CN101466145A - Dual-base-station accurate orientation method based on neural network - Google Patents

Abstract

The invention belongs to the wireless positioning field, which relates to a dual-basic station accurate positioning method based on the neural network. The method comprises the following steps: firstly, evaluating scattering body information around a mobile terminal according to the positioning data measured by the basic station, then utilizing the neural network to remove the fuzzification of the scattering body error information, and finally, using scattering body positioning information which restrains the non line of sight error as a virtual basic station to realize the accurate position. The provided dual-basic station accurate positioning method based on the neural network can effectively restrain the non line of sight transmission error, and has the advantages of stable performance, low cost and high positioning accuracy.

Description

Double-basis station accurate positioning method based on neural net

Technical field

The present invention relates to the wireless location method in the radio communication, a kind of double-basis station accurate positioning method based on neural net particularly is provided.

Background technology

Wireless location technology is used widely technical having obtained of military and civilian.Current quick increase along with data service and multimedia service, people increase day by day to the demand based on the new business of wireless location technology, especially at the communication environment of complexity, in environment such as airport hall, exhibition room, warehouse, supermarket, library, underground parking, mine, usually need to determine portable terminal or its holder, facility and article in indoor positional information, these have all promoted the further investigation to wireless location technology.Simultaneously, one of benchmark service that accurate localization information become the new generation of wireless communication system is provided to the user, wireless location technology also has been applied to aspects such as emergency relief, auto navigation, intelligent transportation, Team Management, with reference to the actual demand of numerous industries, the development prospect of wireless location technology and positioning service will be very wide.

But be subjected to the restriction of conditions such as positioning time, positioning accuracy and complicated indoor environment, fairly perfect location technology also can't be utilized at present well.Many location technology solutions will increase new hardware as wireless location technologies such as A-GPS (agps system) location technology, Bluetooth technology, infrared technologies on portable terminal, this will bring adverse influence to the size and the cost of mobile radio station.Receive the signal that detection is moved as sent simultaneously by a plurality of base stations, by network travelling carriage is positioned estimation according to the parameter that measures.Utilizing the cellular network of Modern Mobile Communications Systems that navigation, positional information are provided is a kind of new locate mode.It is a series of fixing or carry the mobile base station and the information-related signal of location of mobile station is determined location of mobile station by measure arriving (or from), inexpensive positioning service can be provided, just device fabrication merchant and Virtual network operator actively the thinking problem.But because under the radio propagation environment of complexity such as city, because barrier is more, radio propagation environment is abominable, signal is difficult to directly arrive from the base station travelling carriage, generally to after superrefraction or reflection, produce multipath signal, and the arrival receiving system of non line of sight, the TOA of signal measures and very big error also just occurred, so positioning accuracy can be subjected to very big influence.

When utilizing cellular communications networks that portable terminal is carried out wireless location, NLOS (Non Line Of Sight, non-line-of-sight propagation), factors such as measure error will make positioning performance be seriously influenced.Existing method is subjected to the influence of various interference easily, also need the base station more than 3 could realize the location, and position error is bigger under multipath NLOS environment, and positioning accuracy is not high.

Summary of the invention

The present invention is directed to the deficiencies in the prior art, propose a kind of positioning accuracy height, the double-basis station accurate positioning method based on neural net that cost is low is applicable to various localizing environments, is a kind of high accuracy practicability location technology.And, because the neural net of training can be stored the knowledge of relevant process, can directly from historical control information, learn.Just can then this information and current measurement data be compared, reach in the ability that goes out correct conclusion that has under the noise situations so neural net has filtering noise according to the daily historical data training network of object.So just can come process errors information with neural net.Method of the present invention is exactly the characteristics that have scattered information and bigger propagated error in the communication environment according to reality, the accurate location that utilizes metrical informations such as the electric wave time of advent and angle of arrival to realize portable terminal.

The present invention adopts following technical scheme, and it comprises the following steps:

(1) under the non-line-of-sight propagation environment, the propagation of portable terminal emission electric wave produces multipath by a plurality of scattering object reflections back, and these multipath signals just can arrive the base station with the different time from different directions; Utilize typical cellular network to carry out layout, total arrangement is a cellular system that radius of society is L, and concrete coordinate is distributed as: (0,0), $\left(\mathrm{0,2}L\right),(\sqrt{3}L,L),(\sqrt{3}L,-L),(-\sqrt{3}L,L),(-\sqrt{3}L,L)(-\sqrt{3}L,-L),$ Its total arrangement can come adaptive adjustment according to the environment of reality; It is 2 that the base station number that participates in the location is set, and a dominant base is wherein arranged; Record angle of arrival AOA via the portable terminal of a plurality of scattering objects reflection by the double-basis station, dominant base also records the electric wave TOA time of advent of portable terminal;

(2) to be distributed in the mobile terminal MS point be the center of circle to scattering object, and radius is on the disk of r, establishes coordinate to be: and mobile terminal MS (x, y), base station BS _j(x _j, y _j), scattering object S _i(sx _i, sy _i); BS ₁Be serving BS, i.e. dominant base, the electric wave of MS is through arriving scattering object S with angle beta _iArrive base station BS after the reflection again _j, β goes up to obey evenly at [π, π] and distributes, and MS and BS _jBetween distance greater than the reflection circle radius r, the maximum non-line-of-sight propagation angle that then arrives the base station as can be known behind scattering object expands to: $\mathrm{\α}=\arcsin \left(\frac{r}{\sqrt{(x-x_i)+(y-y_i)}}\right);$

(3) multipath number that records of each base station is the M2 road, with the scattering object number relation is arranged, and is 2 times of scattering object number; The angle of arrival AOA that is reflected back by scattering object that records is θ _Ji(j=1,2, i=1,2 ..., M2), promptly the angle of arrival AOA value of i the scattering object that records in j base station is respectively θ _JiIt is l that the electric wave that dominant base records MS is converted into distance value through the electric wave TOA time of advent of i scattering object _Li

(4) each measured value has comprised non-line-of-sight propagation sum of errors measure error, sets up the TOA error profile model of multiple scattering by above TOA distance value and communication environments, and the model mathematical formulae is: l _Ji=l _Ji ⁰+ l _{NLOS ji}+ l _Nji=l _Ji'+l _Nji, wherein: l _Ji ⁰For portable terminal to the base station apart from true value, l _NjiBe its measure error, obeying average is 0, and standard deviation is σ _LnGaussian Profile, l _{NLOS ji}Be the range error that non-line-of-sight propagation causes, l _Ji' be the radio wave propagation distance; And: θ _Ji=θ _Ji ⁰+ θ _Nji+ θ _{NLOS ji}, wherein: θ _Ji ⁰Be the true value of AOA, θ _NjiBe measure error, θ _{NLOS ji}The additional angle expansion that causes for the non-line-of-sight propagation of each scattering object multipath;

(5) for scattering object S _iBy two base station BSs ₁, BS ₂The angle of arrival AOA that records can get: $\tan {\mathrm{\θ}}_{\mathrm{ji}}=\frac{{\mathrm{sx}}_i-x_j}{{\mathrm{sy}}_i-y_j}$ , can to obtain three scattering object coordinates on disk model be S to a plurality of equations of simultaneous in this way _i(sx _i, sy _i); Scattering object position S _i(sx _i, sy _i) passing through of determining obtain after least square method is estimated, the mathematical operation of body is expressed as follows :-sx _i+ (tan θ _Ji) sy _i=-x _j+ (tan θ _Ji) y _j,

Make Z _i=(sx _i, sy _i) ^T $G_i=\left[\begin{array}{cc}-1& \tan {\mathrm{\θ}}_{1i}\\ -1& \tan {\mathrm{\θ}}_{2i}\end{array}\right],$ $H_i=\left[\begin{array}{c}-x_1+y_1\tan {\mathrm{\θ}}_{1i}\\ -x_2+y_2\tan {\mathrm{\θ}}_{2i}\end{array}\right];$ With Z=[z ₁| z ₂| z ₃] for the system of linear equations of variable be: H=GZ, then AOA measures corresponding error vector and is: Ψ=H-GZ, in the formula: G=[G ₁| G ₂| G ₃] ^T, H=[H ₁| H ₂| H ₃], then its least square solution is: Z=(G ^TG) ^-1G ^TH promptly obtains scattering object information; So, thereby try to achieve scattering object S by the position of above data through each scattering object of just having obtained going out according to a preliminary estimate _iWith base station BS _jAir line distance be: ${l\′}_{\mathrm{ji}}=\sqrt{{({\mathrm{sx}}_i-x_j)}^2+{({\mathrm{sy}}_i-y_j)}^2};$

(6) the electric wave TOA time of advent value of utilizing dominant base to record again deducts the air line distance that scattering object arrives the base station, has been exactly travelling carriage to the measuring distance of scattering object: L _i=l _Li-l _Li, then with scattering object as virtual base station, and L _iPosition estimation as travelling carriage to the electric wave TOA time of advent value of scattering object;

(7) owing to the existence of various errors under the nlos environment, such as the non-line-of-sight propagation error, measure error so often comprises very bigger error in the measured value, adopts effective neural net to handle these errors to obtain better positioning performance; This neural net is a kind of feedforward neural network, is three layers of feedforward network with single hidden layer, and the mathematical expression of its design is as follows: 1) use Gaussian function: $\mathrm{\φ}\left(x\right)=\exp (-\frac{x^2}{{\mathrm{\σ}}^2})$ As hidden layer, i.e. the basic function of hidden layer, ‖ ‖ represents the Euclidean norm.By its basis function vector that constitutes be: $F\left(x\right)=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_n\mathrm{\φ}\left(\left|\right|x-c_n\left|\right|\right),$ λ _nAnd c _nBe the network coefficient; If input layer, hidden layer and output layer neuron node number are respectively M, N, K, then the hidden layer unit is output as: $h_n=\exp (-\frac{{(X-C)}^T(X-C)}{{{\mathrm{\σ}}_n}^2}),$ Wherein, C is the vectormatrix that the center of hidden neuron is formed, σ _nIt is the width of n latent node.Network is output as:

, wherein, w _KnRepresent the connection weight of k output unit to n hidden unit, Be k neuronic threshold value; 2) non-linear mapping capability of neural net is embodied on the hidden layer basic function, and its characteristic is mainly by the basic function center C _n, width and neuron number are determined; Use the gradient descent method that the parameter of neural net is carried out the self adaptation adjustment, embody as follows: $J=\frac{1}{2}\mathrm{\ξ}{(W,k)}^2=\frac{1}{2}{(Y\left(k\right)-Y\′(W,k))}^2,$ Wherein, J is an error function, and Y (k) is the output of expectation, and (W k) is actual output to Y ', and W is the vector that all weights of neural net are formed.Hidden layer to the gradient that output layer links weight is adjusted into:

Hidden layer central value matrix adjustment algorithm is: $C_{k+1}=C_k+{\mathrm{\μ}}_k(-\frac{\∂J}{\∂C})=C_k+{\mathrm{\α}}_k(C_k-C_{k-1});$ The width adjustment algorithm is: ${\mathrm{\σ}}_{k+1}={\mathrm{\σ}}_k+{\mathrm{\μ}}_k(-\frac{\∂J}{\∂\mathrm{\σ}})={\mathrm{\σ}}_k+{\mathrm{\α}}_k({\mathrm{\σ}}_k-{\mathrm{\σ}}_{k-1}),$ In the formula, μ _kBe the learning rate of neural net, α _kFactor of momentum for its algorithm.

(8) train the diagnostic network that obtains expecting with this neural net of training sample set pair of effective quantity;

(9) handle θ with the neural net that obtains _JiAnd the L in the step (6) _iInformation; If neural net input layer and output layer are 9 neurons, with [L _i, θ _Ji] through after the correction of neural net, its output layer is revised angle of arrival AOA and electric wave TOA time of advent measured value is: $Y=[L_i\′,{\mathrm{\θ}}_{\mathrm{ji}}\′].$ Then with θ _Ji＇ brings into to handle in the step (5) and just can obtain revised scattering object coordinate $S_i^{\′}({\mathrm{sx}}_i^{\′},{\mathrm{sy}}_i^{\′});$

(10) with on the disk three not the scattering object of conllinear be basic point, the concrete weighted mass center algorithm of using carries out following location Calculation, the weighted mass center algorithm that adopts is through centroid algorithm having been done improvement, having embodied the influence degree of scattering object to the mobile terminal MS position by weighted factor exactly in classical centroid algorithm; Obtaining each scattering object to the factor of influence of MS is: ${\mathrm{\γ}}_1=\frac{1}{L_2\′+L_3\′},$ ${\mathrm{\γ}}_2=\frac{1}{L_1\′+L_3\′},$ ${\mathrm{\γ}}_3=\frac{1}{L_1\′+L_2\′};$ The MS position can estimate accurately so, and its elements of a fix are as follows: $\left[\begin{array}{c}\hat{x}\\ \hat{y}\end{array}\right]=\frac{1}{{\mathrm{\γ}}_1+{\mathrm{\γ}}_2+{\mathrm{\γ}}_3}\left[\begin{array}{ccc}{\mathrm{sx}}_1^{\′}& {\mathrm{sx}}_2^{\′}& {\mathrm{sx}}_3^{\′}\\ {\mathrm{sy}}_1^{\′}& {\mathrm{sy}}_2^{\′}& {\mathrm{sy}}_3^{\′}\end{array}\right]{\left[\begin{array}{ccc}{\mathrm{\γ}}_1& {\mathrm{\γ}}_2& {\mathrm{\γ}}_3\end{array}\right]}^T,$ L in the formula ₁', L ₂', L ₃' for by the inhibition that obtains after the neural net correction three scattering objects of error to the distance of MS; Finally obtain the result

Promptly obtain the exact position of portable terminal, just realized high-precision wireless location.

The present invention compared with prior art, have following conspicuous outstanding substantial characteristics and remarkable advantage: this inventive method only need utilize the double-basis station just can accurately locate portable terminal by the individual reflection statistical channel model.The locator data that the present invention is recorded by the base station estimates the scattering object information around the portable terminal, utilize neural net to eliminate the ambiguity of scattering object control information then, use the weighted mass center algorithm with the locating information that has suppressed the non line of sight error as virtual base station at last and realize accurate location.Double-basis station accurate positioning method based on neural net provided by the invention can effectively suppress the non-line-of-sight propagation error, and has stable performance, the low high advantage of positioning accuracy that reaches of cost.Because hi-Fix performance of the present invention can make positioning service obtain using more widely.

Description of drawings

Double-basis station accurate positioning method based on neural net of the present invention is provided by the following drawings and exemplifying embodiment.

Fig. 1 is the two dimensional surface schematic diagram of the wireless location modular concept of the inventive method;

Fig. 2 is the neural network structure design drawing of the inventive method;

Embodiment

Below will be described in further detail the double-basis station accurate positioning method based on neural net of the present invention.The specific implementation step of the high precision wireless positioning method based on multipath dispersion information of the present invention is as follows:

(1) under the non-line-of-sight propagation environment, in a typical cellular network, there are two base stations to record the angle of arrival of portable terminal (Angle Of Arrival, AOA) information, one of them dominant base also provides (the Time ofArrival electric wave time of advent of portable terminal, TOA), promptly be that the base station number that participates in locating is 2, a dominant base is wherein arranged;

(2) establishing coordinate is: and mobile terminal MS (x, y), base station BS _j(x _j, y _j), scattering object S _i(sx _i, sy _i); BS ₁Be serving BS, i.e. dominant base, the electric wave of MS is through arriving scattering object S with angle beta _iArrive base station BS after the reflection again _j, β goes up to obey evenly at [π, π] and distributes, and MS and BS _jBetween distance greater than the reflection circle radius r, the maximum non-line-of-sight propagation angle that then arrives the base station as can be known behind scattering object expands to: $\mathrm{\α}=\arcsin \left(\frac{r}{\sqrt{(x-x_i)+(y-y_i)}}\right);$

(3) multipath number that records of each base station is the M2 road, with the scattering object number relation is arranged, and is 2 times of scattering object number; The angle of arrival AOA that is reflected back by scattering object that records is θ _Ji(j=1,2, i=1,2.,, M2), promptly the angle of arrival AOA value of i the scattering object that records in j base station is respectively θ _JiIt is l that the electric wave that dominant base records MS is converted into distance value through the electric wave TOA time of advent of i scattering object _Li

(4) each measured value has comprised non-line-of-sight propagation sum of errors measure error, sets up the TOA error profile model of multiple scattering by above TOA distance value and communication environments, and the model mathematical formulae is: l _Ji=l _Ji ⁰+ l _NLOSji+ l _Nji=l _Ji'+l _Nji, wherein: l _Ji ⁰For portable terminal to the base station apart from true value, l _NjiBe its measure error, obeying average is 0, and standard deviation is σ _LnGaussian Profile, l _{NLOS ji}Be the range error that non-line-of-sight propagation causes, l _Ji' be the radio wave propagation distance; And: θ _Ji=θ _Ji ⁰+ θ _Nji+ θ _{NLOS ji}, wherein: θ _Ji ⁰Be the true value of AOA, θ _NjiBe measure error, θ _{NLOS ji}The additional angle expansion that causes for the non-line-of-sight propagation of each scattering object multipath;

(5) for scattering object S _iBy two base station BSs ₁, BS ₂The angle of arrival AOA that records can get: $\tan {\mathrm{\θ}}_{\mathrm{ji}}=\frac{{\mathrm{sx}}_i-x_j}{{\mathrm{sy}}_i-y_j}$ , can to obtain three scattering object coordinates on disk model be S to a plurality of equations of simultaneous in this way _i(sx _i, sy _i); Scattering object position S _i(sx _i, sy _i) passing through of determining obtain after least square method is estimated, the mathematical operation of body is expressed as follows :-sx _i+ (tan θ _Ji) sy _i=-x _j+ (tan θ _Ji) y _j,

Make Z _i=(sx _i, sy _i) ^T, $G_i=\left[\begin{array}{cc}-1& \tan {\mathrm{\θ}}_{1i}\\ -1& \tan {\mathrm{\θ}}_{2i}\end{array}\right],$ $H_i=\left[\begin{array}{c}-x_1+y_1\tan {\mathrm{\θ}}_{1i}\\ -x_2+y_2\tan {\mathrm{\θ}}_{2i}\end{array}\right];$ With Z=[z ₁| z ₂| z ₃] for the system of linear equations of variable be: H=GZ, then AOA measures corresponding error vector and is: Ψ=H-GZ, in the formula: G=[G ₁| G ₂| G ₃] ^T, H=[H ₁| H ₂| H ₃], then its least square solution is: Z=(G ^TG) ^-1G ^TH promptly obtains scattering object information; So, thereby try to achieve scattering object S by the position of above data through each scattering object of just having obtained going out according to a preliminary estimate _iWith base station BS _jAir line distance be: ${l\′}_{\mathrm{ji}}=\sqrt{{({\mathrm{sx}}_i-x_j)}^2+{({\mathrm{sy}}_i-y_j)}^2};$

(6) the electric wave TOA time of advent value of utilizing dominant base to record again deducts the air line distance that scattering object arrives the base station, has been exactly travelling carriage to the measuring distance of scattering object: L _i=l _Li-l _Li, then with scattering object as virtual base station, and L _iPosition estimation as travelling carriage to the electric wave TOA time of advent value of scattering object;

(7) owing to the existence of various errors under the nlos environment, such as the non-line-of-sight propagation error, measure error so often comprises very bigger error in the measured value, adopts effective neural net to handle these errors to obtain better positioning performance; This neural net is a kind of feedforward neural network, is three layers of feedforward network with single hidden layer, and the mathematical expression of its design is as follows: 1) use Gaussian function: $\mathrm{\φ}\left(x\right)=\exp (-\frac{x^2}{{\mathrm{\σ}}^2})$ As hidden layer, i.e. the basic function of hidden layer, ‖ ‖ represents the Euclidean norm.By its basis function vector that constitutes be: $F\left(x\right)=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_n\mathrm{\φ}\left(\left|\right|x-c_n\left|\right|\right),$ λ _nAnd c _nBe the network coefficient; If input layer, hidden layer and output layer neuron node number are respectively M, N, K, then the hidden layer unit is output as: $h_n=\exp (-\frac{{(X-C)}^T(X-C)}{{{\mathrm{\σ}}_n}^2}),$ Wherein, C is the vectormatrix that the center of hidden neuron is formed, σ _nIt is the width of n latent node.Network is output as:

(k=1 ..., K, n=1 ..., N), wherein, w _KnRepresent the connection weight of k output unit to n hidden unit,

Be k neuronic threshold value; 2) non-linear mapping capability of neural net is embodied on the hidden layer basic function, and its characteristic is mainly by the basic function center C _n, width and neuron number are determined; Use the gradient descent method that the parameter of neural net is carried out the self adaptation adjustment, embody as follows: $J=\frac{1}{2}\mathrm{\ξ}{(W,k)}^2=\frac{1}{2}{(Y\left(k\right)-Y\′(W,k))}^2,$ Wherein, J is an error function, and Y (k) is the output of expectation, and (W k) is actual output to Y ', and W is the vector that all weights of neural net are formed.Hidden layer to the gradient that output layer links weight is adjusted into:

Hidden layer central value matrix adjustment algorithm is: $C_{k+1}=C_k+{\mathrm{\μ}}_k(-\frac{\∂J}{\∂C})=C_k+{\mathrm{\α}}_k(C_k-C_{k-1});$ The width adjustment algorithm is: ${\mathrm{\σ}}_{k+1}={\mathrm{\σ}}_k+{\mathrm{\μ}}_k(-\frac{\∂J}{\∂\mathrm{\σ}})={\mathrm{\σ}}_k+{\mathrm{\α}}_k({\mathrm{\σ}}_k-{\mathrm{\σ}}_{k-1}),$ In the formula, μ _kBe the learning rate of neural net, α _kFactor of momentum for its algorithm.

(8) train the diagnostic network that obtains expecting with this neural net of training sample set pair of effective quantity;

(9) handle θ with the neural net that obtains _JiAnd the L in the step (6) _iInformation; If neural net input layer and output layer are 9 neurons, with [L _i, θ _Ji] through after the correction of neural net, its output layer is revised angle of arrival AOA and electric wave TOA time of advent measured value is: $Y=[L_i\′,{\mathrm{\θ}}_{\mathrm{ji}}\′].$ Then with θ _Ji' bring middle processing of step (5) into just can obtain revised scattering object coordinate $S_i^{\′}({\mathrm{sx}}_i^{\′},{\mathrm{sy}}_i^{\′});$

(10) with on the disk three not the scattering object of conllinear be basic point, the concrete weighted mass center algorithm of using carries out following location Calculation, the weighted mass center algorithm that adopts is through centroid algorithm having been done improvement, having embodied the influence degree of scattering object to the mobile terminal MS position by weighted factor exactly in classical centroid algorithm; Obtaining each scattering object to the factor of influence of MS is: ${\mathrm{\γ}}_1=\frac{1}{L_2\′+L_3\′},$ ${\mathrm{\γ}}_2=\frac{1}{L_1\′+L_3\′},$ ${\mathrm{\γ}}_3=\frac{1}{L_1\′+L_2\′};$ The MS position can estimate accurately so, and its elements of a fix are as follows: $\left[\begin{array}{c}\hat{x}\\ \hat{y}\end{array}\right]=\frac{1}{{\mathrm{\γ}}_1+{\mathrm{\γ}}_2+{\mathrm{\γ}}_3}\left[\begin{array}{ccc}{\mathrm{sx}}_1^{\′}& {\mathrm{sx}}_2^{\′}& {\mathrm{sx}}_3^{\′}\\ {\mathrm{sy}}_1^{\′}& {\mathrm{sy}}_2^{\′}& {\mathrm{sy}}_3^{\′}\end{array}\right]{\left[\begin{array}{ccc}{\mathrm{\γ}}_1& {\mathrm{\γ}}_2& {\mathrm{\γ}}_3\end{array}\right]}^T,$ L in the formula ₁, L ₂, L ₃For by the inhibition that obtains after the neural net correction three scattering objects of error to the distance of MS; Finally obtain the result

Promptly obtain the exact position of portable terminal, just realized high-precision wireless location.

Fig. 1 is the two dimensional surface schematic diagram of the inventive method wireless location model, and as shown in it, two base stations record the AOA of portable terminal, and master reference can also provide TOA.Be that electric wave passes through after wireless space is propagated, also just obtained multipath dispersion information.Portable terminal be MS (x, y), base station BS _j(x _j, y _j), dominant base BS ₁Can be positioned at initial point, scattering object is S _i(sx _i, sy _i).Below only TOA and AOA locator data are exactly that the inventive method will be utilized, so the condition of required location is less, realistic communication environment is with low cost fully.

Be illustrated in figure 2 as the neural network structure design drawing of the inventive method.Among the figure, this neural net mainly comprises input, hidden layer, and linear layer and output layer, wherein the transfer function of hidden layer is, the transfer function of output layer is pure linear function.The hidden layer of network has S1 neuron, and output layer has S2 neuron.X among the figure is the network input, and Y is network output.Its operation principle process can be divided into for two steps.At first, neural net is trained the diagnostic network that obtains expecting based on the training sample set (being commonly referred to " sign-error " data set) of some; Secondly, according to current diagnosis input system is diagnosed, the process of diagnosis is utilizes neural net to position the process of Error processing.Before study and diagnosis, need carry out suitable processing to diagnostic raw data and training sample data usually, comprise preliminary treatment and feature selecting/extraction etc., purpose is to provide suitable diagnosis input and training sample for diagnostic network.

In sum, the invention provides a kind of Dual base stations accurate positioning method based on neutral net, be applicable to various localizing environments, When particularly near the portable terminal of need location a lot of barrier being arranged, scattering environments is serious, under the condition that multipath enriches, such as The environment such as city, lofty mountains is the practical location technologies of a kind of high accuracy. The method is at first according to the actual scattering object information that records Locator data estimate scattered information, and set up one group of virtual base station, through many groups TOA and the AOA information that records, set up One group of virtual base station, next through the operations such as location Calculation of least-squares estimation, Processing with Neural Network error, weighted mass center algorithm The arrival target is pinpoint. Wireless location method energy establishment non-line-of-sight propagation error provided by the invention, and have into This is low, energy consumption low and the quite high advantage of positioning accuracy. Because hi-Fix performance of the present invention can make positioning service get To using more widely.

Claims (6)

1, a kind of double-basis station accurate positioning method based on neural net, be used for orienting the exact position of portable terminal from the radio wave signal that the double-basis station receives, it is characterized in that utilizing the good learning characteristic of neural net to eliminate the ambiguity that causes by the electric wave non-line-of-sight propagation, angle of arrival that the base station is recorded and electric wave are revised and suppress non-line-of-sight propagation sum of errors measure error the times of advent, realize the accurate location of portable terminal in the cellular communications networks; Its concrete operations step is as follows:

(1) under the non-line-of-sight propagation environment, the propagation of portable terminal emission electric wave produces multipath by a plurality of scattering object reflections back, and these multipath signals just can arrive the base station with the different time from different directions; Utilize typical cellular network to carry out layout, it is 2 that the base station number that participates in the location is set, and a dominant base is wherein arranged; Record angle of arrival AOA via the portable terminal of a plurality of scattering objects reflection by the double-basis station, dominant base also records the electric wave TOA time of advent of portable terminal;

(2) to be distributed in the mobile terminal MS point be the center of circle to scattering object, and radius is on the disk of r, establishes coordinate to be: and mobile terminal MS (x, y), base station BS _j(x _j, y _j), scattering object S _i(sx _i, sy _i); BS _lBe serving BS, i.e. dominant base, the electric wave of MS is through arriving scattering object S with angle beta _iArrive base station BS after the reflection again _j, β goes up to obey evenly at [π, π] and distributes, and MS and BS _jBetween distance greater than the reflection circle radius r, the maximum non-line-of-sight propagation angle that then arrives the base station as can be known behind scattering object expands to: $\mathrm{\α}=\arcsin \left(\frac{r}{\sqrt{(x-x_i)+(y-y_i)}}\right);$

(3) multipath number that records of each base station is the M2 road, and the angle of arrival AOA that is reflected back by scattering object that records is θ _Ji(j=1,2, i=1,2 ..., M2), promptly the angle of arrival AOA value of i the scattering object that records in j base station is respectively θ _JiIt is l that the electric wave that dominant base records MS is converted into distance value through the electric wave TOA time of advent of i scattering object _Li

(4) each measured value has comprised non-line-of-sight propagation sum of errors measure error, sets up the TOA error profile model of multiple scattering by above TOA distance value and communication environments, and the model mathematical formulae is: l _Ji=l _Ji ⁰+ l _NLOSji+ l _Nji=l _Ji'+l _Nji, wherein: l _Ji ⁰For portable terminal to the base station apart from true value, l _NjiBe its measure error, obeying average is 0, and standard deviation is σ _LnGaussian Profile, l _{NLOS ji}Be the range error that non-line-of-sight propagation causes, l _Ji' be the radio wave propagation distance; And: θ _Ji=θ _Ji ⁰+ θ _Nji+ θ _{NLOS ji}, wherein: θ _Ji ⁰Be the true value of AOA, θ _NjiBe measure error, θ _{NLOS ji}The additional angle expansion that causes for the non-line-of-sight propagation of each scattering object multipath;

(5) for scattering object S _iBy two base station BSs ₁, BS ₂The angle of arrival AOA that records can get: $\tan {\mathrm{\θ}}_{\mathrm{ji}}=\frac{{\mathrm{sx}}_i-x_j}{{\mathrm{sy}}_i-y_i},$ Can to obtain three scattering object coordinates on disk model be S to a plurality of equations of simultaneous in this way _i(sx _i, sy _i); So by the position of above data through each scattering object of just having obtained going out according to a preliminary estimate, thereby the air line distance of trying to achieve scattering object Si and base station BS j is: ${l\′}_{\mathrm{ji}}=\sqrt{{({\mathrm{sx}}_i-x_j)}^2+{({\mathrm{sy}}_i-y_j)}^2};$

(6) the electric wave TOA time of advent value of utilizing dominant base to record again deducts the air line distance that scattering object arrives the base station, has been exactly travelling carriage to the measuring distance of scattering object: L _i=l _Li-l _Li', then with scattering object as virtual base station, and L _iPosition estimation as travelling carriage to the electric wave TOA time of advent value of scattering object;

(7) owing to the existence of various errors under the nlos environment, such as the non-line-of-sight propagation error, measure error so often comprises very bigger error in the measured value, adopts effective neural net to handle these errors to obtain better positioning performance; This neural net is a kind of feedforward neural network, is three layers of feedforward network with single hidden layer;

(8) train the diagnostic network that obtains expecting with this neural net of training sample set pair of effective quantity;

(9) handle θ with the neural net that obtains _JiAnd the L in the step (6) _iInformation; If neural net input layer and output layer are 9 neurons, with [L _i, θ _Ji] through after the correction of neural net, its output layer is revised angle of arrival AOA and electric wave TOA time of advent measured value is: Y=[L _i', θ _Ji'].Then with θ _Ji' bring middle processing of step (5) into just can obtain revised scattering object coordinate S _i' (sx _i', sy _i');

(10) with on the disk three not the scattering object of conllinear be basic point, specifically use the weighted mass center algorithm and carry out following location Calculation: obtaining each scattering object to the factor of influence of MS is: ${\mathrm{\γ}}_1=\frac{1}{L_2\′+L_3\′},$ ${\mathrm{\γ}}_2=\frac{1}{L_1\′+L_3\′},$ ${\mathrm{\γ}}_3=\frac{1}{L_1\′+L_2\′};$ The MS position can estimate accurately so, and its elements of a fix are as follows: $\left[\begin{array}{c}\hat{x}\\ \hat{y}\end{array}\right]=\frac{1}{{\mathrm{\γ}}_1+{\mathrm{\γ}}_2+{\mathrm{\γ}}_3}\left[\begin{array}{ccc}{\mathrm{sx}}_1\′& {\mathrm{sx}}_2\′& {\mathrm{sx}}_3\′\\ {\mathrm{sy}}_1\′& {\mathrm{sy}}_2\′& {\mathrm{sy}}_3\′\end{array}\right]{\left[\begin{array}{ccc}{\mathrm{\γ}}_1& {\mathrm{\γ}}_2& {\mathrm{\γ}}_3\end{array}\right]}^T,$ L in the formula ₁', L ₂', L ₃' for by the inhibition that obtains after the neural net correction three scattering objects of error to the distance of MS; Finally obtain the result

Promptly obtain the exact position of portable terminal, just realized high-precision wireless location.

2, a kind of double-basis station accurate positioning method as claimed in claim 1 based on neural net, it is characterized in that: the typical cellular network of utilization in the described step (1) carries out layout and is: total arrangement is a cellular system that radius of society is L, concrete coordinate is distributed as: (0,0), (0,2L)

Its total arrangement can come adaptive adjustment according to the environment of reality, and the base station that participates in the location is wherein any 2.

3, a kind of double-basis station accurate positioning method based on neural net as claimed in claim 1, it is characterized in that: the multipath number in the described step (3) is the M2 road, with the scattering object number relation is arranged, and is 2 times of scattering object number.

4, a kind of double-basis station accurate positioning method based on neural net as claimed in claim 1 is characterized in that: the scattering object position S in the described step (5) _i(sx _i, sy _i) passing through of determining obtain after least square method is estimated, the mathematical operation of body is expressed as follows :-sx _i+ (tan θ _Ji) sy _i=-x _j+ (tan θ _Ji) y _j,

Make Z _i=(sx _i, sy _i) ^T, $G_i=\left[\begin{array}{cc}-1& \tan {\mathrm{\θ}}_{1i}\\ -1& \tan {\mathrm{\θ}}_{2i}\end{array}\right],$ $H_i=\left[\begin{array}{c}{-x}_1+y_1\tan {\mathrm{\θ}}_{1i}\\ {-x}_2+y_2\tan {\mathrm{\θ}}_{2i}\end{array}\right];$

With Z=[z ₁| z ₂| z ₃] for the system of linear equations of variable be: H=GZ, then AOA measures corresponding error vector and is: ψ=H-GZ, in the formula: G=[G ₁| G ₂| G ₃] ^T, H=[H ₁| H ₂| H ₃], then its least square solution is: Z=(G ^TG) ^-1G ^TH promptly obtains scattering object information.

5, a kind of double-basis station accurate positioning method as claimed in claim 1 based on neural net, it is characterized in that: will be used for the neural net of process errors in the described step (7), the mathematical expression of its design is as follows:

1) use Gaussian function: $\mathrm{\φ}\left(x\right)=\exp (-\frac{x^2}{{\mathrm{\σ}}^2})$ As hidden layer, i.e. the basic function of hidden layer, ‖ ‖ represents the Euclidean norm.By its basis function vector that constitutes be: $F\left(x\right)=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\λ}}_n\mathrm{\φ}\left(\left|\right|x-c_n\left|\right|\right),$ λ _nAnd c _nBe the network coefficient; If input layer, hidden layer and output layer neuron node number are respectively M, N, K, then the hidden layer unit is output as: $h_n=\exp (-\frac{{(X-C)}^T(X-C)}{{{\mathrm{\σ}}_n}^2})$ , wherein, C is the vectormatrix that the center of hidden neuron is formed, σ _nIt is the width of n latent node.Network is output as: Wherein, W _KnRepresent the connection weight of k output unit to n hidden unit,

Be k neuronic threshold value;

2) non-linear mapping capability of neural net is embodied on the hidden layer basic function, and its characteristic is mainly by the basic function center C _n, width and neuron number are determined; Use the gradient descent method that the parameter of neural net is carried out the self adaptation adjustment, embody as follows: $J=\frac{1}{2}\mathrm{\ξ}{(W,k)}^2=\frac{1}{2}{(Y\left(k\right)-Y\′(W,k))}^2,$ Wherein, J is an error function, and Y (k) is the output of expectation, and (W k) is actual output to Y ', and W is the vector that all weights of neural net are formed.Hidden layer to the gradient that output layer links weight is adjusted into:

Hidden layer central value matrix adjustment algorithm is: $C_{k+1}=C_k+{\mathrm{\μ}}_k(-\frac{\∂J}{\∂C})=C_k+{\mathrm{\α}}_k(C_k-C_{k-1});$

The width adjustment algorithm is: ${\mathrm{\σ}}_{k+1}={\mathrm{\σ}}_k+{\mathrm{\μ}}_k(-\frac{\∂J}{\∂\mathrm{\σ}})={\mathrm{\σ}}_k+{\mathrm{\α}}_k({\mathrm{\σ}}_k-{\mathrm{\σ}}_{k-1}),$ In the formula, μ _kBe the learning rate of neural net, α _kFactor of momentum for its algorithm.

6, a kind of double-basis station accurate positioning method as claimed in claim 1 based on neural net, it is characterized in that: the weighted mass center algorithm that adopts in the described step (10) is through centroid algorithm having been done improvement, having embodied the influence degree of scattering object to the mobile terminal MS position by weighted factor exactly in classical centroid algorithm.

Images