Mobile cellular signal ranging method and positioning system based on machine learning

Abstract

The invention discloses a mobile cellular signal ranging method and positioning system based on machine learning. The method comprises the following steps of: continuously receiving two paths of I / Q original signals sent by a base station, performing time synchronization so as to calculate frame starting positions, performing differential operation on the frame starting positions at previous and later moments, generating a clock skew data set, establishing a time sequence prediction model of the base station, and extracting clock skew information between the base station and a receiver from the time sequence prediction model. When a system starts to carry out distance measurement between a base station and a receiver, a pseudo range and a frame starting position which are obtained at a previous moment are utilized to carry out analysis and difference to serve as input of a time sequence prediction model, and the clock skew of the base station at a subsequent moment is accurately predicted. And the clock skew is compensated to the pseudo-range estimation process, so that the distance between the base station and the receiver is calculated. The invention provides a method for realizing accurate and sustainable distance measurement and positioning of a receiver on the basis of the existing mobile communication network infrastructure.

Description

Technical Field

[0001] The present invention relates to the field of positioning and navigation technology, and in particular to a mobile cellular signal ranging method and positioning system based on machine learning. Background Art

[0002] With the applicability of aircraft systems, such as drones, they play an increasingly important role in completing civilian tasks (such as power inspection, agricultural automation and aerial photography) and drone-based military tasks (such as assault defense, strike, reconnaissance and detection). Existing aircraft system navigation mainly relies on GNSS and inertial navigation system (INS) for navigation and positioning. However, when it flies to a complex environment, the navigation and positioning accuracy will be seriously reduced. The obstruction of buildings and trees in urban environments, the obstruction of mountainous terrain, and electronic interference on the battlefield will cause serious interference to GNSS signals, resulting in a GNSS denial state, resulting in a significant reduction in the combined navigation and positioning capabilities.

[0003] In order to ensure the navigation and positioning accuracy of aircraft systems, the study of ranging and positioning methods as a supplement or alternative source for integrated navigation has recently become a widely studied topic. At present, under GNSS denial conditions, integrated navigation mainly relies on inertial navigation and vision modules. The use of mobile cellular signals for ranging is not perfect enough. There is no complete and reliable ranging technology solution using mobile cellular signals. The ranging error is large and cannot meet the navigation and positioning needs. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a mobile cellular signal ranging method and positioning system based on machine learning, which can obtain time information by analyzing the mobile cellular signal, and perform error analysis on the clock deviation between the receiver and the base station, and then use machine learning methods to predict and compensate for the clock deviation.

[0005] A mobile cellular signal ranging method and positioning system based on machine learning, the method comprising:

[0006] S1, obtain the GPS coordinates of each local base station and use the GNSS antenna to obtain the GPS coordinates of the receiver;

[0007] S2, training phase, using software radio equipment to receive mobile cellular signals transmitted by the base station and obtain receiver I / Q two-way sampling data;

[0008] S3, time synchronization is performed on the collected mobile cellular signals, and the frame start position and cell ID of the FDD base station and the TDD base station are calculated respectively;

[0009] S4, performing a difference between the frame start positions at the kth time and the k+1th time, and calculating the difference in clock deviation between the base station and the receiver;

[0010] S5, constructing a time series prediction data set from the clock deviation difference sequence, and classifying and storing the clock deviation difference sequence according to FDD and TDD base stations to construct different time series prediction data sets;

[0011] S6, for the time series prediction data set constructed in S5, divide it into a training set and a test set according to a certain ratio. Perform model training for the time series prediction model, which includes: first constructing a time series prediction model and initializing its model parameters, and then taking the first n data in the data set as the input of the model. Calculate the difference between the model output result and the n+1th data in the data set, and update the model parameters based on the feedback of the result;

[0012] S7, measurement phase, using GNSS antenna to obtain the GPS coordinates of the receiver. Using software radio equipment to receive mobile cellular signals transmitted by the base station, obtain the receiver I / Q two-way sampling data;

[0013] S8, calculating the pseudo distance between the base station and the receiver based on the GPS coordinates of the base station and the GPS coordinates of the receiver;

[0014] S9, performing a differential on the pseudorange at the kth moment and the k+1th moment, and converting the differential pseudorange into a change value of the starting position of the real frame;

[0015] S10, performing time synchronization on the collected mobile cellular signals, and calculating the frame start positions and cell IDs of the FDD base station and the TDD base station respectively;

[0016] S11, performing a difference between the frame start position at the kth moment and the k+1th moment, and calculating a change value of the frame start position with an error;

[0017] S12, comparing the real frame start position change value and the frame start position change value with error, and calculating the clock deviation difference between the base station and the receiver;

[0018] S13, determine whether it is necessary to perform distance measurement between the base station and the receiver. If the system does not need to perform distance measurement at this time, the number of GNSS satellites is greater than or equal to 4 and the satellite signal is not spoofed, then repeat steps S7 to S12. If distance measurement is required, proceed to S14;

[0019] S14, when the system needs to perform ranging, S10 to S11 are repeated to calculate the change value of the frame starting position with error;

[0020] S15, using the previously acquired clock deviation difference sequence between the base station and the receiver as an input of a time series prediction model, and the model predicts the clock deviation difference between the base station and the receiver at a subsequent time;

[0021] S16, subtracting the predicted clock deviation difference between the base station and the receiver from the frame start position change value with error obtained in S14, and calculating the actual frame start position change value;

[0022] S17, converting the real frame start position change value into a pseudorange change value, adding it to the pseudorange between the base station and the receiver at the previous moment, to obtain the accurate pseudorange between the receiver and the base station at the current moment;

[0023] S18, whether the system needs to perform ranging again, if not, repeat steps S7 to S12, if yes, go to S14;

[0024] Preferably, in step S2, using software radio equipment to receive mobile cellular signals transmitted by a base station requires continuous and static data collection at the same location.

[0025] Preferably, in step S3, time synchronization of the mobile cellular signal specifically includes four steps, PSS timing synchronization, CP type detection, frequency offset estimation compensation, and SSS detection. First, the cell ID number and 5ms half-frame timing coarse synchronization are obtained through PSS detection, and then the CP type is determined through CP type detection and OFDM symbol timing synchronization is ensured, and then timing fine synchronization and carrier frequency synchronization are performed through frequency offset estimation compensation, and finally the cell ID number and frame start position can be obtained through SSS detection and 10ms frame timing synchronization can be ensured.

[0026] Preferably, in step S4, data differentiation is performed on the frame start position according to the following formula:

[0027] Δτ r (k) = [τ r (k)-τ r (k-1)]

[0028] where τ r (k) and τ r (k-1) is the frame start position parsed by the receiver at time k and time k-1 respectively.

[0029] Preferably, in step S6, the constructed time series prediction model is not limited to a specific machine learning model, and all machine learning models that can perform time series prediction can be used to train and predict the clock deviation difference between the base station and the receiver.

[0030] Preferably, in step S6, different machine learning methods are used for training and prediction in view of the differences between the mechanisms of the FDD base station and the TDD base station.

[0031] Preferably, in step S8, the pseudorange between the base station and the receiver is calculated according to the following formula:

[0032]

[0033] i∈{1,...,N} represents the signal index, where N is the total number. k=0,1,... represents the time index, r r =[x r ,y r ] T is the two-dimensional position vector of the receiver. s,i =[x s,i ,y s,i ] T is the known two-dimensional position vector of the cellular base station. In addition, Δz r,s,i (k) = z r (k)-z s,i is the height difference between the receiver and the base station. If only two-dimensional ranging positioning is required, it can be ignored. If three-dimensional ranging positioning is required, it cannot be ignored.

[0034] Preferably, in step S9, the differential pseudorange and the frame start position change value can be converted into each other according to the following formula:

[0035] Δρ i =Δτ r (k)·T·c

[0036] where Δρ i is the pseudorange after differentiation, Δτ r (k) is the frame start position after differentiation, c is the speed of light, and T is the sampling time of the LTE system with a bandwidth of 20 MHz, which is 32.55 ns.

[0037] Compared with the prior art, the mobile cellular signal ranging method and positioning system based on machine learning provided by the present invention has the following advantages:

[0038] The mobile cellular signal ranging method and positioning system based on machine learning proposed in the present invention can extract time information from the mobile cellular signal using software radio equipment. A complete ranging technical scheme using mobile cellular signals is established, and a practical clock deviation model is established for the clock deviation between the base station and the receiver. The clock deviation model is trained using a machine learning method, which can effectively predict and compensate for the clock deviation. Thereby, the ranging accuracy between the base station and the receiver is improved. The scheme of the present invention has high feasibility and strong universality, and provides a supplement to the ranging technology in the existing combined navigation and positioning system, thereby improving the credibility of the combined navigation positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a principle block diagram of a mobile cellular signal ranging method based on machine learning provided by the present invention;

[0040] Figure 2 It is a schematic diagram of the experimental scene and hardware facilities provided by the present invention;

[0041] Figure 3 The present invention provides a method for obtaining a frame start position;

[0042] Figure 4 It is a clock error trend diagram of the FDD base station provided by the present invention;

[0043] Figure 5 is a clock error trend diagram of a TDD base station provided by the present invention;

[0044] Figure 6 is a schematic diagram of a neural network provided by the present invention; DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0046] like Figure 1 As shown, in an outdoor environment, an LTE signal is selected, and ranging information is extracted through a signal sent by an LTE base station. The method includes:

[0047] (1) Obtain the local base station information database in advance.

[0048] (2) Select the center frequency as 2644.4Mhz, 1850Mhz. Receive LTE signals through software radio device USRP X310. Figure 2As shown, data is collected at two locations, A and B. LTE signals transmitted by eNB1 and eNB2 are collected. Data is collected for at least one day, so that the overall trend of base station clock deviation can be effectively reflected.

[0049] (3) Collect GNSS data through the GNSS antenna.

[0050] (4) Obtain the I / Q sampling data of USRP X310 and perform time synchronization on the data (PSS timing synchronization, CP type detection, frequency offset estimation compensation, SSS detection), such as Figure 3 As shown in the figure, the frame start position and cell ID can be calculated through time synchronization.

[0051] (5) The data is divided into two categories, FDD and TDD, according to the LTE signal type. The frame start position at time k and time k+1 is differentiated to calculate the difference in clock deviation between the base station and the receiver.

[0052] (6) The clock deviation difference is used to construct a machine learning time series prediction data set. The clock deviation difference is classified and stored by FDD and TDD base stations to construct different time series prediction data sets. In order to obtain good training results, at least two base stations of the same type should be collected for both FDD and TDD base stations, and the collection time should exceed one day. The sample size of the data should exceed 5000.

[0053] (7) The data after difference is a one-dimensional column vector with an interval of 10 seconds. Figure 4 and Figure 5 Represent the clock deviation trends of FDD base stations and TDD base stations respectively.

[0054] (8) The training set and the test set are divided into two sets in a ratio of 8:2. Model training is performed for the machine learning time series prediction model. Here, the RBF neural network is selected. The radial basis function (RBF) network consists of three layers: input, hidden, and output layers. The neural network structure is as follows: Figure 6 As shown. The learning rate is set to 0.001, the input layer size is 5, the hidden layer is three layers, the size is 128, and the output layer size is 1. The kernel function uses the Gaussian function. [X1X2....X n ] represents the clock deviation of the previous moments. Output [y1y2...y p] is the clock deviation of the next moment. The hidden layer nodes are described by radial basis functions, and the output layer nodes are usually described by simple linear functions. The activation function of the hidden layer neurons generates a local response to the input signal. The input signal is usually considered to be a kernel function near the center, and the hidden layer nodes can produce a large output. Therefore, the radial basis function network is a local approximation network with a faster learning speed in practical applications. The first 5 data in the data set are used as the input of the model. Calculate the difference between the model output result and the sixth data in the data set, and use the result to feedback and update the model parameters.

[0055] (9) During the measurement phase, the USRP X310 was used to receive mobile cellular signals with center frequencies of 2644.4Mhz and 1850Mhz to obtain the receiver I / Q sampling data. The GNSS antenna was used to obtain the GPS coordinates of the receiver.

[0056] (10) The straight-line distance between the receiver and the base station can be calculated from the GPS position of the receiver and the base station as the true distance. The GPS coordinates of the base station are located at Figure 2 The locations of eNB1 and eNB2.

[0057] (11) The pseudoranges at the kth moment and the k+1th moment are differentiated, and the differentiated pseudoranges are converted into the change value of the starting position of the real frame.

[0058] (12) Time synchronization is performed on the collected mobile cellular signal data, and the frame start position and cell ID of the FDD base station and TDD base station are calculated respectively.

[0059] (13) The frame start position at the kth moment and the k+1th moment are differentiated to calculate the frame start position change value with error.

[0060] (14) Compare the actual frame start position change value with the error frame start position change value to calculate the clock deviation difference between the base station and the receiver.

[0061] (15) Determine whether the system needs to measure the distance between the base station and the receiver. If not, and the number of GNSS satellites is greater than or equal to 4 and is not spoofed, repeat steps 9 to 14. If not, proceed to step 16.

[0062] (16) When the system needs to perform ranging, steps 12 to 13 are repeated to calculate the change value of the frame starting position with error.

[0063] (17) The previously acquired clock deviation difference sequence between the base station and the receiver is grouped into five as the input of the time series prediction model, and the model predicts a future clock deviation difference between the base station and the receiver. This difference can also be used as the input of the next predicted clock deviation difference.

[0064] (18) Subtract the predicted clock deviation difference between the base station and the receiver from the error-containing frame start position change value obtained in step 16 to calculate the actual frame start position change value.

[0065] (19) The actual frame start position change value is converted into a pseudorange, added to the pseudorange between the base station and the receiver at the previous moment, and the accurate pseudorange between the receiver and the base station at the current moment is calculated. The pseudorange can also be used as a true value to predict the situation at the next moment.

[0066] (20) Determine again whether the system needs to perform ranging. If not, repeat steps 9 to 14. If yes, proceed to step 15.

[0067] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A mobile cellular signal ranging method based on machine learning, characterized in that: The following steps are involved: S1, obtain the GPS coordinates of each local base station and use the GNSS antenna to obtain the GPS coordinates of the receiver; S2, training phase, using software radio equipment to statically receive mobile cellular signals transmitted by the base station to obtain receiver I / Q two-way sampling data; S3, time synchronization is performed on the collected mobile cellular signal data, and the frame start position and cell ID of the FDD base station and the TDD base station are calculated respectively; S4, performing a difference between the frame start positions at the kth time and the k+1th time, and calculating the difference in clock deviation between the base station and the receiver; S5, constructing a time series prediction data set from the difference sequence of clock deviations, and classifying and storing the data according to FDD and TDD base stations to construct different time series prediction data sets; S6, for the time series prediction data set constructed in S5, divide it into a training set and a test set according to a certain ratio. Perform model training for the time series prediction model, which includes: first constructing a time series prediction model and initializing its model parameters, and then using the first n data in the data set as the input of the model. Calculate the difference between the model output result and the n+1th data in the data set, and update the model parameters based on the result; S7, measurement phase, using the GNSS antenna to obtain the GPS coordinates of the receiver, and using the software radio equipment to receive the mobile cellular signal transmitted by the base station to obtain the receiver I / Q two-way sampling data; S8, calculating the pseudo distance between the base station and the receiver based on the GPS coordinates of the base station and the GPS coordinates of the receiver; S9, performing a differential on the pseudorange at the kth moment and the k+1th moment, and converting the differential pseudorange into a change value of the starting position of the real frame; S10, performing time synchronization on the collected mobile cellular signals, and calculating the frame start position and cell ID of the FDD base station and the TDD base station respectively; S11, performing a difference between the frame start position at the kth moment and the k+1th moment, and calculating a change value of the frame start position with an error; S12, comparing the real frame start position change value and the frame start position change value with error, and calculating the clock deviation difference between the base station and the receiver; S13, determine whether it is necessary to perform distance measurement between the base station and the receiver. If the system does not need to perform distance measurement at this time, the number of GNSS satellites is greater than or equal to 4 and the satellite signal is not spoofed, then repeat steps S7 to S12. If distance measurement is required, proceed to S14; S14, when the system needs to perform ranging, S10 to S11 are repeated to obtain the change value of the frame starting position with error; S15, using the previously acquired clock deviation difference sequence between the base station and the receiver as an input of a time series prediction model, and the model predicts the clock deviation difference between the base station and the receiver at a subsequent time; S16, subtracting the predicted clock deviation difference between the base station and the receiver from the frame start position change value with error obtained in S14, and calculating the actual frame start position change value; S17, converting the real frame start position change value into a pseudorange change value, adding it to the pseudorange between the base station and the receiver at the previous moment, to obtain the accurate pseudorange between the receiver and the base station at the current moment; S18, determine again whether the system needs to perform ranging. If not, repeat steps S7 to S12. If yes, enter S14.

2. The mobile cellular signal ranging method and positioning system based on machine learning according to claim 1, characterized in that: In step S2, using software radio equipment to receive mobile cellular signals transmitted by a base station requires continuous and static data collection at the same location.

3. The method for mobile cellular signal ranging based on machine learning according to claim 1, characterized in that: In step S3, time synchronization of mobile cellular signals specifically includes four steps: PSS timing synchronization, CP type detection, frequency offset estimation and compensation, and SSS detection. First, the cell ID number and 5ms half-frame timing coarse synchronization are obtained through PSS detection, and then the CP type is determined through CP type detection and OFDM symbol timing synchronization is ensured, and then the timing fine synchronization and carrier frequency synchronization are performed through frequency offset estimation and compensation, and finally the cell ID number and frame start position can be calculated through SSS detection and 10ms frame timing synchronization can be ensured.

4. The method for mobile cellular signal ranging based on machine learning according to claim 1, characterized in that: In step S4, data differentiation is performed on the frame start position according to the following formula: Dt r (k)=[τ r (k)-t r (k-1)] where τ r (k) and τ r (k-1) is the frame start position parsed by the receiver at time k and time k-1 respectively.

5. The method for mobile cellular signal ranging based on machine learning according to claim 1, characterized in that: In step S6, the constructed time series prediction model is not limited to a specific machine learning model, and all machine learning models that can perform time series prediction can be used to train and predict the clock deviation difference between the base station and the receiver.

6. The time series prediction model according to claim 6, characterized in that: Due to the different characteristics of FDD base stations and TDD base stations, different machine learning methods will be used for training and prediction.

7. The method for mobile cellular signal ranging based on machine learning according to claim 1, characterized in that: In step S8, the pseudorange between the base station and the receiver is calculated according to the following formula: i∈{1,...,N} represents the signal index, where N is the total number. k=0,1,... represents the time index, r r =[x r ,y r ] T is the two-dimensional position vector of the receiver. s,i =[x s,i ,y s,i ] T is the known two-dimensional position vector of the cellular base station. In addition, Δz r,s,i (k) = z r (k)-z s,i is the height difference between the receiver and the base station. If only two-dimensional ranging positioning is required, it can be ignored. If three-dimensional ranging positioning is required, it cannot be ignored.

8. The method for mobile cellular signal ranging based on machine learning according to claim 1, characterized in that: In step S9, the differential pseudorange and the frame start position change value can be converted to each other according to the following formula: Dr. i =Dt r (k)·T·c where Δρ i is the pseudorange after differentiation, Δτ r (k) is the frame start position after differentiation, c is the speed of light, and T is the sampling time of the LTE system with a bandwidth of 20 MHz, which is 32.55 ns.

Images