3.2 RSSI-based localisation using multiple SDRs
3.2.3 The implementation of the RSSI-based localisation using mul-
3.2.3.1 RSSI matching based-indoor localisation
The RSSI matching based indoor localisation can be divided into three steps: 1. Estimate the PLE of the environment empirically.
2. In the training phase, obtain a RSSI pattern lookup table. The position of each training point in the area of interest will have a unique RSSI pattern which consists of the normalized RSSI values with the same numbers of receivers used in the localisation.
3. In the localisation phase, obtain RSSI measurements, do normalization and find the best matching RSSI pattern in the lookup table, so the corresponding location of the transmitter can be determined.
In Step 1), to estimate the PLE of the environment, the RSSI measurements at a few corresponding training points need to be obtained firstly. A RSSI-distance plot can be seen in Figure 3.11, in which the RSSI measurements at a distance up to 7m with 0.5m distance interval are obtained. Theoretically, the RSSI value should re- duce gradually with the increasing distance between the emitter and the receiver. However, some fluctuations can be observed, especially at 2m to 3m and 6m to 7m. Even though many tests are implemented at these abnormal points, the results are almost the same. The fluctuation is caused by reflection characteristics of the indoor area. Rooms with different size or sharp will give different fluctuation patterns of the RSSI-distance plot. After a corresponding relationship between RSSI measurements and distance measurements is obtained, a fitting curve is generated to fit the data. The fitting curve is denoted by the pink curve in Figure 3.11. Therefore, the approx- imated PLE of this indoor area is 2.0838. This value is very close to the empirical value in the free space because the emitter and receiver are placed at the same height and in a line-of-sight condition.
In step 2), the indoor area can be divided into a number of grids. The size of the grid depends on the desired localisation accuracy. In theory, the smaller the grid is, the higher the localisation accuracy is as long as the RSSI measurements are accurate. However, more grids mean more points to be surveyed, which is time consuming. Then the receivers can be deployed to cover the area of interest. In our
0 1 2 3 4 5 6 7 −70 −60 −50 −40 −30 −20 −10 0 Distance(m) RSSI(dBm) Experimental data Theoretical value, n= 2.0838
Figure 3.11: PLE estimate of the indoor area
implementation, 3 SDR-based receivers are used to localise the transmitter. After that, find the intersection points of the grid lines and assign a local coordinates on those points. Followed by marking the points by sticking a small piece of paper on it to make their position easy to identify, as shown in Figure 3.12, where the grid size is 0.5m by 0.5m. Next, measure the distances between each marked spots and the position of the SDR-based receivers and then calculate the RSSI of the three receivers by assuming there is a virtual signal transmitter with constant transmission power at each spot according to (3.3). Store the calculated RSSI measurements in a triple. Since the transmission power of the emitter P0 is unknown, the calculated RSSI value is
actually equal to ˆPi−P0. To remove the unknown transmission power, the RSSI triple
can be normalized by substituting the minimum value among the three values. After normalizing, the minimum RSSI value among the three RSSI measurements will be zero while the other two will be positive values. We can demote the normalized RSSI
by < RSSI1test, RSSI2test, RSSI3test >. After the training process described above is
completed, a RSSI pattern lookup table can be obtained.
In Step 3), when they are a real signal transmitter, the three SDR-based receivers are used to acquire the signal and calculate RSSI value. Then, the real RSSI mea- surements are normalized using the same way as in the training step, so a new RSSI triple<RSSI1, RSSI2, RSSI3>can be obtained. After finding the same or most sim-
ilar RSSI pattern in the lookup table, the location of the emitter can be determined. Figure 3.13 shows the results of three localisation experiments using the proposed RSSI matching-based methods, where the triangles denotes true emitter location, the square mark denotes the estimated emitter location, different color is used to
Figure 3.12: Marks on the intersection of grid lines in indoor area
distinguish different experiments. Green color denotes the first experiment, blue color denotes the second experiment and the red color denotes the third experiment. In the first experiment, the true position of the signal emitter is at position (4, 4)
and the estimated emitter location is at (5, 4), so there is a 0.5m localisation error. In the second experiment, the true position of the signal emitter is at position (4, 3)
and the estimated emitter location is at(4, 3), so there is no localisation error. In the third experiment, the true position of the signal emitter is at position (1, 5)and the estimated emitter location is at(1, 5), so there is also no localisation error. From the localisation results, one can see that the RSSI matching-based method can be used to implement indoor localisation and the localisation error is around 1 grid side length. The advantage of this method is that it is influenced by the change of the transmitter power because the P0 is removed through the normalization, but it takes some time
to calculate RSSI pattern lookup table.