EEE ICC 2014-Ad-hoc and Sensor Networking Symposium radius of the major region with the known reader's power by With this knowledge,several methods based on RSSI can linear interpolation method. be used to compute the real position of the reader.The basic 2)Scanning time:When the number of tags becomes large, one is like the previous method,which just replaces the tag the identification time increases.as shown in Fig.5.If we want size ni with total RSSI si in each identified category,and to localize objects in a very short time by reading the tags,then then calculate the weighted mean of the identified categories' the number of tags in the detection region should be limited. positions in x coordinate as.The estimation method 2 For example,when the number of tags equals to 120,the could be expressed as follows. reading time approximates to 1.5s,which cannot be accepted in the real time location system.Therefore,we should reduce 交=k Ti,s十Ti,e =1 the number of identified tags (e.g.90)in the detection region 2 Si to reduce the reading time (e.g.,1s),which can be achieved = by decreasing the scanning power.We get a training data set ∑15 T which shows the relationship between detected tag size and These basic schemes by using default power to read as many scanning time. tags in different categories as possible.As mentioned before, 3)RSSI distribution:In regard to the tags in the same the larger power causes the number of tags in the scanning detection region,their values of RSSI are different from each range to increase,therefore the reader needs more time to other.Fig.6 shows the RSSI value distribution for tags on the finish the scanning process. single row.We can find that the tags in the major detection Besides,in the tag size based baseline solution,the weight region have the higher RSSI values,while the tags in minor is not reliable due to the variances in tag distributions.For detection region have the lower RSSI values.In fact,the RSSI instance,maybe there is a category containing just a small value is mainly affected by the distance and the angle between number of tags,thus,even all the tags in the category are the tag's surface and the antenna's radiation direction towards identified,it always has a small weight in the computing pro- the reader.The shorter distance to the center of the detection cess.In the RSSI based baseline solution,when the categories region,the higher RSSI value we will get.Here,we get a in the major region contain just a small number of tags,but training data set Ta which shows the relationship between the categories in the minor region contain a lot of tags.The RSSI and distance to the center of the detection region. imbalanced numbers of tags will skew the weight and cause large error in the estimation position. V.BASELINE SOLUTIONS Therefore,these limitations lead to the low efficiency of the A.Category Cardinality based Protocol (CCP) baseline solutions In this baseline solution,we set the reader's power to the VI.LOCALIZATION BASED ON IMPRECISE ANCHORS default value(30.7dbm).Then the reader scans the tags to get the categories of identified tags.We average the locations of A.Compute the optimal value of power the detected categories to get the estimated position.We use Our algorithm tries to reasonably adjust the power of the the number of identified tags to calculate the weight.Because reader,while keeping it at a reasonable level to make sure the start point zi.s and the end point zi.e of the category's the delay satisfy the limits t.As shown in Algorithm 1,we range are known,the estimation method could be expressed use four steps to compute the optimal power.First,we use as follows. a pre-scan process to get the identified tag size.By setting the pre-scan power to an empirical value po,we can get an Wi' 工i,s十工i,e identified tag size no.Second,based on the training data set i= 2 T,we can compute the tag density p by an interpolation 心= ∑1n method.(For example,when the power is 26.7dBm and the number of identified tags is 50,the intersection is between 5 Here,k is the number of categories for identified tags.wi tags/m and 10 tags/m,and it is closer to 10tags/m,thus we is the weight factor for category ci,which is ratio of the set it 8 tags/m.)Third,suppose that we take time to in the number of identified tags in category ci to the number of all pre-scan,then we have the time of t-to left.As mentioned the identified tags in the scanning area.In this method,we use in Section IV,there is a relationship between the identified tag the middle value of the category's rangeto represent size and scanning time as shown in Fig.5.Therefore,we can the category's position.Finally,iis the computed position of get an target tag size limit of n'with respect to the remaining the target. time of t-to by an interpolation method based on the training data set T..Finally,from the relationship shown in Fig.4,by B.RSSI based Protocol (RP) using the p and n',we can compute the optimal power p.The In the previous baseline solution,we just use the number algorithm is shown in Algorithm 3. of identified tags in each category.However,the RSSI is a very important factor to measure the distribution of tags.As B.Category Matching based Protocol(CMP) mentioned in Section IV,the higher the RSSI,the closer the 1)Motivation:During the scanning process,we can get the tag to the center of the detection region. number of tags in each identified category.We denote them 144radius of the major region with the known reader’s power by linear interpolation method. 2) Scanning time: When the number of tags becomes large, the identification time increases, as shown in Fig.5. If we want to localize objects in a very short time by reading the tags, then the number of tags in the detection region should be limited. For example, when the number of tags equals to 120, the reading time approximates to 1.5s, which cannot be accepted in the real time location system. Therefore, we should reduce the number of identified tags (e.g. 90) in the detection region to reduce the reading time (e.g., 1s), which can be achieved by decreasing the scanning power. We get a training data set Tc which shows the relationship between detected tag size and scanning time. 3) RSSI distribution: In regard to the tags in the same detection region, their values of RSSI are different from each other. Fig.6 shows the RSSI value distribution for tags on the single row. We can find that the tags in the major detection region have the higher RSSI values, while the tags in minor detection region have the lower RSSI values. In fact, the RSSI value is mainly affected by the distance and the angle between the tag’s surface and the antenna’s radiation direction towards the reader. The shorter distance to the center of the detection region, the higher RSSI value we will get. Here, we get a training data set Td which shows the relationship between RSSI and distance to the center of the detection region. V. BASELINE SOLUTIONS A. Category Cardinality based Protocol (CCP) In this baseline solution, we set the reader’s power to the default value(30.7dbm). Then the reader scans the tags to get the categories of identified tags. We average the locations of the detected categories to get the estimated position. We use the number of identified tags to calculate the weight. Because the start point xi,s and the end point xi,e of the category’s range are known, the estimation method could be expressed as follows. xˆ = k i=1 wi · xi,s + xi,e 2 , wi = ni k i=1 ni Here, k is the number of categories for identified tags. wi is the weight factor for category ci, which is ratio of the number of identified tags in category ci to the number of all the identified tags in the scanning area. In this method, we use the middle value of the category’s range xi,s+xi,e 2 to represent the category’s position. Finally, xˆ is the computed position of the target. B. RSSI based Protocol (RP) In the previous baseline solution, we just use the number of identified tags in each category. However, the RSSI is a very important factor to measure the distribution of tags. As mentioned in Section IV, the higher the RSSI, the closer the tag to the center of the detection region. With this knowledge, several methods based on RSSI can be used to compute the real position of the reader. The basic one is like the previous method, which just replaces the tag size ni with total RSSI si in each identified category, and then calculate the weighted mean of the identified categories’ positions in x coordinate as xi,s+xi,e 2 . The estimation method could be expressed as follows. xˆ = k i=1 wi · xi,s + xi,e 2 , wi = si k i=1 si These basic schemes by using default power to read as many tags in different categories as possible. As mentioned before, the larger power causes the number of tags in the scanning range to increase, therefore the reader needs more time to finish the scanning process. Besides, in the tag size based baseline solution, the weight is not reliable due to the variances in tag distributions. For instance, maybe there is a category containing just a small number of tags, thus, even all the tags in the category are identified, it always has a small weight in the computing pro￾cess. In the RSSI based baseline solution, when the categories in the major region contain just a small number of tags, but the categories in the minor region contain a lot of tags. The imbalanced numbers of tags will skew the weight and cause large error in the estimation position. Therefore, these limitations lead to the low efficiency of the baseline solutions. VI. LOCALIZATION BASED ON IMPRECISE ANCHORS A. Compute the optimal value of power Our algorithm tries to reasonably adjust the power of the reader, while keeping it at a reasonable level to make sure the delay satisfy the limits tl. As shown in Algorithm 1, we use four steps to compute the optimal power. First, we use a pre-scan process to get the identified tag size. By setting the pre-scan power to an empirical value p0, we can get an identified tag size n0. Second, based on the training data set Tb , we can compute the tag density ρ by an interpolation method.(For example, when the power is 26.7dBm and the number of identified tags is 50, the intersection is between 5 tags/m and 10 tags/m, and it is closer to 10tags/m, thus we set it 8 tags/m.) Third, suppose that we take time t0 in the pre-scan, then we have the time of tl − t0 left. As mentioned in Section IV, there is a relationship between the identified tag size and scanning time as shown in Fig.5. Therefore, we can get an target tag size limit of n with respect to the remaining time of tl−t0 by an interpolation method based on the training data set Tc. Finally, from the relationship shown in Fig.4, by using the ρ and n , we can compute the optimal power pˆ. The algorithm is shown in Algorithm 3. B. Category Matching based Protocol (CMP) 1) Motivation: During the scanning process, we can get the number of tags in each identified category. We denote them IEEE ICC 2014 - Ad-hoc and Sensor Networking Symposium 144