A high-resolution broadband spatial domain non-stationary channel parameter estimation method

Abstract

The present invention discloses a high-resolution broadband spatial non-stationary channel parameter estimation method, which can extract multipath information from channel measurement data, and is divided into four steps, specifically including: 1) parameter initialization, the initialization parameters include the arrival angle, departure angle and delay of the multipath; 2) parameter refinement, this step is based on the spherical wave model to perform a refined search for the initialization parameters; 3) multipath spatial domain birth and death identification, this step is used to identify the birth and death of multipath in the array domain; 4) multipath screening, this step is used to eliminate the multipath estimation results with low signal-to-noise ratio. The algorithm proposed by the present invention adopts spherical wavefront, takes antenna polarization into consideration, and introduces the birth and death of multipath in the array dimension for the first time in the maximum likelihood estimation. Channel measurement data verification shows that this algorithm can extract more effective multipaths than traditional algorithms. In addition, the algorithm supports parameter estimation of arbitrary array shapes, and can support single-polarization channel parameter estimation by simplifying the estimation of multipath polarization components.

Description

Technical Field

[0001] The invention belongs to the field of large-scale MIMO wireless channel measurement and parameter estimation, and relates to a high-resolution broadband spatial domain non-stationary channel parameter estimation method. Background Art

[0002] Applications drive technological innovation. With the development of mobile communication technology, many new application scenarios have emerged, such as millimeter wave, ultra-large-scale MIMO, and integrated communication and perception. New application scenarios bring new channel characteristics. The channel bandwidth of the millimeter wave frequency band is generally large, which brings broadband characteristics; ultra-large-scale MIMO brings spherical wave characteristics due to its large array size; in order to obtain high-resolution delay and angle information, the integrated interawareness application scenario needs to adopt a broadband ultra-large-scale MIMO array, so it has both broadband and spherical wave characteristics. In order to establish a channel model that can accurately describe these new application scenarios, it is first necessary to carry out channel measurements for the corresponding scenarios, and then extract the channel model parameters through the channel parameter estimation algorithm.

[0003] However, the new channel characteristics have brought challenges to traditional channel parameter estimation algorithms. The far-field assumption and narrowband assumption commonly used in traditional channel parameter estimation algorithms will no longer be applicable in broadband ultra-large-scale MIMO measurements. The specific reasons are: 1) The far-field plane wave assumption means that the electromagnetic waves leaving or arriving at the antenna array through multipath can be regarded as plane wave fronts. This assumption is valid when the distance between the transmitter and receiver is greater than the Rayleigh distance. However, as the number of antennas increases, the aperture of the antenna array increases, and the distance between the transmitter and receiver may be less than the Rayleigh distance of the array. At this time, the plane wave assumption will no longer apply, and a spherical wave front needs to be used. In the spherical wave assumption, the scatterer is abstracted as a point, and the time delay and angle of the multipath leaving or arriving at the array need to be calculated based on the geometric relationship between the scatterer and the array; 2) The narrowband assumption is valid when G / λ<<f c / B, where G represents the antenna aperture, λ represents the wavelength of the propagating signal, and f c represents the center frequency, and B represents the signal bandwidth. When the signal bandwidth increases to the point where the narrowband assumption is no longer satisfied, the delay and angle of the multipath are no longer independent parameters and cannot be estimated separately in parameter estimation; 3) In addition, a large number of channel measurement results show that the ultra-large-scale MIMO array has the phenomenon of multipath birth and death in the spatial domain, which is also called the spatial non-stationary characteristic.

[0004] Existing parameter estimation algorithms can be divided into three types: spectral estimation algorithm, parameter subspace estimation algorithm and maximum likelihood estimation algorithm. Parameter estimation algorithms based on spectral estimation are generally of low complexity, but the resolution of spectral estimation algorithms is also low. Estimation of signal parameter via rotational invariance techniques (ESPRIT) and multiple signal classification (MUSIC) are two classic parameter estimation algorithms based on parameter subspace. However, both algorithms are based on the plane wave assumption and cannot be applied to ultra-large-scale MIMO scenarios. The expectation maximization (EM) algorithm is a classic maximum likelihood estimation algorithm with the characteristics of high resolution and high complexity. In order to reduce the complexity of the EM algorithm, the space-alternating generalized expectation-maximization (SAGE) algorithm was proposed. This algorithm reduces the search complexity by dividing the parameters into multiple subsets and estimating them separately. The SAGE algorithm is widely used in measurement data processing due to its high resolution and acceptable complexity. However, although a large number of measurements have found that there is a phenomenon of spatial multipath birth and death in broadband ultra-large-scale MIMO, the current SAGE algorithm still does not take this channel characteristic into account. The mismatch between this model and the measurement results will lead to estimation errors, causing the algorithm to estimate many pseudo-paths, and may even cause the algorithm to fail to converge. In order to make up for the shortcomings of the existing channel parameter estimation algorithm, the present invention proposes a high-resolution channel parameter estimation algorithm based on maximum likelihood, which takes into account the spherical wavefront and broadband effects. In addition, the present invention introduces a spatial non-stationary factor in the traditional spherical wave channel model. By estimating the spatial non-stationary factor of each multipath, the birth and death of multipath in the spatial domain can be identified. Summary of the invention

[0005] The present invention proposes a high-resolution broadband spatial domain non-stationary channel parameter estimation method, wherein the algorithm takes into account the birth and death of multipath in the spatial domain, spherical wavefront, large bandwidth and antenna polarization.

[0006] The specific technical solutions of the present invention are as follows:

[0007] Step S1: introduce a broadband spatial non-stationary channel model.

[0008] Step S2: introduce the continuous interference elimination method.

[0009] Step S3: Give the specific process of initializing the parameters of the channel parameter estimation algorithm.

[0010] Step S4: Give a specific process of refining the parameters of the channel parameter estimation algorithm.

[0011] Step S5: provide a specific process of spatial multipath generation and death identification of the channel parameter estimation algorithm.

[0012] Step S6: Give the specific process of multipath screening of the channel parameter estimation algorithm.

[0013] Specifically, step S1 includes the following steps:

[0014] Step S101: Assume that an antenna array receiving end has M r antennas, the transmitter has M t antennas, then under the assumption of narrow-band plane waves, the transmission function of the lth multipath at the mth antenna at the receiving end and the nth antenna at the transmitting end is:

[0015]

[0016] in Indicates M f transmission frequencies, [·] T represents the transposition operation, <·> represents the inner product operation, P Rx and P Tx denote the polarization of the receiving and transmitting antennas, respectively. Indicates that the lth multipath is at the receiving end P Rx Polarization and Transmitter P Tx Polarization complex gain, Ω Rx,l and Ω Tx,l denote the arrival angle and departure angle of the lth multipath, respectively. and They represent the mth antenna at the receiving end at P Rx The complex gain of the polarization and the nth antenna at the transmitting end at P Tx Polarization complex gain. Under the narrowband assumption, the effect of frequency on antenna gain is ignored, and only the antenna gain at the center frequency is considered. Rx,m and r Tx,n Respectively represent the position vectors of the mth antenna at the receiving end and the nth antenna at the transmitting end, τ l represents the delay of the lth multipath.

[0017] Step S102: In the broadband ultra-large-scale MIMO array, the narrowband plane wave assumption is no longer applicable. In this case, the spherical wave assumption needs to be adopted, and the relationship between antenna gain and frequency also needs to be considered. In this case, the lth multipath is at the fth k The transfer function of the frequency point is:

[0018]

[0019] Among them, Ω Rx,m,l represents the arrival angle of the lth multipath at the mth antenna at the receiving end, Ω Tx,n,l represents the departure angle of the lth multipath at the nth antenna at the transmitting end, Indicates that the mth antenna at the receiving end is at the kth frequency and Ω Rx,m,l The complex gain in the direction, Indicates that the nth antenna at the transmitting end is at the kth frequency and Ω Tx,n,l Complex gain in direction, τ m,n,l It represents the delay of the lth multipath from the mth antenna at the receiving end to the nth antenna at the transmitting end.

[0020] The broadband spherical wave model in step S103 and step S102 can be rewritten in matrix form, as follows:

[0021]

[0022] in, represents the parameter set constituting the lth multipath, d Tx,l represents the distance from the lth first-hop cluster to the reference antenna of the transmitter, d Rx,l represents the distance from the lth last hop cluster to the reference antenna at the receiving end, θ Tx,l and θ Rx,l denote the elevation departure angle and elevation arrival angle of the lth multipath, φ Tx,l and φ Rx,l They represent the azimuth departure angle and azimuth arrival angle of the lth multipath, ⊙ represents the multiplication of the corresponding elements of the matrix, Ψ l represents the delay matrix of the lth multipath, A l Represents the complex amplitude matrix of the lth multipath.

[0023] Step S104: The broadband spherical wave model does not consider the birth and death of multipath in the array domain. By introducing the array birth and death coefficient ζ Rx and Tx , we can get the broadband spatial non-stationary channel transfer function:

[0024]

[0025] Among them, Rx,l and Tx,l Respectively represent the birth and death coefficients of the lth multipath at the receiving end and the transmitting end. Rx,l The mth element in is denoted as ζ Rx,m,l , represents the birth and death coefficient of the lth multipath at the mth antenna at the receiving end, and its value range is 0 and 1, ζ Rx,m,l = 0 means that the lth multipath disappears at the mth antenna at the receiving end, ζ Rx,m,l =1 means that the lth multipath survives at the mth antenna at the receiving end.Tx,n,l It represents the birth and death coefficient of the lth multipath at the nth antenna of the transmitter, and its value range is {0,1}. Its specific meaning is the same as ζ Rx,m,l The same explanation will not be repeated here.

[0026] Step S105: The channel transfer function Y(f) actually measured is:

[0027]

[0028] Among them, Θ l ={Γ l ,ζ Rx,l ,ζ Tx,l} represents the parameter set that constitutes the lth multipath in the broadband spatial non-stationary channel model, L represents the number of multipaths, n(f) represents the noise vector, Represents the power of the noise.

[0029] The continuous interference elimination method can effectively reduce the interference of high-power multipath to low-power multipath. The present invention adopts the interference elimination method. Step S2 will elaborate on the process of the method in detail:

[0030] Step S201: Based on the spatial non-stationary channel model and the estimated l-th multipath parameter set Reconstruct the channel transmission function H(f; Θ) of the lth multipath l ).

[0031] Step S202: Estimate the transmission function Y of the lth multipath based on the successive interference cancellation method l (f):

[0032]

[0033] Steps S3-S6 mainly describe the detailed process of the algorithm. The algorithm is divided into four steps: parameter initialization, parameter refinement, multipath birth and death identification, and multipath screening. Step S3 will describe the parameter initialization step in detail.

[0034] Step S301: In the parameter initialization step, the parameter estimation is based on the narrowband plane wave channel model, so it is necessary to first select some frequency points near the center frequency from the broadband measurement data to ensure that the extracted channel measurement data meets the narrowband assumption. Suppose the extracted frequency point is f narr , then the result of the initialization of the l-th multipath delay is:

[0035]

[0036] Step S302: The initialization result of the 1st multipath arrival angle is:

[0037]

[0038] where ||·|| represents the Frobenius norm, C Rx,l,init Indicates the receiving end steering vector during parameter initialization. The steering vector here is calculated based on the narrowband plane wave assumption. The specific calculation method is:

[0039] Represents a three-dimensional matrix For the matrix slice of the first dimension, the (m,n)th element in the matrix X is:

[0040]

[0041] Step S303: The initialization result of the l-th multipath departure angle is:

[0042]

[0043] in, g in f k The frequency point is calculated as:

[0044]

[0045] in,[·] * Represents the conjugate operation. and Represents the receiving end P Rx Polarization at f k The steering vector corresponding to the frequency point and the transmitter P Tx Polarization at f k The steering vector corresponding to the frequency point. During the parameter initialization process, and Use and , and the antenna gain at the center frequency is used when calculating the steering vector. The matrix D is k The frequency point is calculated as:

[0046]

[0047] in, represents the Kronecker product. C Rx / Tx,k,l Indicates that the lth multipath is at f k The steering vector of the receiving / transmitting antenna array at the frequency point, during the parameter initialization process, C Rx,k,l and C Tx,k,l Use C Rx,l,init and C Tx,l,init Instead, the antenna gain at the center frequency is used when calculating the steering vector.

[0048] Step S4 is used for parameter refinement. The parameter estimation in the parameter refinement step is based on the spherical wave assumption. The detailed steps are as follows:

[0049] Step S401: The method for refining the arrival angle of the first multipath is:

[0050]

[0051] Here the guiding vector C Rx,l The calculation is based on the spherical wave model, and the antenna gain at the center frequency is used to calculate the steering vector.

[0052] Step S402: The method for refining the l-th multipath departure angle is:

[0053]

[0054] in, It is based on an estimate Calculated using the spherical wave model, the steering vector C Tx,l The calculation is also based on the spherical wave model, and the antenna gain at the center frequency is used to calculate the steering vector.

[0055] Step S403: Delay refinement method:

[0056]

[0057] in, It is based on an estimate Calculated using the spherical wave model, It is based on an estimate Calculated using the spherical wave model. In addition, in order to improve the delay resolution, all frequencies of the broadband signal need to be considered during the delay refinement process, so the steering vector for each frequency needs to be calculated separately. Since the arrival angle, departure angle, and delay are preliminarily estimated during the parameter initialization process, the parameter search range in the parameter refinement step can be greatly reduced, thereby reducing the complexity of the algorithm.

[0058] Step S404: lth multipath amplitude α l Estimation method:

[0059]

[0060] Step S5 is used to estimate the birth and death coefficients of the multipath in the array domain, and iteratively estimate the results of the previous parameter estimation until the estimation result of the lth multipath converges. The specific steps are as follows:

[0061] Step S501: Estimate the birth and death coefficient ζ of the lth multipath at the mth antenna at the receiving end Rx,m,l

[0062]

[0063] in, Y m,:,l (f) represents the matrix slice of the three-dimensional matrix Y(f) for the first dimension.

[0064] Step S502: Estimate the birth and death coefficient ζ of the lth multipath at the nth antenna at the transmitting end Tx,n,l

[0065]

[0066] in, Y :,n,l (f) represents the matrix slice of the three-dimensional matrix Y(f) for the second dimension

[0067] Step S503: iteratively estimate the l-th multipath parameter, and record the l-th multipath parameter set estimated for the first time as The parameter set estimated in the i-th iteration is recorded as In this step, the parameter iteration is updated in sequence according to the arrival angle, departure angle, delay, complex amplitude, receiving end array birth and death coefficient, and transmitting end array birth and death coefficient until the final parameters converge. The specific steps are as follows:

[0068]

[0069] in, represents the cost function calculated based on the broadband spatial non-stationary spherical wave model, and the calculation method is:

[0070]

[0071] Since the number of multipaths obtained from actual channel measurement is an unknown parameter, a larger number of multipaths is generally set in the channel parameter estimation algorithm to ensure that all multipath information can be extracted. However, among the estimated multipaths, some multipath estimation results are pseudo-spectra produced by algorithm errors. Therefore, it is necessary to delete these unreliable multipath estimation results to ensure the reliability of the channel parameter estimation results. Step S6 is multipath screening, in which unreliable multipath estimation results are deleted. The specific steps are as follows:

[0072] Step S601: In this algorithm, a multipath parameter is estimated each time It is necessary to calculate its likelihood function, and then calculate the difference between the multipath likelihood function and the previous multipath likelihood function, and add γ as an overfitting penalty term based on the calculated difference. If the calculated overall likelihood function difference is greater than 0, the multipath is regarded as an overfitting result, the multipath estimation result is deleted, and the delay point is skipped in the next search. Here, the value of γ is 2, which means that the multipath estimation result with a signal-to-noise ratio lower than 0dB will be deleted. The calculation method for the difference between the two multipath likelihood values ​​is:

[0073]

[0074] Step S602: Delete the multipath estimation results when the number of visible antennas at the transceiver end is less than half of the total number of arrays.

[0075] Beneficial Effects

[0076] The present invention proposes a high-resolution broadband spatial non-stationary channel parameter estimation algorithm, which can simultaneously consider broadband characteristics, spherical wave characteristics, multipath generation and death in the spatial domain, and antenna polarization, and can be simplified to different application scenarios through flexible configuration. In addition, the algorithm also effectively reduces the complexity of the algorithm through parameter initialization and refined estimation steps; and removes unreliable multipath estimation results through multipath screening steps to ensure the reliability of the algorithm's multipath estimation results.

[0077] In order to verify the accuracy of the algorithm, the parameters of the indoor office scene channel measurement results are extracted based on the algorithm proposed in the present invention. The channel measurement includes a line-of-sight (LOS) position point and a non-LOS (NLOS) position point. The measurement data are processed using the traditional spherical wave SAGE algorithm and the broadband spatial non-stationary parameter estimation algorithm proposed in the present invention, and it is found that the algorithm proposed in the present invention can estimate more effective multipath information. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 A flowchart of a high-resolution broadband spatial domain non-stationary channel parameter estimation algorithm proposed by the present invention;

[0079] Figure 2-1 It is the time delay-azimuth arrival angle power spectrum measured at the LOS position point in the present invention;

[0080] Figure 2-2 The delay-azimuth arrival angle power spectrum estimated by the spherical wave SAGE algorithm at the LOS position point in the present invention;

[0081] Figure 2-3 The delay-azimuth arrival angle power spectrum estimated by the algorithm of the present invention for the LOS position point in the present invention;

[0082] Figure 3-1 It is the time delay-azimuth arrival angle power spectrum measured at the NLOS position point in the present invention;

[0083] Figure 3-2 The delay-azimuth arrival angle power spectrum of the NLOS position point estimated based on the spherical wave SAGE algorithm in the present invention;

[0084] Figure 3-3 It is the delay-azimuth arrival angle power spectrum estimated by the algorithm of the present invention for the NLOS position point in the present invention. DETAILED DESCRIPTION

[0085] In order to better understand the purpose, structure and function of the present invention, a high-resolution broadband spatial domain non-stationary channel parameter estimation algorithm of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0086] The specific technical solutions of the present invention are as follows:

[0087] Step S1: introduce a broadband spatial non-stationary channel model.

[0088] Step S2: introduce the continuous interference elimination method.

[0089] Step S3: Give the specific process of initializing the parameters of the channel parameter estimation algorithm.

[0090] Step S4: Give a specific process of refining the parameters of the channel parameter estimation algorithm.

[0091] Step S5: provide a specific process of spatial multipath generation and death identification of the channel parameter estimation algorithm.

[0092] Step S6: Give the specific process of multipath screening of the channel parameter estimation algorithm.

[0093] Specifically, step S1 includes the following steps:

[0094] Step S101: Assume that an antenna array receiving end has M r antennas, the transmitter has M t antennas, then under the assumption of narrow-band plane waves, the transmission function of the lth multipath at the mth antenna at the receiving end and the nth antenna at the transmitting end is:

[0095]

[0096] in Indicates M f transmission frequencies, [·] T represents the transposition operation, <·> represents the inner product operation, P Rx and P Tx denote the polarization of the receiving and transmitting antennas, respectively. Indicates that the lth multipath is at the receiving end P Rx Polarization and Transmitter P Tx Polarization complex gain, Ω Rx,l and Ω Tx,l denote the arrival angle and departure angle of the lth multipath, respectively. and They represent the mth antenna at the receiving end at P Rx The complex gain of the polarization and the nth antenna at the transmitting end at P Tx Polarization complex gain. Under the narrowband assumption, the effect of frequency on antenna gain is ignored, and only the antenna gain at the center frequency is considered. Rx,m and r Tx,n Respectively represent the position vectors of the mth antenna at the receiving end and the nth antenna at the transmitting end, τ l Represents the delay of the lth multipath.

[0097] Step S102: In the broadband ultra-large-scale MIMO array, the narrowband plane wave assumption is no longer applicable. In this case, the spherical wave assumption needs to be adopted, and the relationship between antenna gain and frequency also needs to be considered. In this case, the lth multipath is at the fth k The transfer function of the frequency point is:

[0098]

[0099] Among them, Ω Rx,m,l represents the arrival angle of the lth multipath at the mth antenna at the receiving end, Ω Tx,n,l represents the departure angle of the lth multipath at the nth antenna at the transmitting end, Indicates that the mth antenna at the receiving end is at the kth frequency and Ω Rx,m,l The complex gain in the direction, Indicates that the nth antenna at the transmitting end is at the kth frequency and Ω Tx,n,l Complex gain in direction, τ m,n,l It represents the delay of the lth multipath from the mth antenna at the receiving end to the nth antenna at the transmitting end.

[0100] The broadband spherical wave model in step S103 and step S102 can be rewritten in matrix form, as follows:

[0101]

[0102] in, represents the parameter set constituting the lth multipath, d Tx,l represents the distance from the lth first-hop cluster to the reference antenna of the transmitter, d Rx,l represents the distance from the lth last hop cluster to the reference antenna at the receiving end, θ Tx,l and θ Rx,l denote the elevation departure angle and elevation arrival angle of the lth multipath, φ Tx,land φ Rx,l They represent the azimuth departure angle and azimuth arrival angle of the lth multipath, ⊙ represents the multiplication of the corresponding elements of the matrix, Ψ l represents the delay matrix of the lth multipath, A l Represents the complex amplitude matrix of the lth multipath.

[0103] Step S104: The broadband spherical wave model does not consider the birth and death of multipath in the array domain. By introducing the array birth and death coefficient ζ Rx and Tx , we can get the broadband spatial non-stationary channel transfer function:

[0104]

[0105] Among them, Rx,l and Tx,l Respectively represent the birth and death coefficients of the lth multipath at the receiving end and the transmitting end. Rx,l The mth element in is denoted as ζ Rx,m,l , represents the birth and death coefficient of the lth multipath at the mth antenna at the receiving end, and its value range is 0 and 1, ζ Rx,m,l = 0 means that the lth multipath disappears at the mth antenna at the receiving end, ζ Rx,m,l =1 means that the lth multipath survives at the mth antenna at the receiving end. Tx,n,l It represents the birth and death coefficient of the lth multipath at the nth antenna of the transmitter, and its value range is {0,1}. Its specific meaning is the same as ζ Rx,m,l The same explanation will not be repeated here.

[0106] Step S105: The channel transfer function Y(f) actually measured is:

[0107]

[0108] Among them, Θ l ={Γ l ,ζ Rx,l ,ζ Tx,l} represents the parameter set that constitutes the lth multipath in the broadband spatial non-stationary channel model, L represents the number of multipaths, n(f) represents the noise vector, Represents the power of the noise.

[0109] The continuous interference elimination method can effectively reduce the interference of high-power multipath to low-power multipath. The present invention adopts the interference elimination method. Step S2 will elaborate on the process of the method in detail:

[0110] Step S201: Based on the spatial non-stationary channel model and the estimated l-th multipath parameter set Reconstruct the channel transmission function H(f; Θ) of the lth multipath l ).

[0111] Step S202: Estimate the transmission function Y of the lth multipath based on the successive interference cancellation method l (f):

[0112]

[0113] Steps S3-S6 mainly describe the detailed process of the algorithm. The algorithm is divided into four steps: parameter initialization, parameter refinement, multipath birth and death identification, and multipath screening. Step S3 will describe the parameter initialization step in detail.

[0114] Step S301: In the parameter initialization step, the parameter estimation is based on the narrowband plane wave channel model, so it is necessary to first select some frequency points near the center frequency from the broadband measurement data to ensure that the extracted channel measurement data meets the narrowband assumption. Suppose the extracted frequency point is f narr , then the result of the initialization of the l-th multipath delay is:

[0115]

[0116] Step S302: The initialization result of the 1st multipath arrival angle is:

[0117]

[0118] where ||·|| represents the Frobenius norm, C Rx,l,init Indicates the receiving end steering vector during parameter initialization. The steering vector here is calculated based on the narrowband plane wave assumption. The specific calculation method is:

[0119] Represents a three-dimensional matrix For the matrix slice of the first dimension, the (m,n)th element in the matrix X is:

[0120]

[0121] Step S303: The initialization result of the l-th multipath departure angle is:

[0122]

[0123] in, g in f k The frequency point is calculated as:

[0124]

[0125] in,[·] * Represents the conjugate operation. and Represents the receiving end P Rx Polarization at f k The steering vector corresponding to the frequency point and the transmitter P Tx Polarization at f k The steering vector corresponding to the frequency point. During the parameter initialization process, and Use and , and the antenna gain at the center frequency is used when calculating the steering vector. The matrix D is k The frequency point is calculated as:

[0126]

[0127] in, represents the Kronecker product. C Rx / Tx,k,l Indicates that the lth multipath is at f k The steering vector of the receiving / transmitting antenna array at the frequency point, during the parameter initialization process, C Rx,k,l and C Tx,k,l Use C Rx,l,init and C Tx,l,init Instead, the antenna gain at the center frequency is used when calculating the steering vector.

[0128] Step S4 is used for parameter refinement. The parameter estimation in the parameter refinement step is based on the spherical wave assumption. The detailed steps are as follows:

[0129] Step S401: The method for refining the arrival angle of the first multipath is:

[0130]

[0131] Here the guiding vector C Rx,l The calculation is based on the spherical wave model, and the antenna gain at the center frequency is used to calculate the steering vector.

[0132] Step S402: The method for refining the l-th multipath departure angle is:

[0133]

[0134] in, It is based on an estimate Calculated using the spherical wave model, the steering vector C Tx,l The calculation is also based on the spherical wave model, and the antenna gain at the center frequency is used to calculate the steering vector.

[0135] Step S403: Delay refinement method:

[0136]

[0137] in, It is based on an estimate Calculated using the spherical wave model, It is based on an estimate Calculated using the spherical wave model. In addition, in order to improve the delay resolution, all frequencies of the broadband signal need to be considered during the delay refinement process, so the steering vector for each frequency needs to be calculated separately. Since the arrival angle, departure angle, and delay are preliminarily estimated during the parameter initialization process, the parameter search range in the parameter refinement step can be greatly reduced, thereby reducing the complexity of the algorithm.

[0138] Step S404: The lth multipath amplitude α l Estimation method:

[0139]

[0140] Step S5 is used to estimate the birth and death coefficients of the multipath in the array domain, and iteratively estimate the results of the previous parameter estimation until the estimation result of the lth multipath converges. The specific steps are as follows:

[0141] Step S501: Estimate the birth and death coefficient ζ of the lth multipath at the mth antenna at the receiving end Rx,m,l

[0142]

[0143] in, Y m,:,l (f) represents the matrix slice of the three-dimensional matrix Y(f) for the first dimension.

[0144] Step S502: Estimate the birth and death coefficient ζ of the lth multipath at the nth antenna at the transmitting end Tx,n,l

[0145]

[0146] in, Y :,n,l (f) represents the matrix slice of the three-dimensional matrix Y(f) for the second dimension

[0147] Step S503: iteratively estimate the l-th multipath parameter, and record the l-th multipath parameter set estimated for the first time as The parameter set estimated in the i-th iteration is recorded as In this step, the parameter iteration is updated in sequence according to the arrival angle, departure angle, delay, complex amplitude, receiving end array birth and death coefficient, and transmitting end array birth and death coefficient until the final parameters converge. The specific steps are as follows:

[0148]

[0149] in, represents the cost function calculated based on the broadband spatial non-stationary spherical wave model, and the calculation method is:

[0150]

[0151] Since the number of multipaths obtained from actual channel measurement is an unknown parameter, a larger number of multipaths is generally set in the channel parameter estimation algorithm to ensure that all multipath information can be extracted. However, among the estimated multipaths, some multipath estimation results are pseudo-spectra produced by algorithm errors. Therefore, it is necessary to delete these unreliable multipath estimation results to ensure the reliability of the channel parameter estimation results. Step S6 is multipath screening, in which unreliable multipath estimation results are deleted. The specific steps are as follows:

[0152] Step S601: In this algorithm, a multipath parameter is estimated each time It is necessary to calculate its likelihood function, and then calculate the difference between the multipath likelihood function and the previous multipath likelihood function, and add γ as an overfitting penalty term based on the calculated difference. If the calculated overall likelihood function difference is greater than 0, the multipath is regarded as an overfitting result, the multipath estimation result is deleted, and the delay point is skipped in the next search. Here, the value of γ is 2, which means that the multipath estimation result with a signal-to-noise ratio lower than 0dB will be deleted. The calculation method for the difference between the two multipath likelihood values ​​is:

[0153]

[0154] Step S602: Delete the multipath estimation results when the number of visible antennas at the transceiver end is less than half of the total number of arrays.

[0155] Compare the difference between the delay-azimuth arrival angle joint power spectrum estimated by the two algorithms and the measurement results. Figure 2-1 , Figure 2-2 and Figure 2-3 It represents the comparison between the delay-azimuth angle of arrival joint power spectrum measured at the LOS position and the estimation results of the two algorithms. Figure 3-1 , Figure 3-2 and Figure 3-3 The figure shows the comparison between the delay-azimuth arrival angle joint power spectrum measured at the NLOS position point and the estimation results of the two algorithms. By comparing the estimation results of the two algorithms, it can be found that the parameter estimation results of the algorithm proposed by the present invention are closer to the measurement results, indicating that the algorithm proposed by the present invention has better performance.

[0156] In summary, the high-resolution broadband spatial non-stationary channel parameter estimation algorithm proposed in the present invention can extract multipath information from channel measurement results, and its performance is better than that of the traditional spherical wave SAGE algorithm. The multipath parameters extracted by the algorithm can be used for channel modeling or channel characteristic analysis, which is of great significance for establishing an accurate channel model.

[0157] It can be understood that the present invention is described by real channel measurement data, and those skilled in the art know that various changes or equivalent substitutions can be made to these features and embodiments without departing from the present invention. In addition, under the guidance of the present invention, these features and embodiments can be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the scope protected by the present invention.

Claims

1. A high-resolution broadband spatial domain non-stationary channel parameter estimation method, characterized in that: The steps include: Construct a broadband spatial non-stationary channel model; Estimate the transfer function of the lth multipath; Channel parameter initialization, including the arrival angle, departure angle and delay of multipath; Channel parameter refinement: This step performs a refined search on the initialized channel parameters based on the spherical wave model; Multipath spatial domain birth and death identification, this step is used to identify the birth and death of multipath in the array domain; Multipath screening is used to eliminate multipath estimation results with low signal-to-noise ratio; The specific steps of constructing a broadband spatial non-stationary channel model include: Assume that an antenna array receiver has M r antennas, the transmitter has M t antennas, then under the assumption of narrow-band plane waves, the transmission function of the lth multipath at the mth antenna at the receiving end and the nth antenna at the transmitting end is: in Indicates M f transmission frequencies, [·] T represents the transposition operation, <·> represents the inner product operation, P Rx and P Tx denote the polarization of the receiving and transmitting antennas, respectively. Indicates that the lth multipath is at the receiving end P Rx Polarization and Transmitter P Tx Polarization complex gain, Ω Rx，l and Ω Tx，l denote the arrival angle and departure angle of the lth multipath, respectively. and They represent the mth antenna at the receiving end at P Rx The complex gain of the polarization and the nth antenna at the transmitting end at P Tx Polarization complex gain; r Rx，m and r Tx，n Respectively represent the position vectors of the mth antenna at the receiving end and the nth antenna at the transmitting end, τ l represents the delay of the lth multipath; The lth multipath is at the fth k The transfer function of the frequency point is: Among them, Ω Rx，m，l represents the arrival angle of the lth multipath at the mth antenna at the receiving end, Ω Tx，n，l represents the departure angle of the lth multipath at the nth antenna at the transmitting end, Indicates that the mth antenna at the receiving end is at the kth frequency and Ω Rx，m，l The complex gain in the direction, Indicates that the nth antenna at the transmitting end is at the kth frequency and Ω Tx，n，l The complex gain in the direction, τ m，n，l represents the time delay of the lth multipath from the mth antenna at the receiving end to the nth antenna at the transmitting end; The broadband spherical wave model is rewritten into matrix form as follows: in, represents the parameter set constituting the lth multipath, d Tx，l represents the distance from the lth first-hop cluster to the reference antenna of the transmitter, d Rx，l represents the distance from the lth last hop cluster to the reference antenna at the receiving end, θ Tx,l and θ Rx，l denote the elevation departure angle and elevation arrival angle of the lth multipath, φ Tx,l and φ Rx,l They represent the azimuth departure angle and azimuth arrival angle of the lth multipath, ⊙ represents the multiplication of the corresponding elements of the matrix, Ψ l represents the delay matrix of the lth multipath, A l represents the complex amplitude matrix of the lth multipath; Broadband spherical wave model introduces array birth and death coefficient ζ Rx and Tx , we get the broadband spatial non-stationary channel transfer function: Among them, Θ l ={Γ l , Rx,l , Tx，l } represents the parameter set constituting the lth multipath in the broadband spatial non-stationary channel model; ζ Rx,l and Tx，l denote the birth and death coefficients of the lth multipath at the receiving and transmitting ends respectively; ζ Rx,l The mth element in is denoted as ζ Rx,m，l , represents the birth and death coefficient of the lth multipath at the mth antenna at the receiving end, ζ Rx,m,l = 0 means that the lth multipath disappears at the mth antenna at the receiving end, ζ Rx,m，l =1 means that the lth multipath survives at the mth antenna at the receiving end; ζ Tx，n，l represents the birth and death coefficient of the lth multipath at the nth antenna at the transmitting end; The channel transfer function Y(f) actually measured is: Where L represents the number of multipaths, n(f) represents the noise vector, represents the power of the noise; The specific steps of estimating the transmission function of the lth multipath include: Multipath estimation is performed based on the continuous interference elimination method. For the first multipath transmission function, the measured transmission function Y(f) is directly used for estimation. After estimating the first multipath parameter set, the transmission function of the first multipath is reconstructed, and the transmission function of the first multipath is subtracted from Y(f). Then, the estimation of the next multipath is continued, and the above operation is repeated until all L multipaths are estimated. The transmission function Y of the lth multipath is estimated. l (f) The specific steps of channel parameter initialization include: Select a frequency point f near the center frequency from the broadband measurement data narr , the result of initializing the l-th multipath delay is: The initialization result of the lth multipath arrival angle is: where ||·|| represents the Frobenius norm, C Rx，l，init Indicates the receiving end steering vector during parameter initialization. The steering vector is calculated based on the narrowband plane wave assumption. The specific calculation method is: Represents a three-dimensional matrix For a matrix slice of the first dimension, the (m, n)th element in the matrix X is: The initialization result of the lth multipath departure angle is: in, g in f k The frequency point is calculated as: in,[·] * represents the conjugate operation; and Represents the receiving end P Rx Polarization at f k The steering vector corresponding to the frequency point and the transmitter P Tx Polarization at f k The steering vector corresponding to the frequency point; during the parameter initialization process, and Use and Instead, the antenna gain at the center frequency is used when calculating the steering vector; the matrix D is k The frequency point is calculated as: in, represents the Kronecker product; C Rx / Tx，k，l Indicates that the lth multipath is at f k The steering vector of the receiving / transmitting antenna array at the frequency point, during the parameter initialization process, C Rx，k，l and C Tx，k，l Use C Rx，l，init and C Tx，l，init Instead, the antenna gain at the center frequency is used when calculating the steering vector; The specific steps of parameter refinement include: The method for refining the arrival angle of the lth multipath is: Steering vector C Rx，l The calculation is based on the spherical wave model, and the antenna gain at the center frequency is used to calculate the steering vector; The method for refining the lth multipath departure angle is: in, It is based on an estimate Calculated using the spherical wave model, the steering vector C Tx，l The calculation is also based on the spherical wave model, and the antenna gain at the center frequency is used to calculate the steering vector; Delay refinement method: in, It is based on an estimate Calculated using the spherical wave model, It is based on an estimate Calculated using the spherical wave model; the steering vector for each frequency point needs to be calculated separately; The lth multipath amplitude α l Estimation method: The specific steps of multipath spatial domain birth and death identification include: Estimate the birth and death coefficient ζ of the lth multipath at the mth antenna at the receiving end Rx，m，l in, Y m，：，l (f) represents the matrix slice of the three-dimensional matrix Y(f) for the first dimension; Estimate the birth and death coefficient ζ of the lth multipath at the nth antenna at the transmitter Tx，n，l in, Y ：，n，l (f) represents the matrix slice of the three-dimensional matrix Y(f) for the second dimension; The lth multipath parameter is iteratively estimated, and the lth multipath parameter set estimated for the first time is recorded as The parameter set estimated in the i-th iteration is recorded as The parameter iteration is updated in sequence according to the arrival angle, departure angle, delay, complex amplitude, receiving end array birth and death coefficient, and transmitting end array birth and death coefficient until the final parameters converge; the specific steps are as follows: in, represents the cost function calculated based on the broadband spatial non-stationary spherical wave model, and the calculation method is: The specific steps of multipath screening include: Each time a multipath parameter is estimated Calculate its likelihood function, and then calculate the difference between its likelihood function and the previous multipath likelihood function, and add γ as an overfitting penalty term based on the calculated difference. If the calculated overall likelihood function difference is greater than 0, the multipath is regarded as an overfitting result, the multipath estimation result is deleted, and the delay point is skipped in the next search; the calculation method of the difference between the two multipath likelihood values ​​is: Delete the multipath estimation results when the number of visible antennas at the transceiver end is less than half of the total number of arrays.

Images