Radio receiving device and method

Abstract

To provide a radio receiving device, method, program, radio unit (RU) and distributed unit (DU) that reduce a processing time for multiuser detection in the DU while reducing fronthaul bandwidth between the RU and DU.SOLUTION: A radio receiving device decomposes an estimated N'×M' channel matrix into N'×B' weight matrix W (B' is an integer equal to or less than N'-1 and equal to or more than M'), each column of which is orthogonal, and a B'×M' matrix containing two or more non-zero matrix elements in each column, generates the weight matrix W or generates a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which equals the weight matrix W, performs reception beamforming for received signals of N' receiving antennas using the weight matrix W or the sub-weight matrix W1, and executes BP algorithm using the signals after the reception beamforming.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present disclosure relates to wireless communication systems, and more particularly to receive signal processing. [Background technology]

[0002] Massive multi-user multi-input multi-output (MIMO) is a well-known technique for achieving high-capacity communications. It uses a base station with multiple antennas to communicate with multiple users. Massive multi-user MIMO is also known as massive MIMO. Massive multi-user MIMO can be used, for example, in the uplink of multiple-access cellular systems such as fifth-generation mobile communications (5G) systems. Massive multi-user MIMO receivers require multi-user detection to separate multi-user signals from the received signal. A 5G base station consists of a distributed unit (DU) and a central unit (CU), which primarily handle digital baseband signal processing, and a radio unit (RU), which primarily handles radio frequency (RF) signal processing for transmitting and receiving wireless signals. Generally, multi-user detection is performed by the DU. The received signal at the RU must then be transmitted to the DU via the fronthaul, which connects the RU and DU. However, in massive multi-user MIMO, wireless signals are received by a large number of antennas, which means that a large number of received signals must be transmitted via the fronthaul, and this requires a wide bandwidth for the fronthaul connecting the RU and DU.

[0003] One well-known multiuser detection algorithm is the Gaussian Belief Propagation (GaBP) algorithm (see, for example, Non-Patent Documents 1 and 2). The GaBP algorithm propagates a quality value (called a belief) representing the reliability of the detected symbol between iterations, thereby gradually improving detection accuracy. The GaBP algorithm is a low-computation detection algorithm because it uses a Gaussian approximation based on the central limit theorem, which assumes a large system. Detectors using the GaBP algorithm include a soft interference canceller (IC), a belief generator (BG), and a soft replica generator (RG). The soft interference canceller cancels interference components from the received signal using replicas of each transmitted symbol obtained in the previous iteration. The belief generator generates beliefs based on the canceled signal. The soft replica generator generates replicas of the transmitted signal based on the beliefs.

[0004] Techniques for improving the performance of soft-decision detection or decoding using the GaBP algorithm include damping, scaling, and node selection. Damping uses a weighted average of beliefs generated in past iterations and beliefs generated in the current iteration as a new belief, thereby suppressing belief oscillations that cause poor convergence (see Non-Patent Document 1). The damping coefficient determines the weight of the weighted average used in damping. Scaling takes into account the low reliability of beliefs at the beginning of an iteration and adjusts the absolute value of the belief to gradually increase as the iterations increase (see Non-Patent Document 2). The scaling coefficient is a parameter used to adjust the absolute value of beliefs during scaling. In the case of MIMO detection, node selection is used to counter fading spatial correlation (correlation between receive antennas) (see Non-Patent Document 2). Specifically, node selection divides a set of receive antenna elements into multiple subsets. Each subset consists of spatially separated (low-correlated) receive antenna elements. The GaBP algorithm with node selection updates only one subset of beliefs in one GaBP iteration, and then updates other subsets of beliefs sequentially in subsequent GaBP iterations.

[0005] While the GaBP algorithm has excellent detection characteristics, it requires knowledge of all received signals to detect the transmitted signal, which requires a wide fronthaul bandwidth to transmit all radio signals received by all receive antennas of the RU from the RU to the DU.

[0006] Another known multiuser detection algorithm is the QR-decomposed Gaussian belief propagation (QR-GA-BP) algorithm (see Non-Patent Document 3). QR-GA-BP decomposes an N'×M' (i.e., N' rows and M' columns) channel matrix defined between one or more M' transmitting antennas of a wireless transmitting device and N' receiving antennas of a wireless receiving device into an N'×M' matrix Q, whose columns are orthogonal, and an M'-order upper triangular matrix using QR decomposition. Furthermore, QR-GA-BP performs receive beamforming by taking the matrix product of the Hermitian conjugate of the generated matrix Q and an N'×1 received signal vector consisting of the received signals of the N' receiving antennas. QR-GA-BP then applies the Gaussian Belief Propagation algorithm to detect transmitted signals from the received signals after beamforming. QR-GA-BP utilizes the fact that the channel matrix after receive beamforming becomes an upper triangular matrix to sequentially generate beliefs from received signals containing fewer transmitted signal components. In the first step, beliefs and replicas are generated using the GaBP algorithm for a received signal containing only one transmitted signal component. In the subsequent m-th step, beliefs and replicas are generated using the GaBP algorithm using the (m-1) received signals used in the (m-1)th step, a received signal containing m transmitted signal components, and the beliefs and replicas generated in the (m-1)th step.

[0007] Due to the properties of upper triangular matrices, the mth received signal contains only the first through mth transmitted signal components. Therefore, when m is small, the accuracy of the Gaussian approximation based on the central limit theorem deteriorates. Therefore, a method of simultaneously generating beliefs for all received signals cannot obtain accurate beliefs. Therefore, QR-GA-BP cancels the interference component of the mth transmitted signal component using the first through (m-1)th transmitted signal components, which are accurately generated by sequentially detecting the first received signal, which does not contain interference components. This accurately cancels interference from the first through (m-1)th transmitted signal components, allowing QR-GA-BP to accurately generate beliefs and replicas for the mth transmitted signal component. QR-GA-BP performs the above steps for all received signals, and by repeating this set of steps, gradually improves detection accuracy. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] P. Som, T. Datta, A. Chockalingam and BS Rajan, "Improved large-MIMO detection based on damped belief propagation," 2010 IEEE Information Theory Workshop on Information Theory (ITW 2010, Cairo), pp.1-5, 2010 [Non-patent document 2] T. Takahashi, S. Ibi and S. Sampei, "Design of criterion for adaptively scaled belief in iterative large MIMO detection," IEICE Transaction on Communications, vol. E102.B, no.2, pp.285-297, 2019 [Non-patent document 3] Shogo Tanabe, Koji Ishibashi, "A Study on Low-Complexity Belief Propagation Demodulation Method Using Sorted QR Decomposition for Massive MIMO," IEICE Technical Report, pp.83-88, 2016 Summary of the Invention [Problem to be solved by the invention]

[0009] QR-GA-BP performs multiuser detection after receive beamforming, reducing the number of received signals used for detection from the number of receive antennas to the number of transmitted signals while still achieving high-accuracy detection. This reduces the number of signals transmitted from the RU to the DU, thereby reducing the required fronthaul bandwidth. However, QR-GA-BP requires longer processing time than standard GaBP. This is because interference cancellation, belief generation, and replica generation are performed sequentially for each received signal, which is not possible for all received signals simultaneously, as in the previously described GaBP. Note that QR-GA-BP cannot be applied to standard GaBP instead of sequential belief generation for each received signal. This is because the equivalent channel matrix after receive beamforming is an upper triangular matrix, and therefore the multiple received signals after receive beamforming each contain different numbers of transmitted signals (components). In other words, the multiple received signals differ in the number of transmitted signals (components) they contain. This degrades the accuracy of the Gaussian approximation based on the central limit theorem for some received signals. Therefore, if beliefs are generated simultaneously for all received signals using GaBP, the accuracy of the generated beliefs will vary, and the algorithm will not converge correctly.

[0010] The GaBP algorithm requires a wideband fronthaul between RU and DU for massive multi-user MIMO using a large number of antennas, while QR-GA-BP has the problem of requiring long processing time for multi-user detection due to its sequential processing.

[0011] That is, the GaBP algorithm can simultaneously update beliefs for all received signals, enabling multiuser detection with short processing times. However, the GaBP algorithm performs multiuser detection using received signals equal to the number of receive antennas, which increases the required bandwidth for the fronthaul. On the other hand, the QR-GA-BP algorithm reduces the number of received signals used for multiuser detection to the number of transmitted signals by using the Q matrix obtained by QR decomposition of the channel matrix for receive beamforming. While this reduces the required bandwidth for the fronthaul, it also requires long processing times due to the need for sequential interference cancellation, belief generation, and replica generation for multiuser detection.

[0012] One of the objectives to be achieved by the embodiments disclosed herein is to provide an apparatus, a method, and a program that contribute to reducing the processing time required for multi-user detection at the DU while reducing the required bandwidth of the fronthaul between the RU and the DU. It should be noted that this objective is only one of multiple objectives to be achieved by the multiple embodiments disclosed herein. Other objectives or problems and novel features will become apparent from the description in this specification or the accompanying drawings. [Means for solving the problem]

[0013] In a first aspect, a wireless receiving device includes at least one beamforming weight generator, a first beamformer, and a BP detector. The at least one beamforming weight generator is configured to generate the weight matrix W by decomposing an estimated N'×M' (i.e., N' rows and M' columns) channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, where B' is an integer between N'-1 and M', and an B'×M' matrix, each column of which includes two or more non-zero matrix elements. Alternatively, the at least one beamforming weight generator is configured to generate a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W. The first beamformer is configured to perform receive beamforming on received signals of the N' receiving antennas using the weight matrix W or the sub-weight matrix W1. The BP detector is configured to execute a BP algorithm using a signal output from the first beamformer and transmitted via a fronthaul.

[0014] In a second aspect, a method performed by a wireless receiving device includes: (a) generating a weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; (b) performing first receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; and (c) Executing a BP algorithm using the signal on which the first receive beamforming has been performed and transmitted via the fronthaul.

[0015] In a third aspect, an RU included in a wireless receiving device and coupled to a DU of the wireless receiving device via a fronthaul includes at least one beamforming weight generator and a first beamformer. The at least one beamforming weight generator is configured to generate the weight matrix W by decomposing an estimated N'xM' channel matrix defined between M' transmit antennas of one or more wireless transmitting devices and N' receive antennas coupled to the wireless receiving device into an N'xB' weight matrix W, whose columns are orthogonal, and a B'xM' matrix, where B' is an integer between N'-1 and M', and each column includes two or more non-zero matrix elements. Alternatively, the at least one beamforming weight generator is configured to generate a combination of an N'xB' sub-weight matrix W1 and at least one second sub-weight matrix, whose product is equal to the weight matrix W. The first beamformer is configured to perform receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1, and provide the beamformed signals to the DU via the fronthaul.

[0016] In a fourth aspect, a method performed by a RU included in a wireless receiving device and coupled to a DU of the wireless receiving device via fronthaul includes: (a) generating a weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; (b) performing receive beamforming on the received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; and (c) providing a beamformed signal to the DU via the fronthaul;

[0017] In a fifth aspect, a DU included in a wireless receiving device and coupled to an RU of the wireless receiving device via a fronthaul includes at least one beamforming weight generator and a beamforming detector. The at least one beamforming weight generator is configured to generate the weighting matrix W by decomposing an estimated N'xM' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'xB' weighting matrix W, whose columns are orthogonal, and a B'xM' matrix, where B' is an integer between N'-1 and M', and each column includes two or more non-zero matrix elements. Alternatively, the at least one beamforming weight generator is configured to generate a combination of an N'xB' sub-weighting matrix W1 and at least one second sub-weighting matrix, whose product is equal to the weighting matrix W. The BP detector is configured to execute a BP algorithm using signals generated by the RU and transmitted via the fronthaul by performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1.

[0018] In a sixth aspect, a method performed by a DU included in a wireless receiving device and coupled to an RU of the wireless receiving device via fronthaul includes: (a) generating a weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; (b) providing the weight matrix W or the sub-weight matrix W1 to the RU; and (c) performing a BP algorithm using signals generated by the RU and transmitted via the fronthaul by performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1.

[0019] In a seventh aspect, a program comprises a set of instructions (software code) that, when loaded into a computer, causes the computer to perform the method according to the second, fourth or sixth aspect. [Effects of the Invention]

[0020] According to the above-described aspects, it is possible to provide an apparatus, a method, and a program that contribute to reducing the processing time required for multi-user detection in the DU while reducing the required bandwidth for the fronthaul between the RU and the DU. [Brief explanation of the drawings]

[0021] [Figure 1] 1 is a diagram illustrating a wireless communication system according to a first embodiment. [Figure 2] FIG. 2 is a diagram illustrating a system model according to the first embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of the configuration of a base station according to the first embodiment. [Figure 4] FIG. 2 is a diagram illustrating an example of the configuration of a processor of a base station according to the first embodiment. [Figure 5] FIG. 2 is a diagram illustrating an example of the configuration of a processor of a base station according to the first embodiment. [Figure 6] FIG. 2 is a diagram illustrating an example of the configuration of a processor of a base station according to the first embodiment. [Figure 7] FIG. 2 is a diagram illustrating an example of the configuration of a processor of a base station according to the first embodiment. [Figure 8] FIG. 2 is a diagram illustrating an example of the configuration of a processor of a base station according to the first embodiment. [Figure 9] FIG. 2 is a diagram illustrating an example of the configuration of a BP detector according to the first embodiment. [Figure 10] FIG. 3 is a diagram illustrating an example of a lookup table according to the first embodiment. [Figure 11] FIG. 3 is a diagram illustrating an example of a lookup table according to the first embodiment. [Figure 12] 6 is a flowchart showing an example of an operation of the base station according to the first embodiment. [Figure 13] FIG. 10 is a diagram illustrating an example of the configuration of a BP detector including an initial value generator according to a second embodiment. [Figure 14] 10 is a flowchart showing an example of initial value generation according to the second embodiment. [Figure 15] FIG. 10 is a conceptual diagram showing deep deployment of a BP detector according to a third embodiment. [Figure 16] FIG. 11 is a diagram illustrating an example of the configuration of a parameter learner according to the third embodiment. [Figure 17] FIG. 10 is a diagram illustrating an example of a computer system according to a third embodiment. [Figure 18] 11 is a flowchart showing an example of parameter learning according to the third embodiment. [Figure 19] FIG. 10 is a diagram showing the bit error rate performance of the BP detector according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0022] Hereinafter, specific embodiments will be described in detail with reference to the drawings. In each drawing, the same or corresponding elements are designated by the same reference numerals, and for clarity of explanation, duplicate explanations will be omitted as necessary.

[0023] First Embodiment FIG. 1 illustrates an exemplary configuration of a wireless communication system according to several embodiments, including the present embodiment. Referring to FIG. 1, a base station 1 provides wireless access to several wireless terminals 2. The base station 1 may be referred to as an access point, a transmission / reception point (TRP), or another name. The base station 1 may be, for example, a gNB in ​​a 5G system, or a system including a RU and a DU. The RU may perform RF signal processing and some physical layer or Layer 1 (e.g., Low PHY) signal processing. The physical layer signal processing performed by the RU may include, for example, digital beamforming. The DU may perform the remaining physical layer (e.g., High PHY) signal processing and Layer 2 signal processing. The RU and the DU are connected via a fronthaul. The fronthaul is also referred to as a mobile fronthaul or a (mobile) fronthaul network. For example, the fronthaul may be a wired fronthaul using one or more optical fibers, or a wireless fronthaul using millimeter-wave radio transmission.

[0024] In some implementations, the wireless communication system may utilize a multi-user MIMO technique for uplink transmission from multiple wireless terminals 2 to the base station 1. In this case, the base station 1 may receive reference signals from the multiple wireless terminals 2, estimate MIMO channels between the multiple wireless terminals 2 and the base station 1 using the received reference signals, and detect transmission signals using data signals received from the multiple wireless terminals 2 and the estimated channels. That is, the base station 1 may perform MIMO detection to separate the multi-user signals of the multiple wireless terminals 2.

[0025] FIG. 2 shows an example of a system model for uplink multi-user MIMO transmission. In FIG. 2, multiple transmitters 20 of multiple wireless terminals 2 communicate with a receiver 10 of a base station 1 via a channel (propagation path) 30. In the example of FIG. 2, each of the M' transmitters 20 has one transmit antenna. Alternatively, each transmitter 20 may have two or more transmit antennas. The receiver 10 of the base station 1 has N' receive antennas. Assume that the total number of transmit antennas, M', is smaller than the total number of receive antennas, N'.

[0026] In the following description, for simplicity, it is assumed that the transmission signal from each wireless terminal 2 (user) is a single-carrier transmission, and that the propagation path between each wireless terminal 2 and the base station 1 is flat fading. On the other hand, even in a multipath fading environment when the transmission signal from each user uses Orthogonal Frequency Division Multiplexing (OFDM) or Single Carrier-Frequency Division Multiple Access (SC-FDMA), the transmission path can be assumed to be flat fading on a subcarrier basis by inserting a cyclic prefix of an appropriate length into the transmission signal. Therefore, this embodiment may be applied to OFDM and SC-FDMA.

[0027] Assume that a total of M' transmit antennas of multiple wireless terminals 2 transmit signals modulated by Quadrature Amplitude Modulation (QAM) and that the signals are received by a base station 1 equipped with N' receive antennas. In this case, the complex signal model in the equivalent low-pass representation is expressed by Equation 1:

number

[0028] The number of modulation symbols for QAM modulation is Q', and for example, Q' is 4 for Quadrature Phase Shift Keying (QPSK) and 16 for 16QAM. Regarding the amplitude of the modulation symbols, the amplitudes of the I and Q axes are {+c, -c} for QPSK and {+c, -c, +3c, -3c} for 16QAM. Here, c is expressed by the following equation 2. E s is the average signal power. The power of the complex noise at each receiving antenna is N0.

number

[0029] 3 shows an example configuration of the base station 1. Referring to FIG. 3, the base station 1 includes an RU 11-1 and a DU 11-2. The RU 11-1 includes a Radio Frequency (RF) transceiver 13, a processor 14-1, and a memory 15-1, and is connected to the antenna array 12. The DU 11-2 includes a processor 14-2, a memory 15-2, and a network interface 16.

[0030] The RF transceiver 13 performs analog RF signal processing to communicate with multiple wireless terminals 2. The RF transceiver 13 may include multiple transceivers. The RF transceiver 13 is coupled to the antenna array 12, the processor 14-1 in the RU 11-1, and the processor 14-2 in the DU 11-2. The RF transceiver 13 receives modulation symbol data from the processor 14-1 in the RU 11-1, generates a transmit RF signal, and provides the transmit RF signal to the antenna array 12. The RF transceiver 13 also generates a baseband receive signal based on the receive RF signal received by the antenna array 12, and provides the receive signal to the processor in the DU 11-2 if the receive signal is a reference signal, or to the processor 14-1 in the RU 11-1 if the receive signal is a data signal.

[0031] The processors 14-1 and 14-2 perform digital baseband communication processing (data plane processing) and control plane processing for wireless communication. The processors 14-1 and 14-2 may include multiple processors. For example, the processors 14-1 and 14-2 may include a modem processor (e.g., a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), or a Digital Signal Processor (DSP)) that performs digital baseband signal processing and a protocol stack processor (e.g., a Central Processing Unit (CPU) or a Micro Processing Unit (MPU)) that performs control plane processing.

[0032] For example, the digital baseband signal processing by the processors 14-1 and 14-2 may include signal processing of the Service Data Adaptation Protocol (SDAP) layer, the Packet Data Convergence Protocol (PDCP) layer, the Radio Link Control (RLC) layer, the Medium Access Control (MAC) layer, and the Physical (PHY) layer. Also, the control plane processing by the processors 14-1 and 14-2 may include processing of Non-Access Stratum (NAS) messages, Radio Resource Control (RRC) messages, Medium Access Control (MAC) Control Elements (CEs), and Downlink Control Information (DCI).

[0033] The network interface 16 is used to communicate with network nodes (e.g., other base stations, a Centralized Unit (CU), and a core network node). The network interface 16 may include, for example, a network interface card (NIC) that complies with the IEEE 802.3 series.

[0034] Each of the memories 15-1 and 15-2 is configured with a combination of volatile memory and nonvolatile memory. The volatile memory may be, for example, Static Random Access Memory (SRAM), Dynamic RAM (DRAM), or a combination thereof. The nonvolatile memory may be, for example, Mask Read Only Memory (MROM), Electrically Erasable Programmable ROM (EEPROM), flash memory, or a hard disk drive, or any combination thereof. The memories 15-1 and 15-2 may include storage located remotely from the processors 14-1 and 14-2. In this case, the processors 14-1 and 14-2 may access the memories 15-1 and 15-2, respectively, via the network interface 16 or another I / O interface.

[0035] The memories 15-1 and 15-2 may include computer-readable media that store one or more software modules (computer programs) that include instructions and data for performing at least a portion of the processing by the base station 1. In some implementations, the processors 14-1 and 14-2 may be configured to read and execute the software modules from the memories 15-1 and 15-2, respectively, to perform at least a portion of the processing by the base station 1 described in the above embodiments.

[0036] According to this embodiment, the processors 14-1 and 14-2 can cause the base station 1 to perform receive signal processing for receive beamforming and multi-user detection (MIMO detection). To that end, as shown in Fig. 4, the processor 14-1 in the RU 11-1 can include a beamformer 143-1, and the processor 14-2 in the DU 11-2 can include a channel estimator 141, a beamforming weight generator 142, and a BP detector 144. Additionally or alternatively, as shown in Figs. 5, 6, and 7, the processor 14-1 may include the beamforming weight generator 142, and as shown in Figs. 7 and 8, the processor 14-2 may include the beamformer 143-2.

[0037] The channel estimator 141 estimates a MIMO channel for generating beamforming weights based on a reference signal received from the RF transceiver 13. The channel estimator 141 is coupled to the beamforming weight generator 142 and generates an estimated channel matrix H c to the beamforming weight generator 142.

[0038] The beamforming weight generator 142 uses the channel matrix H received from the channel estimator 141 c The beamforming weight generator 142 is coupled to the beamformer 143-1 and supplies the weight matrix W to the beamformer 143-1. As shown in Equation 3, the N'×M' channel matrix H c is the N'×B' matrix W and the B'×M' matrix H' c and W is the weight matrix, H' c is the equivalent channel matrix after receive beamforming. B' is the number of received signals after receive beamforming, and is an integer between N'-1 and M'. The weight matrix W is a matrix in which each column corresponding to each element of the received signal vector after receive beamforming is orthogonal, and is a square matrix H' cDue to the properties of the BP detector 144 (described later), each column must contain two or more non-zero matrix elements. As shown in Equation 4, the matrix W, whose columns are orthogonal, is such that the product of the Hermitian conjugate of matrix W and matrix W is an identity matrix. Furthermore, weight matrix W may be a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to weight matrix W. For example, as shown in Equation 5, weight matrix W may be the product of an N'×B' sub-weight matrix W1 and a B'-order square matrix (sub-weight matrix W2). In this case, sub-weight matrix W1 is supplied to beamformer 143-1 in RU11-1, and sub-weight matrix W2 is supplied to beamformer 143-2 in DU11-2.

number

number

number

[0039] In a first implementation of the beamforming weight generator 142, let B' be equal to M' and define the channel matrix H c is decomposed into an N'×M' matrix Q and an M'×M' upper triangular matrix R by QR decomposition as shown in Equation 6. By the definition of QR decomposition, each column of matrix Q is orthogonal. In this case, as shown in Equation 7, matrix W is the product of matrix Q and M'-th order unitary matrix U', and matrix H' c is the product of the Hermitian conjugate of matrix U' and the upper triangular matrix R. Here, the unitary matrix U' must be a unitary matrix with two or more non-zero matrix elements in the first column. This is the square matrix H' cThis is because each column of must have two or more non-zero matrix elements. For example, a Discrete Fourier Transform (DFT) matrix can be used as a unitary matrix. In addition, the GaBP algorithm, which is one of the BP algorithms described below, operates on the premise that the observation noise is white. Therefore, by making each column of the weight matrix W orthogonal, the noise after receive beamforming also remains white, making it possible to detect it using the GaBP algorithm.

number

number

number

[0040] In a second implementation, let B' be equal to M' and define the channel matrix H c is decomposed by singular value decomposition into an N'×M' matrix U consisting of M' left singular vectors, an M'-th diagonal matrix Σ in which the matrix element in the i-th row and i-th column is the i-th singular value, and the Hermitian conjugate of an M'-th unitary matrix V consisting of M' right singular vectors, as shown in Equation 9. At this time, due to the properties of the left singular vectors, each column of matrix U is orthogonal, and the product of matrix Σ and the Hermitian conjugate of matrix V is an M'-th square matrix. At this time, as shown in Equation 9, matrix U is a weight matrix W, and the product of matrix Σ and the Hermitian conjugate of matrix V is a square matrix H' c Furthermore, as shown in Equation 10, the matrix W may be the product of the matrix U obtained by singular value decomposition and an arbitrary M'th order unitary matrix U'. In this case, the square matrix H' c is the product of the Hermitian conjugate of the unitary matrix U', the matrix Σ, and the Hermitian conjugate of the matrix V.

number

number

[0041] Alternatively, as shown in Equation 11, the matrix U consisting of left singular vectors may be the sub-weight matrix W1, and the unitary matrix U' may be the sub-weight matrix W2.

number

[0042] In a third implementation, let B' be equal to M' and define the channel matrix H c As shown in Equation 12, the matrix H c and matrix H c and the matrix H c The Gram matrix is ​​decomposed into the matrix H c Hermitian conjugate of and matrix H c In this case, as shown in Equation 13, the matrix H c and matrix H c The product of the Gram matrix and the -1 / 2 power of the matrix W and the matrix H c The 1 / 2 power of the Gram matrix is ​​the matrix H' c As shown in Equation 14, it can be confirmed that each column of the weight matrix W is indeed orthogonal, and the product of the Hermitian conjugate of matrix W and matrix W is an identity matrix.

number

number

number

[0043] Also, as shown in Equation 15, the channel matrix H c is the sub-weight matrix W1, and the channel matrix H c The -1 / 2 power of the Gram matrix may be the sub-weight matrix W2.

number

[0044] Furthermore, as shown in Equation 16, the matrix W is c and matrix H c It may be a product of the −1 / 2 power of the Gram matrix of and any M′-th order unitary matrix U′.

number

[0045] Also, as shown in Equation 17, the channel matrix H c is the sub-weight matrix W1 and the channel matrix H c The product of the −1 / 2 power of the Gram matrix and an arbitrary M′-th order unitary matrix U′ may be the sub-weight matrix W2. Furthermore, as shown in Equation 18, the channel matrix H c and H c The product of the Gram matrix and the -1 / 2 power of the Gram matrix may be the sub-weight matrix W1, and any M'th order unitary matrix U' may be the sub-weight matrix W2.

number

number

[0046] In the fourth implementation example, the channel matrix H c As shown in Equation 19, K N' k ×M' submatrix H c k and each submatrix H c k As shown in Equation 20, each column is orthogonal to N' k ×B' k matrix W k and B' k ×M' matrix H' c k and W as shown in Equation 21. k The weight matrix is ​​a block diagonal matrix W whose submatrix is ​​a matrix of B'. Here, k is an integer between 1 and K. k is N' kand M', which is the smaller value. In this case, the weight matrix W is a matrix of N' × B', and B' is the sum of B'1 to B' as shown in Equation 23. K The equivalent channel matrix after receiving beamforming is the submatrix H' as shown in Equation 22. c k and the matrix H c , W, H' c are related as shown in Equations 3 and 4. The decomposition of the matrix in Equation 20 may be performed using the decompositions in the above-described implementation examples 1 to 3.

number

number

number

number

number

[0047] These are some examples of implementations of the beamforming weight generator 142, but the implementation is not limited to these examples.

[0048] The beamformer 143-1 generates a received signal vector y consisting of received signals from N' receiving antennas supplied from the RF transceiver 13. c The receive beamforming is performed on the received signal vector y. The receive beamforming uses the weight matrix W or the sub-weight matrix W1 received from the beamforming weight generator 142. More specifically, the receive beamforming is performed by using the Hermitian conjugate of the weight matrix W and the received signal vector y as shown in Equation 24. c The beamformer 143-1 is coupled to the BP detector 144, which receives the signal vector y′ via the fronthaul. cFurthermore, if the DU11-2 includes a beamformer 143-2, the beamformer 143-1 in the RU11-1 supplies a signal to the beamformer 143-2 in the DU11-2 via the fronthaul, and the beamformer 143-2 in the DU11-2 supplies a signal vector y' c supply.

number

[0049] The BP detector 144 detects the received signal vector y′ after receive beamforming supplied from the beamformer 143-1 or 143-2. c The BP algorithm is executed for a total number of iterations T for the received signal. Here, the BP algorithm includes interference cancellation processing using a replica of the transmitted signal, and is an algorithm constructed assuming that the residual interference noise components contained in the received signal after interference cancellation approximate a Gaussian distribution according to the central limit theorem. Possible BP algorithms include GaBP, Generalized Approximate Message Passing (GAMP), and Expectation Propagation. Furthermore, the transmitted signal is detected using the equivalent channel matrix H' after receive beamforming shown in Equation 25. c is used.

number

[0050] Equivalent channel matrix H' after receive beamforming c may be estimated using a demodulation reference signal received separately from the reference signal used to generate the beamforming weight. For the sake of simplicity, the complex equivalent low-pass expression is expressed as a received signal model replaced with an equivalent real signal model shown in Equations 26 to 30:

number

number

number

number

number

number

[0051] The following explanation will be given using the equivalent real number model.

[0052] The BP detector 144 executes a GaBP algorithm, which is one of the BP algorithms, and the GaBP algorithm may use a parameter set. The parameter set may include at least one of a scaling factor, a damping factor, and a weighting factor. In some implementations, the GaBP algorithm uses a first parameter set. The first parameter set includes multiple scaling factors and multiple damping factors. The BP detector 144 uses each of the multiple scaling factors in different iterations of the GaBP algorithm. Similarly, the BP detector 144 uses each of the multiple damping factors in different iterations of the GaBP algorithm. Therefore, the total number of the multiple scaling factors and the total number of the multiple damping factors may be equal to the total number of iterations of the GaBP algorithm.

[0053] In another implementation, the GaBP algorithm executed by the BP detector 144 uses a second parameter set. The second parameter set includes multiple scaling coefficients and multiple weighting coefficients. The BP detector 144 uses these multiple scaling coefficients in different iterations of the GaBP algorithm. Similarly, the BP detector 144 uses these multiple weighting coefficients in different iterations of the GaBP algorithm. Note that, as will be described later, the GaBP algorithm may use multiple weighting coefficients for each iteration. In this case, the second parameter set may include a set of weighting coefficients for each iteration.

[0054] The first parameter set (or the second parameter set) is stored in one or both of memories 15-1 and 15-2 of the base station 1. As shown in Figures 4, 5, 6, 7, and 8, the first parameter set (or the second parameter set) may be saved in the memory 15-2 as a lookup table (LUT) 151.

[0055] Next, a configuration example of the BP detector 144 will be described below. Fig. 9 shows a configuration example of the BP detector 144 that executes the GaBP algorithm. Referring to Fig. 9, the BP detector 144 includes B soft interference cancellers 1441-1 to 1441-B, a belief generator 1442, and B soft replica generators 1443-1 to 1443-B. The soft interference cancellers 1441-1 to 1441-B cancel the received signals y'1 to y' B For example, the soft interference canceller 1441-1 receives the first received signal y1' after receive beamforming. In addition, to perform the t-th iteration, the soft interference canceller 1441-1 receives the soft replicas x hat of all the transmitted signals generated in the previous (t-1) iteration. 1,1 (t-1) ~xHat 1,M (t-1) Here, the x symbol means a superscript ^ on the letter x. Then, the soft interference canceller 1441-1 receives the canceled received signal y' tilde 1,1 (t) ~y' tilde 1,M (t) where the y' tilde means a superscript ~ on the letter y'.

[0056] The belief generator 1442 reads out a plurality of damping coefficients (or a plurality of sets of weighting coefficients) included in the above-mentioned first parameter set (or second parameter set) from the LUT 151. The belief generator 1442 generates the post-cancellation received signal y' 1,1 (t) ~y' tilde 1,M (t) Similarly, the belief generator 1442 receives the canceled received signal y′ tilde generated from the other soft interference canceller 1441-b (where b is 2 to B). b,1 (t) ~y' tilde b,M (t)The belief generator 1442 then uses the damping factor (or set of weighting factors) for the t-th iteration to generate the belief r associated with the first received signal. 1,1 (t) ~r 1,M (t) Similarly, belief generator 1442 generates beliefs associated with the remaining second to Bth received signals.

[0057] The soft replica generator 1443-1 reads out a plurality of scaling coefficients included in the first parameter set (or the second parameter set) from the LUT 151. The soft replica generator 1443-1 generates a belief r associated with the first received signal. 1,1 (t) ~r 1,M (t) from the belief generator 1442. Then, the soft replica generator 1443-1 uses the scaling factor for the t-th iteration to generate the soft replica x hat 1,1 (t) ~xHat 1,M (t) and the soft replica power p 1,1 (t) ~p 1,M (t) Generate.

[0058] After completing the GaBP process for a total number of iterations T, the belief generator 1442 generates the estimates r1 of the separated M transmitted signals. (T) ~r M (T) to the decision and demodulator 145

[0059] The processing performed by soft interference cancellers 1441-1 to 1441-B, belief generator 1442, and soft replica generators 1443-1 to 1443-B will be described in more detail below.

[0060] (1) Soft interference cancellers 1441-1 to 1441-B In the first iteration, a soft replica has not yet been generated. Therefore, the soft interference cancellers 1441-1 to 1441-B supply the first to bth received signals to the belief generator 1442 without performing cancellation processing. In the second or subsequent t-th iteration, the soft interference canceller 1441-b associated with the bth received signal cancels the M-1 transmitted signal components other than the mth transmitted signal from the bth received signal, and generates the received signal y' (tilde) after cancellation. b,m (t) Generate the received signal y' after cancellation. b,m (t) is expressed as follows in Equation 32:

number

[0061] (2) Belief Generator 1442 Belief generator 1442 generates a belief using the received signal after cancellation processing. First, belief generator 1442 generates a belief using the b-th received signal y′ b The received signal y' after cancellation processing is b,m (t) The process expressed by the following equation 33 is performed using b,m (t) get.

number

number

number

[0062] Transmitted signal component s b,m (t) The true transmitted signal x contained in m Equalization gain ω for b,m (t) is expressed as Equation 36 and is used for normalization during scaling:

number

[0063] Here, Equations 33 and 36 are obtained by Gaussian approximation of the residual interference noise components in the cancellation process, assuming a large system. Therefore, when the number of transmission signals included in the residual interference components is small, the accuracy of the Gaussian approximation deteriorates, and the accuracy of the belief deteriorates. For this reason, the channel matrix H' after beamforming c It is desired that each column of contains two or more non-zero matrix elements.

[0064] Next, the belief generator 1442 generates the transmitted signal component s b,m (t) Using belief r b,m (t) The belief generator 1442 uses a damping process or a node selection process. First, the damping process will be explained. The damping process uses a damping coefficient η (t)The weighted average of the transmitted signal component obtained in the previous (t-1)th iteration and the transmitted signal component obtained in the current tth iteration is calculated using:

number

number

[0065] Next, the node selection process will be described. The node selection process is performed according to Equation 39 to obtain s', which is a combination of the transmitted signal components related to the received signals of each beam region in the most recent K iterations. b,m (t) get:

number

[0066] Weighting coefficient η i,t-k (t) Using s' b,m (t) The equalization gain contained in is given by Equation 40:

number

[0067] The belief generator 1442 uses s' obtained by any of the processes. b,m (t) ω' b,m (t) Normalize it with , and normalize the belief r b,m (t) and supplies this to the soft replica generators 1443-1 to 1443-B. b,m (t) is expressed as follows in Equation 41:

number

[0068] (3) Soft replica generators 1443-1 to 1443-B The soft replica generators 1443-1 to 1443-B are configured to use a scaling coefficient a (t) By Belief R b,m (t) Scale the soft replica x hat according to equations 42 and 43 below. b,m (t) and the soft replica power p b,m (t) Generates:

number

number

number

[0069] The decision threshold s' is the set of decision thresholds S Q The set S can take any of the values Q is {0} for QPSK and {0, +2c, -2c} for 16QAM. The tanh function is the hyperbolic tangent function. These equations are used to calculate the soft replica x hat by combining belief information around the decision threshold s'. b,m (t) and the soft replica power p b,m (t) It indicates that

[0070] (4) Output of BP detector 144 After T iterations, the belief generator generates the estimates r of the separated M transmitted signals. m (T) is fed to the decision and demodulator 145. The estimated value r m (T) is expressed as follows in Equation 45:

number

[0071] FIG. 10 shows an example of a first parameter set, i.e., a set of scaling and damping coefficients. These parameters may be stored in the memory 15-2 as a LUT 151. The table shown in FIG. 10 stores parameters for three values ​​(i.e., 3, 8, and 16) of the total number of GaBP iterations T. In this case, the processor 14-2 (e.g., the BP detector 144) of the base station 1 may use a subset of parameters corresponding to the set (or selected) number of iterations T in the GaBP algorithm. The parameters for the number of iterations T include the scaling coefficient a for each iteration. T (t) and damping coefficient η T (t)The subscripts of these parameters represent the total number of iterations. For example, if the total number of iterations T is 3, the parameter is the scaling factor a3 for the first iteration. (1) and damping coefficient η3 (1) Set of scaling factors a3 for the second iteration (2) and damping coefficient η3 (2) set of , as well as the scaling factor a3 for the third iteration (3) and damping coefficient η3 (3) Includes a set of.

[0072] Figure 11 shows an example of the second parameter set, i.e., a combination of scaling and weighting factors. Similar to that of Figure 10, the table shown in Figure 11 stores parameters when the total number of GaBP iterations T is 3, 8, and 16. The parameters for the number of iterations T are the scaling coefficients a T (t) and a set of weighting coefficients {η T,t-k (t)} and

[0073] 10 and 11 are merely examples. For example, the BP detector 144 may be operated only with a fixed total number of iterations. In this case, only the parameters for this total number of iterations may be provided to the base station 1. Alternatively, parameters for total numbers of iterations with values ​​of 4 or more may be provided to the base station 1.

[0074] 12 shows an example of the operation of the base station 1. In step 41, the processor 14-2 (e.g., the channel estimator 141) in the DU 11-2 receives a reference signal received by an antenna via the RF transceiver 13. In step 42, the processor 14-2 (e.g., the beamforming weight generator 142) in the DU 11-2 generates an estimated N'×M' channel matrix H c Let's consider the N'×B' matrix W and the B'×M' matrix H'. cand the matrix W is the beamforming weight. Here, the weight matrix W is a matrix in which each column corresponding to each element of the received signal vector after receiving beamforming is orthogonal, and the matrix H' c must be a matrix with at least two non-zero matrix elements in each column.

[0075] In step 43, the processor 14-1 (e.g., beamformer 143-1) in the RU 11-1 receives a data signal received by an antenna via the RF transceiver 13, performs receive beamforming using the generated beamforming weight W, and transmits the beamformed signal to the processor 14-2 (e.g., BP detector 144) in the DU 11-2 via the fronthaul. The processor 14-2 (e.g., beamformer 143-2) in the DU 11-2 may perform receive beamforming. In this case, the processors 14-1 and 14-2 perform receive beamforming using the beamforming weights W1 and W2, respectively.

[0076] In step 44, the processor 14-2 (e.g., BP detector 144) in the DU 11-2 performs the GaBP algorithm on the beamformed signal to obtain estimates r1 to r2 of the separated M transmitted signals. M In one example, the processor 14-2 (e.g., the BP detector 144) in the DU 11-2 may retrieve the scaling coefficients and damping coefficients (or the scaling coefficients and weighting coefficients) from the memory 15-2. Then, the processor 14-2 (e.g., the BP detector 144) in the DU 11-2 performs the GaBP algorithm using the scaling coefficients and damping coefficients (or the scaling coefficients and weighting coefficients) to generate estimates r1 to r2 of the separated M transmitted signals. MThe processor 14-2 (e.g., BP detector 144) in the DU 11-2 may perform the GaBP algorithm while updating the scaling coefficients and damping coefficients (or the scaling coefficients and weighting coefficients) for each iteration. After that, the processor 14-2 (e.g., decision and demodulator) in the DU 11-2 may generate the estimated values ​​r1 to r2. M The transmitted signals of all M' users are decoded based on the above.

[0077] This embodiment can provide the following advantages. First, in this embodiment, the base station 1 performs receive beamforming using the above-described weight matrix W. This reduces the number of received signals transmitted over the fronthaul, thereby reducing the required fronthaul bandwidth. Second, because the equivalent channel matrix after receive beamforming is a square matrix with two or more non-zero matrix elements in each column, the variation in accuracy among multiple beliefs generated by the BP algorithm is reduced. This makes it possible to simultaneously update beliefs for all received signals, thereby reducing processing time.

[0078] Furthermore, when the GaBP algorithm is used in this embodiment, the following effects can be obtained: Since the GaBP algorithm is based on white noise, the noise after receive beamforming becomes white when each column of the weight matrix W is orthogonal, making detection by the GaBP algorithm possible.

[0079] <Second embodiment> In the second embodiment, in addition to the first embodiment, the BP detector 144 includes an initial value generator 1444. Fig. 13 shows an example of the configuration of the BP detector 144 in the second embodiment. According to Fig. 13, the initial value generator 1444 is coupled to the soft interference cancellers 1441-1 to 1441-B and the belief generator 1442, and outputs the soft replica x hat to the soft interference cancellers. b,m (0) and the soft replica power p b,m (0)Supply: Soft replica x hat b,m (0) and the soft replica power p b,m (0) is generated from the received signal vector y' after receive beamforming supplied from the beamformer 143-1 or 143-2. ​​As shown in Figures 7 and 8, the processor 14-2 in the DU 11-2 may include the beamformer 143-2. ​​In this case, the initial value generator 1444 generates the received signal vector y' supplied from the beamformer 143-1 in the processor 14-1 in the RU 11-1 shown in Equation 46. RU Soft replica x hat b,m (0) and the soft replica power p b,m (0) may be generated.

number

[0080] Soft replica x hat b,m (0) and the soft replica power p b,m (0) is the initial value generator matrix W of B'×M' init Generated using Soft Replica x Hat b,m (0) and the soft replica power p b,m (0) is the initial value generator matrix W init and the received signal vector after receive beamforming (y' or y' RU ) where the soft replica x b,m (0) and the soft replica power p b,m (0) is expressed as an equivalent real number model. In the first iteration, the soft interference cancellers 1441-1 to 1441-B use the soft replica x hat b,m (0)In the first iteration, the belief generator uses the soft replica power supplied from the initial value generator 1444 and the post-cancellation received signal y' tilde supplied from the soft interference cancellers 1441-1 to 1441-B. b,m (1) Generate beliefs using

[0081] As implementation examples of the initial value generator 1444, first to fourth implementation examples described in detail below are possible. In the first implementation example, the beamformer 143-1 or 143-2 supplies the beamformed received signal vector y' to the initial value generator 1444. In the second to fourth implementation examples, the processor 14-2 in the DU 11-2 includes the beamformer 143-2, and the beamformer 143-1 supplies the received signal vector y' RU is supplied to the initial value generator 1444. Implementation examples 1 to 3 are implementation examples when B' is equal to M', and implementation example 4 is an implementation example when B' is an arbitrary integer not larger than N'-1 and not smaller than M'.

[0082] In the first implementation example, the initial value generator matrix W init is the matrix shown in Equation 47.

number

[0083] In the second implementation example, the initial value generator matrix W init is the matrix shown in Equation 48.

number

number

[0084] In this implementation example, when a matrix U consisting of left singular vectors generated by singular value decomposition of the channel matrix is ​​used as the sub-weight matrix W1, the initial value generating matrix W init The Hermitian conjugate of is the product of a matrix V consisting of right singular vectors and the inverse matrix of a matrix Σ consisting of singular values, as shown in Equation 50.

number

[0085] When a matrix Q generated by QR decomposing the channel matrix is ​​used as the sub-weight matrix W1, the Hermitian conjugate of the initial value generating matrix Winit is the inverse matrix of the upper triangular matrix R, as shown in Equation 51.

number

[0086] When the channel matrix is ​​used as the sub-weight matrix W1, the initial value generator matrix W init The Hermitian conjugate of and the sub-weight matrix W2 are the inverse of the Gram matrix of the channel matrix and the −1 / 2 power of the Gram matrix of the channel matrix, respectively, as shown in Equations 52 and 53.

number

number

[0087] In the third implementation example, the sub-weight matrix W1 is a Zero Forcing (ZF) weight, and the initial value generator matrix W init is an identity matrix. The sub-weight matrix W1 may be an MMSE weight. In this case, the beamformer 143-2 included in the processor 14-2 in the DU 11-2 uses the matrix shown in Equation 49 as the sub-weight matrix W2.

[0088] In the fourth implementation example, the initial value generator matrix W init is the matrix shown in Equation 54.

number

[0089] The implementation of the initial value generator 1444 is not limited to these implementation examples.

[0090] 14 shows an example of the operation of the initial value generator 1444. In step 51, the initial value generator 1444 receives a received signal after receive beamforming from the beamformer 143-1 or 143-2. ​​In step 52, the initial value generator 1444 generates a soft replica and a soft replica power based on the received signal after receive beamforming. The soft replica and the soft replica power may be soft decision values ​​of the product of an initial value generating matrix and the received signal after receive beamforming. In step 53, the initial value generator 1444 supplies the generated soft replica and soft replica power to the soft interference canceller and belief generator, respectively, for the first iteration of the GaBP algorithm, which is one of the BP algorithms.

[0091] This embodiment can provide the following effects. First, in this embodiment, the initial value generator 1444 generates the soft replica x hat using the received signal after the receive beamforming. b,m (0) and the soft replica power p b,m (0) and performs belief generation for the first iteration. Since the DU only uses the received signal after receive beamforming, it is only necessary to transmit the received signal after receive beamforming from the RU to the DU, thereby reducing the required fronthaul bandwidth. Secondly, the initial value generator 1444 can generate accurate beliefs for the first iteration of the GaBP algorithm. Therefore, the base station 1 can perform highly accurate detection using the GaBP algorithm with a small number of iterations.

[0092] Furthermore, when the GaBP algorithm is used in this embodiment, the following effect can be obtained: The noise contained in the received signal supplied from the beamformer 143-1 or 143-2 to the soft interference cancellers 1441-1 to 1441-B is white noise, so that performance degradation of the GaBP algorithm can be avoided.

[0093] <Third embodiment> In the third embodiment, the base station 1 has a parameter learner 17 in addition to the configuration of the first or second embodiment. The parameter learner 17 may be included in the processor 14-2. The parameter learner 17 jointly learns the first parameter set (or the second parameter set) using deep unfolding and outputs the jointly learned parameter set. The output parameter set may be stored in the memory 15-2.

[0094] FIG. 15 is a conceptual diagram showing deep layer expansion for multi-user detection based on the GaBP algorithm, which is an example of a BP algorithm. Deep layer expansion is a technique in which an iterative algorithm is expanded in the iterative direction, the resulting processing flow graph is regarded as a deep neural network (DNN), and a deep learning scheme is applied. When the GaBP algorithm executed by the BP detector 144 is expanded in the iterative direction, the GaBP network shown in FIG. 15 is obtained. Each GaBP iteration corresponds to one layer of the DNN. This makes it possible to learn the parameters included in the GaBP network. Specifically, the learnable (trainable) parameters are the scaling coefficients a (t) and damping coefficient η (t) (or the scaling coefficient a for each iteration (each layer) (t) and a set of weighting factors {η i,t-k (t) For example, learning is performed based on the gradient method, and parameters are adjusted together in the direction that reduces the set cost.

[0095] FIG. 16 shows an example of the parameter learner 17. The learning data set 171 includes a transmit signal data set 1711 and a receive signal data set 1712 after receive beamforming. The transmit signal data set 1711 may be generated randomly. A receive signal data set received by multiple receive antennas corresponds to the transmit signal data set 1711 and is generated using the transmit signal data set 1711 and a given channel matrix. The channel matrix may be generated randomly or based on a propagation path model defined in the 3rd Generation Partnership Project (3GPP) specifications or the like. Alternatively, the channel matrix may be generated based on measurement results in an actual environment where the base station 1 is installed. The receive signal data set 1712 after receive beamforming is generated by performing receive beamforming on multiple receive signals received by multiple receive antennas using the same implementation as the receive beamforming used in the base station 1. As a specific implementation of receive beamforming, the method described in the first embodiment can be used.

[0096] The parameter learning system 172 includes a BP detector module 1721 and a learning module 1722. The BP detector module 1721 emulates the processor 14-2 or the BP detector 144 of the base station 1 described in the first or second embodiment. The BP detector module 1721 can execute a BP algorithm implemented in the base station 1. The learning module 1722 trains the BP detector module 1721 using the learning dataset 171. The learning module 1722 may apply one or more deep learning algorithms.

[0097] In one example, the learning module 1722 may use an update algorithm that follows a gradient method. The gradient method update algorithm used may be, for example, an Adaptive Moment Estimation (Adam) optimizer algorithm. In addition, the learning module 1722 may use mini-batch learning. The number of learning iterations may be set to an appropriate number taking into account overfitting to the training data. To update the learning rate, a step algorithm may be used that gradually narrows the update step relative to the number of learning iterations. The cost function may be the mean square error (MSE).

[0098] The learning module 1722 outputs a parameter set 173 that has been jointly trained through deep learning. The jointly trained parameter set 173 includes multiple scaling coefficients and multiple damping coefficients (or multiple scaling coefficients and multiple weighting coefficients). The output parameter set 173 may be stored in the memory 15-2 as the LUT 151, as in FIGS. 8 and 9.

[0099] Parameter learner 17 may be a computer system as shown in FIG. 17. FIG. 17 shows an example of the configuration of a computer system. The computer system executes a computer program including a set of instructions, thereby performing, for example, the method for parameter learner 17. Parameter learner 17 may be a standalone computer or may include one or more networked computers. The computer system may be a server or a client, or both, in a server-client environment. The computer system may be a personal computer, a tablet computer, or a smartphone.

[0100] In the example of FIG. 17 , the computer system includes one or more processors 61, memory 62, and mass storage 63, which communicate with each other via a bus 67. The one or more processors 61 may include, for example, a Central Processing Unit (CPU) or a Graphics Processing Unit (GPU), or both. The computer system may also include other devices, such as one or more output devices 64, one or more input devices 65, and one or more peripherals 66. The one or more output devices 64 include, for example, a video display and a speaker. The one or more input devices 65 include, for example, a keyboard, a mouse, a keypad, a touchpad, or a touchscreen, or any combination thereof. The one or more peripherals 66 include, for example, a printer, a modem, or a network adapter, or any combination thereof.

[0101] One or both of memory 62 and mass storage 63 may include computer-readable media having stored thereon one or more sets of instructions, which may be located partially or completely within processor memory 62. When executed in processor 61, these instructions cause processor 61 to perform, for example, the process shown in FIG.

[0102] FIG. 18 shows an example of the operation of the parameter learner 17. In this example, the GaBP algorithm is used as the BP algorithm. In step 71, the parameter learning system 172 receives a training data set 171. Here, the training data set 171 includes a transmit signal 1711 and a received signal 1712 after receive beamforming using beamforming weights implemented in the base station 1. In step 72, the parameter learning system 172 executes a GaBP algorithm using scaling coefficients and damping coefficients on the training data set 171 to train the GaBP algorithm using deep learning techniques. The trainable parameters are the scaling coefficients and damping coefficients. In step 73, the parameter learning system 172 stores the trained scaling coefficients and damping coefficients 173 in a memory (e.g., memory 15-2).

[0103] Alternatively, in step 72, the parameter learning system 172 may execute the GaBP algorithm using the scaling and weighting factors on the training data set 171. The trainable parameters are the scaling and weighting factors. In this case, in step 53, the parameter learning system 172 stores the trained scaling and weighting factors 173 in a memory (e.g., memory 15-2).

[0104] FIG. 19 shows a comparison of bit error rate (BER) performance between a GaBP detector after receive beamforming according to this embodiment and a conventional linear minimum mean square error (MMSE) detector and GaBP detector that do not perform receive beamforming. In the conventional methods, receive beamforming is not performed. These are simulation results for a multi-user MIMO configuration where M' is the number of terminals and N' is the number of receive antenna elements, and (N', M') = (64, 16). The total number of iterations (T) of the GaBP detectors in this embodiment and the conventional method is 32, and a Low-Density Parity-Check (LDPC) code with a coding rate of approximately 0.6 is used as the error correction code. Graph 81 shows the BER of a GaBP detector that uses a learned parameter set of scaling coefficients and weighting coefficients after receive beamforming using the product of the Q matrix and a unitary matrix described as the first implementation example of the first embodiment. Here, the unitary matrix is ​​a DFT matrix. Graph 82 is a comparative example, showing the BER of a GaBP detector using a parameter set of learned scaling coefficients and weighting coefficients when receive beamforming is not performed. Graph 83 is a comparative example, showing the BER of an MMSE detector when receive beamforming is not performed. This confirms that the GaBP detector of this embodiment can reduce the required fronthaul bandwidth without degradation compared to GaBP detectors using existing methods. Furthermore, the GaBP detector of this embodiment has a BER of 10 -3 An improvement of more than 5 dB is observed.

[0105] In this embodiment, the first parameter set or the second parameter set is learned by deep deployment using the received signal after receive beamforming, so that multiuser detection can be performed by the BP detector 144 using suboptimal parameter values ​​that take into account the receive beamforming and the characteristics of the BP detector 144. This makes it possible to improve the detection characteristics of the BP detector 144.

[0106] <Other embodiments> As described above, in some implementations, the processors 14-1 and 14-2 of the base station 1 execute one or more programs containing instructions for causing a computer to perform the algorithms described in the present embodiment. Additionally, the parameter learner 17 executes one or more programs containing instructions for causing a computer to perform deep learning. The programs contain instructions (or software code) that, when loaded into a computer, cause the computer to perform one or more functions described in the present embodiment. The programs may be stored on a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disk (DVD), Blu-ray® disc or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage, or other magnetic storage device. The programs may also be transmitted on a transitory computer-readable medium or a communication medium. By way of example, and not limitation, transitory computer-readable or communication media include electrical, optical, acoustic, or other forms of propagated signals.

[0107] The parameter sets (e.g., at least one of the scaling coefficients, damping coefficients, and weighting coefficients) used in the GaBP algorithm in the first and second embodiments may not be obtained by deep learning. In some implementations, the scaling coefficients for each iteration may be the same value for all iterations. Additionally or alternatively, the damping coefficients for each iteration may be the same value for all iterations. Additionally or alternatively, the set of weighting coefficients for each iteration may follow a periodic selection pattern of a subset of receive antennas. In this case, the possible values ​​of the weighting coefficients may be 0 or 1.

[0108] The above-described embodiments are merely examples of the application of the technical ideas obtained by the inventors of the present invention; in other words, the technical ideas are not limited to the above-described embodiments, and various modifications are possible.

[0109] For example, some or all of the above embodiments may be described as, but are not limited to, the following supplementary notes.

[0110] (Appendix 1) A wireless receiving device, at least one beamforming weight generator configured to generate the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix, each column of which includes two or more non-zero matrix elements, or to generate a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; a first beamformer configured to perform receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; a Belief Propagation (BP) detector configured to perform a Belief Propagation (BP) algorithm using signals output from the first beamformer and transmitted via a fronthaul; A wireless receiving device comprising: (Appendix 2) a Radio Unit (RU) and a Distributed Unit (DU) connected via the fronthaul; the first beamformer is included in the RU; the BP detector is included in the DU; the at least one beamforming weight generator is included in at least one of the RU and the DU; 2. The wireless receiving device according to claim 1. (Appendix 3) The receive beamforming includes multiplying a Hermitian conjugate of the weight matrix W or a Hermitian conjugate of the sub-weight matrix W1 by a receive signal vector formed of received signals of the N' receive antennas. 3. The wireless receiving device according to claim 1 or 2. (Appendix 4) further comprising a second beamformer; the at least one beamforming weight generator is configured to generate a combination of the sub-weight matrix W1 and a B'×B' sub-weight matrix W2, the product of which is equal to the weight matrix W, and to provide the sub-weight matrix W1 to the first beamformer and the sub-weight matrix W2 to the second beamformer; the first beamformer is configured to perform receive beamforming by taking a product of a Hermitian conjugate of the sub-weight matrix W1 and a receive signal vector formed of receive signals of the N' receive antennas; the second beamformer is configured to perform receive beamforming by multiplying a Hermitian conjugate of the sub-weight matrix W2 by a signal output from the first beamformer and transmitted via the fronthaul, and to provide the beamformed signal to the BP detector; the BP detector is configured to perform the BP algorithm on signals provided by the second beamformer; 4. The wireless receiving device according to claim 1. (Appendix 5) The at least one beamforming weight generator converts the N'×M' channel matrix into K N' k ×M' submatrix, and k The smaller value of M' is B' k Each submatrix is ​​orthogonal to each column of N' k ×B' k matrix W k and B' with two or more non-zero matrix elements in each columnk ×M' matrix, and the matrix W k as diagonal components, as the weight matrix W. 5. The wireless receiving device according to any one of Supplementary notes 1 to 4. (Appendix 6) the at least one beamforming weight generator is configured to output, as the weight matrix W, a product of an N'×M' Q matrix obtained by QR decomposing the N'×M' channel matrix and an M'-th order unitary matrix, or to output the Q matrix as the sub-weight matrix W1 and the M'-th order unitary matrix as the sub-weight matrix W2. 5. The wireless receiving device according to any one of Supplementary notes 1 to 4. (Appendix 7) the at least one beamforming weight generator is configured to output, as the weight matrix W, an N'×M' matrix consisting of left singular vectors obtained by singular value decomposition of the N'×M' channel matrix, or a product of the N'×M' matrix and an M'-th order unitary matrix, or to output the N'×M' matrix as the sub-weight matrix W1 and the M'-th order unitary matrix as the sub-weight matrix W2. 5. The wireless receiving device according to any one of Supplementary notes 1 to 4. (Appendix 8) The at least one beamforming weight generator (a) outputting, as the weight matrix W, a product of the N'×M' channel matrix and the −½ power of a Gram matrix of the N'×M' channel matrix, or a product of the N'×M' channel matrix, the −½ power of a Gram matrix of the N'×M' channel matrix, and an M'-th order unitary matrix; (b) outputting the channel matrix as the sub-weight matrix W1 and outputting the −½ power of the Gram matrix as the sub-weight matrix W2; (c) outputting the channel matrix as the sub-weight matrix W1 and outputting the product of the −½ power of the Gram matrix and an M′-th order unitary matrix as the sub-weight matrix W2; or (d) outputting the product of the channel matrix and the −½ power of the Gram matrix as the sub-weight matrix W1 and outputting an M′-th order unitary matrix as the sub-weight matrix W2; 5. The wireless receiving device according to any one of Supplementary notes 1 to 4. (Appendix 9) further comprising at least one memory configured to store a first parameter set including a plurality of scaling coefficients and a plurality of damping coefficients or a second parameter set including a plurality of scaling coefficients and a plurality of weighting coefficients; the BP detector is configured to perform a Gaussian Belief Propagation (GaBP) algorithm using the first parameter set or the second parameter set; The BP detector comprises: a soft interference canceller configured to subtract all components of the transmission signals except for the m'th transmission signal component from an mth reception signal among a plurality of reception signals using replicas of all transmission signals except for the m'th transmission signal generated in the t-1th iteration, thereby generating a canceled mth reception signal; a belief generator configured to generate a belief associated with the m-th received signal based at least on the damping coefficient or the weighting coefficient and the m-th received signal after the cancellation; a soft replica generator configured to generate a replica of the m' transmitted signal at the t iteration based at least on the scaling factor and the belief; 9. The wireless receiving device according to any one of Supplementary notes 1 to 8. (Appendix 10) At least one of the first parameter set and the second parameter set includes parameters jointly learned using deep learning techniques based on receive beamformed received signals and transmit signals. 10. The wireless receiving device according to claim 9. (Appendix 11) the BP detector includes an initial value generator; the initial value generator is configured to generate soft decision values ​​for a product of a Hermitian conjugate of an initial value generator matrix of B'×M' and a received signal after receive beamforming supplied from the first beamformer as a soft replica and a soft replica power, and to supply the soft replica and the soft replica power to a first iteration of the BP algorithm executed in the BP detector; 11. The wireless receiving device according to any one of Supplementary notes 1 to 10. (Appendix 12) The initial value generating matrix is ​​the equivalent channel matrix H' after receiving beamforming. c and the matrix H' c and the inverse matrix of the product of the matrix H' c is the product of Hermitian conjugates of 12. The wireless receiving device according to claim 11. (Appendix 13) the sub-weight matrix W1 is an N'×M' matrix consisting of left singular vectors obtained by singular value decomposition of the N'×M' channel matrix, The sub-weight matrix W2 is an arbitrary M'th-order unitary matrix, the initial value generating matrix is ​​a product of an M'-th order diagonal matrix whose diagonal elements are the reciprocals of the singular values ​​obtained by the singular value decomposition, and a Hermitian conjugate of an M'-th order square matrix made up of right singular vectors; 12. The wireless receiving device according to claim 11, which directly or indirectly cites claim 4. (Appendix 14) the sub-weight matrix W1 is the N'×M' channel matrix, the sub-weight matrix W2 is the −½ power of the product of the Hermitian conjugate of the channel matrix and the channel matrix, the initial value generator matrix is ​​an inverse matrix of a product of the Hermitian conjugate of the channel matrix and the channel matrix; 12. The wireless receiving device according to claim 11, which directly or indirectly cites claim 4. (Appendix 15) A method performed by a wireless receiving device, comprising: generating the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; performing first receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; and performing a Belief Propagation (BP) algorithm using the first receive beamformed signal transmitted over the fronthaul; A method for providing the above. (Appendix 16) performing the first receive beamforming includes performing the first receive beamforming in a Radio Unit (RU); executing the BP algorithm includes executing the BP algorithm in a Distributed Unit (DU) coupled to the RU via the fronthaul. The method described in Appendix 15. (Appendix 17) the first receive beamforming includes taking a product of a Hermitian conjugate of the weight matrix W or a Hermitian conjugate of the sub-weight matrix W1 and a receive signal vector formed of receive signals of the N' receive antennas; 17. The method according to claim 15 or 16. (Appendix 18) performing a second receive beamforming; generating a combination of the sub-weight matrix W1 and a B'×B' sub-weight matrix W2, the product of which is equal to the weight matrix W; performing the first receive beamforming includes multiplying a Hermitian conjugate of the sub-weight matrix W1 by a receive signal vector formed of receive signals from the N' receive antennas; performing the second receive beamforming includes multiplying a Hermitian conjugate of the sub-weight matrix W2 by a signal vector generated by the first receive beamforming; executing the BP algorithm includes executing the BP algorithm on a signal obtained by the second receive beamforming. The method according to any one of appendices 15 to 17. (Appendix 19) The generating step converts the N'×M' channel matrix into K N' k ×M' submatrix, and k The smaller of M' and B' k Each submatrix is ​​orthogonal to each column of N' k ×B' k matrix W k and B' with two or more non-zero matrix elements in each column k ×M' matrix, and the matrix W k as diagonal components of a block diagonal matrix as the weight matrix W. 19. The method according to any one of appendices 15 to 18. (Appendix 20) The generating includes outputting, as the weight matrix W, a product of an N'×M' Q matrix obtained by QR decomposing the N'×M' channel matrix and an M'-th order unitary matrix, or outputting the Q matrix as the sub-weight matrix W1 and the M'-th order unitary matrix as the sub-weight matrix W2. 19. The method according to any one of appendices 15 to 18. (Appendix 21) The generating includes outputting, as the weight matrix W, an N'×M' matrix consisting of left singular vectors obtained by singular value decomposition of the N'×M' channel matrix, or a product of the N'×M' matrix and an M'-th order unitary matrix, or outputting the N'×M' matrix as the sub-weight matrix W1 and the M'-th order unitary matrix as the sub-weight matrix W2. 19. The method according to any one of appendices 15 to 18. (Appendix 22) The generating step comprises: (a) outputting, as the weight matrix W, a product of the N'×M' channel matrix and the −½ power of the Gram matrix of the N'×M' channel matrix; (b) outputting the channel matrix as the sub-weight matrix W1 and the −½ power of the Gram matrix as the sub-weight matrix W2; (c) outputting the channel matrix as the sub-weight matrix W1 and outputting the product of the −½ power of the Gram matrix and an M′-th order unitary matrix as the sub-weight matrix W2; or (d) outputting the product of the channel matrix and the −½ power of the Gram matrix as the sub-weight matrix W1 and outputting an M′-th order unitary matrix as the sub-weight matrix W2; 19. The method according to any one of appendices 15 to 18. (Appendix 23) Executing the BP algorithm includes: Reading a first parameter set including a plurality of scaling coefficients and a plurality of damping coefficients or a second parameter set including a plurality of scaling coefficients and a plurality of weighting coefficients from a memory; and executing a GaBP algorithm using the first parameter set or the second parameter set; The method according to any one of appendices 15 to 22. (Appendix 24) At least one of the first parameter set and the second parameter set includes parameters jointly learned using deep learning techniques based on receive beamformed received signals and transmit signals. 24. The method described in Appendix 23. (Appendix 25) generating soft decision values ​​for a product of a Hermitian conjugate of an initial value generator matrix of B'×M' and the received signal after the received beamforming as a soft replica and a soft replica power; The method further includes supplying the generated soft replica and soft replica power to a first iteration of the BP algorithm. The method according to any one of appendices 15 to 24. (Appendix 26) The initial value generating matrix is ​​the equivalent channel matrix H' after receiving beamforming. c and the matrix H' c and the inverse matrix of the product of the matrix H' c is the product of Hermitian conjugates of The method described in Appendix 25. (Appendix 27) the sub-weight matrix W1 is an N'×M' matrix consisting of left singular vectors obtained by singular value decomposition of the N'×M' channel matrix, The sub-weight matrix W2 is an arbitrary M'th-order unitary matrix, the initial value generating matrix is ​​a product of an M'-th diagonal matrix whose diagonal elements are the reciprocals of the singular values ​​obtained by the singular value decomposition and a Hermitian conjugate of an M'-th square matrix consisting of right singular vectors; 25. The method of claim 18, which directly or indirectly cites claim 18. (Appendix 28) the sub-weight matrix W1 is the N'×M' channel matrix, the sub-weight matrix W2 is the −½ power of the product of the Hermitian conjugate of the channel matrix and the channel matrix, the initial value generator matrix is ​​an inverse matrix of a product of the Hermitian conjugate of the channel matrix and the channel matrix; 25. The method of claim 18, which directly or indirectly cites claim 18. (Appendix 29) A program for causing a computer to perform a method for a wireless receiving device, the method comprising: generating the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; performing first receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; and performing a Belief Propagation (BP) algorithm using the first receive beamformed signal transmitted over the fronthaul; A program that includes: (Appendix 30) A Radio Unit (RU) included in a wireless receiving device and coupled to a Distributed Unit (DU) of the wireless receiving device via a fronthaul, at least one beamforming weight generator configured to generate the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix, each column of which includes two or more non-zero matrix elements, or to generate a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; a first beamformer configured to perform receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1, and to provide the beamformed signals to the DU via the fronthaul; RU equipped with. (Appendix 31) the beamformed signals are used by the DU to perform a Belief Propagation (BP) algorithm using the beamformed signals; RU as described in Appendix 30. (Appendix 32) The receive beamforming includes multiplying a Hermitian conjugate of the weight matrix W or a Hermitian conjugate of the sub-weight matrix W1 by a receive signal vector formed of received signals of the N' receive antennas. RU as described in Appendix 30 or 31. (Appendix 33) 1. A method performed by a Radio Unit (RU) included in a wireless receiving device and coupled to a Distributed Unit (DU) of the wireless receiving device via a fronthaul, comprising: generating the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; and providing a beamformed signal to the DU via the fronthaul; A method for providing the above. (Appendix 34) A program for causing a computer to perform a method for a Radio Unit (RU) included in a wireless receiving device and coupled to a Distributed Unit (DU) of the wireless receiving device via a fronthaul, the method comprising: generating the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; and providing a beamformed signal to the DU via the fronthaul; A program that includes: (Appendix 35) A Distributed Unit (DU) included in a wireless receiving device and coupled to a Radio Unit (RU) of the wireless receiving device via a fronthaul, at least one beamforming weight generator configured to generate the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix, each column of which includes two or more non-zero matrix elements, or to generate a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; a Belief Propagation (BP) detector configured to perform a Belief Propagation (BP) algorithm using signals generated by the RU and transmitted via the fronthaul by performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W; DU equipped with. (Appendix 36) 1. A method performed by a Distributed Unit (DU) included in a wireless receiving device and coupled to a Radio Unit (RU) of the wireless receiving device via a fronthaul, comprising: generating the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; providing the weight matrix W or the sub-weight matrix W1 to the RU; and performing a Belief Propagation (BP) algorithm using signals generated by the RU and transmitted via the fronthaul by performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; A method for providing (Appendix 37) A program causing a computer to execute a method for a Distributed Unit (DU) included in a wireless receiving device and coupled to a Radio Unit (RU) of the wireless receiving device via a fronthaul, the method comprising: generating the weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating a combination of an N'×B' sub-weight matrix W1 and at least one second sub-weight matrix, the product of which is equal to the weight matrix W; providing the weight matrix W or the sub-weight matrix W1 to the RU; and performing a Belief Propagation (BP) algorithm using signals generated by the RU and transmitted via the fronthaul by performing receive beamforming on received signals of the N' receive antennas using the weight matrix W or the sub-weight matrix W1; A program that includes: [Explanation of symbols]

[0111] 1 base station 2. Wireless terminals 10 Receivers 11-1 Radio Unit (RU) 11-2 Distributed Unit (DU) 12 Antenna Array 13 RF Transceiver 14 processors 15 memory 16 Network Interfaces 141 Channel Estimator 142 Beamforming Weight Generator 143-1 Beamformer in RU 143-2 Beamformer in DU 144 Belief Propagation (BP) detector 145 Decision and Demodulator 151 Lookup Tables 1441-m soft interference canceller 1442 Belief Generator 1443-m Soft Replica Generator 1444 Initial Value Generator 17 Parameter Learner 171 training datasets 1711 Transmitted Signal 1712 Received Signal 172 Parameter Learning System 1721 BP detector module 1722 Learning Modules 173 Learned parameter set 20 Transmitter 30 channels (transmission path)

Claims

1. A wireless receiving device, The weight matrix W is generated by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; and The weight matrix W or the sub-weight matrix W 1 a first beamformer configured to perform receive beamforming on received signals of the N' receive antennas using a Belief Propagation (BP) detector configured to perform a Belief Propagation (BP) algorithm using signals output from the first beamformer and transmitted via a fronthaul; Equipped with the B'×M' matrix is ​​a channel matrix after receive beamforming, and is used in detecting transmitted symbols using the BP algorithm; the at least one second sub-weight matrix is ​​used for receive beamforming in a second beamformer for signals transmitted via the fronthaul when the first beamformer uses the sub-weight matrix W 1 ; Radio receiving device.

2. a Radio Unit (RU) and a Distributed Unit (DU) connected via the fronthaul; the first beamformer is included in the RU; the BP detector is included in the DU; the at least one beamforming weight generator is included in at least one of the RU and the DU; 2. The radio receiving device according to claim 1.

3. The receive beamforming is performed by using the Hermitian conjugate of the weight matrix W or the sub-weight matrix W 1 and a received signal vector formed by the received signals of the N' receive antennas, 3. The radio receiving device according to claim 1 or 2.

4. Further comprising the second beamformer, The at least one beamforming weight generator generates the sub-weight matrix W whose product is equal to the weight matrix W. 1 and the B'×B' sub-weight matrix W 2 and generate a combination of the sub-weight matrix W 1 to the first beamformer and the sub-weight matrix W2 to the second beamformer; the sub-weight matrix W 2 corresponds to the at least one second sub-weight matrix; The first beamformer uses the sub-weight matrix W 1 and a received signal vector formed by the received signals of the N' receiving antennas, the second beamformer is configured to perform receive beamforming by taking a product of a Hermitian conjugate of the sub-weight matrix W2 and a signal output from the first beamformer and transmitted via the fronthaul, and to provide the beamformed signal to the BP detector; the BP detector is configured to perform the BP algorithm on signals provided by the second beamformer.

4. The radio receiving device according to claim 1.

5. The at least one beamforming weight generator converts the N'×M' channel matrix into K N' k ×M' submatrix, and k The smaller value of M' is B' k Each submatrix is ​​orthogonal to each column of N' k ×B' k matrix W k and B' with two or more non-zero matrix elements in each column k ×M' matrix, and the matrix W k as diagonal components, as the weight matrix W. The radio receiving device according to any one of claims 1 to 4.

6. The at least one beamforming weight generator outputs a product of an N'×M' Q matrix obtained by QR decomposing the N'×M' channel matrix and an M'-th order unitary matrix as the weight matrix W, or outputs the Q matrix as the sub-weight matrix W 1 and convert the M'-th order unitary matrix into a sub-weight matrix W 2 and configured to output as the sub-weight matrix W 2 corresponds to the at least one second sub-weight matrix; The radio receiving device according to any one of claims 1 to 4.

7. The at least one beamforming weight generator outputs an N'×M' matrix consisting of left singular vectors obtained by singular value decomposition of the N'×M' channel matrix, or a product of the N'×M' matrix and an M'-th order unitary matrix, as the weight matrix W, or outputs the N'×M' matrix as the sub-weight matrix W 1 and convert the M'-th order unitary matrix into a sub-weight matrix W 2 and configured to output as the sub-weight matrix W 2 corresponds to the at least one second sub-weight matrix; The radio receiving device according to any one of claims 1 to 4.

8. The at least one beamforming weight generator (a) outputting the product of the N'×M' channel matrix and the −½ power of the Gram matrix of the N'×M' channel matrix as the weight matrix W; (b) converting the channel matrix into the sub-weight matrix W 1 and the Gram matrix is ​​converted to the sub-weight matrix W 2 output as, (c) converting the channel matrix into the sub-weight matrix W 1 and the product of the −1 / 2 power of the Gram matrix and the M′-th order unitary matrix is ​​output as a sub-weight matrix W 2 or (d) The product of the channel matrix and the −½ power of the Gram matrix is ​​calculated as the sub-weight matrix W 1 and the M'th order unitary matrix is ​​output as the sub-weight matrix W 2 It is configured to output as the sub-weight matrix W 2 corresponds to the at least one second sub-weight matrix; The radio receiving device according to any one of claims 1 to 4.

9. further comprising at least one memory configured to store a first parameter set including a plurality of scaling coefficients and a plurality of damping coefficients or a second parameter set including a plurality of scaling coefficients and a plurality of weighting coefficients; the BP detector is configured to perform a Gaussian Belief Propagation (GaBP) algorithm using the first parameter set or the second parameter set; The BP detector comprises: a soft interference canceller configured to subtract all components of the transmission signals except for the m'th transmission signal component from an mth reception signal among a plurality of reception signals using replicas of all transmission signals except for the m'th transmission signal generated in the t-1th iteration, thereby generating a canceled mth reception signal; a belief generator configured to generate a belief associated with the m-th received signal based at least on the damping coefficient or the weighting coefficient and the m-th received signal after the cancellation; a soft replica generator configured to generate a replica of the m' transmitted signal at the t iteration based at least on the scaling factor and the belief; The radio receiving device according to any one of claims 1 to 8.

10. At least one of the first parameter set and the second parameter set includes parameters jointly learned using deep learning techniques based on receive beamformed received signals and transmit signals.

10. The radio receiving device according to claim 9.

11. the BP detector includes an initial value generator; the initial value generator is configured to generate soft decision values ​​for a product of a Hermitian conjugate of an initial value generator matrix of B'×M' and a received signal after receive beamforming supplied from the first beamformer as a soft replica and a soft replica power, and to supply the soft replica and the soft replica power to a first iteration of the BP algorithm executed in the BP detector. The radio receiving device according to any one of claims 1 to 10.

12. The initial value generating matrix is ​​the equivalent channel matrix H' after receiving beamforming. c and the matrix H' c and the inverse matrix of the product of the matrix H' c is the product of Hermitian conjugates of 12. The wireless receiving device according to claim 11.

13. The sub-weight matrix W 1 is an N'×M' matrix consisting of left singular vectors obtained by singular value decomposition of the N'×M' channel matrix, The sub-weight matrix W 2 is any M'th order unitary matrix, the initial value generating matrix is ​​a product of an M'-th order diagonal matrix whose diagonal elements are the reciprocals of the singular values ​​obtained by the singular value decomposition, and a Hermitian conjugate of an M'-th order square matrix made up of right singular vectors; 12. The radio receiving device according to claim 11, which directly or indirectly derives from claim 4.

14. The sub-weight matrix W 1 is the N'×M' channel matrix, The sub-weight matrix W 2 is the −½ power of the product of the Hermitian conjugate of the channel matrix and the channel matrix, the initial value generator matrix is ​​an inverse matrix of a product of the Hermitian conjugate of the channel matrix and the channel matrix; 12. The radio receiving device according to claim 11, which directly or indirectly derives from claim 4.

15. A method performed by a wireless receiving device, comprising: Decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, to generate the weight matrix W, or generating an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; The weight matrix W or the sub-weight matrix W 1 performing first receive beamforming on the received signals of the N' receive antennas using performing a Belief Propagation (BP) algorithm using the first receive beamformed signal transmitted over the fronthaul; Equipped with the B'×M' matrix is ​​a channel matrix after receive beamforming, and is used in detecting transmitted symbols using the BP algorithm; the at least one second sub-weight matrix is ​​used for second receive beamforming for signals transmitted via the fronthaul when the first receive beamforming uses the sub-weight matrix W 1 ; method.

16. A program for causing a computer to perform a method for a wireless receiving device, the method comprising: Decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, to generate the weight matrix W, or generating an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; The weight matrix W or the sub-weight matrix W 1 performing first receive beamforming on the received signals of the N' receive antennas using performing a Belief Propagation (BP) algorithm using the first receive beamformed signal transmitted over the fronthaul; Equipped with the B'×M' matrix is ​​a channel matrix after receive beamforming, and is used in detecting transmitted symbols using the BP algorithm; the at least one second sub-weight matrix is ​​used for second receive beamforming for signals transmitted via the fronthaul when the first receive beamforming uses the sub-weight matrix W 1 ; program.

17. A Radio Unit (RU) included in a wireless receiving device and coupled to a Distributed Unit (DU) of the wireless receiving device via a fronthaul, The weight matrix W is generated by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; and The weight matrix W or the sub-weight matrix W 1 a first beamformer configured to perform receive beamforming on received signals of the N' receive antennas using Equipped with The B'×M' matrix is ​​a channel matrix after receiving beamforming, and is used in detecting transmitted symbols using a Belief Propagation (BP) algorithm in the DU; The at least one second sub-weight matrix is ​​used for receive beamforming of a signal transmitted via the fronthaul in a second beamformer arranged in the DU when the first beamformer uses the sub-weight matrix W 1. R.U.

18. 1. A method performed by a Radio Unit (RU) included in a wireless receiving device and coupled to a Distributed Unit (DU) of the wireless receiving device via a fronthaul, comprising: generating a weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W; 1 and at least one second sub-weight matrix; The weight matrix W or the sub-weight matrix W 1 performing first receive beamforming on the received signals of the N' receive antennas using providing a beamformed signal to the DU via the fronthaul; Equipped with The B'×M' matrix is ​​a channel matrix after receiving beamforming, and is used in detecting transmitted symbols using a Belief Propagation (BP) algorithm in the DU; The at least one second sub-weight matrix is ​​used for second receive beamforming in the DU for signals transmitted via the fronthaul when the first receive beamforming uses the sub-weight matrix W 1 . method.

19. A program for causing a computer to perform a method for a Radio Unit (RU) included in a wireless receiving device and coupled to a Distributed Unit (DU) of the wireless receiving device via a fronthaul, the method comprising: generating a weight matrix W by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or generating an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W; 1 and at least one second sub-weight matrix; The weight matrix W or the sub-weight matrix W 1 performing first receive beamforming on the received signals of the N' receive antennas using providing a beamformed signal to the DU via the fronthaul; Equipped with The B'×M' matrix is ​​a channel matrix after receiving beamforming, and is used in detecting transmitted symbols using a Belief Propagation (BP) algorithm in the DU; The at least one second sub-weight matrix is ​​used for second receive beamforming in the DU for signals transmitted via the fronthaul when the first receive beamforming uses the sub-weight matrix W 1 . program.

20. A Distributed Unit (DU) included in a wireless receiving device and coupled to a Radio Unit (RU) of the wireless receiving device via a fronthaul, The weight matrix W is generated by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; and The weight matrix W or the sub-weight matrix W 1 a BP detector configured to perform a Belief Propagation (BP) algorithm using signals generated by the RU and transmitted via the fronthaul by performing first receive beamforming on received signals of the N′ receive antennas using Equipped with the B'×M' matrix is ​​a channel matrix after receive beamforming, and is used in detecting transmitted symbols using the BP algorithm; The at least one second sub-weight matrix is ​​used for second receive beamforming in the DU for signals transmitted via the fronthaul when the first receive beamforming in the RU uses the sub-weight matrix W 1 . DU.

21. 1. A method performed by a Distributed Unit (DU) included in a wireless receiving device and coupled to a Radio Unit (RU) of the wireless receiving device via a fronthaul, comprising: The weight matrix W is generated by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; The weight matrix W or the sub-weight matrix W 1 providing the RU with The weight matrix W or the sub-weight matrix W 1 performing a Belief Propagation (BP) algorithm using signals generated by the RU and transmitted via the fronthaul by performing a first receive beamforming on received signals of the N′ receive antennas using Equipped with the B'×M' matrix is ​​a channel matrix after receive beamforming, and is used in detecting transmitted symbols using the BP algorithm; The at least one second sub-weight matrix is ​​used for second receive beamforming in the DU for signals transmitted via the fronthaul when the first receive beamforming in the RU uses the sub-weight matrix W 1 . method.

22. A program causing a computer to execute a method for a Distributed Unit (DU) included in a wireless receiving device and coupled to a Radio Unit (RU) of the wireless receiving device via a fronthaul, the method comprising: The weight matrix W is generated by decomposing an estimated N'×M' channel matrix defined between M' transmitting antennas of one or more wireless transmitting devices and N' receiving antennas coupled to the wireless receiving device into an N'×B' weight matrix W, each column of which is orthogonal, where B' is an integer between N'-1 and M', and a B'×M' matrix containing two or more non-zero matrix elements in each column, or an N'×B' sub-weight matrix W, the product of which is equal to the weight matrix W. 1 and at least one second sub-weight matrix; The weight matrix W or the sub-weight matrix W 1 providing the RU with The weight matrix W or the sub-weight matrix W 1 performing a Belief Propagation (BP) algorithm using signals generated by the RU and transmitted via the fronthaul by performing a first receive beamforming on received signals of the N′ receive antennas using Equipped with the B'×M' matrix is ​​a channel matrix after receive beamforming, and is used in detecting transmitted symbols using the BP algorithm; The at least one second sub-weight matrix is ​​used for second receive beamforming in the DU for signals transmitted via the fronthaul when the first receive beamforming in the RU uses the sub-weight matrix W 1 . program.

Images