CN115052296B

CN115052296B - Intelligent Rank downlink rate optimization method for 6G

Info

Publication number: CN115052296B
Application number: CN202210572129.4A
Authority: CN
Inventors: 朱文进; 王玉梁; 薛希俊; 徐俊华; 刘少卿; 黄春林
Original assignee: China Telecom Digital Intelligence Technology Co Ltd
Current assignee: China Telecom Digital Intelligence Technology Co Ltd
Priority date: 2022-05-25
Filing date: 2022-05-25
Publication date: 2023-09-05
Anticipated expiration: 2042-05-25
Also published as: WO2023226216A1; CN115052296A

Abstract

The invention discloses a method for optimizing an intelligent Rank downlink rate of 6G, which comprises the following steps: classifying the influence factors of the Rank, obtaining a prediction value of the probability of failure of the downlink rate of the drive test Rank based on the region prediction model, and obtaining a factor classification mark for influencing the Rank; based on a random forest model, analyzing factor classification identifiers in four scene data and Rank identifiers covered by an antenna to obtain the most probable Rank fault maximum factor in a scene/sub-scene, positioning the area where the UE is located and generating an area identifier, and generating the factor identifier with the greatest Rank fault of the scene/sub-scene; and constructing a Rank optimization program and correlating with an antenna coverage scene to realize self-healing adjustment and optimization after a 5G Rank link fails in the drive test downlink rate optimization process of the UE. The invention can support the optimization of the 5G wireless downlink rate to the emerging 6G wireless downlink rate in the 5G-6G transition process.

Description

Intelligent Rank downlink rate optimization method for 6G

Technical Field

The invention belongs to the technical field of wireless communication, and particularly relates to a method for optimizing an intelligent Rank downlink rate of 6G.

Background

With the gradual progress of 5G network construction, the current gap between the 5G network scale and the 4G network is smaller and smaller, more 4G users migrate to the 5G network, and how the 5G network provides users with experience superior to that of the 4G network becomes a main problem facing current operators. The 5G network is an extension of the 4G network, and the advantage of the 5G network value cannot be reflected by the simple coverage, and the 5G terminal has more antenna port support than the 4G terminal (the 5G commercial mainstream is configured as 2T 4R).

In a communication system, a UE (terminal) performs channel estimation according to a wireless CSI-RS reference signal, calculates the maximum number of streams with minimum downlink channel coherence, called "Rank", and translates chinese to "Rank", and reports RI (Rank indicator) to a base station through CSI. While 5G Rank (flow) has a great influence on the user rate, a low Rank under good coverage will directly result in a low rate, the user perception is not different from 4G, and the downlink rate can be hundreds of megabits if Rank is more than 1 flow.

By utilizing the characteristic that the 5G terminal can support more flows, the Rank flow of the 5G user is promoted to be the key of 5G network optimization, and the quality of network experience finally depends on the gap of the Rank. Because the Rank is greatly influenced by factors such as wireless environment, terminal antenna, AAU antenna, channel correlation and the like, and single-point analysis in the process of improving Rank optimization consumes time and labor, how to quickly and accurately improve 5G Rank becomes a main research object for current and future optimization.

Disclosure of Invention

Aiming at the defects of the prior art, the technical problem to be solved by the invention is to provide a method for optimizing the downlink rate of the intelligent Rank of 6G, which can support the optimization of the downlink rate from 5G wireless downlink to emerging 6G wireless downlink in the transition process from 5G to 6G.

In order to achieve the technical purpose, the invention adopts the following technical scheme:

a method for intelligent Rank downlink rate optimization for 6G, comprising:

classifying influence factors of the Rank by combining log data, obtaining a failure probability prediction value of a downlink rate of the drive test Rank based on a region prediction model, obtaining a factor classification identifier of the influence Rank, and storing the factor classification identifier and the failure probability prediction value into the Rank identifier;

step two, analyzing four scene data covered by an antenna and factor classification identifiers in Rank identifiers stored in a log of the gNB based on a random forest model to obtain the most probable Rank fault maximum factor in a scene/sub-scene, positioning an area where the UE is located and generating an area identifier, finally generating the factor identifier with the greatest Rank fault of the scene/sub-scene, and storing the area identifier, the scene identifier and the factor identifier into the Rank identifier;

and thirdly, constructing a Rank optimization program according to 5G Rank problem optimization measures and Rank identification analysis, and correlating the Rank optimization program with an antenna coverage scene/sub-scene to realize self-healing adjustment and optimization after a 5G Rank link fails in the process of optimizing the downlink speed of the drive test by the UE.

In order to optimize the technical scheme, the specific measures adopted further comprise:

analyzing the history log data stored on the base station equipment to find out the data of the factors affecting the Rank and classifying according to hardware, UE placement, base station RF and algorithm;

inputting the most important factor data in each category into a constructed regional prediction model for analysis to obtain sub-category fault probability prediction indexes of each category of the factors affecting the Rank, and generating corresponding factor category identifiers.

The above step one, the region prediction model is as follows:

P(A|B)＝(P(B|A)*P(A))/P(B|A)P(A)+P(B|A')P(A')

p (A|B) is the probability of abnormality of the factor affecting Rank;

p (B|A) is the probability of the result of the total abnormal number of factors affecting Rank/the total number of historical database in the continuous learning process of the regional prediction model;

p (A) is the total number of abnormal data/the total number of historical data of factors affecting Rank, neglecting other factors;

p (B|A') is the probability that the total number of data anomalies of the factor affecting Rank appears in the historical database;

P(A')＝1-P(A)。

in the first step, the sub-classification fault probability prediction index and the corresponding factor classification identifier of each classification are specifically as follows:

the prediction index and the factor classification identification are specifically as follows:

factor one: hardware, corresponding factor classification identification: 2-1;

prediction index: whether the channel is passing or not is judged by the corresponding factor classification mark: 2-1-1;

factor II: UE placement, corresponding factor classification identification: 2-2;

prediction index 1: RSRP equalization between UE antennas, correspondence factor classification identifier 1:2-2-1;

prediction index 2: the placement position and the method of the UE, and factor classification identification 2:2-2-2;

factor three: base station RF, correspondence factor classification identification: 2-3;

prediction index 1: towards the building, the direction angle of the reflection path is increased, and the corresponding factor classification mark 1:2-3-1;

prediction index 2: the downtilt angle of ground reflection is increased in open scene, and corresponding factor classification sign 2:2-3-2;

factor four: algorithm, corresponding factor classification identification: 2-4;

prediction index 1: and (3) day selection terminal: SRS weight, correspondence factor classification identification 1:2-4-1;

prediction index 2: non-day-select terminals: vam+pmi right, correspondence factor classification identity 2:2-4-2.

The first step is to store the factor classification mark and the failure probability prediction value into the Rank mark by using the # symbol combination, and the format is as follows: the factor classification identifies the # occurrence probability prediction value.

The random forest model is built in the second step as follows:

wherein n is the number of sampling scenes, and 4 is taken to represent that the sampling scenes have scenes one, two, three and four;

the i Di i/D i refers to probabilities of scenes one, two, three, four;

h (i) is the total feature quantity of the first scene, the second scene, the third scene and the fourth scene.

The first scene, the second scene, the third scene and the fourth scene are identified as follows:

scene one: the wireless environment is single, the building is rare, and the main scene sign: 1-1;

the field is as follows: road is narrow, and two sides buildings are clustered, and main scene marks: 1-2;

scene III: multilane crossroads, buildings are clustered, road space is wide, and main scene marks are formed: 1-3;

scene four: multilane roads, single row group building, tree shading, scene identification: 1-4.

Combining the area identifier, the scene identifier and the factor identifier by using # coincidence, and putting the combined area identifier, the scene identifier and the factor identifier into a Rank identifier in the format of: the region identification # scene identification # factor classification identification # occurrence probability prediction value.

And step three, in the process of carrying out drive test downlink rate optimization, when an alarm occurs in a Rank link, the processing is specifically as follows:

s1, splitting a Rank identifier to obtain an area where a positioning UE is located;

s2, splitting a Rank identifier to obtain a corresponding scene/sub-scene;

s3, extracting fault keywords according to the drive test data report abnormal content, comparing the fault keywords with factor identifications obtained by splitting Rank identifications, and if the fault keywords appear in the abnormal content for 1 time or more times, confirming that the abnormality is true and effective;

s4, splitting the Rank identification to obtain a predicted value of the probability of occurrence of the fault, and executing a corresponding preset Rank optimization program to realize self-healing or optimization of the fault if the predicted value of the probability is greater than 50%.

The invention has the following beneficial effects:

in the 5G downlink rate optimization process, the Rank optimization thought is to increase multipath and reduce spatial correlation. The optimization of the wireless environment is a key means for improving the Rank, and the invention combines the 5G base station scheduling parameter and the 6G new characteristic parameter, thereby reducing the interference among beams and increasing the whole capacity of the system while improving the Rank to the greatest extent.

In the 6G space scene, the wide application coverage environment of terahertz and millimeter waves is greatly improved compared with that of 5G, and the invention further solves the problem that 5G Rank links are failed in the process of optimizing the downlink speed of the road test by combining an artificial intelligent model on the premise that the current 6G is not practically applied, and the high efficiency of optimizing and adjusting the terahertz, millimeter waves and the scene in the low-altitude area of the 6G is compensated by the high-efficiency application of the artificial intelligent model. Thereby solving the difficulty of optimizing the 5G Rank for the space coverage environment.

Drawings

FIG. 1 is a graph of sub-class fault probability predictors for each class of the present invention;

FIG. 2 is a schematic diagram of the design of a random forest model according to the present invention;

FIG. 3 is a schematic diagram of an intelligent Rank downlink rate optimization method applicable to 6G.

Detailed Description

Embodiments of the present invention are described in further detail below with reference to the accompanying drawings.

As shown in fig. 3, the method for optimizing the downlink rate of the intelligent Rank of the 6G mainly comprises three parts:

namely, constructing an area prediction model, and combining the downlink rate occurrence probability of each classified drive test Rank by the influence factors of Rank of the UE area positioning prediction area: and constructing a regional prediction model, optimizing 5G Rank by using the space coverage environment, classifying the influence factors of the Rank by combining the constructed regional prediction model with log data when the UE performs drive test downlink rate optimization and encounters the condition of low Rank, and obtaining factor classification influencing the Rank and a predictive value of the probability of failure of the downlink rate of the drive test Rank, and generating factor classification identifiers. And combining the factor classification mark and the occurrence probability prediction value into a Rank mark by using a # symbol, wherein the format is as follows: the factor classification identifies the # occurrence probability prediction value.

The specific description is as follows:

analyzing historical log data stored on base station equipment to find out data of factors affecting Rank, and classifying according to hardware, UE placement, base station RF and algorithm;

inputting the most important factor data in each category into a constructed regional prediction model for analysis to obtain sub-category fault probability prediction indexes of each category of the factor affecting Rank, and generating corresponding factor category identification, as shown in figure 1.

Construction of regional prediction model

The formula: p (a|b) = (P (b|a) ×p (a))/P (b|a) P (a) +p (b|a ') P (a')

Prior probability=p (a) [ conditional probability ] P (B) [ adjustment factor ] =p (b|a) x P (a)

P (a) is the total number of data anomalies/the total number of historical data for factors affecting Rank, for example, ignoring other factors: 40%;

p (A') 1-P (A), here 60%;

p (B|A) [ regional prediction model ] in the continuous learning process, the probability of the result of the total abnormal number of factors affecting Rank/the total historical database number of factors is 50%;

the probability that the total abnormal data count of the factors affecting Rank of P (B|A') appears in the historical database is 100% of the normal energy consumption defaults if the historical database is the normal energy consumption defaults;

p (B) is an abnormal use probability formula which directly considers the factors affecting Rank by neglecting other factors

P (B) =p (b|a) P (a) +p (b|a ') P (a'), here 0.5×0.4+1×0.6=0.8;

then it can be calculated according to Bayesian formula, that is

P(A|B)＝(0.5*0.4)/(0.8)＝0.25

0.25 is the probability of abnormality of the factor affecting Rank, thereby completing the whole operation (regional prediction model).

The sub-classification fault probability prediction index and the corresponding factor classification identification of each classification are specifically as follows:

factor one: hardware (factor classification mark: 2-1)

Prediction index: whether the channel is passing or not is corrected.

Factor classification identification: 2-1-1

Factor II: UE placement (factor classification identification: 2-2)

Prediction index 1: RSRP equalization between UE antennas.

Factor classification identification 1:2-2-1

Prediction index 2: placement location and method of UE.

Factor classification identification 2:2-2-2

Factor three: base station RF (factor classification identification: 2-3)

Prediction index 1: direction angle (towards building, increase reflection diameter)

Factor classification identification 1:2-3-1

Prediction index 2: down tilt angle (open scene increasing ground reflection)

Factor classification identification 2:2-3-2

Factor four: algorithm (factor classification mark: 2-4)

Prediction index 1: and (3) day selection terminal: SRS rights

Factor classification identification 1:2-4-1

Prediction index 2: non-day-select terminals: vam+pmi weights

Factor classification identification 2:2-4-2

Step two, analyzing four scene data covered by an antenna and factor classification identifiers in a Rank identifier stored in a log of the gNB based on a random forest model to obtain a Rank fault maximum factor which is most likely to occur in a scene/sub-scene;

positioning the area where the UE is located, generating an area identifier, and finally generating a factor identifier with the largest scene/sub-scene Rank fault;

storing the area identifier, the scene identifier and the factor identifier into a Rank identifier (combining the area identifier, the scene identifier and the factor identifier by using # coincidence, and putting the area identifier, the scene identifier and the factor identifier into the Rank identifier, wherein the format is # [ area identifier # [ scene identifier # ] [ factor identifier # [ factor classification identifier ]

Step two, constructing a random forest model, and analyzing the factor with the greatest weight for lifting Rank in a common scene;

the details are described below in conjunction with fig. 2:

[ random forest model ] formula:

parameter description:

1. a constant n is set as how many sampling scenarios there are.

2. Where |Di|/|D| refers to the probability of scene one, two, three, four, and the total number of entries at the time of calculating Hi is the number of scenes one, two, three, four. Deriving hi=anomaly probability for the scene for each feature.

For example: the history log data brought into scene one has |d| bars, and the history log data conforming to scene one has |di| bars (the declination angle of the AAU machine is not between 10 and 15 degrees).

The random forest model operation flow is as follows:

1. first, input as sample set |D|

2. Randomly selecting a trained data set and sample characteristics for |Di| round training

2-1, carrying out ith random sampling on the training set, and collecting n times to obtain a sampling set containing n samples

2-2 training an nth decision tree model Hi with a sampling set Di,

when training the nodes of the decision tree model, selecting a part of sample features from all sample features on the nodes, and selecting an optimal feature from the randomly selected part of sample features to make a left subtree and right subtree division result Hi of the decision tree

3. Hj is equal to the weighted average of all probability predictions Hi for the scene for which an anomaly probability weighted average occurs.

AAU, active Antenna Unit, active antenna element. AAU is the primary equipment of 5G base stations and is an implementation of a large-scale antenna array. The AAU can be seen as a combination of RRU and antenna, integrating multiple T/R units. The T/R unit is a radio frequency transceiver unit and is used for phased array radar in military first.

Common scenarios include:

scene one: the wireless environment is single, and the buildings are rare.

Main scene identification: 1-1

The solution method comprises the following steps: the downward inclination angle of the AAU machine is 10-15 degrees, the upper beam covers the reflecting surface of the building as much as possible, and the lower beam covers the road, so that multipath is easier to generate, and the speed is improved.

The field is as follows: the road is narrow, and the buildings at two sides are clustered.

Main scene identification: 1-2

The solution method comprises the following steps: the AAU mechanical downtilt angle is 10 degrees+narrow beams, the antenna positions are aligned to the optimal reflecting surfaces of the buildings, beam signals are reflected back and forth between the buildings in groups, a good multipath environment is created, and Rank and speed are improved.

Scene III: multi-lane crossroads, clustered buildings and wide road space.

Main scene identification: 1-3

The solution method comprises the following steps: the AAU covering direction selects the optimal reflecting surfaces of buildings at two sides of the intersection as far as possible, the buildings can not be covered along the road, multi-surface reflection is built as much as possible, and Rank is improved.

Scene four: multi-lane roads, single-row group buildings and trees shadow.

Scene identification: 1-4

The solution method comprises the following steps: the AAU coverage direction selects a single row of buildings or ground, and multipath is created through building, ground and traffic flow reflection, so that Rank and speed are improved.

And thirdly, constructing a Rank optimization program according to the 5G Rank problem optimization measure and Rank identification analysis, and correlating the Rank optimization program with the antenna coverage scene/sub-scene.

From the above, the Rank identification format is: region identifier # [ scene identifier # [ factor classification identifier # ] # occurrence probability prediction value.

In the third step, the downlink speed of the drive test is optimized, and when an alarm occurs in a Rank link, the processing is specifically as follows:

s1, splitting a Rank identifier to obtain an area where the positioning UE is located.

S2, splitting the Rank identification to obtain a corresponding scene/sub-scene.

S3, extracting fault keywords according to the drive test data report abnormal content, comparing the fault keywords with factor identifications obtained by splitting Rank identifications, and if the fault keywords appear in the abnormal content for 1 time or more. The abnormality is confirmed to be true and effective.

[ Rank optimization procedure ] concrete description:

and analyzing the Rank abnormal data in the historical drive test abnormal log data to obtain an optimization measure affecting the Rank problem, and presetting a corresponding program according to a commonly-used solving optimization method.

The 5G Rank problem optimization measure and Rank identification analysis are shown in Table 1.

TABLE 1 5G Rank problem optimization measures and Rank identification analysis Table

Abbreviations and key terms used in the present invention are defined as follows:

rank: the number of space division multiplexed streams. A simple understanding is the same time-frequency resource, which is transmitted simultaneously in several parts in space. Codewords are mapped to each stream by layer mapping (the number of codewords is less than or equal to the number of streams less than or equal to the number of antenna ports). Under the condition that the time-frequency resource is unchanged, the higher the Rank is, the higher the actual throughput rate is.

The principles of Rank and computational mode Rank are in the field of communications, and spatial multiplexing refers to the transmission of different data over different antennas, also called spatial multiplexing. The standard for measuring spatial multiplexing is how many different data can be transmitted at most per moment of a system, which is called "degree of freedom", that is, rank is larger, and the multiplexing gain is larger. Codewords are mapped to the respective streams by layer mapping, the more layers the higher the rate, and Rank determines the number of layers.

UE: user equipment (UserEquipment), i.e. mobile communication terminal equipment, such as a mobile phone.

When the UE accesses the mobile communication network, the base station and the core network need to allocate resources to the UE, and the resources may be divided from the following two angles: network resources: if the base station allocates air channel resources and GTPU transmission resources to the UE, the core network allocates GTPU transmission resources to the UE; system resources: such as threads, processes, single boards, virtual machines, etc. used by the base station or core network device to serve the UE. The 3GPP mobile communication standard defines a RESET process between a base station and a core network, wherein the purpose of the process is to inform the base station to initialize the system resources occupied by the UE through the RESET process when the system resources in the core network fail to influence the UE, and release the network resources related to the UE; otherwise, when the system resource in the base station fails to affect the UE, the core network is informed to initialize the system resource occupied by the UE through the RESET process, and the network resource related to the UE is released.

The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above examples, and all technical solutions belonging to the concept of the present invention belong to the protection scope of the present invention. It should be noted that modifications and adaptations to the invention without departing from the principles thereof are intended to be within the scope of the invention as set forth in the following claims.

Claims

1. A method for intelligent Rank downlink rate optimization for 6G, comprising:

2. The method for optimizing the downlink rate of an intelligent Rank of 6G according to claim 1, wherein the step one analyzes the historical log data stored on the base station device to find out the data of the factor affecting the Rank and classifies the data according to hardware, UE placement, base station RF, algorithm;

3. The method for intelligent Rank downlink rate optimization for 6G according to claim 2, wherein said step-said region prediction model is:

P(A|B)＝(P(B|A)*P(A))/P(B|A)P(A)+P(B|A')P(A')

p (A|B) is the probability of abnormality of the factor affecting Rank;

P(A')＝1-P(A)。

4. the method for optimizing the downlink rate of the intelligent Rank of 6G according to claim 3, wherein in the first step, the sub-classification fault probability prediction index and the corresponding factor classification identifier of each classification are specifically as follows:

factor one: hardware, corresponding factor classification identification: 2-1;

5. The method for optimizing the downlink rate of an intelligent Rank of 6G according to claim 1, wherein the step one includes storing the factor classification identifier and the failure probability prediction value in the Rank identifier in the form of a # symbol combination: the factor classification identifies the # occurrence probability prediction value.

6. The method for optimizing the downlink rate of the intelligent Rank of 6G according to claim 1, wherein the step two constructs a random forest model as follows:

the i Di i/D i refers to probabilities of scenes one, two, three, four;

7. The method for intelligent Rank downlink rate optimization for 6G according to claim 6, wherein the first, second, third, fourth scenario and scenario identification thereof are:

8. The method for optimizing the downlink rate of the intelligent Rank of 6G according to claim 1, wherein the step two combines the area identifier, the scene identifier and the factor identifier with # coincidence, and puts the combined area identifier, the scene identifier and the factor identifier into the Rank identifier in the format of: the region identification # scene identification # factor classification identification # occurrence probability prediction value.

9. The method for optimizing the downlink rate of the intelligent Rank of the 6G according to claim 1, wherein the step three is to perform the downlink rate optimization of the drive test, and when the Rank link generates an alarm, the processing is specifically as follows:

s2, splitting a Rank identifier to obtain a corresponding scene/sub-scene;