The present invention is related to co-pending and commonly assigned U.S. patent application Ser. No. 10/635,367 entitled “Location Positioning in Wireless Networks,” filed Aug. 6, 2003, the disclosure of which is hereby incorporated herein by reference.
The present invention is directed toward wireless communications and, more particularly, to refining location positioning determinations for wireless devices.
It is sometimes desirable to locate the position of a station operable within a wireless, e.g., radio frequency (RF), network. For example, the United States Federal Communications Commission (FCC) has decreed that cellular telephone systems must implement systems to provide mobile telephone position information for use in emergency response, e.g., enhanced 911 (E911) emergency response. Additionally, the position of a station may be important for providing particular services, such as, for example, identifying subscribers and non-subscribers, resource allocation, network security, and location-sensitive content delivery, among other services.
In order to estimate a station's location, a system typically measures a metric that is a function of distance. A typical measured metric is signal strength, which decays logarithmically with distance in free space. Time information, such as time of arrival of a signal or time difference of arrival of a signal at diverse antennas, may be utilized as a measured metric from which distance information may be determined. Typically, several reference points are used with distance information derived from the measured metric in estimating location.
The use of global positioning system (GPS) receivers, which operate in conjunction with a network of middle earth orbit satellites orbiting the Earth to determine the receiver's position, has almost become ubiquitous in navigational applications. In such a GPS network, the aforementioned reference points are the satellites and the measured metric is the time of arrival of the satellite signal to the GPS receiver. The time of arrival of the satellite signal is typically directly proportional to the distance between the satellite and the GPS receiver due to a clear line of sight between the GPS receiver and satellite. By measuring the time of arrival associated with three satellites, a GPS receiver can calculate the longitude and latitude of the GPS receiver. By using time of arrival information with respect to a fourth satellite, a GPS receiver can also determine altitude.
In the aforementioned cellular networks, techniques including signal strength measurements and/or time difference of arrival have been implemented for location determination. For example, U.S. Pat. No. 6,195,556, the disclosure of which is incorporated herein by reference, teaches the use of signal strength measurements in combination with the time difference of arrival of a station's signal in determining the location of the station. Additionally, U.S. Pat. No. 6,195,556 teaches the use of mapping of received signal characteristics associated with particular positions (e.g., receive “signature” associated with each of a plurality of remote station locations) for use in determining a station's location. In the case of the aforementioned cellular network, the base transceiver stations (BTSs) are generally relied upon as the reference points from which distance determinations are made.
Wireless local area network (WLAN) location determination systems have been implemented in two phases: the offline phase and the online phase. In the offline phase, prediction or measurement of the fingerprint (e.g., signal strength, multipath characteristics, etcetera) of wireless access points at particular locations within the service area may be carried out. Location fingerprints may be predicted or measured off-line, such as when a network is being deployed, and are stored in a database resulting in a so-called radio map to relate the wireless signal information and coordinates of the known locations. In the online phase, the fingerprint associated with a remote station at an unknown location is measured during later operation of the network, and compared to the entries in the database. A location estimation algorithm is then applied to infer the location estimate for the unknown location. Location estimation algorithms include, for example but not limited to, triangulation, nearest neighborhood, K-nearest neighbor averaging, and history-based shortest path.
Previously, developing an accurate radio map for location determination required manual calibration throughout the network environment, meaning that before a location determination could be made, an engineer would actually have to physically go out and make calibration measurements at some specified points over the area covered by the network. Based on the manual measurements, the system would construct the radio map, and then make a location determination. This is known as supervised calibration or supervised training. Making manual calibration measurements is expensive and consumes significant manpower. Furthermore, because the wireless environment is constantly changing, the measured parameters are also changing, and repeating calibration to update the measurements is impractical and inefficient. Supervised training, requiring manual calibration, provides relatively accurate resolution, but over time, the accuracy fails as the networks parameters change. It is, therefore, desirable to eliminate the need for making costly and time consuming manual measurements.
The present invention is directed to systems and methods which monitor a network environment, collect client information available online, and refine location determinations of individual clients based on observed information as well as online information. More particularly, embodiments of the present invention comprise to systems and methods which monitor the wireless network, such as by collecting online receive signal strength indicator (RSSI) information observations from client users, to provide location determinations without requiring knowledge of those clients' precise locations.
Embodiments of the present invention are additionally directed to systems and methods to enhance the accuracy of the location determinations in a network, based on observed client information such as, for example, signal strength references. In one embodiment of the present invention, the method employs online received signal strength observations from multiple clients, with known or unknown locations, together with the original observed or estimated signal strength database to refine a radio map of the network environment. Online RSSI observations from client users may be compared with the original observed or estimated signal strength database and the radio map may be refined based on unsupervised training capabilities. Unsupervised system training according to embodiments of the present invention reduces or eliminates the need for live calibration of the network, and instead, existing measurements online can be used to calibrate and fine tune the radio map of the network environment. Additionally, according to embodiments of the present invention, collected RSSI information may be obtained from the normal network transmissions and therefore, does not require any extra overhead to obtain and use the information in location determination.
It is an object of embodiments of the present invention to create an original radio map of mobile station location without requiring manual calibration, by comparing online observations with a generic model estimation and following iterations through until the radio map is within a certain degree of accuracy.
It is a further object of embodiments of the present invention to update an existing radio map of mobile station location created by supervised training, without manually re-measuring network parameters to update calibrations.
It is a yet another object of embodiments of the present invention to use unsupervised training to update an existing radio map of mobile station location that was created by supervised training without expending additional money and manpower.
The foregoing has outlined rather broadly the features and technical advantages of the present invention in order that the detailed description of the invention that follows may be better understood. Additional features and advantages of the invention will be described hereinafter which form the subject of the claims of the invention. It should be appreciated that the conception and specific embodiment disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present invention. It should also be realized that such equivalent constructions do not depart from the invention as set forth in the appended claims. The novel features which are believed to be characteristic of the invention, both as to its organization and method of operation, together with further objects and advantages will be better understood from the following description when considered in connection with the accompanying figures. It is to be expressly understood, however, that each of the figures is provided for the purpose of illustration and description only and is not intended as a definition of the limits of the present invention.
For a more complete understanding of the present invention, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, in which
One embodiment of the present invention involves constantly monitoring a network environment, such as, for example, a wireless network, by collecting the information for client users, such as RSSI information, and making the information available online. Using this information made available online, the unsupervised learning theory may be used to refine a radio map of the network environment and result in more accurate location determinations.
The theory of unsupervised learning in pattern classification is generally summarized here. For example, D={x1, x2, . . . , xn} denotes the set of n unlabeled feature observations drawn independently from a known number c of clusters w={w1, w2, . . . , wc}, according to the mixture density according to the mixture density
where the forms for the cluster-conditional probability of the feature p(x|wj, θj) may be known (e.g. multi-variant Gaussian distribution), but the values for the c parameters θ={θ1, θ1, . . . , θc} may be unknown. The prior probabilities P(wj) may also be included among the unknown parameters. The objective is to estimate the parameters θ and P(wj) with j=1, 2, . . . c using the unlabeled observation set D. The maximum-likelihood estimations of θ and P(w) are the values that maximizes the joint density p(D|θ), represented by the equation:
subject to the constraints that P(wj)≧0, and
In a multi-variant Gaussian distribution case, each parameter θj consists of the components of mean vector μj and covariance matrix Σj, and p(x|wj, θj) is given by
where d is the dimension of the feature vector, |Σj| and Σj−1 are the determinate and inverse, respectively, of Σj, and (x-μ)T is the transpose of x-μ. If the unknown quantities are μj and P(wj), the solution to equation (2) is governed by the following equations:
While these equations appear to be rather formidable, the interpretation is actually quite simple and shows that the maximum-likelihood estimate for μj is merely a weighted average of the samples; the weight for the k-th sample is an estimate of how likely it is that xk belongs to the j-th cluster. In the extreme case where {circumflex over (P)}(wj|xk, {circumflex over (μ)}) is 1.0 when xk is from cluster wj and 0.0 otherwise, {circumflex over (P)}(wj) is the fraction of samples from wj, and {circumflex over (μ)}j is the mean of those samples.
If fairly accurate initial estimations {circumflex over (μ)}j(0) and {circumflex over (P)}0(wj) are available, equations (4-6) indicate an iterative scheme for improving the estimations, according to the equations:
This is, generally, a gradual procedure for maximizing the likelihood function. If the overlap between cluster-conditional densities is small, then the coupling between clusters will be small and converge will be fast. Application of this theory of unsupervised learning allows one to correct, refine or update the accuracy of a radio map through iteration, rather than re-measurement of the network environment and manual re-calibration.
Embodiments of the present invention employ unsupervised learning theory applied directly in location determination technology to create the received signal strength references. Accordingly, location determination is regarded as a pattern classification problem. In specific, the clusters are the particular points in the service area of a network, and the feature space is the RSSI information of a wireless station as experienced by wireless access nodes in the network. Assuming that the received signal strength in a wireless environment follows a log-normal shadowing model, RSSI samples in dB scale from each location candidate are modeled as a multi-variant Gaussian distribution. Further assuming that the standard deviation of the shadowing effects is fixed and known, equations (7-9) can be used in a straight manner to iteratively update the signal strength references μ={μ1, μ1, . . . , μc} at candidate points w={w1, w2, . . . , wc}.
Embodiments of the invention utilize an initial estimation of the signal strength references. For example, signal strength references obtained according to the method disclosed in United Stated patent application Ser. No. 10/635,367 entitled “Location Positioning in Wireless Networks,” may serve to provide an initial estimation on μ according to one embodiment of the present invention. An initial estimation on μ may alternatively be generated according to one embodiment of the present invention, as will be discussed. With sufficient RSSI observation samples, the signal strength reference at each grid point converges to a more accurate value.
Directing attention to
In the embodiment illustrated in
APs 101-103 of the illustrated embodiment provide RF illumination of a service area using multiple antenna patterns. For example, APs 101-103 may implement smart antenna configurations employing phased arrays and/or antenna beam switching to provide multiple antenna patterns. Commercially available APs adapted to provide multiple antenna patterns include, for example, the 2.4 GHz Wi-Fi switches available from Vivato, Inc., San Francisco, Calif.
The illustrated embodiment shows a configuration in which each AP has 10 approximately 36° directional antenna patterns and one omni-directional (approximately 360°) antenna pattern associated therewith. Specifically, AP 101 has directional antenna patterns 110-119 and omni-directional antenna pattern 11 associated therewith. Similarly, AP 102 has directional antenna patterns 120-129 and omni-directional antenna pattern 12 associated therewith and AP 103 has directional antenna patterns 130-139 and omni-directional antenna pattern 13 associated therewith.
It should be appreciated that the directional antenna patterns of the illustrated embodiment are disposed to provide wave fronts along different azimuthal angles, thereby providing directional coverage throughout a portion of the service area around each corresponding AP. However, it should also be appreciated that operation of the present invention is not limited to the particular antenna pattern configuration represented in
As shown in
As shown in
Irrespective of the particular antenna patterns implemented, the APs provide information communication links with respect to remote stations disposed within the service area of the wireless network. Referring again to
It should be appreciated that the antenna patterns illustrated in
As previously mentioned, an initial estimation on signal strength references may be obtained according to the method disclosed in United Stated patent application Ser. No. 10/635,367 entitled “Location Positioning in Wireless Networks.” Additionally or alternatively, a database providing an initial estimation on signal strength references may be constructed as follows. For example, an indoor wireless channel propagation model may be used to obtain received signal strength references for construction of a radio map according to the following generic log path loss model:
where P(d0) represents the power (in dB scale) received at a reference distance d0 from a radiating transmit antenna and β is the path loss exponent. The values of the parameters P(d0) and β depend on the practical environment and radiation power.
Directing attention to
Step 302 of the embodiment illustrated in
Step 303 of the embodiment illustrated in
where d(wi, APk) denotes the geometrical distance between the j-th point, wi, and the k-th AP, APk, and θ (wi, APk) is the angle between wi and APk with respect to the antenna panel direction of APk.
Typically, in an embodiment of the present invention, P(d0) can be calculated, given the transmission power, using the Friis free space equation. In some environments, however, P(d0) may also be obtained empirically. For example, P(d0=1.7 m)=−36 dBm in a semi-open environment using a Lucent Orinoco WLAN Card. The path loss exponent β=3 in an office environment with typical cubicles.
Equation (11) may be repeated until μi[k,p] has been computed for all k, p and j, thereby constructing a radio map using a generic propagation model together with multiple antenna radiation patterns.
Steps of the embodiment illustrated in
A refining process implemented according to an embodiment of the present invention may be used to increase the accuracy of the radio map constructed as discussed in reference to the embodiment of the present invention illustrated in
Step 401 of the embodiment illustrated in
where c is a constant for normalization, and μj is averaged signal strength, i.e. received signal strength reference.
Step 402 of the embodiment illustrated in
It should be appreciated that the algorithm described in Step 402 of the embodiment illustrated in
Step 403 of the embodiment illustrated in
Steps of the embodiment illustrated in
The radio-map refining algorithm in
An online location determination phase may run concurrently with the previously discussed iterative process, above, although location estimates will be more accurate after many iterations of the previously discussed process. In determining the location of a remote station within the service area of the network, one or more APs will use multiple antenna patterns to collect information with respect to the received signal strength of the target remote station. This information is preferably sent to a processor-based system and compared to the received signal strength reference stored in the database on various techniques. For example, the distance approach disclosed in U.S. patent application Ser. No. 10/635,367 entitled “Location Positioning in Wireless Networks,” may be employed. In one embodiment of the present invention, k-nearest neighbor weighted averaging and history-based shortest path approaches may be selected for determining the location of a stationary user and a moving user, respectively.
Directing attention to
As shown in step 501 of the embodiment illustrated in
As shown in step 502 of the embodiment shown in
As shown in step 503 of the embodiment shown in
where d0 is a small real value used to avoid division by zero.
While the algorithm in
In one embodiment of the online location tracking phase for moving clients of the present invention, only one pattern for each antenna panel is used to collect RSSI information due to real-time constraints. As many as 3 APs may be needed, however, based on the well-known triangulation method to estimate a location. By switching the antenna patterns more rapidly, more precise results may be achieved by using multiple patterns as used in the location determination phase.
When tracking a target client, embodiments of the present invention employ the current and past RSSI observations from the client to the audible APs. The user's location at any given instant in time is likely to be near the location for the previous instant. By tracking the user continuously, signal strength information is complemented with the physical contiguity constant to continually improve the accuracy of location estimation.
Directing attention to
As shown in step 601 of the embodiment shown in
As shown in the embodiment of the algorithm shown in
In the dynamic case with a moving target, it is possible to take the same approach as in the static case and estimate each position independently. Since the target is moving, however, a more accurate location estimation can be achieved, particularly given that the static-case estimate may contain noise, by taking into account the “velocity” or “speed” of the moving target.
As shown in step 602 of the embodiment shown in
As shown in step 603 of the embodiment shown in
For example, the 8 nearest neighboring points (in either signal space or physical space) of each estimated individual location for each instant in time, i.e., the 9 best guesses of the station's location for each time instance, may be chosen. Therefore, a history of depth h of such 9 neighbors, according to the earlier example, may be generated. The collected data of the exemplary 9 by h matrix can be viewed as a trellis tree. There are transitions only between columns containing consecutive sets (one set has 9 neighbors, for example). Each transition may be assigned a weight to model the likelihood of the user transitioning in successive instants in time between the locations represented by the two endpoints of the transition path. The larger the weight, the less likely the transition. The Euclidean distance between the two physical locations, calculated according to a simple metric, determines a weight. Each time the trellis tree (the matrix) is updated with the most 9 recent neighbors (and the deletion of the oldest set of neighbors), the shortest path between stages in the oldest and newest sets may be computed. According to embodiments of the present invention, the shortest path represents the most probabilistic movement of the station.
Once the shortest path is determined, the station's location may be estimated as the point at the start of the path, as shown in Step 605 of the embodiment shown in
Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the invention as defined by the appended claims. Moreover, the scope of the present application is not intended to be limited to the particular embodiments of the process, machine, manufacture, composition of matter, means, methods and steps described in the specification. As one will readily appreciate from the disclosure, processes, machines, manufacture, compositions of matter, means, methods, or steps, presently existing or later to be developed that perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein may be utilized. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, compositions of matter, means, methods, or steps.
Number | Name | Date | Kind |
---|---|---|---|
4054881 | Raab | Oct 1977 | A |
4558418 | Keearns | Dec 1985 | A |
5068838 | Klausner et al. | Nov 1991 | A |
5293642 | Lo | Mar 1994 | A |
5327144 | Stilp et al. | Jul 1994 | A |
5893033 | Keskitalo et al. | Apr 1999 | A |
6026304 | Hilsenrath et al. | Feb 2000 | A |
6104344 | Wax et al. | Aug 2000 | A |
6108557 | Wax et al. | Aug 2000 | A |
6148211 | Reed et al. | Nov 2000 | A |
6167274 | Smith | Dec 2000 | A |
6195046 | Gilhousen | Feb 2001 | B1 |
6195556 | Reudink et al. | Feb 2001 | B1 |
6236849 | Reudink et al. | May 2001 | B1 |
6263208 | Chang et al. | Jul 2001 | B1 |
6266514 | O'Donnell | Jul 2001 | B1 |
6362783 | Sugiura et al. | Mar 2002 | B1 |
6496701 | Chen et al. | Dec 2002 | B1 |
6560531 | Joshi | May 2003 | B1 |
6564065 | Chang et al. | May 2003 | B1 |
6697630 | Corwith | Feb 2004 | B1 |
6757544 | Rangarajan et al. | Jun 2004 | B2 |
6782265 | Perez-Breva et al. | Aug 2004 | B2 |
6830213 | Gehlot et al. | Dec 2004 | B1 |
6871077 | Kennedy, Jr. | Mar 2005 | B2 |
6985745 | Quaid | Jan 2006 | B2 |
7020475 | Bahl et al. | Mar 2006 | B2 |
7035651 | Schreiner et al. | Apr 2006 | B2 |
7038584 | Carter | May 2006 | B2 |
7116988 | Dietrich et al. | Oct 2006 | B2 |
20010022558 | Karr et al. | Sep 2001 | A1 |
20020055360 | Chen et al. | May 2002 | A1 |
20030008668 | Perez-Breva et al. | Jan 2003 | A1 |
20030050089 | Ogino et al. | Mar 2003 | A1 |
20030129992 | Koorapaty et al. | Jul 2003 | A1 |
20040072577 | Myllymaki et al. | Apr 2004 | A1 |
20050032531 | Gong et al. | Feb 2005 | A1 |
20050040968 | Damarla et al. | Feb 2005 | A1 |
20050136845 | Masuoka et al. | Jun 2005 | A1 |
20050261004 | Dietrich et al. | Nov 2005 | A1 |
Number | Date | Country |
---|---|---|
1455350 | Nov 2003 | CN |
WO 03069367 | Aug 2003 | WO |
Number | Date | Country | |
---|---|---|---|
20050246334 A1 | Nov 2005 | US |