Colocated method based on EKF and PF under the conditions of nonlinear and non-Gaussian
Technical field
The present invention relates to indoor positioning tracking technique field, be specially under the conditions of a kind of nonlinear and non-Gaussian based on EKF and
The colocated method of PF.
Background technology
Along with the fast development of microelectric technique, communication technology and computer technology, wireless location is as Sensor Network and thing
The important application of networking, is increasingly paid close attention to by people.Currently, localization method based on range finding mainly has RSSI (to arrive letter
Number intensity), AOA (arriving signal angle), TOA (time of arrival (toa)), TDOA (signal step-out time) etc..Wherein, RSSI is
The most more satisfactory a kind of algorithm, is converted into distance by propagation loss, but in actual location environment, electromagnetic wave propagation
There is dynamic characteristic, range measurement is caused severe jamming, bigger range error can be produced;Location side based on AOA range finding
Hardware system equipment required for formula is complicated, relatively costly;TOA and TDOA positioning precision is higher, and positioning time is short, and system realizes
Simply, there is certain ability of anti-multipath, so having preferable application prospect.Wherein, the method for estimation of TDOA typically has two
Kind: a kind of is that the difference of the time of arrival (toa) TOA asking two base stations is to obtain TDOA value;Another kind is to use cross-correlation technique,
The signal that the signal received one base station and another base station receive carries out computing cross-correlation and calculates TDOA value.
Indoor environment is the most complex, and the signal between destination node and base station is propagated and is likely that there are NLOS error
(non-line-of-sight propagation).Use TOA and TDOA technology time destination node is carried out location estimation, TOA value can produce one positive
Additional excessive delay, TDOA value also corresponding can produce an error component.Should by this TOA or the TDOA value with bigger error
For the location estimation of destination node, necessarily cause being remarkably decreased of location algorithm performance, make estimation position that bigger error to occur.
Therefore, how to differentiate to become the key factor improving positioning precision with restraining NLOS error.To this, Wylie differential method
Being a kind of relatively effective method, first it determine whether line-of-sight propagation, if not regarding from the variance measuring parameter error
Away from propagation, then reconstruct line-of-sight propagation value, and the sighting distance value of reconstruct is filtered, then estimate mobile station with sighting distance location algorithm
Position.
Kalman filtering is to utilize minimum mean square error criterion to carry out target dynamic estimation in the case of linear Gauss
Excellent filtering method.In the case of but for nonlinear and non-Gaussian, in addition it is also necessary to make improvements and extend, and spreading kalman filter
Ripple (EKF, Extended Kalman Filter) is a kind of relatively effective improved method.Particle filter (PF, Particle
Filter) being a kind of non-linear, non-Gaussian filtering filtering method based on Monte Carlo thought, it breaches Kalman's filter completely
Ripple theoretical frame, process noise and measurement noise to system do not have any restriction, are processing non-linear, non-gaussian time-varying system
Parameter estimation and state filtering problem aspect have uniqueness advantage and wide prospect.
In actual application, EKF is a kind of the more commonly used non-linear filtering method, but it is only applicable to filtering
Error and the least situation of forecast error, otherwise may result in filtering instability and even dissipate;It is the most linear that PF breaks away from understanding
During filtering problem, random quantity must is fulfilled for the restriction condition of Gauss distribution, but this algorithm there is also some problems, and it relies on
Substantial amounts of sample data could the posterior probability density of preferably approximation system, amount of calculation is bigger.
In summary, EKF and PF has respective advantage, but is also respectively arranged with deficiency.Therefore, the present invention proposes a kind of non-linear non-
Colocated method based on EKF and PF under Gauss conditions.
Summary of the invention
It is an object of the invention to for the deficiencies in the prior art, it is provided that under the conditions of a kind of nonlinear and non-Gaussian based on EKF and
The colocated method of PF.
Step (1), utilize nanoLOC Development kit 3.0 development kit dispose in located space target joint
Point and base station, read destination node respectively with the TOA initial data of each base station, be calculated between destination node and each base station
Distance, wherein M is the total number in base station;
Step (2), Wylie differential method is utilized to judge rmWhether there is NLOS error;
Step (3), TOA value is done difference obtain TDOA value, and the range difference r to all TDOA values correspondencem1It is reconstructed;
Step (4), use EKF TDOA algorithm estimation tkThe position coordinates of moment destination node also judges;
Step (5), use PF TDOA algorithm estimation tkThe position coordinates of moment destination node also judges;
Step (6), to step (4), two position coordinateses of (5) judge, residual weighted obtains tkMoment final
Estimated value;
Step (7), all elements of a fix data obtaining step (6) are weighted smoothing, and obtain destination node
Estimate position coordinates eventually(x, y) compares, the location of method used herein with the true coordinate of destination node
Precision is better than the positioning precision of the localization method that nanoLOC development kit carries, also superior to being used alone EKF algorithm or PF algorithm
Positioning precision.
The method specifically comprises the following steps that
CSS (Chirp Spread Spectrum, linear frequency modulation spreads) is to open based on IEEE802.15.4a agreement
The wireless communication technology sent out.The positioning experiment platform of the present invention uses Nanotron company nanoLOC Development kit
3.0 development kit, this development kit, based on nanoLOC TRX radio frequency chip, can be used for developing communication based on CSS technology, survey
The wireless application such as away from, location.The present invention carries out building of alignment system by using this set development kit.
The alignment system of the present invention utilizes nanoLOC Development kit 3.0 development kit in the middle part of located space
Administration's destination node and base station, use one destination node of M architecture;Wherein, M base station is ordered respectively according to sequence counter-clockwise
Entitled A1, A2..., Am..., AM, the named Tag of destination node, measured obtain actual coordinate value for (x, y).In conjunction with Fig. 1
The step that is embodied as of the present invention is described:
Step one: first read the TOA initial data of destination node and base station, is calculated destination node and M base station
Between distance rm(m=1,2 ..., M).
Step 2: utilize Wylie differential method to judge distance r between destination node and base stationmWhether there is NLOS error.
Wylie differential method idiographic flow is as follows:
The location moment is designated as tk=0, t1,…,tK, wherein tKFor total positioning time, T is the interval of two adjacent moment
Time, it is assumed that destination node is smoothed by fitting of a polynomial with the range measurements of each base station, i.e.
Wherein, the exponent number of J-1 representative polynomial;rm(tk) represent that destination node is at tkMoment and the distance of m-th base station.
Method of least square is utilized to solve unknowm coefficientMeasured value after can being smoothed is
With Sm(tk) as actual distance reference value, calculate rm(tk) distance measure criteria deviation, be represented by
Assume that the distance measure criteria deviation under LOS environment is σm, in actual applications, σmCan be previously according to experimental field
Measure and obtain.
Under NLOS environment, owing to NLOS error exists with the measurement error under LOS environment simultaneously, following knot can be obtained
Really:
1. under LOS environment, it is known that
The most in a nlos environment, it is known that
Gained standard deviationThe biggest, show to be affected by NLOS the biggest.
Step 3: the signal between destination node and all base stations can be identified by step 2 and whether propagate by NLOS shadow
Ring.The distance measure criteria deviation that will recordSort from small to large, select all rmIn by NLOS affected minimum value, by it
It is set to r1, respective base station is as reference base station, and coordinate is (x1,y1);Distance between destination node and remaining base station is according to base station
Distribution sequence counterclockwise be set to r2,…,rM, respective base station coordinate is set to (x2,y2),…,(xM,yM)。
The conversion of TOA value is generated TDOA value, and the range difference obtaining correspondence is:
rm1=rm-r1,2≤m≤M (4)
If being measured statistics by reality it follows that destination node and base station are LOS propagation, then the distance that average delay is corresponding
For ηLOS.Propagate if destination node and base station are NLOS, then the distance that average delay is corresponding is ηNLOS.Corresponding to all TDOA values
Range difference rm1Value be reconstructed, method is as follows:
If 1. rmAnd r1It is LOS to propagate or NLOS propagation, then rm1=rm-r1;
If 2. rmPropagate for NLOS, r1Propagate for LOS, then rm1=(rm-ηNLOS)-(r1-ηLOS);
Note: because all rmValue has utilized Wylie differential method judge and sort, so r1Distance measure criteria poor
One definite proportion rmLittle, therefore there is not rmThe r for LOS propagates1Situation about propagating for NLOS.
Step 4: the range difference r corresponding by TDOA valuem1After carrying out sighting distance reconstruct, calculated by Kalman TDOA algorithm
To destination node at tkThe theoretical coordinate value in momentThen, first threshold δ is set1, according to inequality group:
Judging whether to meet formula (5), if meeting, then retaining this positioning resultContinue next step;Otherwise
Abandon next step to calculate, terminate this position fixing process, return step one and re-read original TOA data.
Step 5: through aforementioned four step, it is believed that this time location data used are accurate data;Utilizing should
Data, are calculated destination node at t by PF TDOA algorithmkThe theoretical coordinate value in momentSpecific as follows:
5.1 state equations setting up destination node motion and observational equation are as follows:
Xtk=F Xt(k-1)+S·Wt(k-1) (6)
Ztk=f (Xtk)+Vtk (7)
In formula, F is state-transition matrix;S is interference transfer matrix;Wt(k-1)And VtkIt is respectively process noise and observation is made an uproar
Sound;ZtkFor destination node at tkThe range difference that the TDOA observation of moment and all base stations is converted to;For destination node at moment tkStatus information, Xt(k-1)For destination node at moment tk-1State letter
Breath, whereinFor tkThe position coordinates of moment destination node,For tk(this speed can for the speed of moment destination node
Obtained by the acceleration transducer in the equipment of location, direction sensor estimation, or by pre-setting the initial speed of destination node
Degree and acceleration calculation obtain), transfer matrix is:
5.2 state equation moved according to destination node and observational equations, set up likelihood probability density function, see formula
(9):
Wherein,For destination node at tkThe range difference that the TDOA predictive value of moment and m-th base station is converted to,
For destination node at tkThe range difference that the TDOA observation of moment and m-th base station is converted to, σvFor observation noise VtkSide
Difference.Represent tkThe status information of moment i-th particle, i=1,2 ..., N, N are total number of particles;
After 5.3 have obtained state equation and observational equation and likelihood probability density function by 5.1 and 5.2, in conjunction with
PF TDOA algorithm filtering, obtains positioning result renewal step as follows:
(1) initialize: tk=0
By prior distribution p (x0) produce populationAll particle weights are 1/N.Wherein, p (x0) by destination node
Known initial state information bonding state equation obtains.
(2)for tk=t1,…,tK
1. at tkMoment, more new particle weights
And normalization
2. resampling, the problem solving sample degeneracy, i.e. carrying out step by step along with iteration, the weights of a lot of particles can become
Obtaining the least, the only particle weights of minority are relatively big, the problem that the number of effective particles in state space reduces.
Utilize effectively sampling yardstick NeffWeigh the degree of degeneration of particle weights
Set an effective sample number NthresholdAs threshold value, if Neff< Nthreshold, then carry out resampling, obtain new
PopulationAll particle weights are set to 1/N.
3. state estimation
Destination node is at tkThe status information in moment is
Obtain destination node at tkThe theoretical coordinate value in moment
The most more new state information
The t that will obtain in step 6kThe final elements of a fix value (x of moment destination nodetk,ytk) be assigned to
end
Step 6: Second Threshold δ is set2, utilize inequality:
Judge step 4, five twice estimated result obtainedWithThe most close, if meeting inequality,
Residual weighted formula (15) is then utilized to obtain tkThe final elements of a fix value (x of moment destination nodetk,ytk), utilize this coordinate figure
Update the status information of destination node previous moment in EKF TDOA algorithm and PF TDOA algorithm respectively;If being unsatisfactory for inequality,
Then terminate this position fixing process, return step one and read initial data.
Wherein,Corresponding residual sum of squares (RSS) is
Corresponding residual sum of squares (RSS) is
Threshold value δ1、δ2Choose and mainly initially select according to the theoretical precision of location equipment, then through actual reality
Test examination is finely adjusted, and realizes filtering the data affected by NLOS error with this, improves positioning precision.
Step 7, pass through step 6, it will obtain α (α≤K) individual elements of a fix value (xtk,ytk), give each data and add
Weight coefficient, is weighted smooth obtaining final estimated resultWherein, the weight coefficient of each data with get this number
According to time relevant, data acquisition must be got over early, and weight coefficient is the least, obtains the most late, and weight coefficient is the biggest.
For the evaluation of localization method performance, positioning precision is one of important indicator.In the present invention, positioning precision is by managing
Opinion estimates position coordinatesWith true location coordinate (x, y) between degree of closeness weigh:
According to the explanation of equipment service manual, under LOS environment under, nanoLOC Development kit 3.0 develops
External member can reach indoor 2 meters, the positioning precision of outdoor 1 meter.The colocated method that the present invention proposes is applied to this exploitation set
In part, it is possible to obtain positioning precision better than it, also will be better than being used alone existing EKF or PF algorithm.
The invention has the beneficial effects as follows: utilize Wylie differential method be identified NLOS error and suppress, preliminary eliminating is not
Just data;Consider the positioning performance of two kinds of TDOA algorithm for estimating of EKF and PF, under the conditions of proposing a kind of nonlinear and non-Gaussian simultaneously
Colocated method based on EKF and PF, the method compared with being more suitable for the localizing environment of nonlinear and non-Gaussian for EKF, relatively PF
For preferably avoid the use to incorrect data, decrease amount of calculation.This colocated method effectively reduces NLOS by mistake
The impact of difference, in conjunction with the advantage of EKF and PF, overcomes both deficiencies, it is achieved more accurate location simultaneously.
Accompanying drawing explanation
Fig. 1 is the positioning flow figure of the present invention;
Fig. 2 is the sensing node TDOA location model figure of the present invention.
Detailed description of the invention
The present invention relates to indoor positioning tracking technique field, be specially under the conditions of a kind of nonlinear and non-Gaussian based on EKF and
The colocated method of PF, decreases NLOS error, in conjunction with the advantage of EKF and PF, overcomes and is used alone one of which
The deficiency of algorithm.
Below in conjunction with accompanying drawing, the present invention will be further described.
CSS (Chirp Spread Spectrum, linear frequency modulation spreads) is to open based on IEEE802.15.4a agreement
The wireless communication technology sent out.The positioning experiment platform of the present invention uses Nanotron company nanoLOC Development kit
3.0 development kit, this development kit, based on nanoLOC TRX radio frequency chip, can be used for developing communication based on CSS technology, survey
The wireless application such as away from, location.The present invention carries out building of alignment system by using this set development kit.
The alignment system of the present invention utilizes nanoLOC Development kit 3.0 development kit in the middle part of located space
Administration's destination node and base station, use one destination node of four architectures, as shown in Figure 2;Wherein, four base stations are according to the inverse time
Pin order is respectively designated as A1, A2, A3, A4, the named Tag of destination node, measured obtain actual coordinate value for (x, y).Knot
Close Fig. 1 and the step that is embodied as of the present invention be described:
Step one: first read the TOA initial data of destination node and base station, be calculated destination node and four base stations
Between distance rm(m=1,2,3,4).
Step 2: utilize Wylie differential method to judgeWhether value has NLOS error.
Wylie differential method idiographic flow is as follows:
The location moment is designated as tk=0, t1,…,tK, wherein tKFor total positioning time, T is the interval of two adjacent moment
Time, it is assumed that the range measurements between each base station and destination node is smoothed by fitting of a polynomial, i.e.
Wherein, the exponent number of J-1 representative polynomial;rm(tk) represent that destination node is at tkMoment and the distance of m-th base station.
Method of least square is utilized to solve unknowm coefficientMeasured value after can being smoothed is
With sm(tk) as actual distance reference value, calculate rm(tk) distance measure criteria deviation, be represented by
Assume that the distance measure criteria deviation under LOS environment is σm, in actual applications, σmCan be previously according to experimental field
Measure and obtain.
Under NLOS environment, owing to NLOS error exists with the measurement error under LOS environment simultaneously, following knot can be obtained
Really:
1. under LOS environment, it is known that
The most in a nlos environment, it is known that
Gained standard deviationThe biggest, show to be affected by NLOS the biggest.
Step 3: can identify whether the propagation between destination node and all base stations is affected by NLOS by step 2.
The distance measure criteria deviation that will recordSort from small to large, select all rmIn by NLOS affected minimum value, set
For r1, respective base station is as reference base station, and coordinate is (x1,y1);Distance between destination node and remaining base station is according to base station
Distribution sequence is set to r counterclockwise2,r3,r4, respective base station coordinate is set to (x2,y2),(x3,y3),(x4,y4)。
The conversion of TOA value is generated TDOA value, and the range difference obtaining correspondence is:
rm1=rm-r1,2≤m≤4 (4)
If being measured statistics by reality it follows that destination node and base station are LOS propagation, then the distance that average delay is corresponding
For ηLOS.Propagate if destination node and base station are NLOS, then the distance that average delay is corresponding is ηNLOS.Corresponding to all TDOA values
Range difference rm1Value be reconstructed, method is as follows:
If 1. rmAnd r1It is LOS to propagate or NLOS propagation, then rm1=rm-r1;
If 2. rmPropagate for NLOS, r1Propagate for LOS, then rm1=(rm-ηNLOS)-(r1-ηLOS);
Note: because all rmValue has utilized Wylie differential method judge and sort, so r1Distance measure criteria poor
One definite proportion rmLittle, therefore there is not rmThe r for LOS propagates1Situation about propagating for NLOS.
Step 4: the range difference r corresponding by TDOA valuem1After carrying out sighting distance reconstruct, calculated by Kalman TDOA algorithm
To destination node at tkThe theoretical coordinate value in momentThen, first threshold δ is set1, according to inequality group:
Judging whether to meet formula (5), if meeting, then retaining this positioning resultContinue next step;Otherwise
Abandon next step to calculate, terminate this position fixing process, return step one and re-read original TOA data.
Step 5: through aforementioned four step, it is believed that this time location data used are accurate data;Utilizing should
Data, are calculated destination node at t by PF TDOA algorithmkThe theoretical coordinate value in momentSpecific as follows:
5.1 state equations setting up destination node motion and observational equation are as follows:
Xtk=F Xt(k-1)+S·Wt(k-1) (6)
Ztk=f (Xtk)+Vtk (7)
In formula, F is state-transition matrix;S is interference transfer matrix;Wt(k-1)And VtkIt is respectively process noise and observation is made an uproar
Sound;ZtkFor destination node at tkThe range difference that the TDOA observation of moment and all base stations is converted to;For destination node at moment tkStatus information, Xt(k-1)For destination node at moment tk-1State letter
Breath, whereinFor tkThe position coordinates of moment destination node,For tk(this speed can for the speed of moment destination node
Obtained by the acceleration transducer in the equipment of location, direction sensor estimation, or by pre-setting the initial speed of destination node
Degree and acceleration calculation obtain), transfer matrix is:
5.2 state equation moved according to destination node and observational equations, set up likelihood probability density function, see formula
(9):
Wherein,For destination node at tkThe range difference that the TDOA predictive value of moment and m-th base station is converted to,
For destination node at tkThe range difference that the TDOA observation of moment and m-th base station is converted to, σvFor observation noise VtkSide
Difference.Represent tkThe status information of moment i-th particle, i=1,2 ..., N, N are total number of particles;
After 5.3 have obtained state equation and observational equation and likelihood probability density function by 5.1 and 5.2, in conjunction with
PF TDOA algorithm filtering, obtains positioning result renewal step as follows:
(3) initialize: tk=0
By prior distribution p (x0) produce populationAll particle weights are 1/N.Wherein, p (x0) by destination node
Known initial state information bonding state equation obtains.
(4)for tk=t1,…,tK
1. at tkMoment, more new particle weights
And normalization
2. resampling, the problem solving sample degeneracy, i.e. carrying out step by step along with iteration, the weights of a lot of particles can become
Obtaining the least, the only particle weights of minority are relatively big, the problem that the number of effective particles in state space reduces.
Utilize effectively sampling yardstick NeffWeigh the degree of degeneration of particle weights
Set an effective sample number NthresholdAs threshold value, if Neff< Nthreshold, then carry out resampling, obtain new
PopulationAll particle weights are set to 1/N.
3. state estimation
Destination node is at tkThe status information in moment is
Obtain destination node at tkThe theoretical coordinate value in moment
The most more new state information
The t that will obtain in step 6kThe final elements of a fix value (x of moment destination nodetk,ytk) be assigned to
end
Step 6: Second Threshold δ is set2, utilize inequality:
Judge step 4, five twice estimated result obtainedWithThe most close, if meeting inequality,
Residual weighted formula (15) is then utilized to obtain tkThe final elements of a fix value (x of moment destination nodetk,ytk), utilize this coordinate figure
Update the status information of destination node previous moment in EKF TDOA algorithm and PF TDOA algorithm respectively;If being unsatisfactory for inequality,
Then terminate this position fixing process, return step one and read initial data.
Wherein,Corresponding residual sum of squares (RSS) is
Corresponding residual sum of squares (RSS) is
Threshold value δ1、δ2Choose and mainly initially select according to the theoretical precision of location equipment, then through actual reality
Test examination is finely adjusted, and realizes filtering the data affected by NLOS error with this, improves positioning precision.
Step 7 passes through step 6, it will obtain α (α≤K) individual elements of a fix value (xtk,ytk), give each data weighting
Coefficient, is weighted smooth obtaining final estimated resultWherein, the weight coefficient of each data with get this data
Time relevant, data acquisition must be got over early, and weight coefficient is the least, obtains the most late, and weight coefficient is the biggest.
For the evaluation of localization method performance, positioning precision is one of important indicator.In the present invention, positioning precision is by managing
Opinion estimates position coordinatesWith true location coordinate (x, y) between degree of closeness weigh:
According to the explanation of equipment service manual, under LOS environment, nanoLOC Development kit 3.0 develops set
Part can reach indoor 2 meters, the positioning precision of outdoor 1 meter.The colocated method that the present invention proposes is applied to this development kit
In, it is possible to obtain positioning precision better than it, also will be better than being used alone existing EKF or PF algorithm.
In a word, it is contemplated that reduce the impact on TDOA value of the NLOS error, EKF and PF is be combined with each other, makes full use of
Respective advantage, overcomes the deficiency being used alone a kind of algorithm so that localization method is more suitable for nonlinear and non-Gaussian simultaneously
Situation, reduces algorithm amount of calculation while improving positioning precision.
Above by with reference to accompanying drawing, the present invention is done special displaying and explanation, it will be apparent to those skilled in the art that
Should be understood that and make various modifications and changes in the form and details, all under without departing substantially from the thought of the present invention and scope
It will be the infringement to patent of the present invention.Therefore the present invention real thought to be protected and scope are limited by appending claims
Fixed.