A MIMO detection method and system based on a memory-computing integrated device, a processing device and a storage medium

Abstract

The application relates to a MIMO detection method and system based on a memory-computing integrated device, a processing device and a storage medium, which comprises the following steps: acquiring a channel matrix and a probability graph model of a to-be-detected MIMO detection task, and determining the initial probability of the state value of each variable node in the probability graph model according to a multi-antenna mixed signal observed by a receiving end and noise variance; constructing a node state matrix corresponding to each undirected edge in the probability graph model; constructing a driving vector corresponding to each undirected edge in the probability graph model; performing memory-computing integrated device configuration based on the node state matrix corresponding to each undirected edge to obtain a memory-computing integrated device subarray corresponding to each undirected edge; and determining the optimal estimation of a transmission symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task according to the memory-computing integrated device subarray corresponding to each undirected edge, the constructed driving vector and the confidence of message transmission between each variable node and each check node connected through the undirected edge. The application can be widely used in the technical field of signal processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal processing, in particular to a detection method and system based on a memory-computing integrated MIMO (Multiple-Input Multiple-Output), a processing device and a storage medium. BACKGROUND

[0002] In the early stage of commercial deployment of the fifth generation mobile communication technology (5G), the high power consumption and short standby time of terminal devices quickly become prominent, becoming one of the key bottlenecks restricting the improvement of user experience, hindering the large-scale popularization of 5G, and limiting its application in battery-sensitive scenarios such as the Internet of Things (IoT) and wearable devices. Users generally feedback that the battery endurance of 5G terminal devices has decreased significantly compared to mature 4G (LTE) terminals, and the devices are more prone to heat, which not only reduces user satisfaction, but also increases the engineering complexity and cost of device heat dissipation. The fundamental reason for the significantly higher power consumption of 5G terminals than 4G terminals is mainly due to the key technical characteristics adopted to achieve revolutionary performance improvement: 1) Bandwidth increases dramatically: The 5G channel bandwidth is significantly higher than that of 4G, with a typical bandwidth increase of 5 times or even more. A wider bandwidth means that the amount of data that needs to be processed by the terminal baseband processing unit in a unit of time increases geometrically, directly leading to a significant increase in baseband processing power consumption. 2) The number of antennas increases dramatically: To achieve high speed, high reliability and large-scale connection, 5G widely adopts MIMO detection methods, and terminal devices are generally equipped with 4 times or even more antennas, which means more RF front-end components and more complex antenna tuning and beam management circuits, resulting in a significant increase in RF front-end power consumption. Since the 5G bandwidth has increased by 5 times and the number of antennas has increased by 4 times, the communication power consumption has increased by 50% to 100%.

[0003] At present, the mainstream technical path in the industry to cope with the performance requirements of 5G is largely limited by the performance boundaries drawn by the classical information theory framework. In order to approach these theoretical limits while meeting the performance indicators of actual deployment, existing communication system design often tends to adopt a "stacked" optimization strategy, trading off performance gains by increasing algorithm complexity. And due to hardware conditions, small sensors and ocean buoys still face serious energy supply shortages.

[0004] Therefore, energy supply shortage and huge energy consumption make information communication networks face energy consumption constraints, and it is urgent to break through the constraints of classical communication theory, break the development mode of technology stacking, break through the power consumption bottleneck of digital circuits, and achieve a cross-order improvement in the energy efficiency of terminal baseband signal processing. SUMMARY

[0005] To solve the above problems, the application aims to provide a MIMO detection method, system, processing device and storage medium based on a memory-compute integrator, which can realize a cross-order improvement in the energy efficiency of terminal baseband signal processing.

[0006] To achieve the above object, the application adopts the following technical scheme: in a first aspect, a MIMO detection method based on a memory-compute integrator is provided, comprising:

[0007] obtaining a channel matrix and a probabilistic graph model of a to-be-detected MIMO detection task, and determining the initial probability of the state value of each variable node in the probabilistic graph model according to the multi-antenna mixed signal observed by the receiving end and the noise variance;

[0008] for any check node in the probabilistic graph model, constructing a node state matrix corresponding to each undirected edge in the probabilistic graph model according to each target undirected edge between the check node and each target variable node connected thereto, the channel matrix of the to-be-detected MIMO detection task, and the multi-antenna mixed signal observed by the receiving end;

[0009] constructing a driving vector corresponding to each undirected edge in the probabilistic graph model according to the confidence of each variable node in the probabilistic graph model in transmitting its state value to the check node, wherein in the initial stage, the initial probability of the state value of each variable node in the probabilistic graph model is taken as the confidence of the variable node in transmitting its state value to each check node connected through the undirected edge;

[0010] configuring a memory-compute integrator based on the node state matrix corresponding to each undirected edge to obtain a memory-compute integrator subarray corresponding to each undirected edge;

[0011] determining the confidence of each check node in the probabilistic graph model in transmitting its state value to the variable node according to the memory-compute integrator subarray corresponding to each undirected edge and the driving vector constructed, and further updating the confidence of each variable node in transmitting its state value to the check node until a set iteration stopping condition is met, to determine the optimal estimation value of the transmission symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task.

[0012] In a second aspect, a MIMO detection system based on a memory-compute integrator is provided, comprising:

[0013] an obtaining unit configured to obtain a channel matrix and a probabilistic graph model of a to-be-detected MIMO detection task, and determine the initial probability of the state value of each variable node in the probabilistic graph model according to the multi-antenna mixed signal observed by the receiving end and the noise variance;

[0014] A node state matrix construction unit is configured to, for any check node in the probabilistic graph model, construct a node state matrix corresponding to each undirected edge in the probabilistic graph model according to each target undirected edge between the check node and each target variable node connected to the check node, a channel matrix of the MIMO detection task to be measured, and a multi-antenna mixed signal observed at the receiving end.

[0015] A driving vector construction unit is configured to construct a driving vector corresponding to each undirected edge in the probabilistic graph model according to a confidence level at which each variable node in the probabilistic graph model transmits a state value of the variable node to a check node connected to the variable node via an undirected edge.

[0016] A configuration unit is configured to configure a memory-compute integrator based on the node state matrix corresponding to each undirected edge to obtain a memory-compute integrator subarray corresponding to each undirected edge.

[0017] An optimal estimation value determination unit is configured to determine a confidence level at which each check node in the probabilistic graph model transmits a state value of the check node to a variable node connected to the check node via an undirected edge according to the memory-compute integrator subarray corresponding to each undirected edge and the constructed driving vector, and further update the confidence level at which each variable node transmits a state value of the variable node to a check node connected to the variable node via an undirected edge until a set iteration stop condition is met, to determine an optimal estimation value of a transmitted symbol corresponding to the multi-antenna mixed signal in the MIMO detection task to be measured.

[0018] In a third aspect, a processing device is provided, which includes computer program instructions, and when the computer program instructions are executed by the processing device, the computer program instructions are configured to implement steps corresponding to the above-mentioned memory-compute integrator-based MIMO detection method.

[0019] In a fourth aspect, a computer readable storage medium is provided, which stores computer program instructions, and when the computer program instructions are executed by a processor, the computer program instructions are configured to implement steps corresponding to the above-mentioned memory-compute integrator-based MIMO detection method.

[0020] The present application has the following advantages due to the above technical solutions.

[0021] 1. The present application takes advantage of the memory-compute integrator, loads static variables in the probabilistic graph iterative reasoning process into the memory-compute integrator, to overcome energy consumption caused by memory access in the iterative process, and uses the Kirchhoff circuit law to transfer discrete fixed-point number operations involved in the iterative process to the analog domain, to break through the power consumption bottleneck of digital circuits, to realize a cross-order improvement in the energy efficiency of terminal baseband signal processing, and to enable the estimation accuracy of the transmitted symbol in the MIMO detection task to be measured to approach the performance limit.

[0022] 2、The application can overcome the power consumption overhead caused by a large number of discrete fixed-point number operations and break through the power consumption bottleneck of digital circuits by transferring the fixed-point number operation of traditional digital circuits to the analog domain.

[0023] 3、The traditional digital MAC unit is composed of a multiplier and an adder, occupies a significant area, and requires multiple memory accesses, three read operations and one write operation each time, resulting in a large amount of energy and delay overhead. The storage and computing integrator based on the unified probability graph adopted by the application completes matrix-vector multiplication in one step, thereby greatly reducing the access frequency and energy overhead.

[0024] 4、The application has a small area, and the occupied area is several tenths of that of a traditional CMOS (Complementary Metal Oxide Semiconductor) circuit.

[0025] In summary, the application can be widely applied in the field of signal processing technology. BRIEF DESCRIPTION OF DRAWINGS

[0026] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Throughout the drawings, the same reference numerals are used for the same components. In the drawings:

[0027] Figure 1 is a method flowchart provided by an embodiment of the application;

[0028] Figure 2 is a unified signal model diagram provided by an embodiment of the application;

[0029] Figure 3 is a connection relationship diagram between check nodes and each variable node in the probability graph model provided by an embodiment of the application;

[0030] Figure 4 is a parallel processing architecture diagram provided by an embodiment of the application;

[0031] Figure 5 is a serial processing architecture diagram provided by an embodiment of the application. DETAILED DESCRIPTION

[0032] It is to be understood that the terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. The terms "comprises", "comprising", "includes", "including" and "has" are inclusive and therefore specify the presence of stated features, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof. The method steps, processes, and operations described herein are not to be construed as necessarily requiring their performance in the particular order

[0033] Although the terms first, second, third, etc. can be used herein to describe various elements, components, regions, layers and / or sections, these elements, components, regions, layers and / or sections should not be limited by these terms. These terms can be only used to distinguish one element, component, region, layer or section from another region, layer or section. Terms such as "first", "second", and other numerical terms when used herein do not imply a sequence or order unless clearly indicated by the context. Thus, a first element, component, region, layer or section discussed below could be termed a second element, component, region, layer or section without departing from the teachings of the example embodiments.

[0034] Spatially relative terms, such as "inner", "outer", "beneath", "below", "lower", "above", "upper", and the like, can be used herein for ease of description to describe one element or feature's relationship to another element(s) or feature(s) as illustrated in the figures. The spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientations depicted in the figures.

[0035] Currently, the mainstream technical path of the industry to cope with the performance requirements of 5G is largely restricted by the performance boundary drawn by the classical information theory framework. In order to approach these theoretical limits while meeting the performance indicators of actual deployment, existing communication system design often tends to adopt a "stacked" optimization strategy, which trades off performance gain for algorithm complexity. And due to the constraints of hardware conditions, small sensors and ocean buoys also face serious energy supply shortages. Therefore, the embodiment of the present application provides a MIMO detection method based on a memory and computing integrated device, which comprises: acquiring the channel matrix and the probabilistic graph model of the to-be-detected MIMO detection task, and determining the initial probability of the state value of each variable node in the probabilistic graph model according to the multi-antenna mixed signal observed by the receiving end and the noise variance; for any check node in the probabilistic graph model, according to each target undirected edge between the check node and each target variable node connected thereto, the channel matrix of the to-be-detected MIMO detection task and the multi-antenna mixed signal observed by the receiving end, a node state matrix corresponding to each undirected edge in the probabilistic graph model is constructed; according to the confidence of each variable node in the probabilistic graph model for transmitting its state value to the check node, a driving vector corresponding to each undirected edge in the probabilistic graph model is constructed, wherein in the initial stage, the initial probability of the state value of each variable node in the probabilistic graph model is taken as the confidence of the variable node for transmitting its state value to each check node connected through the undirected edge; based on the node state matrix corresponding to each undirected edge, a memory and computing integrated device is configured to obtain a memory and computing integrated device subarray corresponding to each undirected edge; according to the memory and computing integrated device subarray corresponding to each undirected edge and the driving vector constructed, the confidence of each check node in the probabilistic graph model for transmitting its state value to the variable node is determined, and then the confidence of each variable node for transmitting its state value to the check node is updated until a set iteration stop condition is met, and the optimal estimation value of the transmission symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task is determined. The present application stores part of the static variables in the memory and computing integrated device such as the memristor, so that part of the traditional digital operation is transferred to the analog domain, and the Kirchhoff circuit law is used for operation, thereby breaking through the power consumption bottleneck of digital circuits.

[0036] Exemplary embodiments of the present application will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present application can be more thoroughly understood, and the scope of the present application can be accurately conveyed to those skilled in the art. The extremely low power consumption general baseband signal processing method and hardware circuit structure proposed by the present application are suitable for different types of application scenarios, and will be described in detail in conjunction with specific embodiments.

[0037] Embodiment 1

[0038] As Figure 1As shown, this embodiment provides a MIMO detection method based on a memory computing device, including the following steps:

[0039] S1. Obtain the channel matrix and probabilistic graphical model of the MIMO detection task to be tested, and determine the initial probability of the state value of each variable node in the probabilistic graphical model based on the variance of the multi-antenna mixed signal and noise observed at the receiver.

[0040] In-memory computing devices are integrated devices that combine storage and computing capabilities, such as resistive random access memory (RRAM), phase-change memory (PCM), magnetic memory (MRAM), and flash memory.

[0041] The MIMO detection task under test is used to observe multi-antenna mixed signals from the receiver. Accurately extract the transmitted symbols to be estimated from the transmitter. The number of states of the transmitted symbols to be estimated at the transmitter is ,Right now , Indicates the transmitted symbol to be estimated The A symbolic vector, Indicates the transmitted symbol to be estimated The A state, The state set space is defined by the number of states, which varies depending on the modulation scheme in the MIMO detection task. Commonly used modulation schemes include QPSK, 8PSK, 16QAM, 64QAM, and 256QAM, with corresponding state numbers... The values ​​are 4, 8, 16, 64 and 256.

[0042] Among them, the probabilistic graphical model is a unified probabilistic graphical model corresponding to the channel matrix, used for the MIMO detection task under test.

[0043] The specific process of step S1 above is as follows:

[0044] S101. Obtain the channel matrix of the MIMO detection task to be tested. .

[0045] Specifically, the channel matrix of the MIMO detection task under test It is a dimension of The matrix, the number of rows of the MIMO channel matrix The number of antennas and columns at the receiving end. The number of antennas at the transmitting end, and the channel matrix. Each element in the table represents the channel characteristics of the corresponding transmit-receive antenna pair.

[0046] Specifically, the transmitted symbol to be estimated Multi-antenna mixed signal and channel matrix There exists a unified signal model as ,like Figure 2 As shown (where Rx is the receiver and Tx is the transmitter). For noise, continuous, continuous, For unknown discrete variables.

[0047] S102. A unified signal model based on the MIMO detection task, according to the channel matrix of the MIMO detection task under test. A probabilistic graphical model for the MIMO detection task to be tested is constructed.

[0048] Specifically, the probabilistic graphical model includes variable nodes. Verification Node And undirected edges, where variable nodes Unknown variables, in this invention, represent transmitted symbols to be estimated. That is, each variable node corresponds to a transmitted symbol to be estimated. A symbol vector in the form of a check node; The system represents the coupling relationship between the transmitted symbols to be estimated at the transmitter, the multi-antenna mixed signal observed at the receiver, and the channel matrix; undirected edges are used to connect variable nodes and check nodes that have a connection relationship. Each variable node uniquely corresponds to a channel matrix. A column vector in the channel matrix, where each verification node uniquely corresponds to a channel matrix. A row vector within the channel matrix. The Middle Row vector representation , Channel matrix No. Line 1 Column elements, representing validation nodes. With variable nodes The correlation between them, if the channel matrix middle If the element is 1, then the check node is valid. With variable nodes They have a connection relationship, if the channel matrix middle If the element is 0, then the check node is validated. With variable nodes No connection exists.

[0049] To better illustrate the relationship between the channel matrix and the probabilistic graphical model, as well as the relationship between variable nodes, check nodes, and undirected edges, the following Example 1 is presented and combined with... Figure 3An introduction is made.

[0050] Example 1: when the channel matrix is a 6x6 matrix, the row vector of the third row is [0, 1, 1, 1, 0, 0], the probability graph model includes 6 check nodes, 6 variable nodes (at this time, 6 is 6), the connection relationship between the third check node (at this time, 6 is 3) and each variable node is as shown in , wherein the 2nd, 3rd and 4th elements in the row vector are 1, and the 1st, 5th and 6th elements are 0, so the 3rd check node has a connection relationship with the 2nd, 3rd and 4th variable nodes and is connected through an undirected edge, and the 3rd check node does not have a connection relationship with the 1st, 5th and 6th variable nodes. It should be noted that Figure 3 , in addition to the 3rd check node, the connection relationship between other check nodes and variable nodes is not drawn, and the connection relationship between other check nodes and variable nodes can be understood with reference to the connection relationship between the 3rd check node and the variable node. Figure 3

[0051] S103, according to the multi-antenna mixed signal observed by the receiving end, the channel matrix of the to-be-measured MIMO detection task and the noise variance, determine the state value initial probability of each variable node in the probability graph model, that is, the initial probability of each variable node corresponding to the to-be-estimated sending symbol .

[0052] Specifically, the determination manner of the state value initial probability can be:

[0053] 1) according to the multi-antenna mixed signal observed by the receiving end, the channel matrix and the noise variance, calculate the initial estimation result of all variable nodes, wherein the initial estimation result of all variable nodes , denotes the transpose of the channel matrix , and denotes the noise variance.

[0054] 2) calculate the distance from the initial estimation result of all variable nodes to the state set space of the variable node, wherein the initial estimation result of the variable node to the state set space is .

[0055] ​​3) obtaining the initial probability of the state value corresponding to each variable node according to the distance of the initial estimation result of all variable nodes to the state set space, wherein the variable node the initial probability of the state value corresponding to each variable node .

[0056] S2, for any check node in the probabilistic graph model, constructing a node state matrix corresponding to each undirected edge in the probabilistic graph model according to each target undirected edge between the check node and each target variable node connected thereto, the channel matrix of the MIMO detection task to be measured, and the multi-antenna mixed signal observed by the receiving end.

[0057] wherein the target variable node is a variable node connected to a certain check node and connected to the check node through an undirected edge. The target undirected edge is an undirected edge connected between the check node and the target variable node.

[0058] Specifically, for any check node in the probabilistic graph model, each target variable node connected to the check node through an undirected edge can be determined, and a target undirected edge between the check node and each target variable node can be determined, and a node state matrix corresponding to each target undirected edge can be constructed according to each target undirected edge and the channel matrix.

[0059] Specifically, the node state matrix is a matrix used to configure the state of the node in the storage-computing integrated device.

[0060] It can be understood that the embodiment can construct a node state matrix corresponding to each target undirected edge connected to a certain check node. Therefore, the embodiment can construct a node state matrix corresponding to each undirected edge according to each undirected edge in the probabilistic graph model. The total number of node state matrices corresponds to the total number of undirected edges.

[0061] S3, constructing a driving vector corresponding to each undirected edge in the probabilistic graph model according to the confidence of each variable node in the probabilistic graph model in transmitting its state value to the check node, wherein in the initial stage, the initial probability of the state value of each variable node in the probabilistic graph model is taken as the confidence of the variable node in transmitting its state value to each check node connected through an undirected edge.

[0062] Specifically, for any check node in the probabilistic graph model, a driving vector corresponding to each undirected edge connected to the check node can be constructed according to each undirected edge connected to the check node and the confidence of each variable node in transmitting its state value to the check node.

[0063] It can be understood that the variable node will only transmit messages to the check nodes having a connection relationship and will not transmit messages to the check nodes not having a connection relationship.

[0064] It should be noted that the driving vector is used to input a corresponding size of driving physical quantity into the memory-compute integrated device, and the memory-compute integrated device performs storage and computation based on the memory-compute integrated characteristics, so as to realize MIMO detection.

[0065] It can be understood that the embodiment can construct the driving vector corresponding to each target undirected edge connected to a certain check node. Therefore, the embodiment can construct the driving vector corresponding to each undirected edge according to each undirected edge in the probabilistic graph model. The total number of driving vectors corresponds to the total number of undirected edges.

[0066] S4, configuring the memory-compute integrated device based on the node state matrix corresponding to each undirected edge to obtain the memory-compute integrated device subarray corresponding to each undirected edge.

[0067] Specifically, for the node state matrix corresponding to any undirected edge, the embodiment can configure the state of each node in the initial-state memory-compute integrated device according to the node state matrix corresponding to the undirected edge, to obtain the memory-compute integrated device subarray corresponding to the undirected edge.

[0068] The initial-state memory-compute integrated device can be a memory-compute integrated device whose node state is not set.

[0069] It can be understood that the embodiment can configure a corresponding memory-compute integrated device subarray for any undirected edge in the probabilistic graph model.

[0070] S5, determining the confidence of each check node in the probabilistic graph model in transmitting its state value to the variable node according to the memory-compute integrated device subarray corresponding to each undirected edge and the constructed driving vector, and further updating the confidence of each variable node in transmitting its state value to the check node until a set iteration stop condition is met, to determine the optimal estimation value of the transmitted symbol corresponding to the multi-antenna mixed signal in the to-be-tested MIMO detection task.

[0071] Specifically, the embodiment can determine the optimal estimation value of the transmitted symbol corresponding to the multi-antenna mixed signal in the to-be-tested MIMO detection task according to the memory-compute integrated device subarray corresponding to each undirected edge in the probabilistic graph model and the constructed driving vector, and according to the confidence of message transmission between each variable node and each check node connected through the undirected edge.

[0072] The MIMO detection method based on the storage-computing integrated device provided in the embodiment can configure the storage-computing integrated device subarray and the driving vector corresponding to each undirected edge in the probabilistic graph model according to the channel matrix and the probabilistic graph model of the MIMO detection task to be detected, and determine the optimal estimation value of the transmission symbol corresponding to the multi-antenna mixed signal in the MIMO detection task to be detected based on the confidence of the message passing between each variable node and each check node connected through the undirected edge, the storage-computing integrated device subarray corresponding to each undirected edge, and the driving vector. The application can store part of the static variables involved in the MIMO detection in the storage-computing integrated device, and transfer part of the digital operation to the storage-computing integrated device, and perform operation by using the storage and computing performance of the storage-computing integrated device, thereby breaking through the power bottleneck and significantly reducing the hardware resource consumption and power consumption.

[0073] Based on Figure 1 , the second MIMO detection method based on the storage-computing integrated device is provided in the embodiment, in which the storage-computing integrated device is a memristor, and at this time, the node state matrix in step S2 is a node resistance state matrix.

[0074] Specifically, for the first undirected edge, the node state matrix is:

[0075]

[0076] wherein, is the corresponding node resistance state matrix; represents the first element value in the candidate vector constructed by the first undirected edge (i.e., the undirected edge between the first check node and the first variable node); and , represents all combinations of variable node states of the variable node connected with the check node and having the value of the variable node ; represents the element value corresponding to the variable node in the candidate vector;

[0077] represents the first signal of the multi-antenna mixed signal observed by the receiving end; represents the first row vector in the channel matrix , wherein, in the channel matrix satisfies The corresponding variable node is the check node. There is a connection. Indicates and verifies the node A set of indexes of variable nodes with connections.

[0078] Specifically, the construction process of the above-mentioned nodal resistance matrix includes:

[0079] S201. Obtain the number of target undirected edges and determine the internal order of each target undirected edge among all target undirected edges.

[0080] S202, constructing from Multiple candidate vectors are obtained by permuting and combining n target characters, and the number of candidate vectors is . For any candidate vector, if the element order of the target character in the candidate vector is equal to the internal order of the target undirected edge, then an association is established between the target character in the candidate vector and the target variable node connected by the target undirected edge. Here, the target character and... correspond.

[0081] S203. Iterate through each target undirected edge in turn.

[0082] S204, For the traversed first... There are several undirected edges representing the target vectors, which divide the multiple candidate vectors into several groups to be processed. The candidate vectors, based on each of the nth and nth... Using candidate vectors, correlation relationships, channel matrices, multi-antenna mixed signals observed at the receiver, and noise variance, construct the node resistance state matrix corresponding to the undirected edge of the target; where the th... Among the candidate vectors, the th... The elements are .

[0083] Specifically, according to each The candidate vectors, correlation relationships, channel matrix, and variance of the multi-antenna mixed signal and noise observed at the receiver include:

[0084] (1) For any candidate vector to be processed, construct a zero vector with a total number of elements equal to the total number of variable nodes. Determine the node order of the target variable nodes associated with each target character in the candidate vector to be processed as the unique order to be arranged for each target character. Replace each zero element in the zero vector whose position order is equal to the order to be arranged with the target character uniquely corresponding to the order to be arranged to obtain the processed candidate vector. The candidate vector to be processed is any first-order variable variable node. There are 10 candidate vectors.

[0085] (2) Determine the unique target row vector corresponding to the verification node in the channel matrix.

[0086] (3) Multiply the target row vector by each processed candidate vector to obtain multiple corresponding values, and arrange them as follows: A column vector of resistance states at each node. This represents the number of states.

[0087] (4) The column vectors of the resistive states of each node are horizontally concatenated, and the results are based on the multi-antenna mixed signal observed at the receiver. and noise variance We obtain the node resistance state matrix corresponding to the target undirected edge.

[0088] To better illustrate the above execution process, this embodiment continues to combine Example 1 and... Figure 3 Let me introduce it.

[0089] Example 1: In this embodiment, for any verification node... The verification node can be determined in the probabilistic graphical model. Connecting the target variable nodes and the target undirected edges. For example... Figure 3 As shown, when When the value is 3, the target variable nodes connected by the 3rd verification node are the 2nd, 3rd, and 4th variable nodes, respectively. At this time, the number of undirected edges in the target node is... The value is 3. Each target undirected edge is an undirected edge connecting the 3rd verification node to the 2nd, 3rd, and 4th variable nodes, respectively. The internal order of each target undirected edge among all target undirected edges is 1, 2, and 3.

[0090] The target character includes 1 and 0, at which point .at this time The result is 3. Multiple candidate vectors are constructed by permuting and combining the three target characters. The number of candidate vectors is 2^3, or 8: (0,0,0), (0,0,1), (0,1,0), (0,1,1), (1,0,0), (1,0,1), (1,1,0), and (1,1,1). It can be understood that the number of elements in the candidate vectors is the same as the number of undirected edges in the target vectors. The first, second, and third target characters in the candidate vectors are then associated with the target variable nodes connected by the undirected edges in the internal order 1, 2, and 3, respectively. In other words, the first, second, and third target characters in the candidate vectors are associated with the second, third, and fourth variable nodes, respectively.

[0091] The process iterates through the first, second, and third target undirected edges in sequence. Upon encountering each target undirected edge, the eight candidate vectors are divided into first candidate vector and second candidate vector. For example, when iterating through the third target undirected edge... is equal to 3, if the third element in the candidate vector is 1, it is determined as the first candidate vector, if the third element in the candidate vector is 0, it is determined as the second candidate vector, at this time, the first candidate vector includes (0, 0, 1), (0, 1, 1), (1, 0, 1) and (1, 1, 1), and the second candidate vector includes (0, 0, 0), (0, 1, 0), (1, 0, 0) and (1, 1, 0).

[0092] Each of the above first candidate vector and second candidate vector is taken as a candidate vector to be processed. For example, when the first candidate vector (0, 1, 1) is taken as a candidate vector to be processed, a zero vector (0, 0, 0, 0, 0, 0) with the total number of elements equal to the total number of variable nodes 6 is constructed first, the first, second and third target characters in the first candidate vector have association relationships with the second, third and fourth variable nodes respectively, therefore, the second, third and fourth zero elements in the zero vector can be replaced by the first, second and third target characters in the first candidate vector respectively, so as to obtain a corresponding processed first candidate vector (0, 0, 1, 1, 0, 0).

[0093] At this time, when the third target undirected edge is traversed, four processed first candidate vectors and four processed second candidate vectors can be obtained according to the target undirected edge. Then, the target row vector corresponding to the third check node in the channel matrix, that is, the third row vector in the channel matrix, can be determined. The target row vector is multiplied by the four processed first candidate vectors respectively, so as to obtain four corresponding values, and a first node state column vector is obtained by vertically arranging the four values. The target row vector is multiplied by the four processed second candidate vectors respectively, so as to obtain four corresponding values, and a second node state column vector is obtained by vertically arranging the four values. The first node state column vector and the second node state column vector are horizontally spliced, and the multiple antenna mixed signals observed by the receiving end and noise variance are obtained, so as to obtain the node state matrix corresponding to the third target undirected edge.

[0094] Specifically, when the memory-computing integrated device is a memristor, at this time, the driving vector in S3 is a voltage vector, and the above step S3 includes:

[0095] When the th target undirected edge is traversed, the th Given several elements, we obtain multiple first vectors. We then remove duplicates from these first vectors to obtain multiple second vectors. For any second vector, we determine the target confidence level corresponding to each target character in the second vector. We multiply the target confidence levels corresponding to each target character in the second vector to obtain the product of the second vector. We arrange the products of each second vector to obtain the voltage vector corresponding to the undirected edge of the target. The target confidence level corresponding to the target character in the second vector is: the confidence level of the target variable node associated with the target character, which transmits its own state value equal to the target character to the verification node.

[0096] To better illustrate the above execution process, this embodiment continues to combine Example 1 and... Figure 3 Let me introduce it.

[0097] Example 1: When verifying the internal order of the target undirected edges connected to the node. When the value is 3, the eight candidate vectors created in this embodiment are (0, 0, 0), (0, 0, 1), (0, 1, 0), (0, 1, 1), (1, 0, 0), (1, 0, 1), (1, 1, 0), and (1, 1, 1). In this case, the third element of each candidate vector can be deleted to obtain eight first vectors: (0, 0), (0, 0), (0, 1), (0, 1), (1, 0), (1, 0), (1, 1), and (1, 1). After deduplication, the remaining four second vectors are obtained: (0, 0), (0, 1), (1, 0), and (1, 1). For any of the second vectors, such as the second vector (0, 1), this embodiment can determine the target confidence level corresponding to each target character in the second vector. The confidence level corresponding to the first target character in the second vector is: the confidence level of the target variable node (i.e., the second variable node) associated with the target character transmitting the message with its own state value equal to 0 to the third verification node. The confidence level corresponding to the second target character in the second vector is: the confidence level of the target variable node (i.e., the third variable node) associated with the target character transmitting the message with its own state value equal to 1 to the third verification node. Multiplying the two determined confidence levels yields one product corresponding to the second vector. Four products are calculated for the four second vectors. Arranging these four products vertically yields the voltage vector corresponding to the third undirected edge of the target.

[0098] Specifically, the first The voltage vector corresponding to each undirected edge is denoted as , dimension The first node resistance matrix Voltage vector of the row for:

[0099]

[0100] wherein, represents the probability of selecting the i-th row corresponding to the index set; represents the i-th element in the voltage vector corresponding to the i-th undirected edge ; , the probability is a function of the information transferred between the variable node and the check node, and the specific calculation method includes but is not limited to:

[0101]

[0102] wherein, represents the message about the to-be-estimated sending symbol transferred between the variable node and the check node , represents the element value corresponding to the variable node in the candidate vector, and , represents the variable node taking a value from the variable node index set connected with the check node ; at the beginning of the first iteration, .

[0103] Specifically, the upper limit of the number of voltage vectors input to the memristor subarray is the number of vectors in the set , which is , wherein the set represents the combination of all variable node states connected with the check node except the variable node .

[0104] Specifically, the message about the to-be-estimated sending symbol transferred between the variable node and the check node , i.e., the confidence, is:

[0105]

[0106] wherein, represents the message about the variable node state taking a value transferred between the variable node and the check node , represents the message about the variable node state taking a value​​​ The message is from the verification node. Other than variable nodes Connected verification nodes Provide message generation, The calculation methods include, but are not limited to, Indicates the variable node at the start of the first iteration. to the verification node The information transmitted between them The message; Indicates the verification node To variable node Passing information about variable nodes between nodes Passing information about variable nodes between nodes The news Indicates the verification node From variable nodes A set of check node indexes with connections Take the value from the middle, but verify the node. Do not take ; Represents variable nodes From the verification node A set of indexes of variable nodes with connections The value is taken from the variable node. Do not take .

[0107] Specifically, the verification node To variable node The message passed between them is:

[0108]

[0109] Specifically, when the in-memory computing unit is a memristor, the in-memory computing unit subarray is a memristor subarray. In this case, the node state matrix in the MIMO detection task under test... It is related not only to the undirected edges and the channel matrix, but also to the mixed signals from multiple antennas observed at the receiver. Related to the observed multi-antenna mixed signal. Since the values ​​are different each time a signal arrives at the receiver, and considering the requirements for stable and high-speed signal processing, it is not possible to refresh the array values ​​after each signal arrival. Therefore, it is necessary to update the node state matrix. After the values ​​are statically defined, they are configured in the in-memory compute unit. Each variable is then directly configured... , Separate quantization and recombination will lead to dimensionality explosion, and deploying these discrete results to in-memory computing is very inefficient. Therefore, this invention employs a holistic quantization method, and step S4 above includes:

[0110] S401, elements in the node resistance state matrix are taken as a whole, and the value distribution can be regarded as a Gaussian distribution with a mean of 0.

[0111] S402, the acceptance interval is set according to the performance and complexity compromise .

[0112] S403, the above whole is quantized into quantization states in the acceptance interval , and the quantization step is , denoted as , forming an array with a size of .

[0113] S404, after quantization, the value of the th element in the th undirected edge is , is the th in the quantization state.

[0114] S405, the quantized node resistance state matrix value is configured into the memory-computing integrated device, and the memory-computing integrated device subarray corresponding to each undirected edge is obtained.

[0115] Specifically, the memristor subarray corresponding to the undirected edge can be represented as:

[0116]

[0117] wherein, represents the th element in the th row of the memristor subarray corresponding to the undirected edge between the th check node and the th variable node in the memristor subarray (i.e., the th memristor subarray).

[0118] Specifically, when the memory-computing integrated device is a memristor, at this time, the above step S5 includes:

[0119] S501, for any undirected edge, according to each voltage element in the voltage vector corresponding to the undirected edge, input a voltage equal to the size of the voltage element to each row of nodes in the memristor subarray corresponding to the undirected edge, and read the current value in the memristor subarray .

[0120] Specifically, the voltage vector is input to the memristor subarray ​​First, routing is performed (selecting which row voltage to input), Corresponding voltage input behavior:

[0121]

[0122] If there are multiple If the same row is selected for input, the same row is selected.

[0123] S502, according to the current value read in each memristor subarray, determine the confidence of each check node in the probabilistic graph model transmitting its state value to the variable node, and then update to obtain the new confidence of each variable node transmitting its state value to each check node .

[0124] Specifically, using the memristor subarray, according to The output current .

[0125] S503, determine whether a preset iteration stop condition is reached.

[0126] S504, if the preset iteration stop condition is not reached, the new confidence of each variable node transmitting its state value to each check node is taken as the current confidence of the corresponding variable node transmitting its state value to each check node, and the calculation is restarted in step S3 until the preset iteration stop condition is reached.

[0127] S505, if the preset iteration stop condition is reached, according to the state value initial probability of each variable node in the probabilistic graph model and the confidence of the current corresponding variable node transmitting its state value to each check node, update to obtain the state value new probability of each variable node, and then determine the optimal estimation value of the transmitted symbol corresponding to the multi-antenna mixed signal in the to-be-measured MIMO detection task.

[0128] Specifically, the state value new probability of each variable node is:

[0129]

[0130] Wherein, represents the message about transmitted between the variable node and the check node at the beginning of the first iteration.

[0131] Specifically, the calculation method of the optimal estimation value of the transmitted symbol corresponding to the multi-antenna mixed signal includes but is not limited to wherein, .

[0132] In a preferred embodiment of the present application, a DAC (Digital Analog Converter) module and an ADC (Analog Digital Converter) module are further included, the ADC module is used to convert the current value outputted by the memristor sub-array into a digital signal for the calculation of the message passing between the variable nodes and the check nodes, and the calculation process can be realized by using an existing FPGA (Field Programmable Gate Array), which will not be described here. The DAC module is used to convert the calculation result of the message passing between the variable nodes and the check nodes into a voltage signal for the iterative calculation of the message passing between the check nodes and the variable nodes.

[0133] In a preferred embodiment of the present application, the computing architecture of the memory-compute-in-one device is provided with two modes, including a time-parallel processing architecture and a time-serial processing architecture.

[0134] As shown in Figure 4 , the time-parallel processing architecture is that each sub-array is independently disconnected, for example, each memristor sub-array is used for MIMO detection, and the implementation is that the input voltage is simultaneously inputted to all the sub-arrays, and the current is simultaneously read out for iterative calculation.

[0135] Figure 5 As shown in, the time-serial processing architecture is that each sub-array is connected, for example, each memristor sub-array is spliced into a large memristor array for MIMO detection, and the implementation is that the voltage is inputted to the corresponding sub-array in time slots in turn, each sub-array reads the corresponding input voltage in turn, and after the current is read, the input is inputted again in time slots for iterative calculation.

[0136] It should be noted that the array values and sizes of the two modes are exactly the same, the sub-arrays of the time-parallel processing architecture are independently disconnected, and the sub-arrays of the time-serial processing architecture are continuously connected, and the sub-arrays can be arranged and connected in turn.

[0137] To better illustrate the above execution process, the time-serial processing architecture shown in Figure 4 will be described in detail in this embodiment.

[0138] As shown in Figure 4 , the node resistance state matrix constructed in this embodiment has , the voltage vector has , and the read current has . Figure 4 DAC is an exponential modulus converter, and ADC is an analog-digital converter.

[0139] wherein, This represents the memristor subarray corresponding to the undirected edge between the first check node and the first variable node with a connection relationship. Indicates the first The verifiable node and the first node with a connection relationship The undirected edges between the variable nodes correspond to the memristor subarrays. It should be noted that... and It does not refer to the second node order of the variable node, but rather to the order of the variable node connected to the check node among all the variable nodes connected to that check node.

[0140] This represents the voltage vector corresponding to the undirected edge between the first verification node and the first variable node with a connection relationship. Indicates the first The verifiable node and the first node with a connection relationship The voltage vector corresponding to the undirected edge between each variable node. This represents the current value in the memristor subarray corresponding to the undirected edge between the first verification node and the first variable node with which it is connected. Indicates the first The verifiable node and the first node with a connection relationship The current values ​​in the memristor subarray corresponding to the undirected edges between the variable nodes.

[0141] Corresponding to the same undirected edge and For example, in this embodiment, the undirected edge can be used as a reference. Each voltage element in the array is transmitted via a digital-to-analog converter (DAC) to the memristor subarray corresponding to the undirected edge. Each row of nodes in the array receives a voltage equal to the size of the voltage element; for example, this voltage is input to the memristor subarray. The first row of nodes in the input and... A voltage equal to the magnitude of the first voltage element is applied to the memristor subarray. The second row of nodes in the input is the same as the input. A voltage equal to the second voltage element is applied until the last row of nodes in the memristor subarray is input. The voltage is equal to the value of the last voltage element. Then, in this embodiment, the current value in the memristor subarray can be read via an analog-to-digital converter (ADC). For the node resistance matrix, voltage vector, and current vector corresponding to other undirected edges, this embodiment can refer to the above process for parallel processing, and will not be repeated here.

[0142] The MIMO detection method based on the memory-computing integrated device can store part of static variables in a device such as a memristor having the memory-computing integrated feature, thereby transferring part of digital operation to an analog domain, performing operation by using the Kirchhoff circuit law, breaking through the power consumption bottleneck of a digital circuit, and realizing the innovation of MIMO detection.

[0143] Embodiment 2

[0144] The embodiment provides a MIMO detection system based on a memory-computing integrated device, which comprises:

[0145] An acquisition unit is configured to acquire a channel matrix and a probabilistic graph model of a to-be-detected MIMO detection task, and determine an initial probability of a state value of each variable node in the probabilistic graph model according to a multi-antenna mixed signal observed by a receiving end and a noise variance.

[0146] A node state matrix construction unit is configured to, for any check node in the probabilistic graph model, construct a node state matrix corresponding to each undirected edge in the probabilistic graph model according to each target undirected edge between the check node and each target variable node connected to the check node, the channel matrix of the to-be-detected MIMO detection task and the multi-antenna mixed signal observed by the receiving end.

[0147] A driving vector construction unit is configured to construct a driving vector corresponding to each undirected edge in the probabilistic graph model according to a confidence degree of each variable node in the probabilistic graph model in transmitting a state value of the variable node to a check node, wherein in an initial stage, the initial probability of the state value of each variable node in the probabilistic graph model is taken as the confidence degree of the variable node in transmitting the state value of the variable node to each check node connected through an undirected edge.

[0148] A configuration unit is configured to perform memory-computing integrated device configuration based on the node state matrix corresponding to each undirected edge, to obtain a memory-computing integrated device subarray corresponding to each undirected edge.

[0149] An optimal estimation value determination unit is configured to determine a confidence degree of each check node in the probabilistic graph model in transmitting a state value of the check node to a variable node according to the memory-computing integrated device subarray corresponding to each undirected edge and the constructed driving vector, and then update the confidence degree of each variable node in transmitting the state value of the variable node to a check node until a set iteration stop condition is met, to determine an optimal estimation value of a transmission symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task.

[0150] The system provided in the embodiment is used for executing the above-mentioned method embodiments, and specific processes and detailed contents are referred to the above-mentioned embodiments, which will not be described herein again.

[0151] The MIMO detection system based on the memory-compute integrated device provided in the embodiment can configure the memory-compute integrated device subarray and the driving vector corresponding to each undirected edge in the probabilistic graph model according to the channel matrix and the probabilistic graph model of the MIMO detection task to be detected, and determine the optimal estimation value of the transmission symbol corresponding to the multi-antenna mixed signal in the MIMO detection task to be detected based on the memory-compute integrated device subarray corresponding to each undirected edge, the driving vector, and the confidence of the message passing between each variable node and each check node connected through the undirected edge. The application can store part of the static variables involved in the MIMO detection into the memory-compute integrated device, and transfer part of the digital operation to the memory-compute integrated device, and perform the operation by using the storage and computing performance of the memory-compute integrated device, thereby breaking through the power consumption bottleneck, reserving the MIMO detection capability, and significantly reducing the hardware resource consumption and power consumption.

[0152] Embodiment 3

[0153] The embodiment provides a processing device corresponding to the MIMO detection method based on the memory-compute integrated device provided in the embodiment 1, and the processing device can be applied to the processing device of a client, such as a mobile phone, a notebook computer, a tablet computer, a desktop computer, etc., to execute the method of the embodiment 1.

[0154] The processing device includes a processor, a memory, a communication interface, and a bus, and the processor, the memory, and the communication interface are connected through the bus to complete the communication among each other. The memory stores a computer program that can run on the processing device, and the processing device executes the MIMO detection method based on the memory-compute integrated device provided in the embodiment 1 when running the computer program.

[0155] In some implementations, the memory can be a high-speed random access memory (RAM), and can also include a non-volatile memory, such as at least one disk memory.

[0156] In other implementations, the processor can be a central processing unit (CPU), a digital signal processor (DSP), or various types of general-purpose processors, which are not limited here.

[0157] In addition, the logic instructions in the memory described above can be implemented in the form of software function units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0158] Those skilled in the art can understand that the structure of the computing device described above is only part of the structure related to the present application scheme, and does not constitute a limitation on the computing device to which the present application scheme is applied. The specific computing device can include more or fewer components, or combine certain components, or have a different component arrangement.

[0159] Embodiment 4

[0160] The embodiment provides a computer program product corresponding to the MIMO detection method based on the memory-computing integrated device provided in Embodiment 1. The computer program product can include a computer readable storage medium, which is loaded with computer readable program instructions for executing the MIMO detection method based on the memory-computing integrated device described in Embodiment 1.

[0161] The computer readable storage medium can be a tangible device that maintains and stores instructions for use by an instruction execution device. The computer readable storage medium can be, for example but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0162] The computer readable storage medium provided in the above embodiment has similar implementation principles and technical effects to the above method embodiments, and will not be described here.

[0163] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more flow or flows and / or block diagram block or blocks. Figure 1 one or more flow or flows and / or block diagram block or blocks.

[0164] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more flow or flows and / or block diagram block or blocks. Figure 1 one or more flow or flows and / or block diagram block or blocks.

[0165] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart or flows and / or block diagram block or blocks. Figure 1 one or more flow or flows and / or block diagram block or blocks. Figure 1 one or more flow or flows and / or block diagram block or blocks.

[0166] The above-described embodiments are merely intended to illustrate the present application, and the structure, connection manner, and manufacturing process of each component can be changed. Any equivalent changes and improvements made on the basis of the technical solutions of the present application shall not be excluded from the protection scope of the present application.

Claims

1. A method for MIMO detection based on a memory-compute integrated device, characterized in that, The method comprises the following steps: obtaining a channel matrix and a probabilistic graph model of a to-be-detected MIMO detection task, and determining an initial probability of a state value of each variable node in the probabilistic graph model according to a multi-antenna mixed signal observed by a receiving end and a noise variance; for any check node in the probabilistic graph model, constructing a node state matrix corresponding to each undirected edge in the probabilistic graph model according to each target undirected edge between the check node and each target variable node connected to the check node, the channel matrix of the to-be-detected MIMO detection task and the multi-antenna mixed signal observed by the receiving end; constructing a driving vector corresponding to each undirected edge in the probabilistic graph model according to a confidence degree of each variable node in the probabilistic graph model for transmitting a state value of the variable node to a check node, wherein, in an initial stage, the initial probability of the state value of each variable node in the probabilistic graph model is taken as the confidence degree of the variable node for transmitting the state value of the variable node to each check node connected to the variable node through an undirected edge; performing memory-computing integrator configuration based on the node state matrix corresponding to each undirected edge to obtain a memory-computing integrator subarray corresponding to each undirected edge; determining a confidence degree of each check node in the probabilistic graph model for transmitting a state value of the check node to a variable node according to the memory-computing integrator subarray corresponding to each undirected edge and the constructed driving vector, and then updating the confidence degree of each variable node for transmitting the state value of the variable node to the check node until a set iteration stop condition is met, to determine an optimal estimation value of a transmission symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task.

2. The MIMO detection method based on a compute-in-memory device according to claim 1, wherein, The method comprises the following steps: obtaining a channel matrix and a probabilistic graph model of a to-be-detected MIMO detection task, and determining an initial probability of a state value of each variable node in the probabilistic graph model according to a multi-antenna mixed signal observed by a receiving end and a noise variance; According to the channel matrix of the MIMO detection task to be detected, a probability graph model of the MIMO detection task to be detected is constructed based on a unified signal model of the MIMO detection task, wherein the unified signal model of the MIMO detection task is , is a multi-antenna mixed signal observed by a receiving end, is a sending symbol to be estimated of a transmitting end; is noise, is a channel matrix; is a row number of the channel matrix ; is a column number of the channel matrix ; obtaining a channel matrix of a to-be-detected MIMO detection task; 3. The MIMO detection method based on the compute-in-memory device according to claim 2, wherein, determining an initial probability of a state value of each variable node in the probabilistic graph model according to a multi-antenna mixed signal observed by a receiving end, the channel matrix of the to-be-detected MIMO detection task and a noise variance. The method comprises the following steps: calculating initial estimation results of all variable nodes according to the multi-antenna mixed signal observed by the receiving end, the channel matrix and the noise variance; calculating distances from the initial estimation results of all variable nodes to a state set space of the variable nodes; 4. The MIMO detection method based on the compute-in-memory device according to claim 1, wherein, obtaining the initial probability of the state value corresponding to each variable node according to the distances from the initial estimation results of all variable nodes to the state set space. obtaining a number of target undirected edges determining an internal order of each target undirected edge among all target undirected edges, respectively; Constructing a plurality of candidate vectors obtained by arranging and combining the target characters, the number of the candidate vectors being , , is the number of states of the to-be-estimated sending symbol of the transmitting end; for any candidate vector, if the element order of the target characters in the candidate vector is equal to the internal order of the target undirected edge, an association relationship is established between the target characters in the candidate vector and the target variable nodes connected with the target undirected edge, wherein the target characters and correspond, represents the th state of the to-be-estimated sending symbol. When the memory-computing integrator is a memristor, the node state matrix is a node resistance state matrix, and the method of constructing the node resistance state matrix corresponding to each undirected edge in the probabilistic graph model comprises the following steps: For the traversed number... There are several undirected edges representing the target vectors, which divide the multiple candidate vectors into several groups to be processed. The candidate vectors, based on each of the nth and nth... Using candidate vectors, correlation relationships, channel matrices, multi-antenna mixed signals observed at the receiver, and noise variance, construct the node resistance state matrix corresponding to the undirected edge of the target; where the th... Among the candidate vectors, the th... The elements are .

5. The MIMO detection method based on a compute-in-memory device according to claim 4, wherein, The node resistance matrix is: wherein, is a corresponding node state matrix; denotes the element value in the candidate vector constructed by the th undirected edge; and denotes the combination of all variable node states connected with the check node and the variable node taking the value , denotes the corresponding element value of the variable node in the candidate vector; denotes the noise variance; denotes the th signal of the multi-antenna mixed signal observed by the receiving end; denotes the th row vector in the channel matrix .

6. The MIMO detection method based on a compute-in-memory device according to claim 5, wherein, sequentially traversing each target undirected edge; The elements in the node impedance matrix are as a whole, the value distribution is considered as a Gaussian distribution with a mean of 0; The method of performing memory-computing integrator configuration based on the node state matrix corresponding to each undirected edge to obtain a memory-computing integrator subarray corresponding to each undirected edge comprises the following steps: The value of the first element in the first undirected edge after quantization is: The value of the first element in the first undirected edge after quantization is: The value of the first element in the first undirected edge after quantization is wherein is a quantization step size; is a first of a number of quantization states; is a second of the number of quantization states. quantizing the whole into a plurality of quantization states within a pre-set acceptance interval; configuring the quantized node resistance state matrix value into the memory-computing integrator to obtain the memory-computing integrator subarray corresponding to each undirected edge.

7. The MIMO detection method based on a compute-in-memory device according to claim 6, wherein, When the memory-computing integrated device is a memristor, the driving vector is a voltage vector, and the confidence with which each check node in the probabilistic graph model transmits its state value to a variable node is determined according to the memory-computing integrated device subarray corresponding to each undirected edge and the driving vector constructed, and the confidence with which each variable node transmits its state value to a check node is updated until a preset iteration stopping condition is met, to determine the optimal estimation value of the transmitted symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task, comprising: For any undirected edge, according to each voltage element in the voltage vector corresponding to the undirected edge, a voltage equal in size to the voltage element is input to each row of nodes in the memristor subarray corresponding to the undirected edge, and the current value in each memristor subarray is read respectively; According to the current value in each memristor subarray read, the confidence with which each check node in the probabilistic graph model transmits its state value to a variable node is determined, and the new confidence with which each variable node transmits its state value to each check node is obtained by updating; It is judged whether the preset iteration stopping condition is reached; If the preset iteration stopping condition is not reached, the new confidence of each variable node transmitting its state value to each check node is taken as the confidence of the current corresponding variable node transmitting its state value to each check node, and the calculation is performed again until the preset iteration stopping condition is reached; If the preset iteration stopping condition is reached, the state value new probability of each variable node is obtained by updating according to the state value initial probability of each variable node in the probabilistic graph model and the confidence of the current corresponding variable node transmitting its state value to each check node, and the optimal estimation value of the transmitted symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task is determined.

8. A MIMO detection system based on a memory computing unit, characterized in that, Comprising: An acquisition unit is configured to acquire a channel matrix and a probabilistic graph model of a to-be-detected MIMO detection task, and determine the state value initial probability of each variable node in the probabilistic graph model according to a multi-antenna mixed signal observed by a receiving end and a noise variance; A node state matrix construction unit is configured to, for any check node in the probabilistic graph model, construct a node state matrix corresponding to each undirected edge in the probabilistic graph model according to each target undirected edge between the check node and each target variable node connected thereto, the channel matrix of the to-be-detected MIMO detection task, and the multi-antenna mixed signal observed by the receiving end; A driving vector construction unit is configured to construct a driving vector corresponding to each undirected edge in the probabilistic graph model according to the confidence with which each variable node in the probabilistic graph model transmits its state value to a check node, wherein in an initial stage, the state value initial probability of each variable node in the probabilistic graph model is taken as the confidence with which the variable node transmits its state value to each check node connected thereto through an undirected edge; A configuration unit is configured to perform memory-computing integrated device configuration based on the node state matrix corresponding to each undirected edge, to obtain a memory-computing integrated device subarray corresponding to each undirected edge. The optimal estimation value determination unit is configured to determine the confidence of each check node in the probabilistic graph model in transmitting its state value to a variable node according to the corresponding memory-computing integrator subarray and the constructed driving vector, and to update the confidence of each variable node in transmitting its state value to a check node until a set iteration stop condition is met, so as to determine the optimal estimation value of the transmission symbol corresponding to the multi-antenna mixed signal in the to-be-detected MIMO detection task.

9. A processing device, characterized by The processing device includes a processor, a memory, a communication interface and a bus, the processor, the memory and the communication interface are connected through the bus to complete the communication among each other; the memory stores a computer program running on the processing device, wherein the computer program is executed by the processing device to implement the steps corresponding to the memory-computing integrator-based MIMO detection method in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer program instructions, wherein the computer program instructions are executed by the processor to implement the steps corresponding to the memory-computing integrator-based MIMO detection method in any one of claims 1-7.

Images