A 5G micro base station indoor positioning method based on data and model

Abstract

The present invention discloses a 5G micro base station indoor positioning method based on data and model, the steps are: constructing a 5G micro base station positioning model according to the channel fading model and the measured data collected in the actual scene; generating a data training fingerprint positioning method model through the model, and optimizing the model training process. The positioning method of the present invention can achieve higher positioning accuracy by combining and supplementing the information of the measured data and the model positioning.

Description

Technical Field

[0001] The present invention relates to the field of communication network technology, and mainly designs a 5G micro base station indoor positioning method based on data and models. Background Art

[0002] In recent years, with the advancement of science and technology, location-based services have become increasingly important. At present, satellite-based positioning technologies, such as the Global Positioning System and the Global Satellite Navigation System, have become very mature.

[0003] Although GPS has accurate positioning, the signal quality is greatly reduced due to the obstruction of buildings, making it unable to work properly indoors. The requirements for indoor positioning are often much higher than those for outdoor positioning. Current indoor positioning services often need to achieve decimeter-level and centimeter-level positioning accuracy. Therefore, indoor positioning has become a more important research direction in the current positioning field.

[0004] At present, there are many mature devices and technologies in the field of indoor positioning, such as Bluetooth technology, WIFI technology, sensor-based positioning technology, etc. However, due to the limitations of the technology itself, the positioning effects of the listed technologies are not very accurate; for example, WIFI technology may achieve meter-level positioning effects with the support of some algorithms; while the error of Bluetooth positioning may reach two to three meters, and in the case of complex indoor environments and many obstructions, the positioning effect may be further reduced. Sensor-based technology is generally used for track calculation. The characteristic of this positioning technology is that it can complete the positioning work only by relying on the pending target without external assistance, but the positioning effect is extremely dependent on the sensitivity of the sensor and will produce irreversible error accumulation.

[0005] In addition to the commonly used positioning equipment and positioning technology, some feasible positioning methods are also needed to complete the complete positioning work. At present, the more mature positioning methods include: positioning methods based on geometric measurement calculation, positioning methods based on positioning fingerprints, positioning methods based on pedestrian dead reckoning (PDR), etc.

[0006] The positioning method based on geometric measurement calculation needs to convert the signal parameters of the communication between nodes and obtain the final position through geometric analysis. The premise of using this method is that the data such as the received signal strength or arrival time difference needs to be converted into distance, or the signal transmission angle needs to be obtained, and then the position is calculated; however, due to the existence of many interferences in the actual environment, such as the obstruction and reflection of the indoor environment, the movement of objects and people, etc., which cause the non-line-of-sight transmission or loss of the signal, the distance obtained by the parameter is inaccurate, which in turn causes the estimated position to be biased.

[0007] The positioning method based on positioning fingerprint can obtain high-precision positioning results without relying on the real-time signal quality, but the construction of the fingerprint database is time-consuming and laborious; and the created fingerprint database also has certain environmental fragility. When the positioning environment is switched or the current indoor environment changes to a certain extent, the original fingerprint database will become invalid. In addition, the accuracy of fingerprint positioning is also directly related to the amount of data in the fingerprint database. Too much fingerprint data collection will generate a lot of manpower costs and reduce the collection efficiency in the offline stage; if the collected data is too little, the overall positioning accuracy of fingerprint positioning may decrease.

[0008] At present, the positioning method based on dead reckoning mainly uses various sensors integrated in the terminal device to obtain information such as acceleration and azimuth generated by the target during movement, and to infer and estimate the next position of the target. However, the PDR-based method requires an accurate initial position and high-precision sensors, and errors will accumulate during the positioning process, requiring additional references to correct the calculated positioning results from time to time.

[0009] The various current positioning devices and positioning methods can achieve a certain positioning accuracy, but they all have obvious defects. How to combine or switch the existing positioning methods to ensure positioning accuracy while reducing positioning time is also very valuable research. Summary of the invention

[0010] In order to solve the above technical problems, the present invention provides a 5G micro base station indoor positioning method based on data and model, which combines the existing fading model with actual data to derive an indoor positioning model, and then generates a data training fingerprint positioning method model based on the positioning model to optimize the model training process; finally, the positioning method based on data plus model can achieve higher positioning accuracy.

[0011] The present invention provides a 5G micro base station indoor positioning method based on data and model, comprising the following steps:

[0012] A 5G micro base station indoor positioning method based on data and model, comprising the following steps:

[0013] Step 1: Build a positioning model for 5G micro base stations based on the channel fading model and measured data collected in actual scenarios;

[0014] Step 2: Calculate the distance between micro base stations based on the channel fading model;

[0015] Step 3: Generate wireless indoor positioning data based on the fixed number of 5G micro base stations in the positioning model;

[0016] Step 4: Input the wireless indoor positioning data and measured data into the random forest model for training at the same time to obtain the accurate position of the target to be measured.

[0017] Furthermore, the specific steps of constructing the positioning model of the 5G micro base station in step 1 are:

[0018] Signal fading includes shadow fading and path loss. In the implementation of the positioning model, a fading model based on the CI model is adopted, where the fading expression is shown in the following formula:

[0019]

[0020] Wherein, d represents the distance between micro base stations, which is converted from the signal strength value received between micro base stations or the same type of parameters; f represents the center frequency, which is determined by the specific transmission frequency band of the micro base station; the CI model is optimized based on the measured data to obtain the path loss index n and the shadow fading standard deviation χ σ The value of; d0 is the distance from the target to be measured to the micro base station, dB is the unit, and log is the operator;

[0021] Assuming that the micro base stations are distributed in a random geometric form, the CI model is simulated using a homogeneous Poisson point process, and a two-dimensional homogeneous Poisson point process with a distribution density of λ is obtained:

[0022]

[0023] Where P represents Poisson distribution, N([a,b])=k means the time interval is ba, the number of events N is k, and λ represents the distribution density. Then the expected distance between the micro base station closest to the target is:

[0024]

[0025] The expected distance squared is obtained as follows:

[0026]

[0027] Furthermore, in step 2, according to the channel fading model, the distance between the micro base stations is solved as follows:

[0028] Assuming that the reference node is a micro base station, the channel fading model adopts the following formula:

[0029]

[0030] The path fading formula is:

[0031]

[0032] Where P TxRepresents the power of the signal transmitting end, P Rx represents the power at the receiving end, G Tx is the transmit gain, G Rx is the receiving gain, Lc represents the total loss, and PL is a function based on the center frequency f and the node distance d, which can be obtained:

[0033]

[0034] For wireless positioning, we need to estimate the distance based on the existing models and data, obtain the position of the target to be measured based on the distance and position information given by the reference point, and convert the formula into distance d as the dependent variable and received signal strength P Rx That is, the formula with RSSI value as parameter or independent variable:

[0035]

[0036]

[0037] Get the distance d(P Rx ,f) function:

[0038]

[0039] Further, in step 3, based on the fixed number of 5G micro base stations in the positioning model, the specific steps of generating wireless indoor positioning data are:

[0040] Assume that the current position of the target to be measured is (x0, y0), and the estimated distance E(r n 2 ) is regarded as the actual distance between the Nth micro base station and the target to be measured, then:

[0041] E(r n 2 )=(x n -x0) 2 +(y n -y0) 2

[0042] Where E(r n 2 ) represents the estimated distance between the Nth node base station and the target to be measured, (x n -x0) 2 +(y n -y0) 2 is the distance formula between two points;

[0043] For the N selected base stations, the equation group can be obtained:

[0044]

[0045] Subtract each equation from the first and rewrite the entire system of equations in matrix form:

[0046]

[0047] The whole equation can be simplified to

[0048] Hx=b

[0049] in

[0050]

[0051] Then the solution of the system of equations is:

[0052] x=(H T H) -1 H T b; H T is the transposed matrix of H.

[0053] Furthermore, in step 4, the positioning data generated by the positioning model and the actually collected data are input into the random forest model for training to obtain the precise location, specifically:

[0054] The input of the random forest model is the training set consisting of positioning data and measured data and the number of iterations T, where is the positioning data in the dataset, y i ∈Y is the real coordinate position of the positioning data in the experimental scene, i = 1, 2, 3, ..., N, and the output is the final classifier f(x); X refers to the data set generated by the positioning model, xi refers to a data in X, Y refers to the data set generated by the measured data, yi refers to a data in Y, and Rn refers to the real number set;

[0055] (1) For t = 1, 2, 3, ..., T;

[0056] The training set in the offline phase is sampled t times, and several groups of data are randomly extracted with replacement; this group of positioning data is used to train the t-th model G t (x) Perform training; during the training process, select several sets of data with obvious features for subtree division;

[0057] (2) Coordinate prediction in the online stage;

[0058] The random forest model obtains multiple prediction results, and the multiple results are averaged to obtain the position of the target to be measured output by the model.

[0059] The beneficial effects of the present invention are as follows: because the signal is easily interfered by various factors in the environment during propagation, resulting in unstable transmission, which limits the application of traditional indoor positioning methods in actual scenarios; the present invention combines existing knowledge with actual data to derive an indoor positioning model, and then generates positioning data through the model, and uses the positioning data and data measured in actual scenarios to train the fingerprint positioning method model, thereby optimizing the model training process. Compared with the single positioning method in the prior art, the positioning method described in the present invention can achieve higher positioning accuracy by combining and supplementing the information of the measured data and the model positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of the positioning method described in the present invention.

[0061] Figure 2 This is a flowchart for indoor positioning using 5G micro base stations.

[0062] Figure 3 It is a schematic diagram of wireless positioning. DETAILED DESCRIPTION

[0063] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.

[0064] like Figure 1 As shown, for the problem that indoor positioning is not accurate enough, the present invention proposes a 5G micro base station indoor positioning method based on data and model, comprising the following steps:

[0065] Step 1: Build a positioning model for 5G micro base stations based on the channel fading model and measured data collected in actual scenarios;

[0066] Step 2: Calculate the distance between micro base stations based on the channel fading model;

[0067] Step 3: Generate wireless indoor positioning data based on the fixed number of 5G micro base stations in the positioning model;

[0068] Step 4: Input the wireless indoor positioning data and measured data into the random forest model for training at the same time to obtain the accurate position of the target to be measured.

[0069] 5G positioning model is implemented, and the specific expression is as follows:

[0070] According to 3GPP and other standards, the millimeter wave frequency bands of 5G devices are 28Ghz and 39Ghz, which are the key frequency bands of 5G millimeter waves. In 3GPP TR, some hot research scenarios of 5G indoor positioning were given, including conferences, laboratories, and shopping malls. Therefore, in the subsequent verification and experimental process, since 39Ghz is mainly used in conference rooms, indoor rooms and other scenarios, the 39GHz frequency band is used as the main frequency band in the indoor positioning method experiment.

[0071] Since wireless signals will experience different fading when propagating in space, generally speaking, signal fading includes two parts, namely shadow fading and path loss.

[0072] The commonly used path fading formula is shown in the formula:

[0073]

[0074] PL is the distance-dependent path fading, where P Tx Represents the power of the signal transmitting end, P Rx represents the power at the receiving end, G Tx is the transmit gain, G Rx is the receiving gain and Lc represents the total loss.

[0075] In the implementation of the positioning model, a fading model based on the CI model is adopted, where the fading expression is shown as follows:

[0076]

[0077] In the CI model, the distance d is calculated by the signal strength value or the same type of parameters received between the base station or communication nodes, and the center frequency f is determined by the specific transmission frequency band of the node. Based on the measured data, the model parameters are optimized to obtain the path loss index n and the shadow fading standard deviation χ σ The value of .

[0078] In the preliminary implementation of the positioning model, we assume that the base stations are distributed in a random geometric form to ensure the universality of the results. In general, when the distribution of wireless network base stations is geometrically random, the homogeneous Poisson point process can be used for simulation.

[0079]

[0080] As shown above, assuming a two-dimensional homogeneous Poisson point process with a distribution density of λ, the expectation of the distance between the nth closest node to the origin and the origin is:

[0081]

[0082] The expected distance squared is obtained as follows:

[0083]

[0084] like Figure 2 As shown in the figure, wireless indoor positioning data is generated based on the number of fixed 5G micro base stations. The specific representation is as follows: Assuming that the current position of the target to be measured is (x0, y0), the estimated distance E(r n 2 ) is regarded as the actual distance between the Nth micro base station and the target to be measured, then:

[0085] E(r n 2 )=(x n -x0) 2 +(y n -y0) 2

[0086] Where E(r n 2 ) represents the estimated distance between the Nth node base station and the target to be measured, (x n -x0) 2 +(y n -y0) 2 is the distance formula between two points;

[0087] For the N selected base stations, the equation group can be obtained:

[0088]

[0089] Subtract each equation from the first and rewrite the entire system of equations in matrix form:

[0090]

[0091] The whole equation can be simplified to

[0092] Hx=b

[0093] in

[0094]

[0095] Then the solution of the system of equations is:

[0096] x=(H T H) -1 H T b; H T is the transposed matrix of H.

[0097] like Figure 3As shown in the figure, if the current positions of the three base stations are (x1, y1), (x2, y2), and (x3, y3), the current position of the target to be measured is (x0, y0).

[0098] Assuming that the reference node is a micro base station, the channel fading model adopts the following formula:

[0099]

[0100] PL is a function based on the center frequency f and the node distance d, which can be obtained:

[0101]

[0102] For wireless positioning, we need to estimate the distance based on the existing models and data, obtain the location of the node to be determined based on the distance and location information given by the reference point, and convert the formula into the distance d as the dependent variable and the received signal strength P Rx That is, the formula with RSSI value as parameter or independent variable:

[0103]

[0104]

[0105] Get the distance d(P Rx ,f) function:

[0106]

[0107] Substituting into the H matrix:

[0108]

[0109]

[0110] From the above formula, we can deduce the process of estimating the accurate location of the target by the received signal strength RSSI value within the coverage area of ​​three base stations. If the number of base stations is increased, the original process can be used for deduction.

[0111] The random forest model is a training model in the existing technology. The random forest model builds multiple decision trees and combines them to obtain a more accurate and stable model. It is a combination of bagging ideas and random selection features. The random forest constructs multiple decision trees. When a sample needs to be predicted, the prediction results of each tree in the forest for the sample are counted, and then the final result is selected from these prediction results by voting. The random forest improves the prediction accuracy without significantly increasing the amount of calculation. The random forest is not sensitive to multivariate collinearity, and the results are relatively robust to missing data and unbalanced data. It can well predict the effects of up to thousands of explanatory variables and is known as one of the best algorithms currently.

[0112] We use the random forest model here mainly because it can balance errors for class imbalanced data.

[0113] The positioning data generated by the positioning model and the actual collected data are input into the random forest model for training to obtain the precise location; the specific representation is:

[0114] The input of the random forest model is the training set consisting of positioning data and measured data and the number of iterations T, where is the positioning data in the dataset, y i ∈Y is the real coordinate position of the positioning data in the experimental scene, i = 1, 2, 3, ..., N, and the output is the final classifier f(x); X refers to the data set generated by the positioning model, xi refers to a data in X, Y refers to the data set generated by the measured data, yi refers to a data in Y, and Rn refers to the real number set;

[0115] (1) For t = 1, 2, 3, ..., T;

[0116] The training set in the offline phase is sampled t times, and several groups of data are extracted with replacement; if the data error is too large, replace it, such as (0,0), (1,1) and (100,100), then (100,100) is replaced and the data is re-extracted, and the group of positioning data is used to t-th model G t (x) Perform training; during the training process, select several sets of data with obvious features for subtree division;

[0117] (2) Coordinate prediction in the online stage;

[0118] The random forest model will produce multiple prediction results, and the multiple results will be averaged to obtain the position of the target to be measured output by the model.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to be a further limitation of the present invention. All equivalent changes made using the contents of the present specification and drawings are within the protection scope of the present invention.

Claims

1. A 5G micro base station indoor positioning method based on data and model, characterized in that: The method comprises the following steps: Step 1: Based on the channel fading model and the measured data collected in the actual scenario, a positioning model of the 5G micro base station is constructed, specifically: Signal fading includes shadow fading and path loss. In the implementation of the positioning model, a fading model based on the CI model is adopted, where the fading expression is shown in the following formula: Wherein, d represents the distance between micro base stations, which is converted from the signal strength value received between micro base stations or the same type of parameters; f represents the center frequency, which is determined by the specific transmission frequency band of the micro base station; the CI model is optimized based on the measured data to obtain the path loss index w and the shadow fading standard deviation χ σ The value of; d0 is the distance from the target to be measured to the micro base station, dB is the unit, and log is the operator; Assuming that the micro base stations are distributed in a random geometric form, the CI model is simulated using a homogeneous Poisson point process, and a two-dimensional homogeneous Poisson point process with a distribution density of λ is obtained: Where P represents Poisson distribution, Q([a,b])=k means that the time interval is ba, the number of events Q occurring is k, and λ represents the distribution density. Then the expected distance between the micro base station closest to the target to be measured and the target to be measured is: The expected distance squared is obtained as follows: Step 2: Calculate the distance between micro base stations based on the channel fading model; Step 3: Generate wireless indoor positioning data based on the fixed number of 5G micro base stations in the positioning model; Step 4: Input the wireless indoor positioning data and measured data into the random forest model for training at the same time to obtain the accurate position of the target to be measured.

2. According to a data and model-based 5G micro base station indoor positioning method according to claim 1, it is characterized in that: Step 2: According to the channel fading model, the distance between micro base stations is solved as follows: Assuming that the reference node is a micro base station, the channel fading model adopts the following formula: The path fading formula is: Where P Tx Represents the power of the signal transmitting end, P Rx represents the power at the receiving end, G Tx is the transmit gain, G Rx is the receiving gain, Lc represents the total loss, and PL is a function based on the center frequency f and the node distance d, which can be obtained: The distance is estimated based on the existing model and data, and the position of the target to be measured is obtained based on the distance and position information given by the reference node. The formula is converted into the distance d as the dependent variable and the received signal strength P Rx That is, the formula with RSSI value as parameter or independent variable: Get the distance d(P Rx ,f) function:

3. According to a data and model-based 5G micro base station indoor positioning method according to claim 2, it is characterized in that: In step 3, based on the fixed number of 5G micro base stations in the positioning model, the specific steps of generating wireless indoor positioning data are: Assume that the current position of the target to be measured is (x0, y0), and the estimated distance E(r n 2 ) is regarded as the actual distance between the Mth micro base station and the target to be measured, then: E(r n 2 )=(x n -x0) 2 +(y n -y0) 2 Where E(r n 2 ) represents the estimated distance between the Mth micro base station and the target to be measured, (x n -x0) 2 +(y n -y0) 2 is the distance formula between two points; For the M selected base stations, the equation group can be obtained: Subtract each equation from the first and rewrite the entire system of equations in matrix form: The whole equation can be simplified to Hx=b in Then the solution of the system of equations is: x=(H T H) -1 H T b; Η T is the transposed matrix of H.

4. According to a data and model-based 5G micro base station indoor positioning method according to claim 1, it is characterized in that: In step 4, the positioning data generated by the positioning model and the actual collected data are input into the random forest model for training to obtain the precise location, specifically: The input of the random forest model is the training set consisting of positioning data and measured data and the number of iterations T, where is the positioning data in the dataset, v i ∈V is the real coordinate position of the positioning data in the experimental scene, i = 1, 2, 3, ..., N, and the output is the final classifier f(x); U refers to the data set generated by the positioning model, u i refers to a piece of data in U, V refers to the data set generated by the measured data, and v i Refers to a piece of data in V, R n is the set of real numbers; (1) For t = 1, 2, 3, ..., T; The training set in the offline phase is sampled t times, and several groups of data are randomly extracted with replacement; this group of positioning data is used to train the t-th model G t (x) Perform training; during the training process, select several sets of data with obvious features for subtree division; (2) Coordinate prediction in the online stage; The random forest model obtains multiple prediction results, and the multiple results are averaged to obtain the position of the target to be measured output by the model.

Images