CN113411817A - Wireless system interference neural network prediction method based on wireless interference model - Google Patents

Wireless system interference neural network prediction method based on wireless interference model Download PDF

Info

Publication number
CN113411817A
CN113411817A CN202110509799.7A CN202110509799A CN113411817A CN 113411817 A CN113411817 A CN 113411817A CN 202110509799 A CN202110509799 A CN 202110509799A CN 113411817 A CN113411817 A CN 113411817A
Authority
CN
China
Prior art keywords
interference
user
neural network
wireless
sinr
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110509799.7A
Other languages
Chinese (zh)
Other versions
CN113411817B (en
Inventor
彭涛
段淦元
郭异辰
刘晗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Posts and Telecommunications
Original Assignee
Beijing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Posts and Telecommunications filed Critical Beijing University of Posts and Telecommunications
Priority to CN202110509799.7A priority Critical patent/CN113411817B/en
Publication of CN113411817A publication Critical patent/CN113411817A/en
Application granted granted Critical
Publication of CN113411817B publication Critical patent/CN113411817B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/06—Testing, supervising or monitoring using simulated traffic
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention discloses a wireless system interference neural network prediction method based on a wireless interference model, which comprises the following steps: step 1, performing prediction model training offline: step 1.1, constructing a training data set; step 1.2, inputting a training data set into a neural network model, and obtaining an interference model of the whole wireless system through data training; step 2, predicting the interference intensity on line: step 2.1, inputting interference user resource allocation vectors to be predicted into the trained neural network model; and 2.2, calculating and outputting the interference strength between the base station end users of the service user, namely a predicted value of the uplink signal-to-interference-and-noise ratio. The prediction method of the invention enables the network to utilize the essential information of the interference to carry out more appropriate prediction, simultaneously utilizes a proper amount of wireless resource allocation data and wireless measurement data generated in the scheduling process, provides a complete and accurate uplink user interference modeling scheme through big data analysis and machine learning algorithm, is simple to implement and is more close to the actual network scene.

Description

Wireless system interference neural network prediction method based on wireless interference model
Technical Field
The invention relates to the technical field of wireless communication, in particular to a wireless system interference neural network prediction method based on a wireless interference model.
Background
In order to effectively improve Network capacity, an Ultra-dense Network (UDN) is used in 5G as one of key technologies, and the transmission distance between a base station and a terminal user is reduced by densely deploying the base station, so as to improve spectrum efficiency, but severe Co-channel Interference (CCI) caused by dense deployment of small base stations brings new challenges to a wireless communication Network.
By means of the Centralized processing characteristic of a Central Unit (CU) of a 5G network architecture and by applying a machine learning algorithm, interference information hidden in massive wireless resource distribution data and wireless measurement data generated in the network operation process is mined, and compared with the existing method, the method is more accurate and more comprehensive in interference information, and is more suitable for an actual wireless network.
An interference matrix in resource allocation plays a crucial role, and the existing methods for acquiring the interference matrix mainly include the following two methods: one is to establish an interference matrix based on sweep frequency data, and the other is to establish an interference matrix based on measurement report messages of the mobile phone. The frequency domain information in the sweep frequency data is complete and is provided with longitude and latitude information, the interference condition on a sampling point can be accurately reflected, the interference matrix generated based on the sweep frequency data can not reflect the interference situation at an unknown place, especially under the condition of dense network, the small position change can bring large interference change, the measurement cost is too high, the mobile phone measurement report contains the real interference situation of the user, however, the interference information only contains several neighboring cells with strong signals, so that the interference information is incomplete, the established interference matrix has a certain error, the interference situation is worse and more complicated under the dense network, when the network is in the dense network with more users, the agility and the accuracy of the establishment of the interference matrix have higher requirements, but the non-real-time property and the low efficiency of the current method, the coarse granularity of information and the coarse precision of prediction cannot adapt to the current network situation.
In the existing cellular network, the wireless resource allocation function is completed by the base station, each Cell basically manages and allocates wireless resources independently, in order to cope with Inter-Cell Interference, the existing network performs negotiation and signaling interaction between network units and compensates to a certain extent by using an enhancement technology, for example, by means of X2 interface interaction information between base stations, the Inter-Cell Interference Coordination (ICIC) or enhanced ICIC (enhanced ICIC, eICIC) technology is used to solve the Inter-Cell Interference problem; or by means of a coordinated multiple Points (CoMP) technology, interference is cooperatively processed between different base stations, or interference is avoided, or the interference is converted into a useful signal, so that a higher rate is provided for a user, and the utilization rate of a network is improved; signaling transmission needs time, so that timeliness is seriously affected, meanwhile, a large number of adjacent cells in the UDN cause considerable signaling exchange overhead to affect network performance, and a CoMP technology needs to measure a large number of channels and consume a large number of pilot frequency resources; and a large amount of computing resources are consumed to process and calculate the signals.
In addition, a new method for constructing an interference matrix exists, some technical schemes can accurately and accurately acquire the interference relationship and strength among users based on a method for predicting the interference situation on a sampling point by using operation data to train based on the traditional multilayer neural network without occupying additional hardware or pilot frequency resources, the interference matrix is constructed, a large amount of data is required to train based on the traditional neural network, off-line training is performed, on-line prediction is performed, the prediction is completely based on a large amount of training data sets, the actual situations of interference and environment are not considered, the prediction precision is difficult to guarantee, meanwhile, the data collection time is long due to the large amount of data sets, the convergence is slow and real-time performance is poor, the precision and real-time performance requirements of actual deployment cannot be met, the fourth method is based on linear regression, and a large amount of off-line data is required to train as the traditional neural network method, the method also has the defects of long data set collection time, poor change perception capability and the like, and the difference is that the data are tried to be fitted into a linear mode, the actual situations of interference and environment are not considered, and it is deduced that the interference situation is represented as a nonlinear condition, the accuracy of linear regression prediction is very poor, and the requirement for constructing an interference model cannot be met well.
The establishment of the interference matrix based on the sweep data requires physical equipment deployment and is inconvenient to implement. The interference matrix established based on the mobile phone measurement report message only contains the information of surrounding strong interference base stations, when the interference matrix is in a dense network with a large number of users, the information is not complete enough, so that effective interference avoidance cannot be performed, and the interference matrix cannot be well matched with an actual scene. The existing interference coordination scheme based on signaling exchange (such as ICIC, eICIC) depends on signaling exchange seriously, so that the interference information which can be transmitted by the existing interference coordination scheme is extremely limited, and the granularity of transmitting the interference information is poor; signaling transmission needs time, so that timeliness is seriously affected, meanwhile, a large number of adjacent cells in the UDN cause considerable signaling exchange overhead to affect network performance, and a cooperative interference scheme (such as CoMP) needs to measure a large number of channels and consume a large number of pilot frequency resources; and a large amount of computing resources are consumed to process and calculate the signals, so that the overall cost is huge.
Although it is theoretically possible to calculate the wireless interference based on the terminal geographical position information and the propagation loss model, two problems are obviously present in reality: firstly, the wireless system is difficult to obtain the geographical position information of the mobile terminal at any time in practice; secondly, the propagation loss model is generally only used in scenes such as simulation evaluation, network planning and the like, and cannot truly and accurately reflect the radio wave propagation condition in the actual situation, and it is quite common that the difference between the model calculation value and the measured value in the existing network reaches 5-10 dB.
Based on the technical problems in the prior art, the invention discloses a wireless system interference neural network prediction method based on a wireless interference model.
Disclosure of Invention
In order to solve the technical problems in the prior art, the invention provides a wireless system interference neural network prediction method based on a wireless interference model.
The invention adopts the following technical scheme:
the invention provides a wireless system interference neural network prediction method based on a wireless interference model, which comprises the following steps:
step 1, performing prediction model training offline:
step 1.1, constructing a training data set;
step 1.2, inputting a training data set into a neural network model, and obtaining an interference model of the whole wireless system through data training;
step 2, predicting the interference intensity on line:
step 2.1, inputting interference user resource allocation vectors to be predicted into the trained neural network model;
and 2.2, calculating and outputting a predicted value of the Uplink Signal-to-noise Ratio (UL-SINR) of the interference strength between the base station end users of the service user.
Further, in step 1.1, a training data set is constructed by: user UmWhen using one Resource Block (RB), the data set stores Resource allocation indication variables of all users on the RB and user U measured by the base stationmUL-SINR on this RB is taken as a piece of data, and the data set of this user is noted as Dm＝
Figure BDA0003059898990000031
Wherein w(i)The vector data is allocated for the resources and,
Figure BDA0003059898990000032
is SINR (dB) data, k is the number of users connected in each base station, N is the total number of data, Q is the total number of users, and U is the userQ-k+1、UQ-k+2、...、UQBelong to the base station SBSQ/k。
Further, in step 1.2, the neural network model includes an input layer, one or more hidden layers, and an output layer, each layer being composed of a plurality of units.
Further, in step 1.2, the neural network model adopts a Huber function as a loss metric function, as shown in the following formula (1):
Figure BDA0003059898990000041
wherein: y represents the true value of the data, f (x) represents the predicted value, the parameter δ is a constant between 0.1 and 10, and the Huber function is a loss of square function when the error value is less than the cut point δ and a linear function when the error value is greater than the cut point δ.
Further, in step 1.1, constructing the training data set includes: collecting wireless resource allocation information and wireless measurement information generated in a scheduling process in a wireless network to complete modeling of interference in the network; and cleaning the collected data, removing the relevant data of the users which are scheduled for a few times and cannot be interfered by normal prediction, and then arranging the relevant data into a training data set according to each user.
Further, in step 1.2, the activation function of the neural network model adopts a logarithmic function, and is derived according to the physical meaning of Signal-to-noise Ratio (SINR) as follows:
in the uplink direction of the mobile communication network, the user UmThe transmitted uplink signal arrives at the home cell CjmThe useful signal received power of the base station is the following formula (2):
Figure BDA0003059898990000042
wherein:
Figure BDA0003059898990000043
for user UmTo home cell CjmChannel gain of (P)mRepresents the transmit power of user m;
and user UmOther cell user U occupying same radio resourcenTo cell CjmThe interference signal power of the base station is the following formula (3):
Figure BDA0003059898990000044
wherein:
Figure BDA0003059898990000045
for user UnTo cell CjmChannel gain of (P)nRepresents the transmit power of user n;
user UmIs in the home cell CjmThe signal-to-interference-and-noise ratio of the base station end is as follows (4):
Figure BDA0003059898990000046
wherein: n is a radical ofmIs a user UmSet of other users reusing the same radio resource, wnE {0, 1} indicates whether user n is interfering with user m on the current resource, σ2Is the noise power;
order to
Figure BDA0003059898990000047
Is the inverse of the SINR (signal to interference ratio),
Figure BDA0003059898990000048
for interfering with user UmFor current service user UnThe inverse of the Signal-to-interference Ratio (SIR) causing interference,
Figure BDA0003059898990000049
for serving user UmSince the UL-SINR used in an actual scenario is usually in dB units, and the optimization target of the algorithm is changed to the dB value of the UL-SINR to meet the actual requirement, the above equation (4) is converted into the dB value expression of the SINR as the following equation (5):
Figure BDA0003059898990000051
further, in step 1.2, the input data of the neural network is a resource allocation vector w ═ wnThe activation function is a logarithmic function-10 lg (x), the output gamma of the neural network is a UL-SINR predicted value, the structure of the neural network is a two-layer full-connection single-output structure, and the structure is designed according to the dB value expression of SINR, and the following formula (6) is adopted:
Figure BDA0003059898990000052
the formula is brought into a designed two-layer fully-connected neural network, the input data of the neural network is a resource allocation vector w, and the number of input nodes is wnThe number of users in the vector, the output node is only 1 as SINR prediction, and the w of the first layer input is multiplied by the corresponding weight
Figure BDA0003059898990000053
Post-linear superposition and biasing
Figure BDA0003059898990000054
Obtaining an input value of a second layer node, wherein the second layer node is an SINR predicted value through the output of a logarithmic activation function f (x) -10lg (x), and parameters obtained in the training process through mining also directly correspond to SIR among users and SNR of the users;
adopting dB values of SIR and SNR as parameters of neural network needing learning adjustment, and adjusting the SINR expression to be the following formula (7):
Figure BDA0003059898990000055
wherein:
Figure BDA0003059898990000056
for interference in dB, user UmFor current service user UnThe SIR that is causing the interference is,
Figure BDA0003059898990000057
serving user U in dBmThe SNR of the received signal.
Compared with the prior art, the invention has the following advantages:
the wireless system interference neural network prediction method based on the wireless interference model, disclosed by the invention, designs the activation function, the structure and the parameter adjustment mode of the special neural network according to the interference characteristics, so that the network can carry out more appropriate prediction by using the essential information of interference, and simultaneously provides a complete and accurate uplink user interference modeling scheme by using a proper amount of wireless resource allocation data and wireless measurement data generated in the scheduling process through big data analysis and a machine learning algorithm.
Drawings
FIG. 1 is a schematic diagram of a neural network according to an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating SINR prediction errors of various schemes in different data set sizes in a comparative example according to the present invention;
FIG. 3 is a schematic diagram of the interference source analysis mining performance of various schemes in different data set scales in the comparative example of the present invention;
FIG. 4 is a schematic diagram illustrating the average elapsed time for each user to obtain an interference model in each scenario for different training data set sizes in a comparison example of the present invention;
FIG. 5 is a diagram illustrating average training time consumption of users according to different schemes with different prediction accuracies in the example of the present invention.
Detailed Description
In order that the above objects, features and advantages of the present invention may be more clearly understood, the present invention is described in further detail below with reference to specific embodiments, it should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
Examples
The wireless system interference neural network prediction method based on the wireless interference model comprises the following steps:
step 1, performing prediction model training offline:
step 1.1, constructing a training data set;
step 1.2, inputting a training data set into a neural network model, and obtaining an interference model of the whole wireless system through data training;
step 2, predicting the interference intensity on line:
step 2.1, inputting interference user resource allocation vectors to be predicted into the trained neural network model;
and 2.2, calculating and outputting a predicted value of the interference strength UL-SINR between the base station end users of the service user.
In step 1 of this embodiment, a neural network model is designed according to the interference physical characteristics;
in step 2 of this embodiment, a machine learning algorithm is used to train a prediction model, and input data are historical wireless resource allocation data (such as usage of each resource block) and network measurement data (such as uplink signal to interference plus noise ratio) until the model training is completed;
in step 2 of this embodiment, interference strength prediction is performed on line, corresponding input data under different interference conditions is input into the prediction model, and a corresponding interference prediction result is output.
In the embodiment, the accurate modeling of the interference in the network can be completed by collecting a large amount of data generated in the wireless network without additional hardware deployment;
in step 3, a single interfering user and a plurality of interfering users are included under different interference conditions.
In step 1.1, a training data set is constructed by: user UmEvery time an RB is used, the resource allocation indicating variable of all users on the RB and the user U measured by the base station end are stored in the data setmUL-SINR on this RB is taken as a piece of data, noting the user's data set as
Figure BDA0003059898990000071
Wherein w(i)The vector data is allocated for the resources and,
Figure BDA0003059898990000077
is SINR (dB) data, k is the number of users connected in each base station, N is the total number of data, Q is the total number of users, and U is the userQ-k+1、UQ-k+2、...、UQBelong to the base station SBSQ/k。
In step 1.2, the neural network model adopts a Huber function as a loss measurement function:
wherein y represents the true value of the data, f (x) represents the predicted value, the parameter delta is a constant of 0.1-10, the Huber function is a square loss function when the error value is smaller than the cut point delta and is a linear function when the error value is larger than the cut point delta, the sensitivity problem to the abnormal value is reduced, and the Huber function is derivable everywhere and is more suitable for the parameter learning adjustment of the neural network in the scene, in the prior art, the loss function commonly used in the neural network is a square error loss function, namely, the loss value calculated by the loss function is the square error of the predicted value and the true value output by the network, the parameter is adjusted by back propagation according to the loss value, thereby realizing the effective learning adjustment of the parameter, but the square loss function is very easily influenced by the abnormal value, the abnormal value can cause the calculated loss value to be suddenly very large, so that the parameter learning of the network vibrates violently, and can be reduced to the precision of the parameter learning and is more difficult to converge, however, in an actual wireless network, fluctuation of small-scale fading is often severe, which results in a large number of abnormal values in a data set, and thus, the use of a square error function cannot be well adapted to the characteristic of severe change of a wireless network channel environment.
In step 1.1, constructing a training data set comprises: collecting wireless resource allocation information and wireless measurement information generated in a scheduling process in a wireless network to complete modeling of interference in the network; and cleaning the collected data, removing the relevant data of the users which are scheduled for a few times and cannot be interfered by normal prediction, and then arranging the relevant data into a training data set according to each user.
The activation function of the neural network model adopts a logarithmic function, and is derived according to the physical meaning of the signal to interference and noise ratio (SINR) as follows:
in the uplink direction of the mobile communication network, the user UmThe transmitted uplink signal arrives at the home cell CjmThe useful signal receiving power of the base station is:
Figure BDA0003059898990000072
wherein
Figure BDA0003059898990000073
For user UmTo home cell CjmThe channel gain of (a);
and user UmOther cell user U occupying same radio resourcenTo cell CjmThe interference signal power of the base station is:
Figure BDA0003059898990000074
wherein
Figure BDA0003059898990000075
For user UnTo cell CjmThe channel gain of (a);
user UmIs in the home cell CjmThe signal-to-interference-and-noise ratio of the base station end is as follows:
Figure BDA0003059898990000076
wherein N ismIs a user UmSet of other users reusing the same radio resource, wnE {0, 1} indicates whether user n is interfering with user m on the current resource, σ2Is the noise power;
order to
Figure BDA0003059898990000081
Is the inverse of the SINR (signal to interference ratio),
Figure BDA0003059898990000082
for interfering with user UmFor current service user UnThe inverse of the signal-to-interference ratio that causes interference,
Figure BDA0003059898990000083
for serving user UmBecause the UL-SINR used in an actual scenario is usually in dB, the optimization target of the algorithm is changed into the dB value of the UL-SINR to better meet the actual requirement, and then the above equation is converted into the dB value expression of the SINR:
Figure BDA0003059898990000084
the interference relationship to be mined is all SIR and SNR in the above formula, and interference relationship (namely SIR and SNR) information obtained by data mining is helpful for improving the performance of ICIC and eICIC; the SINR prediction can assist subsequent resource management algorithms, and as can be seen from the above formula, the dB expression of UL-S1NR is linear superposition operation for firstly calculating SIR and SNR between users, and then nonlinear logarithm operation is performed, and the operation logic is similar to that of a neural network, so that the neural network algorithm is considered to be used in the scheme to realize mining of SIR between users and SNR of a service user, and UL-SINR of the service user can be predicted in a given RB allocation mode.
In step 1.2, as shown in fig. 1, the structure of the neural network is a two-layer fully-connected single-output structure, and the neural network model includes oneThe input layer, one or more hidden layers and an output layer, each layer is composed of a plurality of units, and the input data of the neural network is a resource allocation vector w ═ wnThe activation function is a logarithmic function-10 lg (x), the output gamma of the neural network is a UL-SINR predicted value, and the structure is designed according to a dB value expression of SINR:
Figure BDA0003059898990000085
the formula is brought into a designed two-layer fully-connected neural network, the input data of the neural network is a resource allocation vector w, and the number of input nodes is wnThe number of users in the vector, the output node is only 1 as SINR prediction, and the w of the first layer input is multiplied by the corresponding weight
Figure BDA0003059898990000086
Post-linear superposition and biasing
Figure BDA0003059898990000087
Obtaining an input value of a second-layer node, wherein the second-layer node outputs a logarithm activation function f (x) -10lg (x) to be an SINR predicted value, the special neural network designed in the embodiment completely realizes a UL-SINR calculation process, and parameter adjustment in a training process can also mine to obtain SIR and SNR;
because the definition domain of the logarithm function of the neural network is not negative, and the power values are known to be not negative according to the physical meaning, as the parameters needing to be learned and adjusted in the neural network, the parameters are required to be guaranteed to be non-negative in the learning and adjusting process, meanwhile, the corresponding values are considered to be small (generally less than 0.01), the magnitude difference is large, the same parameters are unknown parameters of the inverse SIR, and the inverse SIR value corresponding between the service user and the weak interference user is approximately 10-2～10-3And the inverse SIR value with a strongly interfering user is close to around 1. Compared with 10 of the inverse SNR-7～10-8The neural network learning and adjusting process is difficult to keep the same fine adjustment precision within such a large magnitude range,the accuracy cannot be guaranteed, so that the accuracy of SINR prediction is reduced;
therefore, in this embodiment, the dB values of SIR and SNR are used as the parameters that the neural network needs to learn and adjust, and the SINR expression is adjusted as follows:
Figure BDA0003059898990000091
wherein the content of the first and second substances,
Figure BDA0003059898990000092
the parameter satisfies nonnegativity and the physical meaning is more definite through the processing, and the absolute value of the parameter value after adjustment basically falls to 100～101And the magnitude order enables the neural network to realize high-precision adjustment, so that the SINR prediction is more accurate.
Comparative example
Based on the wireless system interference neural network prediction method based on the wireless interference model provided in the embodiment, the performance of UL-SINR prediction is carried out on Q users in the simulation scene in a performance comparison mode, the comparison scheme is a linear regression algorithm prediction method and a traditional neural network prediction method, and the evaluation indexes are prediction accuracy comparison of the same data volume, training time comparison of the same data volume and prediction accuracy comparison of the same training time.
In the above embodiment, co-channel interference is mainly considered, and the transmission power of the user can be detected at the small base station side, so that the average value of the interference strength and the UE detected at the small base station side can be obtained by counting the multiple average values of the interference signal strength_iThe average value of the received signal strength is obtained by multiple average values of the received signal strength, the UL-SINR actual value is the ratio of the latter to the former, and the predicted average root mean square error can be calculated and obtained according to the UL-SINR actual value.
Wherein, the simulation parameters are shown in the following table:
simulation parameter table
Figure BDA0003059898990000093
Figure BDA0003059898990000101
TABLE 1
As shown in fig. 2, the evaluation index for evaluating the UL-SINR prediction performance is the Average root mean square error (i.e., Average RMSE (dB)) of the predicted value and the true value, in the figure, LRA represents the linear regression algorithm prediction method, MLP _ old represents the conventional neural network prediction method, and MLP _ old represents the neural network prediction method described in the above embodiment, and the prediction accuracies of the three methods are compared with each other as the training data amount increases, and the prediction errors of the three schemes all decrease as the data amount increases, but the neural network scheme in the above embodiment has a better prediction accuracy than the other two schemes under the same training data amount, where the Average RMSE (dB) when the training data amount is twenty thousand is respectively: the linear regression method is 3.21dB, the traditional neural network method is 0.36dB, the new neural network method is 0.19dB, the prediction error of the linear regression scheme is far greater than that of the other two schemes, the prediction accuracy of the traditional neural network scheme is 20 ten thousand pieces of data, the new neural network can achieve the prediction accuracy only by 1 ten thousand pieces of data, the required data size is reduced by one magnitude, the prediction performance of the method in the embodiment is always superior to that of the two comparison schemes, the prediction accuracy is far superior to that of the linear regression, the accuracy is improved by nearly 50% compared with that of the traditional neural network method, and the performance advantage of the SINR prediction accuracy of the new neural network scheme is very outstanding.
As shown in fig. 3, the evaluation index of the interference source identification performance is the Average root mean square error (i.e., Average RMSE (dB)) of the predicted SIR and the true SIR, i.e., the comparison condition of the interference source identification performance of the three methods with the increase of the training data amount, and it can be known from the above figure that the interference source identification performance of the neural network algorithm described in the above embodiment is superior to the conventional neural network algorithm and the linear regression algorithm, and the advantage is increased with the increase of the training data set size, and when the data amount of each user reaches 10000 samples, the prediction error of the interference source identification performance of the neural network algorithm described in the above embodiment can be less than 0.5dB, so as to meet the performance requirement; when the data amount per user reaches 200000 samples, the prediction error of the neural network algorithm described in the above embodiment has reached 0.10dB, which is far better than 0.25dB of the neural network algorithm and 2.59dB of the linear regression algorithm, so the performance advantage of the new version of neural network algorithm in the aspect of interference source identification is very prominent.
The evaluation index for evaluating the UL-SINR prediction time is average per-user training time, the comparison situation of the prediction time consumption of the three methods along with the increase of the training data amount is shown in FIG. 4, the prediction time consumption of the prediction method in the embodiment is always superior to that of the traditional neural network method, although the prediction time consumption is slower than that of the linear regression method, the linear regression method does not need multiple iterations and has poor prediction accuracy, comparability is avoided, the average training time consumption of a new version of neural network scheme is 1 order of magnitude lower than that of the traditional neural network scheme under the same data set scale, the method in the embodiment can realize second-level prediction, and the time performance advantage is very outstanding.
By determining performance indexes (usually prediction errors such as average prediction RMSE), the optimal training data set scale can be determined more accurately, not only can the situation that the prediction accuracy is insufficient due to too small scale be avoided, but also the situation that the time consumption for collecting data and training is increased due to too large scale and the timeliness of a model is influenced can be avoided, under a certain average prediction RMSE of each interference modeling scheme, the training time consumption required by each user is as shown in FIG. 5, and it can be seen that the training time consumption of the prediction method of the embodiment is two orders of magnitude lower than that of the traditional neural network method under the condition of the same prediction accuracy, and is far better than that of the prediction time consumption of the linear regression method, the time consumption of the traditional neural network algorithm is two orders of magnitude higher than that of the new neural network algorithm under the condition that the prediction errors are less than 0.5dB, the training time required by each user is less than 1s on the average, the sub-second-level operation under the precision requirement is realized, and the real-time performance is very strong; however, the conventional neural network algorithm needs about 25.3s, and the linear regression algorithm even cannot meet the precision requirement, which shows that the performance of the method in the above embodiment for UL-SINR prediction is greatly superior to that of the existing method.
The present invention is not limited to the above-described embodiments, which are described in the specification and illustrated only for illustrating the principle of the present invention, but various changes and modifications may be made within the scope of the present invention as claimed without departing from the spirit and scope of the present invention. The scope of the invention is defined by the appended claims.

Claims (7)

1. A wireless system interference neural network prediction method based on a wireless interference model is characterized by comprising the following steps:
step 1, performing prediction model training offline:
step 1.1, constructing a training data set;
step 1.2, inputting a training data set into a neural network model, and obtaining an interference model of the whole wireless system through data training;
step 2, predicting the interference intensity on line:
step 2.1, inputting interference user resource allocation vectors to be predicted into the trained neural network model;
and 2.2, calculating and outputting a predicted value of the uplink signal-to-interference-plus-noise ratio (UL-SINR) of the interference strength between the base station end users of the service user.
2. The method for predicting the interference neural network of the wireless system based on the wireless interference model according to claim 1, wherein in step 1.1, the training data set is constructed by: user UmWhen one resource block RB is used, the resource allocation indicating variable of all users on the RB and the user U measured by the base station end are stored in the data setmUL-SINR on this RB is taken as a piece of data, noting the user's data set as
Figure FDA0003059898980000011
Wherein w(i)The vector data is allocated for the resources and,
Figure FDA0003059898980000012
is UL-SINR (dB) data, k is the number of users connected in each base station, N is the total number of data, Q is the total number of users, and U is the userQ-k+1、UQ-k+2、...、UQBelong to the base station SBSQ/k。
3. The method according to claim 2, wherein in step 1.2, the neural network model comprises an input layer, one or more hidden layers, and an output layer, and each layer comprises a plurality of units.
4. The method for predicting the interference neural network of the wireless system based on the wireless interference model, according to claim 3, wherein in step 1.2, the neural network model adopts a Huber function as the loss metric function, as shown in the following formula (1):
Figure FDA0003059898980000013
where y represents the true value of the data, f (x) represents the predicted value, the parameter δ is a constant between 0.1 and 10, and the Huber function is a loss of square function when the error value is less than the cut point δ and a linear function when greater than the cut point.
5. The method for predicting the wireless system interference neural network based on the wireless interference model according to claim 1, wherein in step 1.1, constructing the training data set comprises: collecting wireless resource allocation information and wireless measurement information generated in a scheduling process in a wireless network to complete modeling of interference in the network; and cleaning the collected data, removing the relevant data of the users which are scheduled for a few times and cannot be interfered by normal prediction, and then arranging the relevant data into a training data set according to each user.
6. The method for predicting the interference neural network of the wireless system based on the wireless interference model according to claim 1, wherein in step 1.2, the activation function of the neural network model adopts a logarithmic function, and is derived according to the physical meaning of the signal to interference plus noise ratio SINR as follows:
in the uplink direction of the mobile communication network, the user UmThe transmitted uplink signal reaches the home cell
Figure FDA0003059898980000021
The useful signal received power of the base station is the following formula (2):
Figure FDA0003059898980000022
wherein:
Figure FDA0003059898980000023
for user UmTo home cell
Figure FDA0003059898980000024
Channel gain of (P)mRepresents the transmit power of user m;
and user UmOther cell user U occupying same radio resourcenReach cell
Figure FDA0003059898980000025
The interference signal power of the base station is the following formula (3):
Figure FDA0003059898980000026
wherein:
Figure FDA0003059898980000027
for user UnTo the cell
Figure FDA0003059898980000028
Channel gain of (P)nRepresents the transmit power of user n;
user UmIs in the home cell
Figure FDA0003059898980000029
The signal-to-interference-and-noise ratio of the base station end is as follows (4):
Figure FDA00030598989800000210
wherein N ismIs a user UmSet of other users reusing the same radio resource, wnE {0, 1} indicates whether user n is interfering with user m on the current resource, σ2Is the noise power;
order to
Figure FDA00030598989800000211
Is the inverse of the SINR (signal to interference ratio),
Figure FDA00030598989800000212
for interfering with user UmFor current service user UnThe inverse of the signal-to-interference ratio SIR causing interference,
Figure FDA00030598989800000213
for serving user UmSince the UL-SINR used in an actual scenario is usually in dB, and the optimization target of the algorithm is changed to the dB value of the UL-SINR to better meet the actual requirement, the above equation (4) is converted into the dB value expression of the SINR as the following equation (5):
Figure FDA00030598989800000214
7. the method of claim 6, wherein the method comprises predicting the interference neural network of the wireless system based on the wireless interference modelIn step 1.2, the input data of the neural network is a resource allocation vector w ═ wnThe activation function is a logarithmic function-10 lg (x), the output gamma of the neural network is a predicted value of UL-SINR, and the structure is designed according to a dB value expression of SINR, and is as follows (6):
Figure FDA00030598989800000215
the formula is brought into a designed two-layer fully-connected neural network, the input data of the neural network is a resource allocation vector w, and the number of input nodes is wnThe number of users in the vector, the output node is only 1 as SINR prediction, and the w of the first layer input is multiplied by the corresponding weight
Figure FDA00030598989800000216
Post-linear superposition and biasing
Figure FDA00030598989800000217
Obtaining an input value of a second layer node, wherein the second layer node is an SINR predicted value through the output of a logarithmic activation function f (x) -10lg (x), and parameters obtained by mining in the training process also directly correspond to SIR between users and SNR of the users;
adopting dB values of SIR and SNR as parameters of neural network needing learning adjustment, and adjusting the SINR expression to be the following formula (7):
Figure FDA0003059898980000031
wherein the content of the first and second substances,
Figure FDA0003059898980000032
for interference in dB, user UmFor current service user UnThe SIR that is causing the interference is,
Figure FDA0003059898980000033
to be composed ofdB denoted, serving user UmThe SNR of the received signal.
CN202110509799.7A 2021-05-11 2021-05-11 Wireless system interference neural network prediction method based on wireless interference model Active CN113411817B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110509799.7A CN113411817B (en) 2021-05-11 2021-05-11 Wireless system interference neural network prediction method based on wireless interference model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110509799.7A CN113411817B (en) 2021-05-11 2021-05-11 Wireless system interference neural network prediction method based on wireless interference model

Publications (2)

Publication Number Publication Date
CN113411817A true CN113411817A (en) 2021-09-17
CN113411817B CN113411817B (en) 2022-08-23

Family

ID=77678138

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110509799.7A Active CN113411817B (en) 2021-05-11 2021-05-11 Wireless system interference neural network prediction method based on wireless interference model

Country Status (1)

Country Link
CN (1) CN113411817B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114244767A (en) * 2021-11-01 2022-03-25 北京邮电大学 Load balancing-based link minimum end-to-end delay routing algorithm
CN115175380A (en) * 2022-06-24 2022-10-11 超讯通信股份有限公司 Intelligent control method and device for 5G small cell and 5G small cell
CN115529618A (en) * 2022-08-02 2022-12-27 北京邮电大学 Uplink interference simplification identification and performance prediction method for wireless system
CN117119508A (en) * 2023-10-20 2023-11-24 江苏云舟通信科技有限公司 System and method for analyzing error code number of wireless communication channel

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110420019A (en) * 2019-07-29 2019-11-08 西安电子科技大学 A kind of depth recurrence heart rate estimation method of ballistocardiography signal
CN110707763A (en) * 2019-10-17 2020-01-17 南京理工大学 AC/DC power distribution network load prediction method based on ensemble learning
CN111225384A (en) * 2018-11-26 2020-06-02 ***通信有限公司研究院 Uplink interference modeling method, interference determining method and device
WO2020244746A1 (en) * 2019-06-05 2020-12-10 Telefonaktiebolaget Lm Ericsson (Publ) Mu-mimo operation control

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111225384A (en) * 2018-11-26 2020-06-02 ***通信有限公司研究院 Uplink interference modeling method, interference determining method and device
WO2020244746A1 (en) * 2019-06-05 2020-12-10 Telefonaktiebolaget Lm Ericsson (Publ) Mu-mimo operation control
CN110420019A (en) * 2019-07-29 2019-11-08 西安电子科技大学 A kind of depth recurrence heart rate estimation method of ballistocardiography signal
CN110707763A (en) * 2019-10-17 2020-01-17 南京理工大学 AC/DC power distribution network load prediction method based on ensemble learning

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114244767A (en) * 2021-11-01 2022-03-25 北京邮电大学 Load balancing-based link minimum end-to-end delay routing algorithm
CN114244767B (en) * 2021-11-01 2023-09-26 北京邮电大学 Link minimum end-to-end delay routing algorithm based on load balancing
CN115175380A (en) * 2022-06-24 2022-10-11 超讯通信股份有限公司 Intelligent control method and device for 5G small cell and 5G small cell
CN115175380B (en) * 2022-06-24 2023-06-02 超讯通信股份有限公司 5G small cell intelligent control method and device and 5G small cell
CN115529618A (en) * 2022-08-02 2022-12-27 北京邮电大学 Uplink interference simplification identification and performance prediction method for wireless system
CN117119508A (en) * 2023-10-20 2023-11-24 江苏云舟通信科技有限公司 System and method for analyzing error code number of wireless communication channel
CN117119508B (en) * 2023-10-20 2024-02-02 江苏云舟通信科技有限公司 System and method for analyzing error code number of wireless communication channel

Also Published As

Publication number Publication date
CN113411817B (en) 2022-08-23

Similar Documents

Publication Publication Date Title
CN113411817B (en) Wireless system interference neural network prediction method based on wireless interference model
Lee et al. Deep power control: Transmit power control scheme based on convolutional neural network
CN110401964B (en) Power control method based on deep learning for user-oriented center network
US10485011B2 (en) Method for scheduling and/or muting of radio resources in a wireless communication system
JP4910018B2 (en) Network interference evaluation method, dynamic channel allocation method and apparatus used in a wireless network
CN113382477B (en) Method for modeling uplink interference between wireless network users
CN113239632A (en) Wireless performance prediction method and device, electronic equipment and storage medium
CN111083724B (en) OCWG-TOPSIS-based user association method in fog computing network
Fernandes et al. Comparison of artificial intelligence and semi-empirical methodologies for estimation of coverage in mobile networks
JP2023535841A (en) Energy-saving processing method and processing device for base station
CN113905446A (en) Method, device and network equipment for estimating downlink modulation coding mode
CN103974264A (en) Frequency point optimal selection method
CN113242066B (en) Multi-cell large-scale MIMO communication intelligent power distribution method
CN106982454A (en) A kind of user's access mechanism based on load
CN116567843A (en) Wireless resource allocation optimization device and method
CN113644946B (en) Deep learning dynamic beam forming method for user-oriented central network
Xing et al. Spectrum efficiency prediction for real-world 5g networks based on drive testing data
CN114285504A (en) Large-scale wireless energy supply backscattering communication network transmission performance research method
Zhang et al. Map2Schedule: An End-to-End Link Scheduling Method for Urban V2V Communications
Jiang et al. User rate and energy efficiency of HetNets based on Poisson cluster process
CN115529618A (en) Uplink interference simplification identification and performance prediction method for wireless system
Busson et al. Impact of resource blocks allocation strategies on downlink interference and SIR distributions in LTE networks: a stochastic geometry approach
Adeel et al. Performance analysis of random neural networks in lte-ul of a cognitive radio system
CN116032392A (en) Method for predicting uplink interference mining among wireless network users based on local data
CN108449718B (en) Method for predicting position of mobile user in ultra-dense heterogeneous network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant