There are two elements, a and b, which represent the shape and scale parameters of the time-varying channel element in its active state. , The shape and scale parameters of the time-varying channel element in its inactive state are represented by s. i for The i-th element in Indicates the channel sparsity. Indicates a Gaussian distribution. In this context, 0 represents the mean of the Gaussian distribution. Let V be the variance of the Gaussian distribution. Represents the gamma distribution. Gamma distribution The parameters.
[0038] Off-network components and Each is modeled as a uniform distribution. and Off-grid components are updated via the grid estimation module. Based on the three-layer structured prior probability model, the joint distribution... Represented as Channel estimation is based on the joint distribution. Approximate the posterior distribution of each latent variable.
[0039] To address the data detection problem and simplify computational complexity, data symbols are emitted from the time-delay sequence domain. The model is an independent and identically distributed Gaussian distribution, i.e. ,in , The average symbol energy.
[0040] Step S2: The receiver constructs a variational Bayesian off-network estimation framework, which includes a channel estimation module, a parameter update module, and a data detection module. The data detection module uses variational Bayesian inference based on two-dimensional OTSM unitary transform to quickly iteratively estimate the transmitted data. The channel estimation module reconstructs the measurement matrix using the data detected by the data detection module and iteratively estimates the time-varying channel using a subspace-constrained variational Bayesian inference method. The parameter update module updates the time delay, Doppler hyperparameters, and measurement matrix related to the time-varying channel by maximizing the log-likelihood function.
[0041] The channel estimation module aims to solve for the set of latent variables. Regarding the observation vector received by the receiver Maximum posterior distribution The problem is transformed into solving for a probability distribution using variational Bayesian inference. and The problem of minimizing KL divergence: According to mean-field theory, Factorized For variational distribution Each element is subject to independent constraints, i.e. The minimization problem is solved using an alternating optimization method, given a set of latent variables. Probability distribution of other variables Variational distribution It can be represented as: ,in, They are variational distributions. The mean and variance, This indicates the operation of diagonalizing a vector into a matrix.
[0042] For variational distribution The above minimization problem simplifies to a problem concerning the variational mean. and variance The problem of quadratic optimization Solving this quadratic optimization problem yields the following results: Closed-form solution .
[0043] In solving the mean This requires inverting a high-dimensional matrix, with a complexity of O(n log n). To avoid inverting high-dimensional matrices and reduce computational complexity, a subspace is defined. for The main energy index set, i.e. subspace Elements outside the subspace are set to 0. The mean is solved iteratively using the gradient descent method. The (t+1)th iteration Represented as Set the initial value of the iteration to The iteration eventually yielded ,in, , , In order to seek the truth, They are respectively and The i-th diagonal element, yes The i-th element of the matrix sum matrix Each is a matrix sum matrix submatrix, The dimension is complex space, Represents the gradient. This indicates the gradient descent step size.
[0044] When using the alternating optimization method to solve the above minimization problem, for the variational distribution... According to the three-layer structured prior probability model, Approximately gamma distribution , where shape parameters Scale parameters Therefore The mean is expressed as , , and They represent The mean and variance of.
[0045] When using the alternating optimization method to solve the above minimization problem, for the variational distribution... According to the three-layer structured prior probability model, Approximates Bernoulli distribution , for The normalized posterior probability is expressed as: , ,therefore, The mean is expressed as .
[0046] When using the alternating optimization method to solve the above minimization problem, for the variational distribution... According to the three-layer structured prior probability model, Approximately gamma distribution Shape parameters Scale parameters Therefore The mean is expressed as Where c represents the prior probability of noise accuracy. In the shape parameter, d represents the prior probability of noise accuracy. The scale parameter in This represents the square of the second norm of a vector.
[0047] Consider fractional delay and fractional Doppler. log-likelihood function For the parameter update module, the optimization problem of maximizing the log-likelihood function is expressed as: ,based on First-order Taylor expansion The optimization problem is decoupled into two parts, each concerning... and The quadratic optimization problem, with For example, the quadratic optimization problem is expressed as: , , Therefore, the required off-grid parameters However, when the number of grid points is large, high-dimensional matrix inversion is still required for off-grid updates. To avoid matrix inversion, parameter updates are only performed on... subspace The gradient descent method is used to iteratively solve for the mean. , and maintain Other off-grid components are 0, that is, Once the iteration converges, it is necessary to limit the off-network components. exist Within the range. For Doppler off-grid parameters The update is similar. Here, C represents a constant. For the constant term of the Taylor expansion, the matrix... sum matrix They are Regarding time delay grid points and Doppler grid points The first derivative.
[0048] To obtain more accurate channel estimation results, a low-complexity data detection module is used to assist in channel estimation. In the data detection module, due to the actual time-delay sequence domain channel matrix... Unknown, only the previous channel estimate can be used. Through the equivalent time delay sequence domain channel Channel matrix for approximate reconstruction of data detection Next, the data detection and channel estimation steps are iteratively performed in the data detection module. It's important to note that the "channel estimation" step in the data detection module is only used to obtain accurate data symbols. Subsequently, these detected symbols will be used as "virtual pilots" in the channel estimation module to optimize the channel estimation results.
[0049] Constructing a channel matrix using the equivalent time delay sequence domain channel. And expressed in the form of eigenvalue decomposition, that is , where the diagonal matrix .according to Eigenvalue decomposition structure, OTSM unitary transform The received signal and the signal to be detected are converted to the time-frequency domain to obtain the following results: and By analyzing dual-observation variables Perform variational Bayesian inference to solve for latent variables. The corresponding maximum posterior distribution , This represents the data to be detected, i.e., the initial time-delay sequence field vector. Represents the Walsh-Hadamard matrix. Represents the Fourier matrix. It represents the Kronecker product.
[0050] Variational distribution : The variational approximation is expressed as follows: By definition , the mean and variance They are represented as follows: It is concluded that, by using the OTSM unitary transform, the... The variational posterior operation is transformed from a large-scale inverse operation to a Hadamard product, reducing the computational complexity from... Reduce to When the calculation is complete Then, it is re-transformed by inverse unitary transformation. Obtain an estimate of the data symbol.
[0051] Step S3 involves alternating and iterative updates of the channel estimation module, parameter update module, and data detection module to obtain the final channel estimation result.
[0052] Step S2 specifically describes the channel support vector in the Bayesian off-network estimation framework. variational distribution Channel precision vector distributed Noise precision vector variational distribution sparse channel vector variational distribution and data detection vector variational distribution The specific update process involves iteratively optimizing the three modules mentioned above until the calculation results converge, and then combining this with the equivalent time delay sequence domain channel. The channel environment is reconstructed, and the channel estimation results of the OTSM system are finally obtained.
[0053] The performance of the invention was verified using Matlab numerical simulation software.
[0054] Simulation parameters: Set the OTSM frame size to... The subcarrier spacing is The satellite's speed is User speed is Set data symbol power Pilot symbol power is higher than data symbol power. ,Right now The NTN-TDL-D satellite-to-ground propagation environment reported in 3GPP TR38.811 is selected as the simulation channel. A channel based on the equivalent delay sequence domain is defined. Normalized mean square error ( The indicator is Define the normalized mean square error index of the observed vector y as: .
[0055] Figure 3 and Figure 4 In the comparison scheme:
[0056] The data-assisted 3LHS-VBI method, which combines variational Bayesian inference with a three-layer prior structure of the data-assisted module, still requires large-scale matrix inversion calculations.
[0057] The 3LHS-VBI method with known whole frame data: The whole frame data is known and directly used as pilots. Channel estimation is performed using variational Bayesian inference with a three-layer prior structure, which requires large-scale matrix inversion.
[0058] The data-assisted SC-VBI method is the subject of this invention.
[0059] like Figure 3The figure shows the normalized mean square error of the observation vector y of the present invention and the comparison scheme as a function of the number of iterations under a signal-to-noise ratio of 12dB. Figure 3 As can be observed, all three methods converge to below -13dB within 15 iterations. The 3LHS-VBI method, with known full-frame data, requires no data detection and has the fastest convergence speed, serving as an upper bound for the channel estimation performance of other methods. The convergence performance of this invention is comparable to that of the data-assisted 3LHS-VBI method, but its computational complexity is lower. This indicates that this invention achieves a good balance between convergence performance and computational complexity.
[0060] like Figure 4 The figure shows the variation curves of the normalized mean square error of the equivalent time delay sequence domain channel under different signal-to-noise ratio conditions, comparing the methods of the present invention and the comparison method. Figure 4 As can be observed, the data-assisted channel estimation method significantly improves system performance compared to the traditional 3LHS-VBI method. When the signal-to-noise ratio (SNR) is relatively high at 25 dB, the data-assisted 3LHS-VBI method achieves a 13.61 dB NMSE gain compared to the method without data assistance, while the data-assisted SC-VBI method only achieves a 9.19 dB NMSE gain. Compared to the 3LHS-VBI method with known full-frame data, the NMSE difference of the data-assisted 3LHS-VBI method under high SNR conditions can be reduced to less than 1 dB, indicating that the three-layer structured prior probability model can accurately capture the potential sparsity characteristics of the physical channel in the time-delay sequence domain. Furthermore, the data-assisted SC-VBI method achieves a channel estimation performance close to -35 dB and achieves lower computational overhead in practical satellite-to-ground channel estimation.
[0061] In this invention, the terms "first," "second," and "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. The use of terms such as "upper," "lower," "left," "right," "front," and "rear" to indicate orientation or positional relationships is based on the orientation or positional relationships shown in the accompanying drawings and is only for the convenience of describing the invention, not to indicate or imply that the device referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation on the scope of protection of this invention. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.
[0062] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0063] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.
Claims
1. A low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM, characterized in that: Includes the following steps: Step S1: Construct a system model based on OTSM. This system model includes a transmitter, a time-varying channel, and a receiver. The time-varying channel is modeled based on a three-layer structured prior probability model. The transmitter performs OTSM transformation on the transmitted data and transmits it through the time-varying channel. Step S2: The receiver constructs a variational Bayesian off-network estimation framework, which includes a channel estimation module, a parameter update module, and a data detection module. The data detection module uses variational Bayesian inference based on two-dimensional OTSM unitary transform to quickly iterate and estimate the transmitted data. The channel estimation module reconstructs the measurement matrix from the data detected by the data detection module and iteratively estimates the time-varying channel using the variational Bayesian inference method with subspace constraints. The parameter update module performs off-network updates on the time-varying channel-related delay and Doppler hyperparameters and measurement matrix by maximizing the log-likelihood function; Step S3 involves alternating and iterative updates of the channel estimation module, parameter update module, and data detection module to obtain the final channel estimation result.
2. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 1, characterized in that: In step S1, the time-varying channel Modeled as a Gaussian distribution ,in, For time-varying channels The precision vector follows a Bernoulli-Gamma distribution. , To represent time-varying channels The support vectors of the activation states of the elements follow a Bernoulli prior distribution. , Indicates the number of grid points for time delay measurement. Indicates the number of grid points in the Doppler measurement. and They are vectors and The There are two elements, a and b, which represent the shape and scale parameters of the time-varying channel element in its active state. , The shape and scale parameters of the time-varying channel element in its inactive state are represented by s. i for The i-th element in Indicates the channel sparsity. Indicates a Gaussian distribution. This represents the gamma distribution.
3. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 2, characterized in that: In step S1, the system model transmits data in each channel coherence period consisting of MN symbols, including pilot symbols and data symbols. The patterns of the pilot symbols and data symbols are represented as follows: ,in, Indicates pilot symbol, This represents the data symbol at position (l, p) in the time-delay sequence domain, where l0 and p0 represent pre-defined positions. max This represents the grid point value of the system's maximum time delay.
4. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 3, characterized in that: In step S2, the channel estimation module aims to solve for the set of latent variables. Regarding the received observation vector Maximum posterior distribution The problem is transformed into solving for a probability distribution using variational Bayesian inference. and The problem of minimizing KL divergence: According to mean-field theory, Factorized The minimization problem is solved using an alternating optimization method, where... , All are variational distributions. This refers to the noise accuracy in a three-layer structured prior probability model. They are variational distributions. The mean and variance, This indicates the operation of diagonalizing a vector into a matrix.
5. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 4, characterized in that: In step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered. The minimization problem simplifies to a quadratic optimization problem involving the variational mean and variance. Solving this quadratic optimization problem yields the following results: Closed-form solution To avoid inverting high-dimensional matrices, the gradient descent method is used to iteratively solve for the mean. The (t+1)th iteration Represented as Set the initial value of the iteration to The iteration eventually yielded ,in, , , yes The i-th diagonal element, They are respectively and The i-th diagonal element, To take off-network latency into account and Doppler components Measurement matrix, subspace for The main energy index set, matrix sum matrix Each is a matrix sum matrix submatrix, subspace Set all external elements to 0. For matrix The inverse matrix, In order to seek the truth, Represents the space of complex numbers. Represents the gradient. This indicates the gradient descent step size.
6. The low-overhead channel estimation method for a low-Earth orbit satellite communication system based on OTSM according to claim 5, characterized in that: In step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered. , Approximately gamma distribution , The mean is expressed as ,in, For shape parameters, For scale parameters, , and They represent The mean and variance of.
7. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 6, characterized in that: In step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered. , Approximates Bernoulli distribution , The mean is expressed as ,in, for The normalized posterior probability.
8. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 7, characterized in that: In step S2, when solving the minimization problem using the alternating optimization method, the variational distribution is considered. , Approximately gamma distribution , The mean is expressed as ,in, For shape parameters, Let c be the scale parameter, and c represent the prior probability of noise accuracy. In the shape parameter, d represents the prior probability of noise accuracy. The scale parameter in This represents the square of the second norm of a vector.
9. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 8, characterized in that: In step S2, for the parameter update module, the optimization problem of maximizing the log-likelihood function is expressed as: ,based on The first-order Taylor expansion decouples the optimization problem into two parts, each with respect to a different order of operations. and The problem is a quadratic optimization problem, where C represents a constant.
10. The low-overhead channel estimation method for low-Earth orbit satellite communication systems based on OTSM according to claim 9, characterized in that: In step S2, the data detection module utilizes OTSM unitary transform. The received signal and the signal to be detected are converted to the time-frequency domain to obtain the following results: and By analyzing dual-observation variables Perform variational Bayesian inference to solve for latent variables. The corresponding maximum posterior distribution , This indicates the data to be tested. Represents the Walsh-Hadamard matrix. Represents the Fourier matrix. It represents the Kronecker product.