The current wireless local area network localization scheme mainly based on the RSSI, we propose a scheme that could combine the advantage of two kinds of traditional RSSI based methods, triangulation method and RSSI fingerprinting method.
If we only care about the accuracy of localization, the RSSI fingerprinting method is more accurate and much more fault-tolerant than RSSI-based triangulation method in average for the time being. Nevertheless, RSSI fingerprinting method wastes too much time to collect all possible RSSI data at the indoor environment in offline training phase. On the contrary, the advantage of triangulation method is exactly no demand for offline training, we only need to do preliminary work once. So it’s high time to design a new scheme to merge the advantage of these two methods; we called it Hybrid Overlapping-based Algorithm (HOF Algorithm).
Hybrid Overlapping-based Algorithm (HOF Algorithm) is based on the signal coverage overlapping to determine the possible area of the wireless devices, and take the three (or more) monitor access point’s RSSI record as the fingerprint value of these possible areas. On the other hand, because we have known the type of wireless devices, we could categorize the wireless devices into two kinds. The first kind is access point and the second kind of wireless devices is moving station, such as notebook, PDA, smart phone and so on. We could simultaneously make use of static analysis for access points and dynamic analysis for moving stations. Static analysis means the wireless access points normally will not move due to the limitation of power supply and network resources. Therefore, even if the fluctuation of RSSI we still could collect the range of the fluctuation corresponding to some region of the environment.
3-1. Preliminary
Because the most significant advantage of RSSI-based triangulation method is that the method has no need to waste any time in training phase. The method only needs to determine a RSSI to geometric distance table. Ideally, if there are no obstacles between the transmitter and receiver, that is to say, no obstacles between the monitor access point and the moving station. But in real environments, the signal strength will be affected by many effects, such as signal refraction, signal diffraction, signal reflection, multipath effect and so forth. In the first place, an indoor noise model which is called Floor Attenuation Factor [8] (FAF) is adopted to eliminate the influence of signal multipath effect, signal attenuation and so on. The equation shows the result of Floor Attenuation Factor.
P(d) : the signal strength at distance d
n : the rate at which the path loss increase with distance
d0: the distance of the reference point
C : the maximum number of obstructions (walls) up to which the attenuation factor makes a difference
nW : the walls between Transmitter to Receiver
This model provides a flexible solution to solve the influence of indoor obstacles.
Although it cannot solve the entire phenomenon that the unstable signal produced, yet it still gives a quantitative method to eliminate the problem. But in real environment, the factor that may have an impact on the RSSI is much more complicated than the
nW * WAF nW < C C * WAF nW >= C
model mentioned before. For example, the signal attenuation of the angle between the transmitter and the receiver cannot be precisely calculated by this model. And the signal attenuation phenomenon by these effects cannot also be measured. By the way, let’s see the two cases in the following paragraphs.
Case 1: Influence of Multiple Floors
Sometimes moving station may in the position that may cause monitor access points to calculate a error position, such as determining at a fault floor. For another case, once the station is near the window, its signal may be received by different floors. It will cause a problem, which floor it located?
Owing to these special cases, a solution to will be proposed to solve such problem. In the first place, we concentrate on the moving station is near the window as the Figure.6 shown.
Figure.6. The Moving Station is Near The Window Proposed Solution
The monitor access points could be deployed in similar shape in order to collect uniform signal strength. Once the wireless devices are close to the window, the wireless device has great probability to be received by multiple floor monitor access points. In order to solve this problem, these monitor wireless devices that have receive this wireless devices of different floors can be ranged by the RSSI. If the monitor access points are on the same floor of the wireless device, the RSSIs of the same floor have more chance to receive stronger RSSIs. After receiving all monitor access points of different floors that can hear the wireless device, these monitor access points could be ranged by RSSI. That is to say, among all monitor APs, select those with RSSI values higher than the average. The floor of the monitor access points that belong to half of the stronger RSSI finally could be chosen.
Case 2: Stair Well
The HOF algorithm will firstly determine which floor the wireless devices are, but in some special cases, the wireless devices are neither in the upper floor nor in the lower floor. That is to say, the wireless devices are in the stair wells of the building.
For monitor access points, if the wireless devices can both be listened by the monitor access points of upper and lower floor, this kind of situation as “In-Between State”
could be defined just like the Figure 7 shown.
Figure.7 Stair Well of the Building
Proposed Solution
Two methods are proposed to solve this kind of problem. The first solution is use the history of previous localization result. All positions of wireless device could be recorded according to the indoor topology. If the previous calculated location of wireless device is near the stairs, then the next time both monitor access point of upper floor and lower floor either cannot detect this wireless device or both can detect
the wireless device with more or less RSSI. Because the previous location is near the stair well, the next position may have great possibility to be in-between state.
Another solution is to place a Monitor AP in all possible stair wells. In other words, a monitor AP could be placed in all stair wells before starting to localization. If the RSSI is especially the biggest than those of other the stair-well APs, the device must be in stair well. But on the contrary, it wastes many hardware resources on the monitor access points.
Oscillations of RSSI Signals
In reality, the RSSI signal represents a range of areas that the wireless devices may be. Ideally, all the possible positions that have equal RSSI to the monitor AP are a circle as the Figure.8 shown
Figure.8 the Most Ideal Situation of Signal Overlapping
But owing to the attenuation of RSSI signal and many obstacles in the indoor environments, the real range of receive signal strength is an abnormal curves as the
Figure.9 shown.
Figure.9 Real Range of RSSI of 3 Monitor APs
Consequently, in order to evaluate the influence that the obstacle will produce, the experiment should be design to measure how much influence that the obstacles made. And the result would result in such abnormal circle. An experiment to measure how much the effect of single wall should be designed. Firstly, a monitor access point place is beside the wall and on the other side of the wall, the wireless devices is place in different angles. For the sake of precisely measuring the signal attenuation by the obstacles through field experiments, the wireless device could be placed at different angles from 0oto 180oevery 10 degrees as the Figure.10 and Figure.11 shown.
Figure.10 Place the Wireless Device at Different Angles with Radius 100 cm
Besides, the effect of different angle could be measured as the Figure.11 shown.
The signal strength of 0ois use as a datum to compare signal strength from 0oto 90o every ten degrees. In addition, we reversely take 90o as a datum to compare signal strength from 90oto 180oevery ten degrees as the Figure.13 shown.
Figure.11 Take the Signal Strength of 90 degree as Datum Point
Figure.12 Take the Signal Strength of 0 degree as Datum Point
After measuring the effect of one single wall made of 27.5cm-concrete wall from different angles, an important conclusion is reached that signal strength through one single wall at different angles will decrease at most 31 %~ 33%. In average, the signal strength will decrease about 12%~15%. The experiment let us know how much effect the single wall made of 27.5cm-concrete wall will not make a great impact on the signal strength. In other words, these walls could be simplified since how much effect it would make has known by experiments.
Because our method is based on the signal overlapping to implement RSSI fingerprinting in indoor environment, the map has to be divided into the appropriate size. Due to the sensitivity of our experiment devices, the least distance that can make the RSSI change is about 1.5 meter. For the simplicity of implementation, the indoor map could be divided into uniform grid that we called Divide Unit (DU) in this paper
as the Figure.13 shown.
Figure.13 Divide the Map into uniform 1.5m x 1.5m Divide Unit
From the reasons given above, many problems have to be solved in real environment to make the more accurate localization results. So, in the following paragraph, a novel scheme according to the phenomenon that the indoor obstacles will produce would be produced. Our algorithm is to solve these problems that the related papers have not totally solved. Our method only needs online phase and let the program to learn without manually collecting the information from the environment.
3-2. Observatoin
The proposed scheme is based on the Triangulation method, so there is one thing that needs to check at first. Is the wireless device is under the intersection of three monitor access points? We have tried to do experiments to verify this idea. Through experiments, devices locate in the intersection of signal range of monitor access points with high possibility (>95%) as the Figure 14-1 shown below.
Figure.14-1 Devices locate in the intersection with high probability
After checking the wireless devices is under the intersection of signal range of the three monitor access points, we could use the fingerprinting method in the intersection with the information of our indoor map to improve the accuracy of localization as the figure 14-2 shown below.
Figure 14-2. Add Fingerprint in the divide units of intersection of 3 monitor APs
3-3. Proposed Scheme Architecture
The picture given above represents the system architecture of proposed scheme, and the arrow means the data flow of our system. RSSI triple will firstly be used to Triangulation and Fingerprinting Matching, the difference between Triangulation and Fingerprinting Matching is according to the previously collected data enough or not.
If the previously collected data is not enough, the system will only work by Triangulation method. Through Triangulation method in most cases, we could reduce the possible area of the devices, which is what we call Region Reduction, then we try to work with Fingerprinting method by the RSSI triple and reduced region together.
We called the step as Signature Feedback as Fingerprint. Once the previously collected data is not enough, the system will work both Triangulation and Fingerprinting Matching based on the Data Base and the latest RSSI triple. By the way, two methods are working simultaneously and going to get two different results.
But we will only adapt the result of Fingerprint methods owing to the method is generally more accurate than the Triangulation method. The last step is output the most possible divide unit of the device.
3-4.Localization Algorithm
Error Distance: ED Moving Station: STA Wireless Access Point: AP
Divide Unit: DU, 1.5 meters X 1.5 meter Square Unit
Input Argument:
RSSIn: RSSI of the target measured by APn
Localization Algorithm (RSSI 1, RSSI2, RSSI 3) Determine which floor the wireless device is Triangulation Method
Region Reduction
if( Intersection region DU < N) // N : Reasonable DU Num of Intersection ex: 10 if(exist DU has enough RSSI) // enough means 20 or more RSSI data
for each (DU that has enough RSSI)
Find the least difference between & received RSSInof stored DU else // Start up phase => inaccurate data
Add RSSI to all divide unit of Intersection of 3 MAP Use original (x, y) as output
else if ( Intersection region DU > N) // too much DU, Signal unstable need adjust //Fail data => need to Adjust
Adjust Algorithm (RSSIn, real (x, y)) // Improve Accuracy by the real pos Input the real (x, y) of wireless devices
Add RSSI to DU corresponding to real (x, y) Record final (x, y) and MAC of STA to DB
// provide admin to manually adjust accuracy
Adjust Algorithm (calculated (x, y), real pos (x, y), expected error distance)
Localization Flow Chart
Floor Determin
RSSI Triple
Start
Triangulation Method
Region Reduction
Signature Feedback
Fingerprint Matching
Output
Data Base
In order to explain the working process of our proposed method Hybrid Overlapping Fingerprinting more precisely, the diagram illustrates the flow of proposed scheme.
In the first place, the possible region is required to decide that the wireless could be. So the possible position could be found by the range of signal coverage. For any specific monitor access point, the RSSI fluctuates continuously with time. In reality, the range of the same signal strength should be an abnormal circle as the Figure15 shown.
Figure.15 Real Range of Same Signal Strength
However, this kind of intersection is hardly to calculate the intersection of the range of signal. But the signal fluctuation range must have its mean, max variation, min variation. Because the signal attenuation by the wall has been measured, the max attenuation of variation range of these indoor obstacles is also recorded. If the region of max variation intersection range is chosen, the high possibility could be derived to
contain the real position of the wireless devices as the Figure.16 shown
Figure.16 the Oscillation Range of RSSI
Figure.17 Ideally Max Intersection of Coverage Range of Monitor APs
But for the simplicity of implementation, it’s hardly to use the original
intersection to record the RSSI data. Consequently, the intersection of coverage range will be decided and translated to the corresponding divide units as the Figure.18 shown.
Figure.18 Possible Range of Wireless Devices in Divide Units
Subsequently, the range of intersection is going to be determined. Because once the intersection is too large (over 10 divide units), the coverage range must have great possibility that may lead to larger error distance. So only the data that the coverage range is smaller than the pre-defined threshold is adapted. Then it can be two possible cases. The first case is these intersection units have no enough RSSI fingerprinting data, so the data in these divide units of intersection range have to be recorded. Since the range that the wireless device may be only known, so the RSSI triple is uniformly added to all divide units of intersection. A data structure is designed to store these online collecting RSSI data, the format of the RSSI data is a triple such as (RSSI1, RSSI2, RSSI3), ex: (20, 41, 15). In the beginning of our Hybrid Overlapping
Fingerprinting method, these data from the range of three monitor access points’
signal overlapping will be collected and stored these triples in the corresponding divide units.
The second case is that once the intersection of some divide unit’s RSSI triple is enough, then many methods could be used to find the most possible position of the wireless device. Just like the Figure.19 shown, the intersection of this time has two divide units that have enough RSSI triple.
Figure.19 the DUs of Intersection Have Enough RSSI Triples
Firstly, in order to match the most possible divide unit from the divide units that have enough RSSI triples from the previously collected data. Here we use K-Nearest-Neighbor [13] algorithm to match the most possible divide unit. The next step, the only one point will be decided from the divide unit. Because the map information and the type of wireless devices have known, some impossible position of the divide unit could be excluded. In fact, a point from the selected divide unit is
randomly selected and then check if the point is on the impossible position such as walls or pillars. If not, the point is our final output. By contraries, if the point is on the impossible position, another point from the selected divide unit has to be randomly selected again until the point is not on the impossible position.
Figure.20 Determine the Divide Unit from DU that has Enough RSSI Triple
3-5. Weight Localization Algorithm
In the previous algorithm, a problem is found that all the divide units will have the same RSSI triple is unreasonable. Because the edge of intersection is only a little range in the intersection, so the probability that the wireless device in different divide unit is not equal. Consequently, a Weighted HOF algorithm should be designed to solve this problem.
As the Figure.20 shown blow, the divide units in the intersection of signal coverage range should initially be determined, then give the weight by the following 4 cases
Case 1: Only Vertex of the DU in Intersection Weight = 1
Case 2: Two Vertices of the DU in Intersection If Centroid is in Intersection
Weight = 2
Else // Centroid is Outside Intersection Weight = 1
Case 3: Three Vertices of the DU in Intersection Weight = 3
Case 4: Four Vertices of the DU in Intersection Weight = 4
Figure.21 The Example of Weighted HOF
The stored RSSI triple is
Figure.22 The Data Structure to Store Weighted Triple (0,2)
Another case is that once the intersection of divide unit is too large, which means the signal strength is highly oscillated by some reason. So under this situation, some improved algorithm has to be done to solve this problem just like the Figure.23 shown.
Figure.23 The Range of Overlapping is Too Large
When the overlapping range is too large, which means the localization result of this time is very inaccurate. Consequently, localization result of this time has to be discarded or use the adjust algorithm to improve the inaccurate result.
Figure.24 The Situation Need Adjust Algorithm Adjust Algorithm
Because the unstable data cannot be judged is an abnormal case or a normal case, this triple discard may be a kind of fault. So once the user could report the real position (x, y) to our server, the triple could be added to the corresponding divide unit of the real (x, y) of the user.
Adjust Algorithm
User Report Real (x,y)
Add the RSSI triple to the DU corresponding to Real (x, y)