WO2020248092A1

WO2020248092A1 - Blind multipath identification method and system for mimo system based on weighted integrated clustering algorithm

Info

Publication number: WO2020248092A1
Application number: PCT/CN2019/090526
Authority: WO
Inventors: 谢宁; 谭杰
Original assignee: 深圳大学
Priority date: 2019-06-10
Filing date: 2019-06-10
Publication date: 2020-12-17

Abstract

The present disclosure relates to a blind multipath identification method for an MIMO system based on a weighted integrated clustering algorithm. The method comprises: multiple users send communication request signals to a base station; the users adjust the transmission power of the users on the basis of response signals fed back by the base station, so that the base station allows the communication requests of the users; the users send information signals to the base station; the base station generates the input signals of basic clustering algorithms on the basis of the separated information signals, and on the basis of the output result of each basic clustering algorithm, the base station adaptively adjusts the weight of the output result of each basic clustering algorithm by means of a weighted integrated clustering algorithm, and identifies the paths of the information signals of the users according to the outputs of the weighted integrated clustering algorithm; the base station performs the maximum-ratio combining of the users and decodes the information signal of each user. According to the present disclosure, multipath identification can be stably and effectively performed on the basis of a weighted integrated clustering algorithm.

Description

Method and system for blind multipath recognition of MIMO system based on weighted ensemble clustering algorithm

Technical field

The present disclosure relates to the field of wireless communication technology, and in particular to a method and system for blind multipath recognition of a MIMO system based on a weighted ensemble clustering algorithm.

Background technique

With the increasing demand of users for wireless communication data transmission, there is an increasing need for the application of high-quality wireless communication technology. In order to achieve high-speed and high-capacity data transmission requirements, multiple-input multiple-output MIMO technology (Multiple-Input Multiple-Output) ) Is widely concerned by the market, especially multi-user MIMO systems. Multi-user MIMO provides a Space Division Multiple Access (SDMA) architecture for wireless communication, and a multi-user MIMO system can provide huge advantages over traditional point-to-point MIMO systems. In SDMA, multiple users use the same frequency channel to transmit simultaneously, thereby increasing the achievable capacity without the need for additional RF spectrum. One of the main tasks of the SDMA receiver is to distinguish the signal sent by the source.

In order to achieve high-speed and reliable communication, channel identification is required. In the prior art, channel identification is usually implemented through three methods. The three methods are sending training sequences, complex precoder technology and using some special properties of the sent signal. In the first method, too many training sequences cause pilot pollution problems, that is, residual interference may be caused by the reuse of pilot sequences in neighboring cells. In the second method, when the transmitter knows the interference in the channel in advance, the code can be designed to compensate, so that the capacity of the channel is the same as that without interference. However, the second method is not suitable for the actual wireless communication environment, because it can be assumed a priori that there is little information on channel state information (CSI).

The third method takes advantage of the cyclostationary characteristics associated with the virtual channel generated by the time and space oversampling of the received signal. For example, the introduction of iterative least squares with projection and iterative least squares with enumeration algorithms, or the use of limited letter properties of binary shift keying (BSK), phase shift keying (PSK) and quadrature amplitude modulation (QAM) ) Digital modulation format. There is also a single input multiple output (SIMO) system identification strategy extended to the MIMO case. However, the third method has two main problems: first, the transmission signal is a special type rather than a general type, and second, it requires a large enough received data sample, which is not suitable for ultra-reliable and low-latency communication (URLLC). Specifically, URLLC is a new service category supported by 5G New Radio (NR). It is aimed at emerging applications, where data messages are time-sensitive and must be securely end-to-end with high reliability and low latency requirements. deliver. Low latency requirements mean that data transmissions that cannot be decoded at the receiver before the deadline are useless and can be discarded from the system, resulting in a loss of reliability. For low-latency communication, that is, an end-to-end delay of approximately 1ms, short data packets are recommended. Therefore, the third method is not suitable for this situation.

In massive MIMO systems, channel identification is increasingly challenged. The reason is that in addition to the base station (BS) equipped with a large antenna, each user (also called the user terminal) is also equipped with many antennas, which generates a large amount of multipath for each user. In a massive MIMO system, effective combination techniques can be used to improve the quality of message reception. If each path can be accurately classified to the corresponding user, physical layer authentication can be applied to improve the security of the entire system by comparing the CSI of each user in the current time slot with the previous time slot. However, because small-scale multipath fading and receiver noise can cause randomness, and each user has a limit on the transmit power, which cannot be arbitrarily adjusted by the base station, some paths may be difficult to classify.

Summary of the invention

The present disclosure is proposed in view of the above situation, and its purpose is to provide a method and system for blind multipath identification of a MIMO system based on a weighted ensemble clustering algorithm that can perform multipath identification stably and effectively.

To this end, the first aspect of the present disclosure provides a blind multipath identification method of a MIMO system based on a weighted ensemble clustering algorithm, which is a method of MIMO system based on a weighted ensemble clustering algorithm of a wireless communication system including a user terminal and a base station. The blind multipath identification method is characterized in that it includes: a plurality of the user terminals send a communication request signal to the base station; the base station feeds back a response signal to the user terminal based on the communication request signal, and the user terminal is based on The response signal determines whether to adjust the transmit power of the user end so that the base station allows the communication request of each user end; when the base station allows the communication request of each user end, multiple user ends send The base station sends an information signal; the base station separates the information signal through a spatial filter, and the base station generates an input signal of a basic clustering algorithm for any path of any user terminal based on the separated information signal, based on Obtaining the output result of the basic clustering algorithm by the input signal and the basic clustering algorithm, obtaining a corresponding feature matrix based on the output results of each basic clustering algorithm, and obtaining a set matrix based on each feature matrix; the base station Obtain the system probability based on the set matrix, and then obtain the objective function of the weighted ensemble clustering algorithm. The base station adaptively adjusts the weight of the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, And according to the output of the weighted ensemble clustering algorithm, the path of the information signal of the user terminal is identified; and the base station obtains the maximum ratio combination of each user terminal based on all the paths of each user terminal, and Decoding the information signal of each user terminal.

In the present disclosure, when the base station receives multiple user terminals to send communication request signals to the base station and allows the user terminals to send communication requests, multiple user terminals send information signals to the base station; the base station separates the information signals through a spatial filter, and the base station The information signal generates the input signal of the basic clustering algorithm, obtains the output result of the basic clustering algorithm based on the input signal and the basic clustering algorithm, obtains the corresponding feature matrix based on the output results of each basic clustering algorithm, and obtains the set matrix based on each feature matrix ; The base station obtains the system probability based on the ensemble matrix, and then obtains the objective function of the weighted ensemble clustering algorithm. The base station adaptively adjusts the weight of the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, and according to the weighted ensemble clustering The output of the algorithm identifies the path of the information signal of the user end; and the base station obtains the maximum ratio combination of each user end based on all the paths of each user end, and decodes the information signal of each user end. In this case, the base station can obtain a weighted ensemble clustering algorithm based on the basic clustering algorithm, and the weighted ensemble clustering algorithm can perform multipath identification stably and effectively.

In the blind multipath identification method involved in the first aspect of the present disclosure, optionally, the base station obtains the K-th input signal x _k of the basic clustering algorithm of the first path of the user terminal through channel estimation _{, l} (t), the input signal x _k,l (t) of the first path of the K-th user terminal satisfies the formula (I):

Wherein, r _k,l (t) represents the output signal of the _kth user end of the l path separated by the spatial filter,

Represents the channel estimation value of the 1th path of the Kth user end. In this way, the input signal of each path of each user terminal can be obtained, which facilitates subsequent multipath identification.

In the blind multipath identification method involved in the first aspect of the present disclosure, optionally, the base station may adaptively adjust the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm. In the weight, the regularization parameter R is introduced to optimize the weight of the output result of the basic clustering algorithm, and the regularization parameter R satisfies the formula (II):

Among them, ω _i represents the weight of the feature matrix Di of the _i-th path, and N _p represents the basic clustering result. As a result, it is possible to avoid overfitting the vector of weights to a feature matrix.

In the blind multipath identification method involved in the first aspect of the present disclosure, optionally, the base station may adaptively adjust the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm. In the weight, the vector of the weight is set to be equal to the preset value, the path-user tendency matrix B is updated through the first update rule, and after the updated path-user tendency matrix B is obtained, the weighted vector is updated based on the second update rule. Vector. In this way, it is possible to optimize the path-user tendency matrix and the weight vector.

In the blind multipath identification method involved in the first aspect of the present disclosure, optionally, a path-user tendency matrix B is established, and the base station recognizes the path and the path through the path-user tendency matrix B based on the set matrix. For the relationship between the user terminals, the system probability is obtained based on the path-user tendency matrix B and the set matrix, and the system probability satisfies formula (III):

Among them, N _p represents the basic clustering result, ω _i represents the weight of the feature matrix D _i of the i-th path,

Represents a vector of weights, E _i,j represents each element of the set matrix, and represents the relationship between the i-th path and the j-th path of the user end, β _i,k represents the _i-th path of the _kth user end The strength of each path, β _j,k represents the strength of the j-th path of the K-th user terminal,

Indicates the possibility that the i-th path and the j-th path of the k-th client are detected to belong to the same client. In this way, the relationship between the path and the client can be identified and the possibility that the two paths belong to the same client can be judged.

The second aspect of the present disclosure provides a blind multipath recognition system of a MIMO system based on a weighted ensemble clustering algorithm, which is a blind multipath recognition system of a MIMO system based on a weighted ensemble clustering algorithm that includes a user device and a receiving device, It is characterized in that it includes: a plurality of the user devices, which are used to send a communication request signal to the receiving device; and the receiving device, which is used to feed back a response signal to the user device based on the communication request signal, The user device determines, based on the response signal, whether to adjust the transmission power of the user device so that the receiving device allows the communication request of each user device, wherein when the receiving device allows each user device When a communication request is made, a plurality of the user devices send information signals to the receiving device, the receiving device separates the information signals through a spatial filter, and the receiving device generates any one of the information signals based on the separated information signals The input signal of the basic clustering algorithm for any path of the user device, the output result of the basic clustering algorithm is obtained based on the input signal and the basic clustering algorithm, and the corresponding output result is obtained based on the output result of each basic clustering algorithm The feature matrix obtains a set matrix based on each of the feature matrices, the receiving device obtains the system probability based on the set matrix, and then obtains the objective function of the weighted ensemble clustering algorithm, and the receiving device is based on the weighted ensemble clustering algorithm The objective function adaptively adjusts the weight of the output results of each of the basic clustering algorithms, and identifies the path of the information signal of the user device according to the output of the weighted ensemble clustering algorithm, and the receiving device is based on each All the paths of the user equipment are combined by obtaining the maximum ratio of each user equipment, and the information signal of each user equipment is decoded.

In the present disclosure, when the receiving device receives a plurality of user devices to transmit a communication request signal to the receiving device and permits the communication request of the user device, the plurality of user devices transmit information signals to the receiving device; the receiving device separates the information signals through a spatial filter The receiving device generates the input signal of the basic clustering algorithm based on the separated information signal, obtains the output result of the basic clustering algorithm based on the input signal and the basic clustering algorithm, and obtains the corresponding feature matrix based on the output results of each basic clustering algorithm. Each feature matrix obtains the set matrix; the receiving device obtains the system probability based on the set matrix, and then obtains the objective function of the weighted ensemble clustering algorithm. The receiving device adaptively adjusts the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm. Weight, and identify the path of the information signal of the user device according to the output of the weighted ensemble clustering algorithm; and the receiving device obtains the maximum ratio combination of each user device based on all the paths of each user device, and decodes each user device’s Information signal. In this case, the receiving device can obtain a weighted ensemble clustering algorithm based on the basic clustering algorithm, and the weighted ensemble clustering algorithm can stably and effectively perform multipath identification.

In the blind multipath identification system involved in the second aspect of the present disclosure, optionally, the receiving device obtains the K-th input signal of the basic clustering algorithm of the first path of the user device through channel estimation x _{k ,l} (t), the input signal x _k,l (t) of the first path of the K-th user device satisfies the formula (I):

Where r _k,l (t) represents the output signal of the _kth user device's lth path separated by the spatial filter,

Represents the channel estimation value of the kth user equipment's first path. In this way, the input signals of each path of each user device can be obtained, which facilitates subsequent multipath recognition.

In the blind multipath recognition system involved in the second aspect of the present disclosure, optionally, the receiving device adaptively adjusts the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm In the weight, the regularization parameter R is introduced to optimize the weight of the output result of the basic clustering algorithm, and the regularization parameter R satisfies the formula (II):

In the blind multipath recognition system involved in the second aspect of the present disclosure, optionally, the receiving device adaptively adjusts the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm In the weights, the vector of weights is set equal to the preset value, the path-user tendency matrix B is updated through the first update rule, and after the updated path-user tendency matrix B is obtained, the vector of weights is updated based on the second update rule. In this way, it is possible to optimize the path-user tendency matrix and the weight vector.

In the blind multipath recognition system involved in the second aspect of the present disclosure, optionally, a path-user tendency matrix B is established, and the receiving device recognizes the path and the path through the path-user tendency matrix B based on the set matrix. For the relationship between the user devices, a system probability is obtained based on the path-user tendency matrix B and the set matrix, and the system probability satisfies the formula (III):

Among them, E represents the set matrix, N _p represents the basic clustering result, ω _i represents the weight of the feature matrix Di of the _i-th path,

Represents a vector of weights, E _i,j represents each element of the set matrix, and represents the relationship between the i-th path and the j-th path of the user equipment, β _i,k represents the k-th user equipment The strength of i paths, β _j,k represents the strength of the j-th path of the K-th user device,

It indicates the possibility of detecting that the i-th path and the j-th path of the k-th user device belong to the same user device. In this way, the relationship between the route and the user device can be recognized and the possibility that the two routes belong to the same user device can be judged.

Compared with the prior art, the examples of the present disclosure have the following beneficial effects:

Because the performance of different clustering algorithms on the same system (such as the same number of multiple clients and the same number of multiple paths) is basically different, and even the same algorithm may have different performance under different channel conditions . In addition, because the clustering results of different clustering algorithms can describe the basic mode of a multi-user MIMO system from different aspects, a single clustering algorithm can usually only capture one aspect of the uplink system. Therefore, the present disclosure proposes a method and system for blind multipath recognition of a MIMO system based on a weighted ensemble clustering algorithm. The blind multipath recognition method involved in the present disclosure can integrate the basic clustering results of different basic clustering algorithms to generate accurate and reliable clustering results. Compared with the traditional weighted ensemble clustering algorithm that processes each basic clustering result equally, the weighted ensemble clustering algorithm involved in the present disclosure can adaptively adjust the optimal weights of different basic clustering results. Thus, the basic clustering results that obtain higher weights may be more reliable and can be regarded as important features. However, the basic clustering result with lower weight may be a less reliable feature, and the cluster that produces the basic clustering result with lower weight may be far from the actual situation. In this case, with these weights, it is possible to perform selection between the elements (base clustering results) and produce a more reliable output. The blind multipath recognition method involved in the present disclosure can perform multipath recognition stably and effectively.

Description of the drawings

FIG. 1 is a schematic diagram showing the signal transmission between the user terminal and the base station of the blind multipath identification method of the MIMO system based on the weighted ensemble clustering algorithm involved in an example of the present disclosure.

FIG. 2 is a schematic flowchart showing a method for blind multipath identification of a MIMO system based on a weighted ensemble clustering algorithm involved in an example of the present disclosure.

FIG. 3 is a schematic diagram showing the blind multipath identification method of the MIMO system based on the weighted ensemble clustering algorithm involved in the example of the present disclosure using the weighted ensemble clustering algorithm for blind multipath identification.

4 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Rayleigh channel under the condition of the difference of the received signal-to-noise ratio in the blind multipath identification method involved in the example of the present disclosure.

5 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Nakagami channel under the condition of different parameters m in the blind multipath identification method involved in the example of the present disclosure.

6 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Rayleigh channel under different initial trade-off parameters in the blind multipath identification method involved in the example of the present disclosure.

FIG. 7 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Rayleigh channel under different scale factors of estimation errors in the blind multipath identification method involved in the example of the present disclosure.

FIG. 8 is a schematic structural diagram showing a blind multipath recognition system of a MIMO system based on a weighted ensemble clustering algorithm involved in an example of the present disclosure.

Detailed ways

Hereinafter, the preferred embodiments of the present disclosure will be described in detail with reference to the drawings. In the following description, the same symbols are assigned to the same components, and repeated descriptions are omitted. In addition, the drawings are only schematic diagrams, and the ratio of dimensions between components or the shapes of components may be different from actual ones.

The present disclosure relates to a blind multipath recognition method of a MIMO system based on a weighted ensemble clustering algorithm, which may be a blind multipath recognition method of a MIMO system based on a weighted ensemble clustering algorithm of a wireless communication system including a user terminal and a base station. In the present disclosure, multipath recognition can be performed stably and effectively based on the weighted ensemble clustering algorithm. The present disclosure will be described in detail below with reference to the accompanying drawings.

FIG. 1 is a schematic diagram showing the signal transmission between the user terminal and the base station of the blind multipath identification method of the MIMO system based on the weighted ensemble clustering algorithm involved in an example of the present disclosure. In some examples, as shown in FIG. 1, the blind multipath identification method of the MIMO system based on the weighted ensemble clustering algorithm can be applied to a signal transmission model including multiple user terminals and a base station. Among them, multiple user terminals may be located in a cell covered by the base station. Multiple users can transmit signals with the base station through wireless communication.

In some examples, in the uplink of the multi-user MIMO system shown in FIG. 1, there may be K multiple user terminals, and K is a natural number. Each user terminal is equipped with multiple antennas. It is assumed that the base station shown in Figure 1 has a large enough antenna to provide a strong spatial resolution capability. There are L _k independent paths between the k-th user terminal and the base station. For example, there are L ₁ independent paths between the first user terminal and the base station. That is, the k-th user terminal has L _k resolvable paths at the base station. The total number of resolvable paths N _{L is} expressed as

In some examples, the user terminal shown in FIG. 1 may include, but is not limited to, user equipment. In some examples, user equipment may include, but is not limited to, smart phones, notebook computers, personal computers (PCs), personal digital assistants (PDAs), mobile Internet devices (MIDs), and wearable devices. Devices (such as smart watches, smart bracelets, smart glasses) and other electronic devices. Among them, the operating system of the user device can include but not limited to Android operating system, IOS operating system, Symbian operating system, BlackBerry (Blackberry) operating system, Windows Phone8 operating system, etc. The base station shown in FIG. 1 may include, but is not limited to, a device that communicates with a wireless terminal through one or more sectors on an air interface in an access network. The base station can be used to convert received air frames and IP packets into each other, and act as a router between the wireless terminal and the rest of the access network, where the rest of the access network can include an Internet Protocol (IP) network. The base station can also coordinate the attribute management of the air interface. For example, the base station can be a base station (BTS, Base Transceiver Station) in GSM or CDMA, a base station (NodeB) in WCDMA, or an evolved base station (NodeB or eNB or e-NodeB, evolutional NodeB) in LTE. B).

In some examples, in the signal transmission model shown in FIG. 1, a short frame structure can be used for signal transmission between the user terminal and the base station. The short frame structure can be transmitted through the wireless channel. The wireless channel can be a memoryless block fading channel. Since the transmission duration of the short frame is relatively short, the channel fading can be kept constant during one data frame, but the channel fading can be different for different data frames and different paths. Among them, channel fading can include large-scale path loss and small-scale complex fading coefficients.

In some examples, in the signal transmission model as shown in FIG. 1, a spatial filter may be provided at the base station. The base station receives the transmitted signals from different users, and the output signal of the spatial filter of the base station satisfies formula (1):

Where α _k represents the large-scale path loss of the k-th user terminal, h _k,l represents the small-scale complex fading coefficient of the l-th path of the k-th user terminal, P _k represents the transmission power of the k-th user terminal, and s _k ( t) represents the transmitted signal of the k-th user terminal, n _k,l (t) is the residual noise of the k-th user terminal's l path through the spatial filter, which satisfies

In some examples, for the large-scale path loss in the output signal of the spatial filter of the base station, the large-scale path loss α _{k at the kth} user end can satisfy

Among them, the path loss index α _d satisfies α _d ≥2. d _k can represent the distance between the user terminal and the base station. The wavelength λ of the transmitted signal satisfies λ=c/f _c , c=3×10 ⁸ m/s, and f _c represents the carrier frequency of the transmitted signal. The large-scale path loss α _{k of} the k-th user end can be determined by the distance d _k , which has nothing to do with a specific path. In this disclosure, it is assumed that the location of each user remains unchanged.

In some examples, for the small-scale complex fading coefficients in the output signal of the spatial filter of the base station, in the additive white Gaussian noise (AWGN) channel, the small-scale path of the k-th user terminal in the channel fading is The complex fading coefficient h _k,l satisfies h _k,l =1.

In some examples, in Rayleigh and Nakagami-m channels, the probability density function (PDF) of the small-scale complex fading coefficient |h| satisfies f _Ray (|h|)=2|h|exp(-| h| ² ) and

Among them, m∈[1/2,∞) and Γ(·) are Gamma functions. In addition, the Nakagami-m fading channel is widely used when modeling wireless communication channels. For example, the Nakagam-m distribution is often used in land mobile and indoor mobile multipath propagation and flashing ionospheric radio links, where the parameter m can be adjusted to represent different scenarios.

In some examples, the channel corresponding to the smaller m value has severe fading. Under the limit m→∞, the Nakagami-m fading channel is close to the non-fading additive white Gaussian noise (AWGN) channel. In addition, the Nakagami-m distribution includes a one-sided Gaussian distribution with m=1/2 and a Rayleigh distribution with m=1. For Rayleigh and Nakagami-m channels, the two channels can be modeled as a uniform distribution between [0,2π].

In some examples, the base station can extract channel fading from each path as the main feature. The present disclosure is not limited to this, and propagation delay may also be an important feature of each path.

In some examples, in order to realize multipath identification without using different pilots of different clients, in the case that each client has no previous CSI and no message is detected in each path, the present disclosure proposes A blind multipath recognition method of MIMO system based on weighted ensemble clustering algorithm. The blind multipath identification method of the MIMO system based on the weighted ensemble clustering algorithm involved in the present disclosure may be simply referred to as the blind multipath identification method.

FIG. 2 is a schematic flowchart showing a method for blind multipath identification of a MIMO system based on a weighted ensemble clustering algorithm involved in an example of the present disclosure. FIG. 3 is a schematic diagram showing the blind multipath identification method of the MIMO system based on the weighted ensemble clustering algorithm involved in the example of the present disclosure using the weighted ensemble clustering algorithm for blind multipath identification.

In some examples, as shown in FIG. 2, the blind multipath identification method includes multiple user terminals sending communication request signals to the base station (step S10). In step S10, based on the signal transmission model shown in FIG. 1, each user terminal may send a communication request signal to the base station. The communication request signal may have a short frame structure. In addition, the communication request signal transmitted by each user terminal can reach the base station through the fading channel without memory block.

In some examples, as shown in FIG. 2, the blind multipath identification method may include the base station feeding back a response signal to the user end based on the communication request signal, and the user end determines whether to adjust the transmit power of the user end based on the response signal, so that the base station allows each A communication request from the user side (step S20). In step S20, based on the signal transmission model shown in FIG. 1, the base station may receive the communication request signal. The base station may include a user registration database. The base station checks whether the communication request signal of each user terminal is legal through the user registration database.

In some examples, when the communication request signal of each user terminal received by the base station is illegal, the base station may interrupt the communication with each user terminal. If the communication request signal of each user end received by the base station is legal, the base station estimates the communication request signal and calculates the received signal-to-noise ratio γ _{k of} each user end. That is, the base station estimates the scaled large-scale path loss P _k |α _k | ^{2 of} each user terminal based on the communication request signal. Specifically, after the communication request signal transmitted by each user terminal reaches the base station, the communication request output signal can be obtained through the spatial filter of the base station. The communication request output signal can be obtained by formula (1), where the transmission signal s _k (t) of the k-th user terminal can be a communication request signal. Hypothesis

Based on the communication request output signal, the scaled large-scale path loss of each user terminal can be obtained to obtain the received signal-to-noise ratio (SNR) of each user terminal. The received signal-to-noise ratio γ _{k of} the kth user terminal satisfies the formula (2): γ _k =P _k |α _k | ² (2), where α _k represents the large-scale path loss of the kth user terminal, and P _k represents the The transmit power of k user terminals. For the large-scale path loss α _{k of} the k-th user terminal, refer to the relevant detailed description in FIG. 1 above. In step S20, the base station can calculate the difference between the received signal-to-noise ratios of any two user terminals. For example, the base station may calculate the difference between the received signal-to-noise ratio γ _k of the k-th user terminal and the received signal-to-noise ratio γ _j of the j-th user terminal, where the k-th user terminal and the j-th user terminal are different users. , That is, k≠j. The difference Δ _k,j between the received signal-to-noise ratio γ _k of the k-th user terminal and the received signal-to-noise ratio γ _j of the j-th user terminal satisfies formula (3): Δ _k,j =|γ _k -γ _j | (3 ). The number of differences Δ _k,j can be multiple.

In step S20, the base station may compare any difference with the set threshold ε _Δ , and feed back a response signal to the user terminal based on the comparison result. The response signal may include a first response signal and a second response signal. The user terminal determines whether to adjust the transmit power based on different response signals. That is, the base station feeds back different response signals to the user terminal based on the comparison result to adjust the transmit power of the user terminal. Specifically, if each difference calculated by the base station is greater than the set threshold ε _Δ , the base station feeds back the first response signal to the user terminal and allows the user terminal to request communication. Each user terminal receives the first response signal and maintains the transmission power. If the difference calculated by the base station is less than or equal to the set threshold ε _Δ , the base station feeds back the second response signal to the user terminal, and the user terminal receives the second response signal and adjusts the transmission power to satisfy that the difference is greater than the set threshold ε _Δ , so that the base station Allow communication requests from the user side. That is, after the user terminal receives the second response signal and adjusts the transmission power, it retransmits the communication request signal to the base station. The base station recalculates each difference and compares it with the set threshold ε _Δ until each difference is greater than the set threshold. Set the threshold ε _Δ , the base station allows the communication request from the user end. In this case, by comparing the difference with the set threshold ε _Δ and meeting the requirements, it can be ensured that each path of each user terminal is correctly identified in the subsequent.

In some examples, the base station can control the power of each user terminal through automatic power control. For example, the radio frequency signal received by the transceiver station of the base station is sequentially input into a filter and a frequency converter with filtering function to obtain an intermediate frequency signal, and then the intermediate frequency signal is input into the automatic power control module of the base station to control the power. Among them, the automatic power control module includes an A/D converter, a DC removing unit, a power estimation unit, and a power feedback adjustment unit. The automatic power control process of the automatic power control module includes: passing the intermediate frequency signal through an A/D converter to obtain a digital signal, the digital signal passes through a variable-point DC removing unit to obtain a zero-average digital intermediate frequency signal, and the digital intermediate frequency signal passes through points The variable power estimation unit obtains the power estimation of the signal. The power estimation value passes through the power feedback adjustment unit to obtain a new gain coefficient value. The new gain coefficient is applied to the limit adjustment process in the next time period, and finally the digital intermediate frequency signal The output is maintained near the stable power.

In some examples, the base station can stabilize the received signal before sending it out, which can effectively reduce or avoid the loss of the communication signal during wireless transmission and ensure the user's communication quality. In some examples, the base station can use frequency division multiplexing to allocate the number of channels used. When the available bandwidth of a physical channel exceeds the bandwidth required by a single information signal, the total bandwidth of the physical channel can be divided into several sub-channels that have the same bandwidth as the transmission of a single information signal. The corresponding information signal is transmitted on each sub-channel to realize the simultaneous transmission of multiple information signals (multi-channel signals) in the same channel. Before the frequency division multiplexing of multiple signals, the spectrum of each signal needs to be moved to different segments of the physical channel spectrum through spectrum shifting technology, so that the bandwidth of each information signal does not overlap with each other. After shifting the spectrum, it is necessary to modulate each signal with a different carrier frequency. Each signal is centered on its corresponding carrier frequency and is transmitted on a sub-channel with a certain bandwidth. In addition, in order to prevent mutual interference, it is necessary to use anti-interference protection measures to isolate each sub-channel.

In some examples, as shown in FIG. 2, the blind multipath identification method may include when the base station allows each user terminal's communication request, multiple user terminals send information signals to the base station (step S30). In step S30, when the base station permits the communication request of each user terminal, all the user terminals simultaneously send message signals to the base station through the same frequency channel. Based on the signal transmission model shown in Figure 1, there are multiple independent paths between each user terminal and the base station, and the base station does not know the number of paths on the user terminal. Each user terminal sends a message signal to the base station through corresponding multiple independent paths.

In some examples, as shown in Figure 2, the blind multipath recognition method may include the base station separating information signals to obtain the input signal of the basic clustering algorithm, and then to obtain the output result of the basic clustering algorithm, based on each basic clustering algorithm The output results obtain the corresponding feature matrix and the set matrix. The base station obtains the system probability based on the set matrix, and then obtains the objective function of the weighted ensemble clustering algorithm, and then adaptively adjusts the weights of the output results of each basic clustering algorithm, and gathers them according to the weighted ensemble. The output of the similar algorithm identifies the path of the information signal on the user side (step S40). Specifically, the base station separates the information signal through the spatial filter. The base station generates the input signal of the basic clustering algorithm for any path of any user terminal based on the separated information signal, and obtains the basic clustering algorithm based on the input signal and the basic clustering algorithm. The output result is based on the output results of each basic clustering algorithm to obtain the corresponding feature matrix, based on each feature matrix to obtain the set matrix, the base station obtains the system probability based on the set matrix, and then obtains the objective function of the weighted ensemble clustering algorithm. The objective function of the class algorithm adaptively adjusts the weight of the output results of each basic clustering algorithm, and identifies the path of the information signal of the user terminal according to the output of the weighted ensemble clustering algorithm.

In step S40, based on the signal transmission model shown in Figure 1, the base station can separate each path through a spatial filter. From the above, it can be seen that the base station has enough large-scale antennas to provide strong spatial resolution capabilities, so most The path is spatially distinguishable. In this case, the base station can separate the information signal and obtain the information signal in each path. The spatial filter of the base station can capture the multipath behind all spatially distinguishable paths.

In some examples, the pilot signal can be used to assist channel estimation. The estimation error can be determined by the received signal-to-noise ratio (SNR). Channel estimate

Satisfy formula (4):

among them,

Represents the estimation error of the l-th path of the Kth client and is modeled as

ρ represents the scale factor of the estimation error. In this way, the channel estimation value of each path of each user end can be obtained. Base station use channel estimate

And the output signal of the spatial filter generates the input signal of the clustering algorithm of each path. That is, the base station can obtain the input signal x _k,l (t) of the clustering algorithm for the l path of the κ-th user terminal through channel estimation, and the input signal x _k,l (t) satisfies the formula (5):

Among them, r _k,l (t) represents the output signal of the l path of the Kth user end separated by the spatial filter, and the output signal can be obtained by formula (1) based on the transmitted signal. The transmitted signal may be the information signal of each path of each user terminal.

Represents the channel estimation value of the l-th path of the Kth user end. In this way, the input signal of each path of each user terminal can be obtained, which facilitates subsequent multipath identification. In some examples, when there is no estimation error, that is, when the scale factor ρ of the estimation error is zero, the input signal can satisfy equation (6):

Based on equation (6), a constellation diagram (not shown) under the condition that each user end has the same number of multipaths can be drawn. Each user terminal has its own distribution area, and there is an overlapping area between each user terminal. Then the problem of multipath recognition can be transformed into a problem of unsupervised learning. Since the classic clustering algorithm cannot be directly applied to a normal constellation diagram with all constellation points, one constellation point in the constellation diagram can be selected as the input signal of the clustering algorithm.

In some examples, as shown in FIG. 3, there may be K user terminals. Each client can have L _k paths. After the base station obtains the input signal of the basic clustering algorithm for each path through the above steps, it inputs the corresponding input signal of each path into the weighted ensemble clustering algorithm, and each input signal passes through the basic clustering algorithm to obtain the basic clustering algorithm The output result. That is, the output result of the basic clustering algorithm is obtained based on the input signal and the basic clustering algorithm.

In some examples, various basic clustering algorithms can be divided into different categories in the weighted ensemble clustering algorithm. The classification can be developed from the perspective of the algorithm designer, focusing on the technical details of the general process of the clustering process. The choice of basic clustering algorithm usually considers three categories. Specifically, the base station can divide various basic clustering algorithms into three types: partition-based, model-based, and layer-based. In the partition-based method, all basic clustering algorithms are quickly determined. For example, K-means clustering algorithm (K-means), K-medoids clustering algorithm (K-medoids) and spectral clustering (SC). In a model-based method, data can be formed by mixing basic probability distributions, such as Gaussian Mixture Model (GMM); in a layer-based method, data can be organized in a hierarchical manner based on proximity media, such as agglomerated hierarchical clustering (AHC). Intermediate nodes can obtain proximity. Stratification-based methods can be condensed (bottom-up) or split (top-down). Agglomerative clustering starts with one object in each cluster and recursively merges two or more most suitable clusters. Split clustering starts from the data set as a cluster, and recursively splits the most suitable cluster.

In some examples, the basic clustering algorithm selected may be spectral clustering (SC), K-means clustering algorithm (such as K-means and K-means (Cityblock)), K-center point clustering algorithm (such as K-medoids And K-medoids (Cityblock)), Gaussian mixture model (GMM), agglomerative hierarchical clustering algorithm (AHC (Complete), AHC (Single), AHC (Average) and AHC (Weighted)). The main features corresponding to the above 10 basic clustering algorithms are the eigenvalues of the similarity matrix, the squared Euclidean distance, the sum of absolute differences, the squared Euclidean distance, the sum of absolute differences, the regularization value 10 ² , the maximum Long distance, shortest distance, unweighted average distance, and weighted average distance. The basic clustering algorithm selected in the examples of the present disclosure is not limited to the basic clustering algorithms listed above. In some examples, multiple basic clustering results can be obtained according to the aforementioned 10 basic clustering algorithms. The number of basic clustering results can be denoted by N _b . Then, multiple basic clustering results (that is, the output results of the basic clustering algorithm) are integrated through the following weighted integrated clustering algorithm and subsequent processing is performed.

In step S40, in the weighted ensemble clustering algorithm, a corresponding feature matrix can be obtained based on the output results of each basic clustering algorithm. Specifically, given the uplink (UL) system shown in FIG. 1 and the basic clustering results N _p , each basic clustering result can be regarded as a feature of the UL system. In order to analyze these features, a feature matrix can be generated from each basic clustering result. The i-th feature matrix D _i (i=1, 2,..., N _p ) may be an adjacency matrix. Wherein the i-th matrix D _i Each entry may indicate whether a respective paths together. The i-th feature matrix _Di may be a block diagonal matrix after arrangement. The set matrix E is obtained based on each feature matrix. The set matrix can satisfy formula (7):

Satisfy

Represents a vector of weights, and D _i represents each feature matrix. In this way, a set matrix can be obtained. The aggregate matrix can effectively use the information provided by each feature matrix.

In some examples, the problem of obtaining the set matrix can be regarded as the problem of learning the best linear combination of the feature matrix. In addition, in order to avoid overfitting the weight vector W to an eigen matrix, that is, the weight coefficient of one eigen matrix is one and the weight coefficients of all other eigen matrices are zero, a regularization parameter R can be introduced. The regularization parameter R satisfies formula (8):

Among them, ω _i represents the weight of the feature matrix Di of the _i-th path, and N _p represents the basic clustering result. The regularization parameter R can represent the sum of the negative entropy of the weight of the feature matrix Di of the _i-th path. The regularization parameter R can penalize the single feature matrix with the largest weight. The acquisition of the above-mentioned feature matrix and set matrix and the introduction of the regularization parameter R can be regarded as the generation stage of the weighted ensemble clustering algorithm.

In step S40, the set matrix E may represent the relationship between the results of the information signal path of the user end. Each entry E _{i,j of the} set matrix E can represent the relationship between the i-th path and the j-th path on the user side. In an uplink system, when there is a stronger interaction between the paths of two information signals, it is more likely that the two information signals come from the same user terminal. Assuming that the uplink system includes K users, for the i-th path observed in the uplink system, a parameter β _i,k can be introduced to represent the value from the kth (k=1, 2,...,K ) Observe the strength of the i-th path at the user end. The i-th path observed in the uplink system may have a higher parameter β _i,k on multiple users. The path-user tendency matrix B is established based on the parameters β _i,k . The base station can identify the relationship between the path and the user terminal through the path-user tendency matrix B based on the aggregate matrix, and the path-user tendency matrix B satisfies the formula (9): B= [β _i,k ](9), where β _i,k represents the strength of the i-th path of the K-th user terminal. In this way, the relationship between the path and the client can be identified.

Indicates the possibility of detecting that the i-th path and the j-th path of the Kth client belong to the same client. As a result, the possibility that the paths of the two information signals belong to the same user terminal can be represented by a matrix. In addition, the parameter E _i,j can represent the possibility that the observed i-th path and the observed j-th path belong to the same user terminal. Therefore, the paired interaction described by the collective matrix E is affected by the unobserved non-negative parameter Q=BB ^T. Each element β _i,k β _j,k can indicate the contribution of the k-th user terminal to the non-negative parameter Q _i,j .

In some examples, assume that the likelihood of parameters E _i,j is determined by

Given, where

Is the Poisson probability density function and Γ(·) is the Gamma function. In some examples, the system probability P(E|B) can be obtained based on the aforementioned set matrix E and path-user tendency matrix B. The system probability P(E|B) satisfies formula (10):

Represents the vector of weight, E _i,j represents each element of the set matrix, and represents the relationship between the i-th path and the j-th path on the user side,

Indicates the possibility that the i-th path and the j-th path of the k-th client are detected to belong to the same client. Thus, it is possible to determine the possibility that the two paths belong to the same client. For the uplink system, the system probability defined by equation (10) can be maximized to estimate the path-user tendency matrix B, and the objective function of the weighted ensemble clustering algorithm can be obtained by taking the negative logarithm and the descending constant. The objective function J(B) satisfies formula (11):

Satisfies _{β i, k ≥0 (11)} , where, N _L represents the number of each client's path, Γ (·) indicates the gamma (the Gamma) function. Thus, the objective function of the weighted ensemble clustering algorithm can be specifically obtained. Among them, the third equal sign can be established because Γ(1)=Γ(2)=1, and the value of the parameter E _i,j is between 0 and 1. In some examples, suppose Γ(E _i,j +1)=1.

In step S40, since the above-mentioned uplink system is a weighted undirected system, and the Bayesian NMF model assumes that the joint membership of two paths in the same user terminal can increase the probability of a link between the two paths. In formula (9) to formula (11), the regularization parameter R is introduced, and formula (8) is substituted for the set matrix E, then the objective function J(B) of formula (11) can be transformed to obtain the transformation objective function J(W ,B), the transformation objective function J(W,B) satisfies:

Satisfy

Among them, λ is a trade-off parameter and satisfies λ≥0. The trade-off parameter can control the balance between the objective function J(B) and the regularization parameter R in equation (11). In some examples, the base station adaptively adjusts the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm. Among them, the process of adaptive adjustment can be regarded as the minimization of the conversion objective function J(W, B) of the constraint equation (12) to form a constrained nonlinear optimization problem. In the optimization process, alternately update the path-user tendency matrix B and the weight vector W, and repeat this backup update process until the result converges.

Specifically, in the base station adaptively adjusting the weights of the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, the vector W of the weights can be made equal to the preset value, and the path-user is updated through the first update rule. Propensity matrix B, the first update rule satisfies formula (13):

Among them, β _i,k represents the strength of the i-th path of the K-th client, β _j,k represents the strength of the j-th path of the K-th client, and β _i,l represents the i-th path of the l-th client _strength, β _{j, l} l represents the strength of the user terminal j-th path, N _L represents the number of each client's path. In this way, the route-user tendency matrix can be optimized. Among them, since the weight vector W is equal to the preset value, J(B) is minimized and the path-user tendency matrix B is obtained. The first update rule may be a multiplicative update rule. After obtaining the updated path-user tendency matrix B, the weight vector W can be updated based on the second update rule, which satisfies formula (14):

Among them, λ represents the trade-off parameter, the non-negative parameter satisfies Q=BB ^T , and D _m represents the characteristic matrix of the m-th path. Thus, the vector of weights can be optimized. In addition, if the path-user tendency matrix B is initialized with a non-negative value, the path-user tendency matrix B during the iteration will always be non-negative. In addition, in the base station adaptively adjust the weight of the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, the regularization parameter R can be introduced to optimize the weight of the output result of the basic clustering algorithm, and the regularization parameter R satisfies Formula (8). As a result, it is possible to avoid overfitting the vector of weights to a feature matrix. The acquisition of the system probability, the objective function and the update process of the path-user tendency matrix B and the weight vector W can be regarded as the path detection stage of the weighted ensemble clustering algorithm.

Based on the weighted ensemble clustering algorithm described in step S40, the input parameters of the weighted ensemble clustering algorithm shown in Figure 3 are independent basic clustering results _Ai (i = 1, 2,..., N _b ), user The number of terminals K, the weighing parameter λ; the output parameters are the path-user tendency matrix B, the weight vector W, and the minimum objective function s. The brief steps of the weighted ensemble clustering algorithm can include: obtaining different clustering results by applying different basic clustering algorithms on the uplink system; initializing the weight vector W and the path-user tendency matrix B; when the stopping criterion is met Under the condition of, construct the set matrix E; update the path-user tendency matrix B according to formula (13); update the weighted vector W according to formula (14); calculate the minimum value s of the objective function of formula (12); return the weighted vector W , Path-user tendency matrix B, minimum objective function s. In addition, the first update rule and the second update rule are used to iteratively update the weighted vector W and the path-user tendency matrix B, and the algorithm stops when the number of iterations reaches 150.

In some examples, in order to avoid local minima, the weighted ensemble clustering algorithm can be repeated 10 times with random initial conditions, and the minimum value s of the objective function can be selected to output. And, when it is determined that the value of each weight satisfies ω _i =1/N _p , then the set matrix E satisfies

In this case, the feature matrix of the i-th path is treated equally.

In some examples, the weighted ensemble clustering algorithm may predefine a trade-off parameter λ. The trade-off parameter λ can control the influence of the regularization parameter R, and the trade-off parameter λ can control the relative difference between the characteristic matrices. For example, setting λ=∞ can force all basic clustering algorithms to have equal weights. Set λ=0, and the regularization parameter is not used at this time. Through the second update rule, it can be determined that the effect of the trade-off parameter λ depends on

In order to facilitate the selection of the trade-off parameter λ, the trade-off parameter λ can be set to

Proportionally. That is, the weighing parameter λ satisfies formula (15):

In some examples, the initial trade-off parameter λ ₀ value is changed and the corresponding performance is evaluated to determine the appropriate trade-off parameter λ value. If the λ ₀ value is large enough, the weights assigned to each basic clustering result are almost equal. Therefore, the performance of the weighted ensemble clustering algorithm will not change significantly as the initial trade-off parameter λ ₀ increases. In other examples, when a part of the basic clustering results are poor, a small part of the initial trade-off parameter λ ₀ changes will cause a large change in performance.

In some examples, the total time of equation (13) and equation (14) in the update process can be calculated. The time to update the path-user preference matrix B is

And the time to update the weighted vector W is

Therefore, the total time of the proposed weighted ensemble clustering algorithm is

Where T is the number of iterations. Because the path-user tendency matrix B is sparse, the real-time is much smaller than

In addition, although the basic clustering result N _p is time-consuming, with the rapid development of computer hardware, a large number of operations can be performed. In addition, the generation process of basic clustering results can be processed in parallel. Through parallel computing, different clustering results can be generated simultaneously on modern multi-core processors, and the running time can be shortened.

In step S40, the path of the information signal of the user terminal is identified according to the output of the weighted ensemble clustering algorithm. The output of the weighted ensemble clustering algorithm is the ensemble clustering result. The result of ensemble clustering can be considered as a set of basic clustering algorithms after adjusting the weights. The base station can divide the output of the weighted ensemble clustering algorithm into clusters. Each cluster contains all the paths of the information signal of each user. Then blind multipath recognition is performed on the cluster to realize the blind multipath recognition of each path of each user end. As shown in Figure 3, the number of clusters M can be determined according to a weighted ensemble clustering algorithm. In some examples, M=K can be set, where K is the number of user terminals available in the registration phase. The number of paths per cluster is unknown. In this case, a suitable performance measurement can be selected to accurately evaluate and fairly compare the performance of various clustering algorithms. In some examples, the cluster corresponding to the path of the information signal of each user end includes at least one path, and each user end finally has only one cluster. These clusters can meet two requirements: one, each group must contain at least one object; second, each object must belong to a group.

In the present disclosure, the performance of the blind multipath recognition method can be evaluated by judging the degree of correspondence between the predicted multipath and the known multipath. Among them, the Jaccard metric can be used to evaluate the similarity between a set of predicted multipaths and a set of reference multipaths. The Jaccard metric can measure the degree to which the predicted multipath corresponds to the reference multipath of the user-path pair level, and consider the number of paths for each user. The value of the Jaccard metric varies between 0 and 1. The higher the value, the better the performance. In an ideal channel, the value of the Jaccard metric is equal to 1, which can show that the base station correctly recognizes all paths. In actual channels, especially under bad channel conditions, the value of the Jaccard metric will decrease. In some examples, the base station can set the path for the base station to identify the user end when the received signal-to-noise ratio of the user end reaches a certain standard. Specifically, the base station can set the threshold of the value of the Jaccard metric, where the threshold ε _J can be set as the lower limit of the path that the base station can correctly identify. If Jaccard≥ε _J , then the performance of blind multipath recognition meets the requirements.

In some examples, as shown in Figure 2, the blind multipath identification method of the MIMO system based on the clustering algorithm may include the base station obtaining the maximum ratio combination of each user terminal based on all paths of each user terminal, and decoding the information of each user terminal Signal (step S50). In step S50, the base station can collect all paths of each user terminal and perform maximum ratio combining (MRC) for each user terminal to improve the received signal-to-noise ratio. In some instances, for example, provided ε _J = 0.95, and when Jaccard≥ε _J, the base station can properly collect more than 95% of the path for each subscriber in order to improve the final properties of the maximum ratio combining (the MRC) (i.e., final Receive signal-to-noise ratio), and the remaining path (less than 5% of the path) can be considered as additional path noise. In addition, the base station can receive the information signal from the user terminal and decode the information signal to complete the uplink transmission of the multi-user MIMO system. Steps S10 to S20 can be regarded as the registration phase in the uplink data transmission method based on the MIMO system. Steps S30 to S50 can be regarded as the message transmission phase in the uplink data transmission method based on the MIMO system.

4 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Rayleigh channel under the condition of the difference of the received signal-to-noise ratio in the blind multipath identification method involved in the example of the present disclosure. 5 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Nakagami channel under the condition of different parameters m in the blind multipath identification method involved in the example of the present disclosure. 6 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Rayleigh channel under different initial trade-off parameters in the blind multipath identification method involved in the example of the present disclosure. FIG. 7 is a schematic diagram showing the waveforms of the Jackard metric value and the received signal-to-noise ratio in the Rayleigh channel under different scale factors of estimation errors in the blind multipath identification method involved in the example of the present disclosure. In some examples, as shown in Figures 4 to 7, the number of user terminals is set to three, and each user terminal has the same number of paths and satisfies L=L ₁ =L ₂ =L ₃ , and set the path loss index α _d = 2 and the carrier frequency of the transmitted signal f _c = 2 GHz.

In some examples, the waveform A shown in FIG. 4 is a waveform at Δ=0.5dB. The waveform B is the waveform at Δ=0.8dB. The waveform C is the waveform at Δ=1dB. The waveform D is the waveform at Δ=2dB. The waveform E is the waveform at Δ=3dB. As shown in Figure 4, when the received signal-to-noise ratio of the user terminal increases, the value of the Jaccard metric of the weighted ensemble clustering algorithm increases. When the received signal-to-noise ratio or the difference Δ of the received signal-to-noise ratio becomes large enough, the increasing trend of the value of the Jaccard metric gradually decreases and finally stabilizes. For example, if γ ₁ ≥ 16dB, the value of the Jaccard metric at Δ=2dB is equal to the value at Δ=3dB. Therefore, for areas with a high received signal-to-noise ratio, the difference Δ of the received signal-to-noise ratio can be appropriately reduced.

In some examples, different waveforms in the waveform diagram shown in FIG. 5 are obtained under different parameters m, specifically, the waveform F is a waveform under m=0.5. The waveform G is a waveform at m=0.8. The waveform H is the waveform at m=1. The waveform M is a waveform at m=1.5. The waveform N is the waveform at m=3. The number of paths for each client is 50. As shown in Figure 5, when the parameter m (also referred to as the order m) increases, the influence of channel fading can be reduced, and the value of the Jaccard metric of the weighted ensemble clustering algorithm increases accordingly. When the received signal-to-noise ratio or parameter m becomes large enough, the increasing trend of the value of the Jaccard metric gradually decreases and finally stabilizes. For example, if γ ₁ ≥ 18dB, the value of the Jaccard metric at m=1.5 is equal to the value at m=3.0.

In some examples, different waveforms in the waveform diagram shown in FIG. 6 are obtained under different initial trade-off parameters λ ₀ , specifically, the waveform O is a waveform under λ ₀ =0.01. The waveform P is a waveform at λ ₀ =0.1. The waveform Q is the waveform at λ ₀ =0.2. The waveform R is a waveform at λ ₀ =0.5. The waveform S is the waveform at λ ₀ =1. The number of paths for each client is 50. As shown in Figure 6, as the initial trade-off parameter λ ₀ increases, the performance of the weighted ensemble clustering algorithm decreases. When the performance reduction gap gradually decreases, for the larger initial trade-off parameter λ ₀ , the weighted ensemble clustering algorithm degenerates to the original ensemble clustering algorithm (that is, there is no process of adjusting the weights. In the original ensemble clustering algorithm, each basic clustering algorithm is The class results are added directly). In other examples, in areas of high received signal-to-noise ratio, the weighted ensemble clustering algorithm gradually converges to the same performance with the initial trade-off parameter λ ₀ . This is because each basic clustering algorithm works better in areas of high received signal-to-noise ratio, and the original ensemble clustering algorithm is also acceptable.

In some examples, different waveforms in the waveform diagram shown in FIG. 7 are obtained under different scale factors ρ of estimation errors. Specifically, the waveform T is a waveform under ρ=0. The waveform U is a waveform at ρ=1/γ. The waveform Y is the waveform at ρ=0.1. The waveform W is the waveform at ρ=0.2. The number of paths for each client is 50. As shown in Figure 7, by comparing the results of ρ=0 (no error) and ρ=1/γ, it is known that the influence of the channel estimation error in the low received SNR region is much greater than that in the high received SNR region. Therefore, the channel estimation error can be ignored for high received signal-to-noise ratio. In some other examples, ρ is set to a constant value. When ρ=0.1, the influence becomes larger, although as the value of the received signal-to-noise ratio increases, the Jaccard value increases. However, when ρ is set to a large constant value, for example, ρ=0.2, the performance of the overall algorithm further deteriorates to a value independent of the received signal-to-noise ratio.

In the present disclosure, because the performance of different clustering algorithms on the same system (for example, the same number of multiple user terminals and the same number of multiple paths) is basically different, and even if the same algorithm is under different channel conditions, May have different performance. In addition, because the clustering results of different clustering algorithms can describe the basic mode of a multi-user MIMO system from different aspects, a single clustering algorithm can usually only capture one aspect of the uplink system. Therefore, the present disclosure proposes a blind multipath recognition method of the MIMO system based on a weighted ensemble clustering algorithm. The blind multipath recognition method involved in the present disclosure can integrate the basic clustering results of different basic clustering algorithms to generate accurate and reliable clustering results. Compared with the traditional weighted ensemble clustering algorithm that processes each basic clustering result equally, the weighted ensemble clustering algorithm involved in the present disclosure can adaptively adjust the optimal weights of different basic clustering results. Thus, the basic clustering results that obtain higher weights may be more reliable and can be regarded as important features. However, the basic clustering result with lower weight may be a less reliable feature, and the cluster that produces the basic clustering result with lower weight may be far from the actual situation. In this case, with these weights, it is possible to perform selection between the elements (base clustering results) and produce a more reliable output. The blind multipath recognition method involved in the present disclosure can perform multipath recognition stably and effectively through the weighted ensemble clustering algorithm, and can accurately evaluate and fairly compare the performance of various clustering algorithms and weighted ensemble clustering algorithms.

FIG. 8 is a schematic structural diagram showing a blind multipath recognition system of a MIMO system based on a weighted ensemble clustering algorithm involved in an example of the present disclosure. In some examples, as shown in FIG. 8, the blind multipath recognition system (blind multipath recognition system for short) 1 of the MIMO system based on the weighted ensemble clustering algorithm may be a wireless communication system having a user device 10 and a receiving device 20. Blind multipath recognition system of MIMO system based on clustering algorithm1. Wherein, the user equipment 10 and the above-mentioned user terminal may have a similar concept, and the receiving apparatus 20 and the above-mentioned base station may have a similar concept. The user device 10 and the receiving device 20 may perform signal transmission through wireless communication.

In some examples, the number of user devices 10 may be multiple. A plurality of user devices 10 can transmit a communication request signal to the receiving device 20. The communication request signal may have a short frame structure. For details, see step S10 above. The user device 10 may include but is not limited to user equipment. In other examples, the receiving device 20 may include but is not limited to a base station.

In some examples, the receiving device 20 may be used to feed back a response signal to the user device 10 based on the communication request signal. Specifically, the receiving device 20 receives the communication request signal. The receiving device 20 can check whether the communication request signal of each user device 10 is legal through the user registration database. When the communication request signal is valid, the receiving device 20 estimates the communication request signal and calculates the received signal-to-noise ratio γ _{k of} each user device 10. The received signal-to-noise ratio γ _k satisfies equation (2). The receiving device 20 may calculate the difference Δ _k,j between the received signal-to-noise ratio γ _k of the k-th user equipment 10 and the received signal-to-noise ratio γ _j of the j-th user equipment 10, and then provide the user with the difference Δ _k,j The device 10 feeds back a response signal. The difference Δ _k,j satisfies equation (3). For details, see step S20 above.

In some examples, the user device 10 may determine whether to adjust the transmission power of the user device 10 based on the response signal, so that the receiving device 20 allows the communication request of each user device 10. The response signal may include a first response signal and a second response signal. If each difference calculated by the receiving device 20 is greater than the set threshold ε _Δ , the receiving device 20 feeds back the first response signal to the user device 10 and allows the user device 10 to make a communication request. After the user device 10 receives the second response signal and adjusts the transmission power, it retransmits the communication request signal to the receiving device 20, and the receiving device 20 recalculates each difference and compares it with the set threshold ε _Δ until each difference is equal to Greater than the set threshold ε _Δ , the receiving device 20 allows the communication request of the user device 10. In addition, the receiving device 20 can control the power of each user device 10 through automatic power control. For details, see step S20 above.

In some examples, when the receiving device 20 allows the communication request of each user device 10, a plurality of user devices 10 may send an information signal to the receiving device 20. Among them, all the user devices 10 can simultaneously send message signals to the receiving device 20 through the same frequency channel. There are multiple independent paths between each user device 10 and the receiving device 20. Each user device 10 sends a message signal to the receiving device 20 through corresponding multiple independent paths.

In some examples, the receiving device 20 may include a filter (eg, a spatial filter). The receiving device 20 can separate the information signal through a spatial filter. The receiving device 20 may generate an input signal of the basic clustering algorithm of any path of any user device 10 based on the separated information signal. Specifically, the receiving device 20 may obtain the input signal x _k,l (t) of the basic clustering algorithm for the first path of the Kth user device 10 based on the separated information signal and through channel estimation. Channel estimate

Satisfy formula (4). The input signal x _k,l (t) of the l-th path of the K-th user equipment 10 satisfies equation (5). Where r _k,l (t) represents the output signal of the lth path of the Kth user equipment 10 separated by the spatial filter,

It represents the channel estimation value of the l-th path of the K-th user equipment 10.

In some examples, based on the input signal and the basic clustering algorithm, the receiving device 20 may obtain the output result of the basic clustering algorithm. That is, the receiving device 20 can input the corresponding input signal of each path into the weighted ensemble clustering algorithm to obtain the output result of the basic clustering algorithm. Among them, the weighted ensemble clustering algorithm can divide various basic clustering algorithms into different categories. The selected basic clustering algorithm can be the 10 basic clustering algorithms mentioned above. For details, see step S40 above.

In some examples, based on the output results of each basic clustering algorithm, the receiving device 20 may obtain a corresponding feature matrix. Based on each feature matrix, the receiving device 20 can obtain an aggregate matrix. The set matrix can satisfy formula (7). The receiving device 20 may obtain the system probability based on the set matrix, and then obtain the objective function of the weighted ensemble clustering algorithm. The system probability P(E|B) satisfies equation (10). The objective function J(B) satisfies equation (11). Among them, E represents the set matrix, N _p represents the basic clustering result, ω _i represents the weight of the feature matrix Di of the _i-th path,

Represents a vector of weights, E _i,j represents each element of the set matrix, and represents the relationship between the i-th path and the j-th path of the user device 10, β _i,k represents the _i-th path of the _k- th user device 10 The strength of the path, β _j,k represents the strength of the j-th path of the K-th user device 10,

It indicates the possibility of detecting that the i-th path and the j-th path of the k-th user device 10 belong to the same user device 10. For details, see step S40 above.

In some examples, in order to avoid overfitting the weight vector W to a feature matrix, a regularization parameter R can be introduced. The regularization parameter R satisfies equation (8). The regularization parameter R can represent the sum of the negative entropy of the weight of the feature matrix Di of the _i-th path. The acquisition of the above-mentioned feature matrix and set matrix and the introduction of the regularization parameter R can be regarded as the generation stage of the weighted ensemble clustering algorithm. In some examples, the path-user tendency matrix B is established, and the receiving device 20 may identify the relationship between the path and the user device 10 based on the set matrix through the path-user tendency matrix B. The receiving device 20 can obtain the system probability based on the path-user tendency matrix B and the aggregate matrix E. For details, see step S40 above.

In some examples, the receiving device 20 may adaptively adjust the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm. The process of adaptive adjustment can be regarded as a constrained nonlinear optimization problem by minimizing the transformation objective function J(W, B). In the optimization process, alternately update the path-user tendency matrix B and the weight vector W, and repeat this backup update process until the result converges. For details, see step S40 above.

In some examples, in the receiving device 20 adaptively adjusts the weights of the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, the vector of the weights can be made equal to the preset value, and the path is updated by the first update rule -User preference matrix B, obtain the updated path-After user preference matrix B, update the weight vector based on the second update rule. The first update rule satisfies equation (13). The second update rule satisfies equation (14). In some examples, in the receiving device 20 adaptively adjusting the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, the regularization parameter R may be introduced to optimize the weight of the output result of the basic clustering algorithm, The regularization parameter R satisfies equation (8). For details, see step S40 above.

In some examples, the path of the information signal of the user device 10 is identified according to the output of the weighted ensemble clustering algorithm. The output of the weighted ensemble clustering algorithm is the ensemble clustering result. The result of ensemble clustering can be considered as a set of basic clustering algorithms after adjusting the weight. The receiving device 20 may divide the output of the weighted ensemble clustering algorithm into clusters. Each cluster contains all the paths of the information signal of each user device 10. Then, blind multipath identification is performed on the cluster to realize the blind multipath identification of each path of each user device 10. For details, see step S40 above.

In some examples, the receiving device 20 may obtain the maximum ratio combination of each user device 10 based on all the paths of each user device 10 and decode the information signal of each user device 10. The receiving device 20 can receive the information signal of the user device 10 and decode the information signal to complete the uplink transmission of the multi-user MIMO system. See step S50 above.

Although the present disclosure has been specifically described above with reference to the drawings and embodiments, it can be understood that the foregoing description does not limit the present disclosure in any form. Those skilled in the art can make modifications and changes to the present disclosure as needed without departing from the essential spirit and scope of the present disclosure, and these modifications and changes fall within the scope of the present disclosure.

Claims

A method for blind multipath identification of a MIMO system based on a weighted ensemble clustering algorithm is a method for blind multipath identification of a MIMO system based on a weighted ensemble clustering algorithm of a wireless communication system including a user terminal and a base station, and is characterized in that:

include:

A plurality of the user terminals send a communication request signal to the base station;

The base station feeds back a response signal to the user terminal based on the communication request signal, and the user terminal determines whether to adjust the transmission power of the user terminal based on the response signal, so that the base station allows the communication request of each user terminal ；

When the base station permits the communication request of each of the user terminals, a plurality of the user terminals send information signals to the base station;

The base station separates the information signal through a spatial filter, and the base station generates an input signal of a basic clustering algorithm for any path of any user terminal based on the separated information signal, based on the input signal and the The basic clustering algorithm obtains the output result of the basic clustering algorithm, the corresponding feature matrix is obtained based on the output results of each basic clustering algorithm, the set matrix is obtained based on each of the feature matrices, and the base station obtains the system based on the set matrix Probability, and then obtain the objective function of the weighted ensemble clustering algorithm, the base station adaptively adjusts the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, and according to the weighted ensemble clustering algorithm The output of the similar algorithm identifies the path of the information signal of the user terminal; and

The base station obtains the maximum ratio combination of each user terminal based on all the paths of each user terminal, and decodes the information signal of each user terminal.
The blind multipath identification method according to claim 1, characterized in that:

The base station obtains the input signal x k,l (t) of the basic clustering algorithm of the first path of the Kth user terminal through channel estimation, and the input signal of the l path of the Kth user terminal x k ,l (t) satisfies formula (Ⅰ):
Wherein, r k,l (t) represents the output signal of the kth user end of the l path separated by the spatial filter,
Represents the channel estimation value of the 1th path of the Kth user end.
The blind multipath recognition method according to claim 1, wherein:

In the base station adaptively adjusting the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, introducing a regularization parameter R to optimize the weight of the output result of the basic clustering algorithm, The regularization parameter R satisfies the formula (II):
Among them, ω i represents the weight of the feature matrix Di of the i-th path, and N p represents the basic clustering result.
The blind multipath recognition method according to claim 1, wherein:

In the base station adaptively adjusting the weights of the output results of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, the vector of the weights is made equal to the preset value, and the path-user is updated by the first update rule The trend matrix B, after obtaining the updated path-user trend matrix B, update the weight vector based on the second update rule.
The blind multipath recognition method according to claim 1, wherein:

A path-user tendency matrix B is established, and the base station recognizes the relationship between a path and the user terminal through the path-user tendency matrix B based on the set matrix, based on the path-user tendency matrix B and the set The matrix obtains the system probability, which satisfies the formula (Ⅲ)
Among them, E represents the set matrix, N p represents the basic clustering result, ω i represents the weight of the feature matrix Di of the i-th path,
Represents a vector of weights, E i,j represents each element of the set matrix, and represents the relationship between the i-th path and the j-th path of the user end, β i,k represents the i-th path of the kth user end The strength of each path, β j,k represents the strength of the j-th path of the K-th user terminal,
Indicates the possibility that the i-th path and the j-th path of the k-th client are detected to belong to the same client.
A blind multipath recognition system of a MIMO system based on a weighted ensemble clustering algorithm is a blind multipath recognition system of a MIMO system based on a weighted ensemble clustering algorithm that includes a user device and a receiving device, and is characterized in that:

include:

A plurality of the user devices, which are used to send a communication request signal to the receiving device; and

The receiving device is configured to feed back a response signal to the user device based on the communication request signal, and the user device determines whether to adjust the transmission power of the user device based on the response signal, so that the receiving device allows every Communication requests from said user equipment,

Wherein, when the receiving device allows the communication request of each user device, a plurality of the user devices send an information signal to the receiving device, and the receiving device separates the information signal through a spatial filter, and the The receiving device generates the input signal of the basic clustering algorithm of any path of any one of the user devices based on the separated information signal, and obtains the output of the basic clustering algorithm based on the input signal and the basic clustering algorithm As a result, the corresponding feature matrix is obtained based on the output results of each basic clustering algorithm, and the set matrix is obtained based on each of the feature matrices. The receiving device obtains the system probability based on the set matrix, and then obtains the objective function of the weighted ensemble clustering algorithm The receiving device adaptively adjusts the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, and provides information about the user device according to the output of the weighted ensemble clustering algorithm The signal path is identified, and the receiving device obtains the maximum ratio combination of each user device based on all the paths of each user device, and decodes the information signal of each user device.
The blind multipath recognition system according to claim 6, characterized in that:

The receiving device obtains the input signal x k,l (t) of the basic clustering algorithm of the first path of the Kth user device through channel estimation, and the input of the first path of the Kth user device The signal x k,l (t) satisfies the formula (Ⅰ):
Where r k,l (t) represents the output signal of the kth user device's lth path separated by the spatial filter,
Represents the channel estimation value of the kth user equipment's first path.
The blind multipath recognition system according to claim 6, characterized in that:

In the receiving device adaptively adjusting the weight of the output result of each basic clustering algorithm based on the objective function of the weighted ensemble clustering algorithm, the regularization parameter R is introduced to optimize the weight of the output result of the basic clustering algorithm , The regularization parameter R satisfies the formula (II):
Among them, ω i represents the weight of the feature matrix Di of the i-th path, and N p represents the basic clustering result.
The blind multipath recognition system according to claim 6, characterized in that:

In the receiving device adaptively adjusting the weights of the output results of each of the basic clustering algorithms based on the objective function of the weighted ensemble clustering algorithm, the vector of the weights is made equal to the preset value, and the path is updated through the first update rule − The user tendency matrix B, after obtaining the updated path-user tendency matrix B, update the weight vector based on the second update rule.
The blind multipath recognition system according to claim 6, characterized in that:

A path-user tendency matrix B is established, and the receiving device recognizes the relationship between a path and the user device through the path-user tendency matrix B based on the set matrix, based on the path-user tendency matrix B and the Collect the matrix to obtain the system probability, which satisfies the formula (Ⅲ):
Among them, E represents the set matrix, N p represents the basic clustering result, ω i represents the weight of the feature matrix Di of the i-th path,
Represents a vector of weights, E i,j represents each element of the set matrix, and represents the relationship between the i-th path and the j-th path of the user equipment, β i,k represents the k-th user equipment The strength of i paths, β j,k represents the strength of the j-th path of the K-th user device,
It indicates the possibility of detecting that the i-th path and the j-th path of the k-th user device belong to the same user device.