AP coverage

The second analysis performed on the WiFi dataset is the computation of the portion of each block covered by WiFi. In this case we take into account both the position of the AP and its signal strength, so we can identify the area, within a given range from the AP, that is covered by its signal.

We know that there are many factors that can interfere with WiFi signal propagation, but our focus is more on exploiting map information than on precise estimation of WiFi coverage, so we decide to use common ranges for indoor APs that have been estimated between 25 and 50 meters as in Berezin et al. (2012). Where 50 meters represents a “reasonable” assumption that we compare with the more restrictive assumption of only 25 meters of coverage.

Figure 6: Access points position and coverage as retrieved from the WiGle dataset (year 2015) for a 500m x 500m block with coordinates 14,12. AP coverage radius is set to 25 meters.

In Figure 8 we show the WiFi coverage resulting from the WiGle dataset for a 500m x 500m block. The circles represent the position and coverage of the 154 APs identified in this area. The unmanaged deployment of the APs is responsible of the inefficient coverage of the area within the block because there is a great overlap between neighbouring APs.

Clearly, to evaluate the diffusion, the deployment and the usefulness of WiFi community networks, the computation of the amount of the block area covered by WiFi connectivity is a better information than the number of APs per block. Nevertheless ISPs always advertise their, so called, “coverage” using only the number of APs available in a region (for example, on the FON website (FON 2016) we read that they count for 4,000,0000 hotspots in France, 1,000,000 in Netherlands, and 950,000 in Japan).

Our tool, unlike most of the online map services that only show the location of the APs, allows the computation of the effective WiFi network coverage of a given area, or, better, of a given public area, that is a public street. As it will be detailed in the Section 5, the “street WiFi coverage value” is of more practical use than a “generic area coverage value” because both traveling people and smart objects in the city tend to be placed on the streets when they try to reach for a WiFi connection. The “area coverage value”, in fact, comprises also building and private areas that are not accessible to traveling people and common IoT devices.

5

Discussion

In this section we discuss the results given by our approach when applied to the WiGLE WiFi dataset of the city of Turin, Italy. From the data on the location of the APs and their coverage range, we compute an estimation of the real coverage of the WiFi network using the percentage of public street areas covered by the APs in a given area (that is supposed to be more meaningful than the usually used raw AP density).

The computation of the portion of the streets in an area that is covered by WiFi can be obtained because our framework uses both the information on the APs position from the WiGle dataset and the information on the underlying map retrieved from OpenStreetMap. Figure 8 shows the portion of the streets within the APs coverage area as a black line. As described in Section 3 we compute the WiFi connectivity of the streets considering non-overlapping streets segments of 10 meters, so it may happen that the portion of the street marked as covered in the map extends a little bit outside the circles. As expected, main streets have a higher percentage of APs because they have been measured (and traveled) more frequently. However, also side streets show a good WiFi connectivity.

For this block our framework reveals an area coverage of 50% and a street coverage of 60%. Street coverage ratio (SCR) is defined as the ratio between the total length of the streets in a block and the total length of the portions of the streets inside the AP coverage circles.

Figure 7. Street coverage ratio cumulative distribution function (CDF) considering square grids of blocks of increasing size around the city center.

Figure 9 shows the SCR cumulative distribution function (CDF) around the city center (block 14,15) considering gridcells of increasing size (see Figure 6). The smallest one (9x9) comprises all the blocks with

Figure 8. Percentage of street in a block (500m x 500m) covered by WiFi versus the number of APs measured in the block.

In order to evaluate the AP density impact on the street coverage percentage, we have plotted the relation between these values in Figure 10 for each block of 500m x 500m. The circles represent the estimate of the AP coverage with a radius of 50m and the crosses the AP coverage with a radius of 25m. The distribution of the points shows a relatively small difference between the two estimates. Fitting the data with a linear regression model reveals that the assumption of a 50m radius gives a street coverage ratio that is 1.23 times better than the ratio given by assuming a 25m radius. This analysis shows that in the case of an unplanned deployment of WiFi access points, the coverage of the surrounding streets increases in a logarithmic fashion. Thus, it is relatively easy to achieve around 50% of street coverage in a block, i.e., a hundred APs may be sufficient, but a great increase above that percentage may be impractical, i.e., we need about a thousand of APs to reach 70-80%. As observed in other works such as (Seufert, 2015), where the authors investigated the relation between the spatial structures of wireless networks and population densities, an intelligent choice of the placement of a few new APs may be a viable option to a significant increase of the overall street, and city, WiFi coverage.

6

Conclusions

In this paper, we have presented a unique framework for mapping wireless measurements on the data provided by the OpenStreetMap service. Given the GPS coordinates, SSID (and if available the MAC address) of WiFi access points, this tools adds to the common practice of representing this data on a map the ability to relate it with other geographical information. The tool is released as open source software on Github, and it is named WiFi Street Coverage Explorer (WiFiStreetCoverageExplorer 2016).

Among all the uses of the tool that can be foreseen, we have shown the results of mapping a WiFi dataset retrieved from the WiGle website on the street information of the city of Turin retrieved from OSM. At first, we have used this information to characterize the different areas of the city by computing a common result for this kind of analysis, that is the access point density. Then, we have improved this initial result by matching the position and coverage range of the hotspots with the map of the streets of the city and by computing the percentage of streets with WiFi connectivity in each area. This value is clearly a better estimation of the real wireless network coverage in a given location of the city.

This estimation can be particularly useful for the increasing number of WiFi Community Networks that, at the present time, only represent a very rough estimate of their coverage advertising just the raw number of APs per area (as we have seen, the real coverage depends also on their location and percentage of overlapping). In addition, the ability to analyze the portion of streets covered by a WiFi network can be used to plan the deployment of “smart devices” in the Internet of Things scenario foreseen for the upcoming “smart cities”. In fact, the connectivity of these devices, like dumpsters, traffic lights and manholes, that are usually placed along

the streets and in public places, heavily depends on the street coverage that we can compute with our framework.

Future work will focus on three tasks. First, on improving the system with more reliable AP information from direct measurements or other open datasets, eventually updated, almost, in real-time (e.g., once a day). In fact the results obtained with the WiGle data does not seem to reflect so accurately the real deployment and position of the wireless hotspots. Second, on extending the results that can be computed by the system, for example, in addition to the street coverage ratio we can compute further details such as the longest segment of a street without coverage (i.e., gap) and how other gaps are distributed. Third, on studying the optimal placement of the minimal amount of hotspots in the city, so that the WiFi street coverage can reach the desired level of service.