Downlink training design method for dual wideband millimeter wave communication system based on channel reciprocity

Abstract

The application relates to a kind of downlink training design methods of double broadband millimeter wave communication system based on channel reciprocity, which utilizes the geometric reciprocity information of path number, path angle and path time delay obtained by uplink detection, combines double broadband channel model to determine downlink angle-time delay support interval;For beam oblique sight effect, design broadband flat full-digital training precoder, and alternately decomposed into mixed training precoder;Based on beam-time delay domain joint support, folding sub-interval is divided, the starting position of coset is optimized, and multiple coset pilot pattern is generated;User end is matched filtering, support detection, bayesian information criterion support refinement and block least square reconstruction, and downlink equivalent channel estimation result is obtained.Compared with prior art, the application has the advantages of reducing pilot and beam scanning overhead, suppressing broadband beam gain mismatch, target angle expansion area power leakage and beam interference, improving downlink channel estimation precision and the like.

Description

Technical Field

[0001] This invention relates to the field of millimeter-wave communication technology, and in particular to a downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity. Background Technology

[0002] Millimeter-wave communication technology is an important technology for high-frequency, high-bandwidth wireless transmission. Compared to traditional low-frequency communication, the millimeter-wave band has richer spectrum resources, enabling higher transmission rates and larger access capacity. As mobile communication systems evolve towards 5G-Advanced and 6G, millimeter-wave communication has significant application value in scenarios such as enhanced mobile broadband, ultra-high-speed access, industrial internet, and immersive services. Due to the significant propagation loss and weak diffraction ability of millimeter-wave signals, communication systems typically require large-scale antenna arrays and highly directional beamforming to improve received signal quality and coverage.

[0003] In practical millimeter-wave communication systems, hybrid precoding architecture is a commonly used beamforming implementation. This architecture connects a large-scale antenna array through a limited number of RF links and combines analog and digital precoding to complete signal transmission. It achieves a good balance between hardware cost, power consumption, and beamforming capabilities, thus having high engineering application value.

[0004] However, in broadband millimeter-wave communication systems, acquiring downlink channel state information still faces high training overhead. Existing systems typically rely on layered beam scanning, beam refinement, and pilot-aided channel estimation to obtain downlink channel information; as antenna size, subcarrier number, and system bandwidth increase, the number of training beams and pilot resource overhead also rise. Especially in dual-broadband systems, the array response varies with the subcarrier frequency, and the same training beam will exhibit pointing offset at different frequencies, i.e., beam squint effect. This effect causes a decrease in broadband beam gain, power leakage outside the target angle extension region, and enhanced mutual interference between training beams, thereby reducing the quality of downlink training observations.

[0005] Traditional OFDM systems commonly use uniform comb pilots and block pilots, which are simple and easy to deploy. However, they struggle to utilize the structured sparsity of millimeter-wave channels in the angle and delay domains, leading to wasted pilot resources. While some non-uniform pilot methods can reduce the number of pilots, their pilot positions are typically globally irregularly distributed, resulting in complex configurations, difficulties in control signaling indication, and inconvenience in pilot reuse, hindering practical deployment in standardized systems. Furthermore, while uplink and downlink exhibit some geometric reciprocity under the same propagation environment, and parameters such as path number, path angle, and path delay can be obtained from uplink probing, the complex path gain is affected by frequency and environmental variations and cannot be directly reused for downlink channel recovery.

[0006] A search of Chinese Patent Publication No. CN107465636A reveals a channel estimation method for a millimeter-wave massive MIMO array space-frequency dual-wideband system. Specifically, the method includes: obtaining initial channel information for each user based on their initial training sequence; acquiring a space-frequency dual-wideband channel model for the millimeter-wave massive MIMO array; using the model to obtain initial angle and initial delay information for each incident path of each user based on the initial channel information; adjusting the transmission time of each user for the next channel estimation step based on the initial angle and initial delay information, ensuring that the angle and delay information for each path of each user meets preset conditions; transmitting a training sequence for channel estimation based on the transmission time determined in the previous step; and after receiving the training sequences from all users, updating the angle, delay, and gain information for each channel path of each user, and reconstructing the uplink channel using the model. This existing patent achieves low-overhead uplink channel estimation through accurate channel modeling and user soft packet scheduling, while the downlink channel is directly obtained from the uplink channel, thereby accurately modeling the space-frequency dual broadband channel and reducing uplink channel estimation overhead. However, it does not address how to solve problems such as high downlink training overhead, reduced training observation quality due to beam squint, and low pilot resource allocation efficiency.

[0007] Therefore, how to utilize the shared uplink and downlink geometric path information to design low-overhead, broadband flat, and leakage-controllable training beams in dual-bandwidth millimeter-wave systems, and configure pilot patterns with clear structures, easy indication, and matching channel support characteristics, in order to reduce downlink training overhead, improve effective observation quality, and enhance downlink equivalent channel estimation accuracy, has become an urgent technical problem to be solved. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a downlink training design method for a dual broadband millimeter-wave communication system based on channel reciprocity. This method determines the downlink channel angle-delay support structure by utilizing the path geometry parameters shared by the uplink and downlink, optimizes the design of the broadband flat training precoder to suppress broadband beam gain attenuation caused by beam squint, power leakage outside the target angle extension region, and mutual interference between different training beams, and configures a multi-coupling pilot pattern that matches the joint delay support, thereby achieving reliable estimation of the downlink equivalent channel under low pilot overhead conditions.

[0009] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, a downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity is provided, characterized in that the method includes the following steps: Step S1: Utilize the reciprocity of path geometry parameters in the uplink and downlink, combined with the channel model, to obtain the angle support interval and delay support interval of each propagation path on the discrete grid of the downlink channel. Step S2: Considering the beam squint effect of the dual broadband system, broadband flattening optimization is performed on the training beams pointing to each propagation path and interference between training beams is suppressed to obtain a broadband flattened all-digital training precoder. Step S3: Considering the unit constant modulus hardware constraint of the hybrid precoder, the broadband flat all-digital training precoder is subjected to alternating optimization decomposition to obtain the broadband flat hybrid training precoder. Step S4: Construct a beam-delay domain equivalent channel based on the broadband flat hybrid training precoder, and consider the leakage effect between training beams to obtain a joint delay support set shared by each training beam. Step S5: Determine the sampling period of the multiple co-sets, divide the joint delay support set into folded sub-intervals, and filter the translation indexes of each folded sub-interval that still fall into the original joint delay support set after translation, to obtain the folded sub-interval support index set. Step S6: Determine the number of cosets that satisfy the full column rank condition of the local sensing matrix based on the set of support indices of the folded sub-intervals, and optimize the set of starting positions of the cosets to obtain the multi-coset pilot pattern; Step S7: The base station sends orthogonal downlink training pilots according to the broadband flat hybrid training precoder and the multi-coset pilot pattern. The user terminal obtains the beam-frequency domain pilot observation results after matched filtering. Step S8: The user terminal estimates the main time delay support intervals of each training beam based on the beam-frequency domain pilot observation results and the multi-coupling pilot pattern, and obtains the estimated joint support set after union processing. Step S9: Construct a local sensing matrix based on the estimated joint support set, and perform block least squares channel reconstruction on each training beam to obtain the downlink equivalent channel estimation result.

[0010] As a preferred technical solution, step S1 specifically includes the following steps: In step S101, the base station estimates the path geometric parameters shared by the uplink and downlink through uplink probing. These path geometric parameters include the number of resolvable paths. , No. Path delay of each propagation path and path angle ,in ; Step S102: Substitute the path geometry parameters into the dual broadband downlink space-frequency domain channel model, and perform a two-dimensional inverse discrete Fourier transform on the obtained channel representation to obtain the downlink angle-delay domain channel representation. Step S103: Based on the downlink angle-delay domain channel characterization, determine the first... The main energy support intervals of the propagation path in the angular and time delay dimensions are combined by Cartesian product to obtain the rectangular block support intervals of the propagation path on the discrete angle-time delay grid of the downlink channel; wherein, the main energy support interval refers to the continuous index interval in which the single-sided Dirichlet kernels coherently superimpose on the discrete grid and form the main lobe energy concentration; if the continuous index interval crosses the boundary of the discrete grid, it is represented by a cyclic index.

[0011] As a preferred technical solution, step S2 specifically includes the following steps: Step S201: Based on the beam squinting characteristics of the array response changing with the subcarrier frequency in the dual broadband channel model, determine the frequency-dependent equivalent angle of each subcarrier in the full bandwidth for each propagation path, and form the target angle extension region corresponding to each propagation path. Step S202: Construct a corresponding array steering vector for each frequency-related equivalent angle, so that each propagation path corresponds to a set of array steering vectors covering the full bandwidth equivalent angle variation; Step S203: The minimum value of the inner product gain of each all-digital training beam and the array steering vector of the corresponding propagation path is taken as the broadband worst gain of the training beam. With the optimization objective of maximizing the sum of the broadband worst gains of all training beams, under the conditions of satisfying the average power leakage constraint outside the target angle extension region, the average correlation constraint between beams, and the beam unity power constraint, the all-digital training beams are jointly optimized to obtain the broadband flat all-digital training precoder.

[0012] As a preferred technical solution, step S3 specifically includes the following steps: Step S301: Using the broadband flat all-digital training precoder as the approximation target, construct a hybrid training precoder formed by cascading analog precoders and digital precoders. Step S302: A broadband flat hybrid training precoder is obtained through alternating optimization. When the analog precoder is fixed, the digital precoder is updated using the least squares method. When the digital precoder is fixed, the analog precoder is updated using the Riemannian manifold optimization method under the unit constant modulus constraint.

[0013] As a preferred technical solution, step S4 specifically includes the following steps: Step S401: Using the broadband flat hybrid training precoder as the training beam matrix, beam domain projection and time delay domain transformation are performed on the downlink space-frequency domain channel to obtain the time delay domain channel representation corresponding to each training beam. Step S402: Based on the main time delay support intervals of the propagation paths corresponding to each training beam and the inter-beam energy leakage caused by non-orthogonal wide beams, the time delay support intervals of each propagation path obtained in step S1 are subjected to union processing to obtain a joint time delay support set shared by each training beam, and this set is used as a shared time delay support reference for the multi-coordinate pilot pattern design in step S6.

[0014] As a preferred technical solution, step S5 specifically includes the following steps: Step S501: Determine the multi-coset sampling period R And based on the total number of subcarriers K Obtain the length of the folded delay grid. Q ,in K=RQ ; Step S502: Represent the joint delay support set as several non-overlapping continuous delay intervals, and define the boundaries of each continuous delay interval according to... Q Take the modulus and set 0 and 0. Q As the first and last boundaries of the folded delay grid are sorted together, the folded delay range between adjacent boundaries is divided into folded sub-intervals; Step S503, for each folded sub-interval, with Q The translation interval is periodically translated on the original time delay grid. The translation indexes that still fall completely into the joint time delay support set after translation are selected to obtain the corresponding folded sub-interval support index set.

[0015] As a preferred technical solution, step S6 specifically includes the following steps: Step S601: Calculate the cardinality of the support index set of each folded sub-interval, and use the maximum value among them as the lower bound of the number of cosets required for full column rank reconstruction of the local sensing matrix, and determine the number of cosets of the multi-coset pilot. Step S602: Using a backward pruning method, while maintaining the full column rank of each local sensing matrix, the candidate coset starting positions are screened based on minimizing the block least squares reconstruction error and the column correlation of the local sensing matrix as optimization criteria, to obtain the coset starting position set. Step S603: Using each position in the set of co-set starting positions as the starting point of the uniform pilot subsequence, select subsequent pilot subcarriers according to the multi-co-set sampling period to generate multiple uniform pilot subsequences with the same period but different starting points, and merge the uniform pilot subsequences to obtain the multi-co-set pilot pattern.

[0016] As a preferred technical solution, step S7 specifically includes the following steps: Step S701: On the pilot subcarrier specified by the multi-coordinate pilot pattern, the base station uses the broadband flat hybrid training precoder to simultaneously excite L training beams and transmit pilot vectors at at least L consecutive pilot OFDM symbol times. The pilot vectors are arranged according to the symbol times to form a pilot symbol matrix, and the pilot sequences corresponding to different training beams are mutually orthogonal. In step S702, the user terminal performs matched filtering on the received pilot signal based on the pilot symbol matrix, separates the channel observation values ​​of each training beam on the pilot subcarrier, and obtains the beam-frequency domain pilot observation results.

[0017] As a preferred technical solution, step S8 specifically includes the following steps: Step S801: The user terminal constructs folded observations based on the beam-frequency domain pilot observation results and the multi-coset pilot pattern, using the time delay domain mode folding relationship corresponding to the multi-coset sampling, and calculates the energy projection metric between the candidate sensing atom and the folded observations to determine the coarse localization atom index of the main time delay energy range of each training beam. Step S802: Map the coarse positioning atom index and its adjacent index to the original time delay grid, construct candidate time delay support intervals, and use the Bayesian information criterion to adjust the interval boundaries to obtain the estimated main time delay support intervals of each training beam. Step S803: Perform union processing on the estimated main time delay support intervals of each training beam to obtain the estimated joint support set.

[0018] As a preferred technical solution, step S9 specifically includes the following steps: Step S901: Based on the estimated joint support set, and according to the folded sub-interval division and translation index filtering method, re-determine the estimated folded sub-interval and its corresponding estimated folded sub-interval support index set. Step S902: Combine the set of coset starting positions and the set of estimated folded sub-interval support indices to construct the local perception matrix corresponding to each estimated folded sub-interval and establish the corresponding local linear observation model. Step S903: Perform block least squares channel reconstruction within each estimated folded sub-interval in each training beam, fill the corresponding delay positions with the reconstructed beam-delay domain channel coefficients, and obtain the downlink equivalent channel estimation result through frequency domain transformation.

[0019] Compared with the prior art, the present invention has the following advantages: 1) This invention utilizes the path geometry parameters shared by the uplink and downlink to determine the downlink angle-delay support structure, transforming downlink training from full-space beam scanning to directional training oriented towards the main propagation path. This reduces the time-frequency resource overhead caused by traditional beam scanning and large-scale pilot search, and improves the efficiency of downlink channel information acquisition. 2) This invention optimizes the broadband flat training precoder for beam squint effect in dual broadband systems. While improving the worst subcarrier gain of each training beam, it suppresses power leakage outside the target angle spread region and mutual interference between training beams, thereby improving the signal-to-interference-plus-noise ratio (SINR) of pilot observation in the beam domain at the receiver. Compared with training beam designs that do not explicitly consider broadband gain flatness, leakage and mutual interference constraints, this invention can obtain more stable broadband training gain. 3) This invention projects the downlink space-frequency domain channel into a low-dimensional beam domain equivalent channel by using a broadband flat training precoder, so that the channel estimation object is transformed from the full-dimensional antenna domain channel into the equivalent channel corresponding to a small number of training beams, which reduces the downlink channel estimation dimension and subsequent reconstruction complexity, and is beneficial to obtaining the main downlink channel information with limited training resources. 4) This invention employs a multi-co-set pilot pattern that matches the joint delay support. It is formed by merging multiple uniform pilot subsequences with the same sampling period but different starting positions to create a periodic non-uniform pilot structure. Compared with traditional uniform pilot schemes, this invention can reduce invalid pilot sampling. Compared with globally irregular pilot schemes, this invention has a clearer parameterized structure, which facilitates pilot position indication, pilot reuse, and engineering deployment. 5) This invention optimizes the coset starting position based on the folded sub-interval support structure and adopts a matched support detection and block least squares reconstruction process at the user end to make the local sensing matrix meet the full column rank reconstruction condition and improve the matrix condition number, thereby reducing noise amplification in the matrix inversion process. Compared with other pilot design and channel estimation schemes, this invention can improve the accuracy of main delay support detection and reduce the normalized mean square error (NMSE) of downlink equivalent channel estimation under the same pilot budget. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the downlink training design method provided in an embodiment of the present invention; Figure 2 A flowchart of the optimization process for a broadband flat all-digital training precoder provided in an embodiment of the present invention; Figure 3 The optimized broadband flat all-digital training beammap provided in this embodiment of the invention; Figure 4 This is a schematic diagram illustrating the construction of folded sub-intervals and their supporting index sets provided in an embodiment of the present invention; Figure 5This is a schematic diagram of a multi-column pilot pattern provided in an embodiment of the present invention; Figure 6 A flowchart of the receiver channel reconstruction process provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] This embodiment uses a millimeter-wave MISO-OFDM downlink communication system with a large-scale uniform linear array configured at the base station and a single-antenna architecture at the user end as an example to illustrate the downlink training design method of the present invention. Please refer to [link to previous document]. Figure 1 This system configuration is for illustrative purposes only and does not constitute a limitation on the scope of protection of this invention. This invention is also applicable to millimeter-wave MIMO-OFDM communication systems.

[0023] The base station employs a hybrid precoding structure consisting of analog and digital precoders. Let the total number of subcarriers in the system be... The carrier frequency is The system bandwidth is The subcarrier spacing is , No. The baseband equivalent frequency of each subcarrier is Assume the geometric path parameters of the wireless propagation environment include the number of resolvable paths. And the path delay corresponding to each propagation path. and path angle direction cosine Its complex gain is denoted as ,in .

[0024] In this embodiment, the base station obtains the above-mentioned geometric path parameters through uplink detection and uses them as prior information for downlink training beam design and pilot pattern design. Since the uplink and downlink frequencies, hardware links, and propagation disturbances may be different, the downlink path complex gain is still re-estimated through the downlink training pilot.

[0025] Step S1: Based on the above path geometry parameters and the dual-bandwidth millimeter-wave downlink channel model, the base station determines the support interval of each propagation path on the downlink channel discrete angle-delay grid.

[0026] Specifically, the first The direction cosine of the path Path delay Substituting into the downlink space-frequency domain channel model, construct the downlink space-frequency domain channel matrix. It can be represented as in, This refers to the number of base station antennas. The guide vector for the transmission array. For frequency domain steering vectors, The frequency-dependent phase shift matrix is ​​introduced by the propagation delay of the array signal. Represents the Hadamard product. For the th root antenna and the first Subcarriers, The elements in can be derived from express.

[0027] right Performing a two-dimensional inverse discrete Fourier transform yields the angle-delay domain channel. For the first The path, whose time delay dimension discrete index and angle dimension discrete index are respectively represented as:

[0028]

[0029] Base station according to and Determine the first Delay support set of each path and angular support set The rectangular block support region of this path on the downlink discrete angle-delay grid is obtained in Cartesian product form. .

[0030] Step S2, the base station according to Figure 2 The process shown illustrates the design of a broadband flat all-digital training precoder to address the beam squint effect in a dual broadband system.

[0031] For the There are several propagation paths, and the base station bases determine the path angle and direction cosine based on these paths. and the frequencies of each subcarrier Determine the frequency-dependent equivalent direction cosine of the path across the full bandwidth. , and by Forming target angle extension area Subsequently, the base station constructs the corresponding array steering vector based on the equivalent direction cosines related to each frequency. Record the first The all-digital training beam is Then the broadband flat all-digital training preencoder is represented as .

[0032] This embodiment aims to improve the worst-case broadband gain of each training beam within the corresponding target angle extension region, and introduces auxiliary variables. Indicates the first If the worst-case gain lower bound of the training beam is found, then the optimization objective can be expressed as: To ensure the flatness of beam gain across the entire bandwidth, for the first... Beam training beam, which extends in the target angular region The beam gain on the equivalent direction cosines of each frequency correlation must satisfy the first type of constraint. To suppress power leakage outside the target angle extension region, in complement region Selecting uniformly from above Sampling points at several angles, and constructing the target external guidance matrix. Introducing auxiliary variables Representing the upper limit of leakage, the second type of constraint can be expressed as: To suppress mutual interference between different training beams, the average correlation between training beams is used to characterize the degree of inter-beam interference, and an auxiliary variable is introduced. The third type of constraint, representing the upper limit of mutual interference, can be expressed as: Each training beam must also satisfy a unity power constraint, the fourth type of constraint which can be expressed as:

[0033] In the specific solution, an iterative optimization method of alternating beam updates is adopted. First, an initial training beam is constructed based on the steering vector of the center frequency of each propagation path; then, the beams to be updated are selected sequentially, and a single-beam optimization subproblem is constructed under the condition that the other beams are fixed. When solving this subproblem, the beam gain lower bound constraint is linearized into an affine constraint; for the target external leakage constraint and the inter-beam interference constraint, the current auxiliary upper bound variable is used respectively. and The constraint residuals are calculated as constraint boundaries, and the corresponding non-negative dual weights are updated based on the constraint residuals. Simultaneously, the constraint residuals are used to calculate constraint residuals. and Perform a tightening update to progressively reduce the upper limits of target leakage and inter-beam interference with each iteration. After updating the current beam, normalize it using its unit norm. Repeat the above process until the convergence criterion is met, and output the optimization results. Please refer to the optimized beam pattern. Figure 3 .

[0034] Step S3, after obtaining Subsequently, considering the hardware constraints of hybrid precoding, it was decomposed into analog precoders. and digital pre-encoder The cascaded form yields a broadband flat hybrid training preencoder. .

[0035] The hybrid decomposition process can be represented as follows: in, The elements in the equation must satisfy the constant modulus constraint.

[0036] In the specific solution, an alternating minimization method is adopted. (Fixed) hour, The update is in, These are the regularization parameters. Done. After the update, power scaling is applied to make... Each column vector satisfies the unity power constraint.

[0037] fixed At that time, the Riemannian manifold descent method is used to update the time under the unit constant modulus constraint. Repeat the alternating update process described above until the convergence condition is met, resulting in a hybrid training preencoder that satisfies the hardware constraints. .

[0038] Step S4: Using the information obtained in step S3 Construct the beam-delay domain equivalent channel and generate a joint delay support set shared by all training beams.

[0039] Specifically, for By sequentially performing beam domain projection and time delay domain transformation, we obtain ,in, Let K be the DFT matrix. The The line represents the first The time-delay domain equivalent channel corresponding to the training beam.

[0040] Considering the inter-beam leakage phenomenon caused by non-orthogonal wide beams, based on the path delay support intervals in step S1... To obtain a shared joint latency support set .

[0041] Step S5, according to Figure 4The method shown applies to the joint delay support set. Perform folded sub-interval partitioning and determine the support index set corresponding to each folded sub-interval. Assume the multi-coset sampling period is... ,and ,in This represents the length of the folded delay grid.

[0042] Will It is represented as several non-overlapping consecutive time delay intervals, i.e. ,in and They represent the first The left and right boundaries of a continuous time delay interval.

[0043] According to the left and right boundaries of all continuous intervals Take the modulus and set 0 and 0. The first and last boundaries of the folded time-delay mesh are sorted together to obtain... This divides the folded delay grid into There are n disjoint folded sub-intervals, where the nth sub-interval is n. The sub-intervals are represented as follows: .

[0044] For any folded sub-interval ,by To generate periodically translated copies of the original time-delay grid at intervals. ,in , This is a translation symbol. If... Completely fall into Then Add the supporting index set to the folded sub-interval, i.e.

[0045] Define the cardinality of the supporting index set and order .

[0046] Step S6: Determine the number of cosets for the multi-coset pilot based on the support index set of the folded sub-intervals, and optimize the set of coset starting positions to obtain... Figure 5 The multi-column pilot pattern shown.

[0047] Let the set of starting positions of the coset be... ,in , If the number of cosets is given, then the pilot subcarrier index set is:

[0048] The total number of pilot subcarriers is To ensure that the local linear model corresponding to each folded sub-interval is recoverable, the number of cosets... At least should meet .

[0049] For each folded sub-interval ,according to and Constructing a local sensing matrix Its elements can be represented as in, express The first in One supporting index, Indicates the first The starting position of each group.

[0050] The set of starting positions of the co-set The optimization is based on the block least squares error and the correlation of the columns of the local sensing matrix, which are expressed as follows:

[0051] In the specific solution, This serves as the initial candidate set. A backward pruning method is used to progressively delete the starting positions of cosets. In each iteration, while maintaining the full column rank of each local perception matrix, several starting positions that minimize the increase in column correlation after deletion are first selected. Then, the position that minimizes the block-level error cost after deletion is chosen from these. The smallest candidate starting position is deleted. Repeat pruning until the set size is reached. This yields the optimized set of coset starting positions. And based on this, a multi-set pilot pattern is generated.

[0052] Step S7: The base station trains the pre-encoder according to the broadband flat hybrid training method obtained in step S3. The multi-coset pilot pattern obtained in step S6 is used to send orthogonal downlink training pilots, and the user end obtains beam-frequency domain pilot observation results through matched filtering.

[0053] Specifically, the base station only Training pilots are transmitted on the designated subcarriers, and using... Simultaneously motivate One training beam. The base station has no fewer than... Each consecutive pilot OFDM symbol transmits mutually orthogonal pilot sequences, forming a pilot symbol matrix. ,satisfy User-side utilization Matched filtering is performed to separate the pilot observations corresponding to each training beam.

[0054] By stacking the matched filtering results on all pilot subcarriers column-wise, the beam-frequency domain pilot observation matrix is ​​obtained. in, Select the pilot subcarrier matrix. This is the beam domain noise matrix.

[0055] Step S8: The user terminal estimates the main time delay support intervals of each training beam based on the beam-frequency domain pilot observation results and the multi-coset pilot pattern, and obtains the estimated joint support set through union processing.

[0056] Specifically, the user terminal uses the optimized coset starting position set. Constructing a global folded sensing matrix , its first Line 1 Column elements are

[0057] For the The training beam is generated based on the folding transformation relationship of multi-coset sampling. Convert the corresponding row observations in the matrix into a folded observation matrix. , of which Line 1 Column elements are in, , , express The Each element.

[0058] During the coarse positioning stage supported by major time delays, Each column is considered a candidate sensing atom. For the first... Candidate sensing atoms Calculate its relationship with The normalized projected energy between them is used, and the atom index with the largest projected energy is used as the initial positioning result, i.e.

[0059] Will The coarse-location atomic index set is obtained by incorporating the indexes of the indexes and their adjacent indices. For any candidate atomic index The corresponding original delay index set is Therefore, we obtain the first... Candidate delay support interval corresponding to the training beam

[0060] User terminal Internal enumeration of continuous candidate intervals that satisfy the minimum length constraint And select the optimal estimation interval based on the Bayesian information criterion. Let , express Zhongyou The specified column submatrix Represents the regularized projection matrix. Indicates the first The pilot observation vector of the training beam, then The selection criterion function is

[0061] User-side selection The smallest candidate interval is used as the first Estimation of the main time delay support interval of the training beam After refining the intervals of each training beam, the estimated joint support set is obtained. .

[0062] Step S9: Based on the estimated joint support set A local sensing matrix is ​​constructed, and block least squares reconstruction is performed in each training beam to obtain the downlink equivalent channel estimation result.

[0063] Specifically, according to the folded sub-interval division method in step S5, for Re-perform boundary modulo, folded sub-interval division, and translation index filtering to obtain the estimated folded sub-intervals. and its corresponding estimated support index set Subsequently, this is combined with the set of starting positions of the coset. Construct each Corresponding local sensing matrix .

[0064] For the Beam training beam and the first Each estimated folded sub-interval is used to extract column indices from the folded observation matrix. observation block And establish a local linear observation model ,in, This is the local beam-delay domain channel coefficient matrix to be recovered.

[0065] Perform block least squares reconstruction to obtain the local beam-delay domain channel coefficient matrix. Then, according to the estimated supporting index set The order of the translation indices in the middle will... The coefficients of each row are sequentially backfilled into the original time delay grid. The position corresponding to its effective translation copy is then determined. The above process is repeated for all estimated folded sub-intervals and all training beams, followed by a frequency domain transformation to obtain the downlink equivalent channel estimation results. For the complete user-side reception and processing flow, please refer to [link / reference needed]. Figure 6 .

[0066] Therefore, this invention, through the above specific embodiments, utilizes the geometric reciprocity information of path number, path angle, and path delay obtained from uplink detection, combined with a dual broadband channel model to determine the downlink angle-delay support interval; designs a broadband flat all-digital training precoder to address the beam squint effect, and alternately decomposes it into a hybrid training precoder; divides folded sub-intervals based on beam-delay domain joint support, optimizes the coset start position, and generates a multi-coset pilot pattern; at the user end, through matched filtering, support detection, Bayesian information criterion support refinement, and block least squares reconstruction, the downlink equivalent channel estimation result is obtained. Compared with the prior art in the background art, the invention has the following specific advantages: 1) Wideband gain flatness of downlink training beams: This invention optimizes the wideband flatness to ensure that each training beam maintains a stable training gain on all subcarriers. Existing patents do not involve any training beam design, and their downlink channels have not undergone a downlink training process in essence.

[0067] 2) Active suppression of inter-beam interference and leakage: The present invention explicitly suppresses inter-beam interference and power leakage outside the target angle extension region in the design of the training precoder, thereby improving the SINR of the receiver beam domain pilot observation. Existing patents do not have this effect.

[0068] 3) Significant reduction in pilot overhead: By using a multi-coset pilot pattern, this invention samples only a portion of the folded sub-intervals in the time delay domain, which significantly reduces pilot resource overhead compared to the full subcarrier pilot scheme of existing patents.

[0069] 4) Improved reconstruction accuracy: This invention optimizes the starting position of the coset to make the local sensing matrix meet the full column rank condition and improve the condition number, thereby reducing noise amplification in the matrix inversion process; Existing patents, due to the use of full subcarrier pilots, do not have the problem of the condition number of the observation matrix, but the resource overhead is unacceptable.

[0070] 5) Engineering deployment friendliness: The multi-coset pilot pattern of the present invention has a clear parameterized structure (sampling period, folding interval, coset start position), which facilitates pilot position indication and pilot reuse in standardized systems; existing patents do not have corresponding pilot designs.

[0071] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity, characterized in that, The method includes the following steps: Step S1: Utilize the reciprocity of path geometry parameters in the uplink and downlink, combined with the channel model, to obtain the angle support interval and delay support interval of each propagation path on the discrete grid of the downlink channel. Step S2: Considering the beam squint effect of the dual broadband system, broadband flattening optimization is performed on the training beams pointing to each propagation path and interference between training beams is suppressed to obtain a broadband flattened all-digital training precoder. Step S3: Considering the unit constant modulus hardware constraint of the hybrid precoder, the broadband flat all-digital training precoder is subjected to alternating optimization decomposition to obtain the broadband flat hybrid training precoder. Step S4: Construct a beam-delay domain equivalent channel based on the broadband flat hybrid training precoder, and consider the leakage effect between training beams to obtain a joint delay support set shared by each training beam; Step S5: Determine the sampling period of the multiple co-sets, divide the joint delay support set into folded sub-intervals, and filter the translation indexes of each folded sub-interval that still fall into the original joint delay support set after translation, to obtain the folded sub-interval support index set. Step S6: Determine the number of cosets that satisfy the full column rank condition of the local sensing matrix based on the set of support indices of the folded sub-intervals, and optimize the set of starting positions of the cosets to obtain the multi-coset pilot pattern; Step S7: The base station sends orthogonal downlink training pilots according to the broadband flat hybrid training precoder and the multi-coset pilot pattern. The user terminal obtains the beam-frequency domain pilot observation results after matched filtering. Step S8: The user terminal estimates the main time delay support intervals of each training beam based on the beam-frequency domain pilot observation results and the multi-coupling pilot pattern, and obtains the estimated joint support set after union processing. Step S9: Construct a local sensing matrix based on the estimated joint support set, and perform block least squares channel reconstruction on each training beam to obtain the downlink equivalent channel estimation result.

2. The downlink training design method for a dual-bandwidth millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S1 specifically includes the following steps: In step S101, the base station estimates the path geometric parameters shared by the uplink and downlink through uplink probing. These path geometric parameters include the number of resolvable paths. , No. Path delay of each propagation path and path angle ,in ; Step S102: Substitute the path geometry parameters into the dual broadband downlink space-frequency domain channel model, and perform a two-dimensional inverse discrete Fourier transform on the obtained channel representation to obtain the downlink angle-delay domain channel representation. Step S103: Based on the energy concentration of the unilateral Dirichlet kernel in the downlink angle-delay domain channel characterization, determine the first... The main energy support intervals of the propagation path in the angular and time delay dimensions are combined by Cartesian product to obtain the rectangular block support intervals of the propagation path on the discrete angle-time delay grid of the downlink channel. When the main energy support intervals cross the corresponding discrete grid boundaries, they are represented by cyclic support intervals that are connected end to end.

3. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S2 specifically includes the following steps: Step S201: Based on the beam squinting characteristics of the array response changing with the subcarrier frequency in the dual broadband channel model, determine the frequency-dependent equivalent angle of each subcarrier in the full bandwidth for each propagation path, and form the target angle extension region corresponding to each propagation path. Step S202: Construct a corresponding array steering vector for each frequency-related equivalent angle, so that each propagation path corresponds to a set of array steering vectors covering the full bandwidth equivalent angle variation; Step S203: The minimum value of the inner product gain of each all-digital training beam and the array steering vector of the corresponding propagation path is taken as the broadband worst gain of the training beam. With the optimization objective of maximizing the sum of the broadband worst gains of all training beams, under the conditions of satisfying the average power leakage constraint outside the target angle extension region, the average correlation constraint between beams, and the beam unity power constraint, the all-digital training beams are jointly optimized to obtain the broadband flat all-digital training precoder.

4. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S3 specifically includes the following steps: Step S301: Using the broadband flat all-digital training precoder as the approximation target, construct a hybrid training precoder formed by cascading analog precoders and digital precoders. Step S302: A broadband flat hybrid training precoder is obtained through alternating optimization. When the analog precoder is fixed, the digital precoder is updated using the least squares method. When the digital precoder is fixed, the analog precoder is updated using the Riemannian manifold optimization method under the unit constant modulus constraint.

5. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S4 specifically includes the following steps: Step S401: Using the broadband flat hybrid training precoder as the training beam matrix, beam domain projection and time delay domain transformation are performed on the downlink space-frequency domain channel to obtain the time delay domain channel representation corresponding to each training beam. Step S402: Based on the main time delay support intervals of the propagation paths corresponding to each training beam and the inter-beam energy leakage caused by non-orthogonal wide beams, the time delay support intervals of each propagation path obtained in step S1 are subjected to union processing to obtain a joint time delay support set shared by each training beam, and this set is used as a shared time delay support reference for the multi-coordinate pilot pattern design in step S6.

6. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S5 specifically includes the following steps: Step S501: Determine the multi-coset sampling period R And based on the total number of subcarriers K Obtain the length of the folded delay grid. Q ,in K =RQ ; Step S502: Represent the joint delay support set as several non-overlapping continuous delay intervals, and define the boundaries of each continuous delay interval according to... Q Take the modulus and set 0 and 0. Q As the first and last boundaries of the folded delay grid are sorted together, the folded delay range between adjacent boundaries is divided into folded sub-intervals; Step S503, for each folded sub-interval, with Q The translation interval is periodically translated on the original time delay grid. The translation indexes that still fall completely into the joint time delay support set after translation are selected to obtain the corresponding folded sub-interval support index set.

7. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S6 specifically includes the following steps: Step S601: Calculate the cardinality of the support index set of each folded sub-interval, and use the maximum value among them as the lower bound of the number of cosets required for full column rank reconstruction of the local sensing matrix, and determine the number of cosets of the multi-coset pilot. Step S602: Using a backward pruning method, while maintaining the full column rank of each local sensing matrix, the candidate coset starting positions are screened based on minimizing the block least squares reconstruction error and the column correlation of the local sensing matrix as optimization criteria, to obtain the coset starting position set. Step S603: Using each position in the set of co-set starting positions as the starting point of the uniform pilot subsequence, select subsequent pilot subcarriers according to the multi-co-set sampling period to generate multiple uniform pilot subsequences with the same period but different starting points, and merge the uniform pilot subsequences to obtain the multi-co-set pilot pattern.

8. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S7 specifically includes the following steps: Step S701: On the pilot subcarrier specified by the multi-coordinate pilot pattern, the base station uses the broadband flat hybrid training precoder to simultaneously excite L training beams and transmit pilot vectors at at least L consecutive pilot OFDM symbol times. The pilot vectors are arranged according to the symbol times to form a pilot symbol matrix, and the pilot sequences corresponding to different training beams are mutually orthogonal. In step S702, the user terminal performs matched filtering on the received pilot signal based on the pilot symbol matrix, separates the channel observation values ​​of each training beam on the pilot subcarrier, and obtains the beam-frequency domain pilot observation results.

9. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S8 specifically includes the following steps: Step S801: The user terminal constructs folded observations based on the beam-frequency domain pilot observation results and the multi-coset pilot pattern, using the time delay domain mode folding relationship corresponding to the multi-coset sampling, and calculates the energy projection metric between the candidate sensing atom and the folded observations to determine the coarse localization atom index of the main time delay energy range of each training beam. Step S802: Map the coarse positioning atom index and its adjacent index to the original time delay grid, construct candidate time delay support intervals, and use the Bayesian information criterion to adjust the interval boundaries to obtain the estimated main time delay support intervals of each training beam. Step S803: Perform union processing on the estimated main time delay support intervals of each training beam to obtain the estimated joint support set.

10. The downlink training design method for a dual-wideband millimeter-wave communication system based on channel reciprocity according to claim 1, characterized in that, Step S9 specifically includes the following steps: Step S901: Based on the estimated joint support set, and according to the folded sub-interval division and translation index filtering method, re-determine the estimated folded sub-interval and its corresponding estimated folded sub-interval support index set. Step S902: Combine the set of coset starting positions and the set of estimated folded sub-interval support indices to construct the local perception matrix corresponding to each estimated folded sub-interval and establish the corresponding local linear observation model. Step S903: Perform block least squares channel reconstruction within each estimated folded sub-interval in each training beam, fill the corresponding delay positions with the reconstructed beam-delay domain channel coefficients, and obtain the downlink equivalent channel estimation result through frequency domain transformation.

Images