Hybrid distributed baseband processing architecture and method suitable for XL-MIMO system
Through the hybrid distributed baseband processing architecture and D-PAPC-WMMSE algorithm, the problems of high complexity and large information interaction of traditional centralized signal processing algorithms are solved, and flexible distributed signal detection in multi-user XL-MIMO systems are realized and precoding reduces the amount of information interaction and improves processing power.
Patent Information
- Application Number
- CN202411048380.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2024-12-03
AI Technical Summary
Traditional centralized signal processing algorithms have high computational complexity and large information interactions in multi-user XL-MIMO systems, making it difficult to adapt to the needs of ultra-large-scale antenna arrays. The existing distributed algorithms can only be applied to specific architectures and cannot be promoted.
A hybrid distributed baseband processing architecture is proposed, including central unit and multiple distributed units. Through bidirectional connection and chain connection, the channel matrix and precoding matrix are decomposed, and the D-PAPC-WMMSE algorithm is adopted for parallel preprocessing and iterative update. , realize distributed signal detection and precoding.
It significantly reduces the amount of information interaction between distributed units, realizes the advantages of star and chain architecture, can complete distributed baseband processing tasks that cannot be completed by a single star or chain architecture, and improves the flexibility and processing capabilities of the system. .
Smart Images

Figure CN119070865A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to mobile communication technology, and in particular to a hybrid distributed baseband processing architecture and method suitable for an XL-MIMO system. Background Art
[0002] In multi-user XL-MIMO systems, signal processing technologies such as downlink precoding and uplink signal detection are key enabling technologies for multi-antenna systems to have high spectral efficiency and energy efficiency. However, traditional centralized signal processing algorithms face two major bottlenecks: high computational complexity and high information interaction. With the application of ultra-large-scale antenna arrays in the future, the dimensions of the channel matrix and precoding matrix will increase significantly, and the difficulty of implementing traditional centralized signal processing algorithms will increase dramatically.
[0003] After searching the prior art, it was found that Xiaotong Zhao et al. proposed a distributed DR-WMMSE precoding algorithm for the downlink multi-user XL-MIMO system under the star-shaped distributed baseband processing architecture in "Communication-Efficient Decentralized Linear Precoding for Massive MU-MIMO Systems" IEEE Transactions on Signal Processing, 2023, 7(4): 491-507. However, the algorithm is only applicable to the downlink distributed star architecture and cannot be extended to other architectures. Kaipeng Li et al. proposed an optimal BCD-MMSE signal detection algorithm for the uplink multi-user XL-MIMO system under the chain distributed baseband processing architecture in "Decentralized linear MMSE equalizer under colored noise for massive MIMO systems." in Proc. IEEE Global Commun. Conf., 2021, 01-06. However, the algorithm is only applicable to the uplink distributed chain architecture and cannot be extended to other architectures. Therefore, the present invention proposes a general distributed baseband processing architecture suitable for uplink signal detection and downlink precoding. Summary of the invention
[0004] The object of the present invention is to provide a hybrid distributed baseband processing architecture and method suitable for XL-MIMO system in view of the deficiencies of the prior art.
[0005] The objective of the present invention is achieved through the following technical solutions:
[0006] A hybrid distributed baseband processing architecture suitable for XL-MIMO system includes a central unit (CU) and multiple (C) distributed units (DUs). The base station antenna is divided into C antenna clusters, each antenna cluster is connected to a DU. In addition, the CU is bidirectionally connected to all DUs, and the DUs are chained.
[0007] In the hybrid distributed baseband processing structure, the channel matrix of user k can be decomposed into Each block matrix is stored in the local DU. The precoding matrix of user k can be structured as Therefore, the signal y received by user k in equation (1) can be k Rewrite as
[0008]
[0009] in is the local channel matrix of user k, is the local precoding matrix of user k.
[0010] A hybrid distributed baseband processing method applicable to an XL-MIMO system, based on the above architecture, comprises the following steps:
[0011] Step 1: Each distributed unit performs parallel preprocessing, which specifically includes the following sub-steps:
[0012] Step 1.1: Each distributed unit first calculates the local channel correlation matrix H c (H c ) H , where H c represents the channel matrix from the cth distributed unit to all users;
[0013] Step 1.2, each distributed unit sends the matrix obtained in step 1.1 to the central unit;
[0014] Step 2: The central unit sums the information obtained by each distributed unit to obtain a channel correlation matrix;
[0015] Step 3, the central unit uses the channel correlation matrix obtained in step 2 to perform R-WMMSE algorithm iteration and outputs the matrix X;
[0016] Step 4: The central unit broadcasts the matrix X output in step 3 to all distributed units;
[0017] Step 5: After receiving the information sent by the central node, each distributed unit calculates the local precoding matrix (H c ) H X.
[0018] Step 6: The central unit updates the variables, including the following sub-steps:
[0019] Step 6.1, calculate the minimum mean square error receiving matrix for each user:
[0020] Step 6.2, calculate the weight matrix for each user:
[0021] Step 6.3: The central unit sets the W of all users k ,U k Broadcast to all distributed units and send Q=HP to distributed unit 1;
[0022] Step 7, each distributed unit serially updates the local precoding matrix in turn, and then updates the matrix Q and sends it to the next distributed unit;
[0023] Step 8: Repeat steps 6-7 until the system capacity converges.
[0024] The beneficial effect of the present invention is that by using the hybrid distributed architecture provided in the present invention, the respective advantages of the star and chain architectures can be effectively utilized, the amount of information interaction between distributed units can be significantly reduced, and distributed baseband processing tasks that cannot be completed by a single star or chain architecture can be realized. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 is a schematic diagram of a hybrid distributed baseband architecture proposed by the present invention;
[0026] Figure 2 is a flow chart of the method of the present invention;
[0027] Figure 3 is a schematic diagram of information interaction in a scenario of coordinated transmission of four distributed units in one embodiment of the present invention;
[0028] Figure 4 This is a performance comparison of a centralized algorithm and a D-PAPC-WMMSE algorithm applicable to a hybrid distributed baseband architecture in an embodiment of the present invention in a large-scale fading Rayleigh channel cooperative transmission scenario with four distributed units. DETAILED DESCRIPTION
[0029] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, and the purpose and effects of the present invention will become more apparent.
[0030] In a multi-user XL-MIMO system, a base station with M antennas sends signals to K users, and each user has N antennas. k , the total number of receiving antennas is The channel matrix between the antenna of user k and the base station is The antenna of user k receives the signal Satisfy the following formula:
[0031]
[0032] in, represents the data flow signal of user k, assuming is the total number of user data flows. is the precoding matrix of user k, represents the Gaussian white noise on user k antenna, represents the Gaussian white noise power.
[0033] From the perspective of all users, the received signal of the user is expressed as:
[0034] y=HPs+n (2)
[0035] in, is the joint channel matrix between all downlink user antennas and base station antennas, is the joint precoding matrix for all downlink users, Indicates the data flow information of all downlink users. is the constellation point set corresponding to the data modulation and demodulation method such as 16-QAM, Represents the Gaussian white noise at the receiving end of all downlink users.
[0036] Among them, "T" represents the transposition of the matrix, and "H" represents the conjugate transposition of the matrix.
[0037] Technical solution of the present invention:
[0038] The first step is to design a hybrid distributed baseband processing architecture, such as Figure 1 As shown, this architecture integrates the characteristics of star and chain topologies.
[0039] A hybrid distributed baseband processing architecture suitable for XL-MIMO system includes a central unit (CU) and multiple (C) distributed units (DUs). The base station antenna is divided into C antenna clusters, each antenna cluster is connected to a DU. In addition, the CU is bidirectionally connected to all DUs, and the DUs are chained.
[0040] The hybrid distributed architecture has the following advantages: it can adapt to a variety of distributed communication signal processing algorithms and select the algorithm execution mode according to the characteristics of the algorithm. This flexible architecture facilitates the execution of distributed algorithms suitable for star structures (such as feedforward algorithms) and chain structures (such as BCD algorithms). In addition, unlike star and chain architectures, which fail when the CU or DU fails, this hybrid architecture can guarantee its baseband processing capabilities even when the DU fails or is shut down.
[0041] In the hybrid distributed baseband processing structure, the channel matrix of user k can be decomposed into Each block matrix is stored in the local DU. The precoding matrix of user k can be structured as Therefore, the signal y received by user k in equation (1) can be k Rewrite as
[0042]
[0043] in is the local channel matrix of user k, is the local precoding matrix of user k. The hybrid distributed baseband processing architecture proposed in the present invention has the advantages of both chain topology and star topology.
[0044] Step 2: Algorithm Design
[0045] like Figure 2 As shown, a hybrid distributed baseband processing method applicable to an XL-MIMO system comprises the following steps:
[0046] Initialization phase:
[0047] Step 1: Each distributed unit performs parallel preprocessing, which specifically includes the following sub-steps:
[0048] Step 1.1: Each distributed unit first calculates the local channel correlation matrix H c (H c ) H , where H c represents the channel matrix from the cth distributed unit to all users;
[0049] Step 1.2, each distributed unit sends the matrix obtained in step 1.1 to the central unit;
[0050] Step 2: The central unit sums the information obtained by each distributed unit to obtain a channel correlation matrix;
[0051] Step 3, the central unit uses the channel correlation matrix obtained in step 2 to perform R-WMMSE algorithm iteration and outputs the matrix X;
[0052] Step 4: The central unit broadcasts the matrix X output in step 3 to all distributed units;
[0053] Step 5: After receiving the information sent by the central node, each distributed unit calculates the local precoding matrix (H c ) H X.
[0054] D-PAPC-WMMSE algorithm formal iteration stage:
[0055] Step 6: The central unit updates the variables, including the following sub-steps:
[0056] Step 6.1, calculate the minimum mean square error receiving matrix for each user:
[0057] Step 6.2, calculate the weight matrix for each user:
[0058] Step 6.3: The central unit sets the W of all users k ,U k Broadcast to all distributed units and send Q=HP to distributed unit 1;
[0059] Step 7, each distributed unit serially updates the local precoding matrix in turn, and then updates the matrix Q and sends it to the next distributed unit;
[0060] Step 8: Repeat steps 6-7 until the system capacity converges.
[0061] The following takes the distributed precoding solution under the single antenna power constraint as an example to illustrate how to implement the algorithm case in the hybrid distributed baseband processing architecture.
[0062] remember That is The kth sub-block of . The R-WMMSE algorithm (existing technology) is a low-complexity and efficient algorithm for total power constraints. Its iterative steps mainly include the following three steps. First, update the receiving weight U of user k according to formula (4): k . Secondly, update the weight matrix W of user k k Finally, update the low-dimensional variable matrix X of user k k Note that the above steps update all users k separately.
[0063]
[0064]
[0065]
[0066] where αk is the rate weight of the kth user, P max is the upper power limit of the base station. Repeat the above three equations (4), (5), (6) for iteration until the difference between two consecutive objective function values is less than a predetermined threshold. Finally, let X = [X1, X2, …, X K ], the precoding matrix is P = H H X is given.
[0067] The proposed D-PAPC-WMMSE algorithm can effectively solve the sum rate maximization problem under the single antenna power constraint in a distributed manner. It includes three main iterative steps. First, the receiving weight matrix U of user k is updated according to formula (7): k .
[0068]
[0069] Secondly, update the weight matrix W of user k according to formula (8): k .
[0070]
[0071] Finally, the local precoding vector p of the mth antenna is updated according to equation (9): m .
[0072]
[0073] in P m is the rated power constraint of the mth antenna. In the D-PAPC-WMMSE algorithm, each distributed unit first performs local preprocessing, calculates the local channel correlation matrix and sends it to the central unit. Then the central unit performs R-WMMSE iteration to obtain a heuristic solution and broadcasts the output matrix to each distributed unit. Each distributed unit calculates the local precoding matrix. At this point, the preprocessing stage is completed. Then a single antenna formal iteration is performed, and the central unit updates W k ,U k , broadcast to all distributed units, and send Q = HP to distributed unit 1. Note that in formula (9) It is denoted as matrix B, which can be calculated. Note that all variables required for updating in equation (9) have been obtained, and each distributed unit serially updates the local precoding matrix p m , then update the matrix And send it to the next distributed unit. Repeat the above steps until the system capacity converges.
[0074] It is noted that the proposed D-PAPC-WMMSE is a distributed algorithm that can effectively solve the sum rate maximization problem under the single antenna power constraint, while the existing literature has not yet studied the distributed single antenna power constraint designed for distributed baseband architecture.
[0075] Example
[0076] like Figure 3 As shown in the figure, four distributed units perform coordinated transmission. Distributed units DU1, DU2, DU3, and DU4 are connected to the central unit in a star topology, and DUs are connected to each other in a chain topology, and all can exchange information. Figure 3 As shown in (a), each distributed unit sends the local channel correlation matrix to the central unit, and the central unit sends the matrix X obtained by R-WMMSE iteration to each distributed unit. Figure 3 As shown in (b), the central unit will W k ,U k Broadcast to all distributed units, and send Q=HP to DU1. Each distributed unit serially updates the local precoding matrix in turn, and then updates the matrix Q and sends it to the next distributed unit.
[0077] Figure 4 The performance comparison of the centralized algorithm C-PAPC-WMMSE and the D-PAPC-WMMSE algorithm applicable to the hybrid distributed baseband architecture in the present invention is shown in the scenario of coordinated transmission of four distributed units in a large-scale fading Rayleigh channel;
[0078] Figure 4 (a) The number of users is 8 and the signal-to-noise ratio is 0dB.
[0079] Figure 4 (b) The number of users is 12 and the signal-to-noise ratio is 10dB.
[0080] like Figure 4 As shown, the distributed D-PAPC-WMMSE algorithm has better rate performance than the centralized algorithm while reducing the amount of information interaction and significantly reducing the number of iterations.
[0081] The present invention is not limited to the above specific implementations. A person skilled in the art can implement the present invention by using other various specific implementations based on the content disclosed in the present invention. Therefore, any design that adopts the design structure and ideas of the present invention and makes some simple changes or modifications falls within the protection scope of the present invention.
Claims
1. A hybrid distributed baseband processing architecture for XL-MIMO systems, It is characterized in that It includes a central unit CU and multiple distributed units DU. The base station antenna is divided into C antenna clusters, each antenna cluster is connected to a DU; the CU is bidirectionally connected to all DUs, and the DUs are chained. In the hybrid distributed baseband processing structure, the channel matrix of user k can be decomposed into Each block matrix is stored in the local DU; the precoding matrix of user k is structured as Therefore, the signal y received by user k is k Rewrite as in is the local channel matrix of user k, is the local precoding matrix of user k.
2. A hybrid distributed baseband processing method for XL-MIMO system, based on the architecture of claim 1, It is characterized in that The following steps are involved: Initialization phase: Step 1: Each distributed unit performs parallel preprocessing, which specifically includes the following sub-steps: Step 1.1: Each distributed unit first calculates the local channel correlation matrix H c (H c ) H , where H c represents the channel matrix from the cth distributed unit to all users; Step 1.2, each distributed unit sends the matrix obtained in step 1.1 to the central unit; Step 2: The central unit sums the information obtained by each distributed unit to obtain a channel correlation matrix; Step 3, the central unit uses the channel correlation matrix obtained in step 2 to perform R-WMMSE algorithm iteration and outputs the matrix X; Step 4: The central unit broadcasts the matrix X output in step 3 to all distributed units; Step 5: After receiving the information sent by the central node, each distributed unit calculates the local precoding matrix (H c ) H X; D-PAPC-WMMSE algorithm formal iteration stage: Step 6: The central unit updates the variables, including the following sub-steps: Step 6.1, calculate the minimum mean square error receiving matrix for each user: Step 6.2, calculate the weight matrix for each user: Step 6.3: The central unit sets the W of all users k ,U k Broadcast to all distributed units and send Q=HP to distributed unit 1; Step 7, each distributed unit serially updates the local precoding matrix in turn, and then updates the matrix Q and sends it to the next distributed unit; Step 8: Repeat steps 6-7 until the system capacity converges.
3. A hybrid distributed baseband processing method suitable for XL-MIMO system as claimed in claim 2, It is characterized in that Under the single antenna power constraint, the R-WMMSE algorithm is as follows: remember That is The kth sub-block of the R-WMMSE algorithm mainly includes the following three steps: First, update the receiving weight U of user k according to formula (4): k ; Secondly, update the weight matrix W of user k k ; Finally, update the low-dimensional variable matrix X of user k k . Note that the above steps update all users k separately; where α k is the rate weight of the kth user, P max is the power upper limit of the base station; repeat the above three equations (4), (5), (6) for iteration until the difference between two adjacent objective function values is less than a predetermined threshold; finally, let X = [X 1 ,X 2 ,…,X K ], the precoding matrix is P = H H X is given.
4. A hybrid distributed baseband processing method suitable for XL-MIMO system as claimed in claim 2, It is characterized in that Under the single antenna power constraint, the D-PAPC-WMMSE algorithm includes three main iterative steps: First, update the receiving weight matrix U of user k according to formula (7): k : Secondly, update the weight matrix W of user k according to formula (8): k : Finally, the local precoding vector p of the mth antenna is updated according to equation (9): m : in P m is the rated power constraint of the mth antenna.
