Working with Bluetooth can be a frustrating experience since different manufacturers may not have a fully consistent interpretation of the specification, different stack implementa-tions support different subsets of the specification and stack stability is still improving.
Consequently Bluetooth systems should never be analysed purely theoretically. This sec-tion demonstrates a number of tracking systems formed using the techniques described above. Although quantitative results are very dependent on a number of factors including the geometry of the base deployment, the layout and construction of the tracking space and the path and time taken during the trials, they nevertheless reveal a lot about the performance and utility of the systems.
Although each of the systems has use cases where it will be most appropriate and the personal energy meter will rely on input from heterogeneous systems deployed in different
buildings, in general TcB and TsB are most likely to scale well to large areas with many users where no additional bootstrapping information is available. These two systems are therefore evaluated further in this section.
5.6.1 TsB tracking evaluation
To test a TsB system seven bases were positioned throughout an office corridor as shown in Figure 5.12(a). As before, the situation was as representative as possible: Bluetooth had to co-exist with WiFi, walls attenuated signals and those not participating continued to work as usual. Each base was set to discoverable with the following parameters (quoted hereafter in 625 µs slots):
Tw inq scan 4,000 Tinq scan 4,096
In effect, this meant that each base was almost continuously listening for inquiries. This prevents them from having meaningful Bluetooth connections themselves, but means re-sponse rates are fast and that the target therefore saves energy by not having to transmit inquiry packets for as long. To have a chance of discovering all the bases, the target needs to inquire until at least the first train change occurs (2.56 s for the specification-required Npage= 256). In the test each scan was run for 5.12 s.
A walking test was performed with a laptop set to continuously scan and to collect all HCI inquiry responses. The ground truth location of the device was determined using a Bat attached to the laptop user.
Figure 5.12 illustrates the results for each of the seven bases. The thin purple line shows the route taken by the user, whilst the thicker blue lines indicate parts of the walk where inquiry events for the relevant base were being received. In practice, however, a target cannot know that it will not receive any further inquiry events from that base until a complete scan has occurred without an event. The thicker green lines show the parts of the walk that a real target would have to consider itself in range of the base.
5.6.2 TcB tracking evaluation
The most scalable TcB mode involves the use of bases with a common spoofed address.
This technique was applied to the arrangement of bases shown in Figure 5.12(a), all of which were set to have the same arbitrary address. As is clear from the TsB results, the bases had overlapping coverage, which is problematic when multiple devices assume the same address. To limit the paging area for each base EMI shielding tape was applied to the Bluetooth adaptor in the laptop. Two layers of 3M 1345 EMI shielding tape were
(a) (b)
(c) (d)
(e) (f)
(g) (h)
Figure 5.12: TsB system results. (a) Positions of the seven bases. (b–h) Tracking results for each base.
sufficient to reduce the range such that overlap was minimal. This tape could instead be applied to the base adaptors if the target does not have an exposed Bluetooth antenna (e.g. on a mobile phone). The shielding increases granularity at the cost of requiring an increased number of base stations to ensure coverage; the appropriate tradeoff will vary depending on the level of energy apportionment required. If building-level apportionment is sufficient (perhaps because no more fine-grained consumption data is available), Class 1 Bluetooth devices with a nominal range of 100 m could be used instead, reducing further the deployment complexity.
A similar test to that for TsB was then performed, setting the base paging parameters to be:
Tw pg scan 18 Tpg scan 2,048
These parameters were chosen because higher duty cycles have two disadvantages in a TcB system using spoofing. Firstly, where coverage areas do overlap, it greatly increases the chance of collisions when multiple bases respond to the same page. To see this the target was placed near to one base, but potentially in range of three others with the same spoofed address. It was then left (without EMI shielding) performing remote name requests on the spoofed address, varying the duty cycle. A failure rate of 0.38% was observed with a Tw pg scan:Tpg scan of 18:2,048, which rose to 4.01% using 18:512. With the shielding tape applied, only the adjacent base was sighted, and the failure rate was only 0.003% using 18:2,048.
The second disadvantage of a high duty cycle for connection-based tracking is that it requires the base radios to be almost constantly listening and therefore unable to main-tain even the remote name request connections long enough for reliable communication.
Neither disadvantage is present for the TsB system because it does not use spoofing nor does it need to maintain connections.
The particular settings used were trialled because they are the default BlueZ settings.
The results were sufficiently good that there was no need to alter the duty cycle further, although other system implementers could optimise the parameters for their particular usage scenario.
Figure 5.13 illustrates the test results, which were collected using remote name requests from the laptop to the special address. A coloured line is drawn from each base to the position of the user at the moment the HCI layer reported the remote name. Black crosses are used to indicate the report of a name request failure. There is a strong spatial locality for each of the bases, and a good update rate when in range of the bases. The connection failures can be attributed to both unintentional areas of overlapping coverage and coverage holes from an unoptimised distribution of bases. Even without these optimisations, as would probably be the case in the majority of deployments intended solely for energy
Figure 5.13: TcB system results.
metering, the connection-based approach provided results comparable to the scan-based equivalent.