A distributed precoding method for large-scale MIMO networks

Abstract

The present invention discloses a distributed precoding method for a large-scale MIMO network. First, each base station obtains the downlink channel matrix of all network users and independently generates the single-cell initial precoding matrix of the users in this cell; then each base station independently calculates the inter-cell virtual interference covariance matrix; each base station exchanges the inter-cell virtual interference covariance matrix through a backhaul link; finally, each base station iterates in parallel to solve the precoding matrix of the users served by each base station, that is, to obtain the distributed precoding matrix of the whole network. When the inter-cell virtual interference covariance is known, the user's virtual interference plus noise sum covariance model and the virtual weighted sum rate model only depend on the user precoding matrix of this cell. The method of the present invention can be calculated and optimized locally, and only the inter-cell virtual interference covariance needs to be exchanged between base stations in the initialization stage, so as to ensure the system and rate performance under the premise of minimizing the inter-cell backhaul link overhead and system synchronization requirements.

Description

Technical Field

[0001] The invention relates to wireless communication downlink precoding, and in particular to a large-scale MIMO distributed precoding method for a network without data sharing between cells. Background Art

[0002] Massive multiple-input-multiple-output (MIMO) is a technology that deploys large-scale antenna arrays on the base station (BS) side to provide efficient communication services to a large number of users. Due to its ability to greatly improve spectrum and energy efficiency, it has been deeply studied and has become a key technology for wireless communication standards. In a large-scale MIMO system with no data sharing between cells, each base station can use distributed precoding to reduce the impact of inter-cell interference.

[0003] However, the use of large-scale antenna arrays has dramatically increased the computational complexity and signaling overhead of precoding in large-scale MIMO systems without data sharing between cells, and has also put forward higher requirements on the network's latency performance. To alleviate these burdens, it is necessary to design a large-scale MIMO precoding design method that only relies on a small amount of information interaction between cells, so that each base station can obtain the precoding matrix of the users it serves locally in parallel and distributed manner. Related literature has studied the distributed network precoding design method that performs serial or parallel network precoding with only a small amount of information interaction between cells under the criterion of minimizing total power, ensuring the signal-to-interference-plus-noise-ratio (SINR) and minimizing total power. So far, there has been no research on the design of large-scale MIMO distributed precoding for parallel execution under the criterion of maximizing system weights and rates. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a large-scale MIMO distributed precoding method for networks without data sharing between cells, so as to overcome the shortcomings of the prior art, approach the optimal weighted and rate performance, and reduce the computational signaling overhead and synchronization requirements.

[0005] Technical solution: A distributed precoding method for a large-scale MIMO network, each cell is equipped with a base station, and the distributed precoding method includes:

[0006] Step 1: Each base station independently generates a single-cell initial precoding matrix for users in the cell, and obtains a downlink channel matrix for all network users;

[0007] Step 2: Based on the single-cell initial precoding matrix and the downlink channel matrix, each base station independently calculates the inter-cell virtual interference covariance matrix for users in other cells outside the cell;

[0008] Step 3: Each base station exchanges the inter-cell virtual interference covariance matrix through a backhaul link;

[0009] Step 4: Each base station iterates in parallel to solve the precoding matrix of the users served by each base station according to the inter-cell virtual interference covariance matrix and the downlink channel matrix of the users in the cell, that is, to obtain the distributed precoding matrix of the whole network.

[0010] Furthermore, each cell in the massive MIMO network is equipped with a base station located at a geometric center, each cell has K users randomly distributed in the cell, and each base station and user is equipped with M and N antennas respectively; in step 4, when iteratively solving the precoding matrix of the user served by each base station, the optimization goal is to maximize the virtual weighted sum rate of each base station;

[0011] For the lth base station, the mathematical expression of maximizing the virtual weighted sum rate is:

[0012]

[0013]

[0014] In the formula, is the virtual weighted sum rate of the lth base station, P k,l is the precoding matrix of the kth user in the lth cell, k = 1, 2, 3...K, the superscript H represents the matrix transpose, P l is the maximum transmit power of the base station in the lth cell;

[0015] in:

[0016]

[0017] In the formula, is the weighted sum rate of the network, P u,r is the precoding matrix of the u-th user in the r-th cell, is the virtual precoding matrix of the u-th user in the r-th cell, is the virtual sum rate of the cell calculated by the lth base station, is the virtual sum rate of the rth cell calculated by the lth base station, w k,l is the weight corresponding to the kth user in the lth cell, I N is the N-order unit matrix, is the covariance matrix of Gaussian noise in the downlink received signal of the kth user in the lth cell, H k,l,lis the channel matrix from the lth base station to the kth user in the lth cell, w u,r is the weight corresponding to the u-th user in the r-th cell, is the covariance matrix of the virtual interference plus noise of the u-th user in the r-th cell calculated by the l-th base station, S u,r,u,r is the virtual effective signal covariance matrix of the u-th user in the r-th cell;

[0018] in:

[0019]

[0020]

[0021] Where σ is the noise power, P u,l is the precoding matrix of the u-th user in the l-th cell, S k,l,u,r is the interference covariance matrix caused by the u-th user in the r-th cell to the k-th user in the l-th cell, H u,r,l is the channel matrix from the lth base station to the uth user in the rth cell, S u,r,k,r is the interference covariance matrix caused by the kth user in the rth cell to the uth user in the rth cell, S u,r,k,j is the interference covariance matrix caused by the kth user in the jth cell to the uth user in the rth cell.

[0022] Furthermore, the MM algorithm is used to transform the problem of maximizing the virtual weighted sum rate into a quadratic optimization problem for iterative solution, and its iterative expression is:

[0023]

[0024] Where:

[0025]

[0026]

[0027]

[0028]

[0029]

[0030] in:

[0031]

[0032]

[0033]

[0034]

[0035] In the formula, the superscripts (d) and (d+1) of the parameters represent the corresponding values ​​of the parameters in the dth and d+1th iterations. It is the intermediate variable of the iterative calculation process; is the dual variable, I M is an M-order unit matrix, is the covariance matrix of the virtual interference plus noise of the kth user in the lth cell calculated by the lth base station, H k,l,r is the channel matrix from the rth base station to the kth user in the lth cell, for and the expected signal of the kth user in the lth cell, is the covariance matrix of the virtual interference plus noise of the u-th user in the r-th cell calculated by the l-th base station, H u,r,l is the channel matrix from the lth base station to the uth user in the rth cell, for The sum of the expected signal of the u-th user in the r-th cell.

[0036] Beneficial effects: Compared with the prior art, the present invention has the following advantages: This design method only needs to perform inter-cell information interaction in the initialization phase after calculating the inter-cell virtual covariance matrix of other cell users through the single-cell initial precoding matrix, and no longer needs to perform inter-base station signaling interaction during the precoding iteration process, which significantly reduces the inter-cell return link overhead and system synchronization requirements, and maintains near-optimal sum rate performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flowchart of the design method for massive MIMO distributed precoding in networks without data sharing between cells;

[0038] Figure 2 A flow chart of a large-scale MIMO distributed precoding system for a network without data sharing between cells;

[0039] Figure 3 A flow chart showing that each base station performs distributed precoding in parallel after initialization;

[0040] Figure 4 This is a diagram of the network MIMO downlink precoding transmission scenario without data sharing between cells. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0042] Consider a network massive MIMO communication system, which consists of L cells, each of which is equipped with a base station at the geometric center, and K users are randomly distributed in the cell. Each base station and user is equipped with M and N antennas respectively. Each base station serves its own set of users, and user data is not shared between base stations in the network.

[0043] The downlink received signal y of the kth user in the lth cell k,l for:

[0044]

[0045] In the formula, H k,l,l is the channel matrix from the lth base station to the kth user in the lth cell; P k,l is the precoding matrix of the kth user in the lth cell, subject to the cell-by-cell power constraint P l is the maximum transmission power of the base station in the lth cell, and the superscript H represents the matrix transposition; d k,l is the data flow of the kth user in the lth cell, satisfying d u,r is the data flow of the u-th user in the r-th cell; P u,l is the precoding matrix of the u-th user in the l-th cell; d u,l is the data flow of the u-th user in the l-th cell; H k,l,r is the channel matrix from the rth base station to the kth user in the lth cell; P u,r is the precoding matrix of the u-th user in the r-th cell; n k,l is the noise. The first term on the right side is k,l,l P k,l d k,l The corresponding signal is the expected signal, the second term and the third Represent the interference within the cell and between cells respectively. For a certain user, the covariance matrix of the sum of its virtual interference plus noise is equal to the sum of the covariance matrix of the noise, the covariance matrix of the interference within the cell, and the covariance matrix of the virtual interference of other cells to the user.

[0046] Assuming that the user side knows perfect CSI, the sum of received interference and noise, that is, the sum of interference within the cell and between cells Considered as the covariance matrix R k,l Gaussian noise, the weighted sum rate of the network is:

[0047]

[0048] The covariance matrix is ​​R k,l for:

[0049]

[0050] The problem of maximizing the network system and rate under the power constraint of each base station is expressed as:

[0051]

[0052] In the formula, w k,l is the weight corresponding to each user; σ is the noise power; I N is an N-order unit matrix.

[0053] Since the precoding matrices of each cell are coupled together, the solution of equation (4) requires joint optimization of the entire network. To decouple it between base stations, for the lth base station, the precoding matrices of other base stations u≠l are fixed to the initial precoding matrix of a single cell That is, the virtual precoding matrix of the u-th user in the r-th cell defines the interference covariance matrix caused by the u-th user in the r-th cell to the k-th user in the l-th cell Virtual rate of the kth user in the lth cell for:

[0054]

[0055] In the formula, The covariance matrix of the virtual interference plus noise of the kth user in the lth cell calculated for the lth base station;

[0056]

[0057] Similarly, the virtual rate of the uth user (u≠l) in the rth cell calculated by the lth base station is for:

[0058]

[0059] In the formula, The virtual covariance matrix of the u-th user in the r-th cell calculated for the l-th base station is:

[0060]

[0061] In the formula, H u,r,l is the channel matrix from the lth base station to the uth user in the rth cell, is the interference covariance matrix caused by the kth user in the rth cell to the uth user in the rth cell, is the interference covariance matrix caused by the kth user in the jth cell to the uth user in the rth cell; H u,r,r is the channel matrix from the r-th base station to the u-th user in the r-th cell, is the virtual precoding matrix of the kth user in the rth cell, H u,r,j is the channel matrix from the jth base station to the uth user in the rth cell, is the virtual precoding matrix of the kth user in the jth cell.

[0062] Therefore, the virtual weighted sum rate calculated by the lth base station is for:

[0063]

[0064] In the formula, is the virtual sum rate of the cell calculated by the lth base station, The virtual sum rate of the rth cell calculated for the lth base station.

[0065] Taking the virtual weighted sum rate as the approximation of the objective function of the original problem, the precoding matrix design problem for the lth base station can be expressed as:

[0066]

[0067] When the virtual covariance matrices of other cells are known, the above virtual weighted rate maximization problem can be decoupled into L independent sub-problems and can be solved separately and in parallel on each BS.

[0068] The virtual weighted sum rate in the formula is about P 1,l ,...,P K,l is a non-concave function, so it is difficult to find a closed-form solution to the problem. To solve this problem, consider the MM algorithm, whose solution can be proved to converge to the local maximum of the original function.

[0069] First, for u≠l, define the following virtual covariance matrix:

[0070]

[0071]

[0072]

[0073]

[0074] In the formula, the superscripts (d) and (d+1) of the parameters represent the corresponding values ​​of the parameters in the d-th and d+1-th iterations; for and the expected signal of the kth user in the lth cell, for and the expected signal of the uth user in the rth cell, H u,r,lis the channel matrix from the lth base station to the uth user in the rth cell

[0075] Then, the following function g l For function exist A minorizing function at:

[0076]

[0077] In the formula, and is a constant;

[0078]

[0079]

[0080] in;

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] In the formula, It is an intermediate variable in the iterative calculation process.

[0088] Using the minorizing function g l , transform the original problem into a sequence of concave quadratic optimization problems to be solved iteratively, and find the stable point of the original problem:

[0089]

[0090] Define its Lagrangian function:

[0091]

[0092] In the formula, μ l >0 is the dual variable, then the KKT condition is:

[0093]

[0094]

[0095]

[0096] μ l >0 (29)

[0097] For a given Minimal for:

[0098]

[0099] like The optimal solution is:

[0100]

[0101] Otherwise, the optimal dual variable Can be searched by binary search.

[0102] In summary, the distributed precoding method can be summarized as follows:

[0103] Step 1: Initialize the single-cell initial precoding matrix Virtual interference covariance matrix between interacting cells Set d = 0;

[0104] Step 2: Calculation and

[0105] Step 3: Calculation and

[0106] Step 4: Update

[0107] Step 5: Repeat steps 2-4 until convergence.

[0108] In the method of the present invention, each cell reduces the impact of interference between cells through distributed precoding. The logical process of distributed precoding includes: establishing a virtual interference covariance matrix between cells and a covariance matrix model of the sum of virtual interference plus noise of users; establishing a virtual weighted sum rate model; establishing a minorizing function model of the virtual weighted sum rate; solving the virtual weighted sum rate maximization problem to obtain a distributed precoding iteration. Among them, the calculation of the parallel precoding iteration stage of each cell includes: S1: for all users in the network, calculate the covariance matrix of the sum of virtual noise plus interference; S2: update the user precoding matrix of this cell; S3: repeat S1-S2 until convergence or reaching the specified number of iterations.

[0109] In the method of the present invention, when the virtual interference covariance between cells is known, the virtual weighted sum rate is only related to the user precoding of the cell; the virtual weighted sum rate maximization precoding design problem is converted into a quadratic optimization problem through the MM algorithm for iterative solution.

[0110] The inter-cell virtual interference covariance matrix, the user's virtual interference plus noise covariance matrix difference model and the virtual weighted sum rate model only rely on the user precoding matrix of the cell. Each base station can perform local calculations and maximize the virtual weighted sum rate to obtain the precoding matrix of the served user. This method only requires inter-cell information interaction during the initialization phase, after calculating the inter-cell virtual covariance matrix of other cell users through the single-cell initial precoding matrix. In the precoding iteration process, signaling interaction between base stations is no longer required, which significantly reduces the inter-cell backhaul link overhead and system synchronization requirements, and maintains near-optimal sum rate performance.

[0111] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A distributed precoding method for a large-scale MIMO network, characterized in that: Each cell is equipped with a base station, and the distributed precoding method includes: Step 1: Each base station independently generates a single-cell initial precoding matrix for users in the cell, and obtains a downlink channel matrix for all network users; Step 2: Based on the single-cell initial precoding matrix and the downlink channel matrix, each base station independently calculates the inter-cell virtual interference covariance matrix for users in other cells outside the cell; Step 3: Each base station exchanges the inter-cell virtual interference covariance matrix through a backhaul link; Step 4: Each base station iteratively solves the precoding matrix of the user served by each base station in parallel according to the inter-cell virtual interference covariance matrix and the downlink channel matrix of the user in the cell, that is, obtains the distributed precoding matrix of the whole network; Each cell in the massive MIMO network is equipped with a base station located at the geometric center, each cell has K users randomly distributed in the cell, and each base station and user is equipped with M and N antennas respectively; in step 4, when iteratively solving the precoding matrix of the user served by each base station, the optimization goal is to maximize the virtual weighted sum rate of each base station; For the lth base station, the mathematical expression of maximizing the virtual weighted sum rate is: In the formula, is the virtual weighted sum rate of the lth base station, P k,l is the precoding matrix of the kth user in the lth cell, k = 1, 2, 3...K, the superscript H represents the matrix transpose, P l is the maximum transmit power of the base station in the lth cell; in: In the formula, is the weighted sum rate of the network, P u,r is the precoding matrix of the u-th user in the r-th cell, is the virtual precoding matrix of the u-th user in the r-th cell, is the virtual sum rate of the cell calculated by the lth base station, is the virtual sum rate of the rth cell calculated by the lth base station, w k,l is the weight corresponding to the kth user in the lth cell, I N is the N-order unit matrix, is the covariance matrix of Gaussian noise in the downlink received signal of the kth user in the lth cell, H k,l,l is the channel matrix from the lth base station to the kth user in the lth cell, w u,r is the weight corresponding to the u-th user in the r-th cell, is the covariance matrix of the virtual interference plus noise of the u-th user in the r-th cell calculated by the l-th base station, S u,r,u,r is the virtual effective signal covariance matrix of the u-th user in the r-th cell; in: Where σ is the noise power, P u,l is the precoding matrix of the u-th user in the l-th cell, S k,l,u,r is the interference covariance matrix caused by the u-th user in the r-th cell to the k-th user in the l-th cell, H u,r,l is the channel matrix from the lth base station to the uth user in the rth cell, S u,r,k,r is the interference covariance matrix caused by the kth user in the rth cell to the uth user in the rth cell, S u,r,k,j is the interference covariance matrix caused by the kth user in the jth cell to the uth user in the rth cell.

2. The distributed precoding method for a large-scale MIMO network according to claim 1, characterized in that: The problem of maximizing the virtual weighted sum rate is transformed into a quadratic optimization problem by the MM algorithm for iterative solution. The iterative expression is: Where: in: In the formula, the superscripts (d) and (d+1) of the parameters represent the corresponding values ​​of the parameters in the dth and d+1th iterations. It is the intermediate variable of the iterative calculation process; is the dual variable, I M is an M-order unit matrix, is the covariance matrix of the virtual interference plus noise of the kth user in the lth cell calculated by the lth base station, H k,l,r is the channel matrix from the rth base station to the kth user in the lth cell, for and the expected signal of the kth user in the lth cell, is the covariance matrix of the virtual interference plus noise of the u-th user in the r-th cell calculated by the l-th base station, H u,r,l is the channel matrix from the lth base station to the uth user in the rth cell, for The sum of the expected signal of the u-th user in the r-th cell.

Images