PROCESO CONSTITUCIONAL ARGENTINO

It is difficult to exactly compare the Durham tracker with others on an entirely even footing, as there is a large spectrum of trackers aimed at many different niches in the market. Our system is modular and takes in a variety of I2C sensors — it’s highly configurable while remaining a pre-set system that could be used off the shelf as a personal, vehicle or asset tracker. The trackers listed above may not all be exactly comparable, but are all potential alternatives for some applications. Most are not directly suitable for use with a UAV, for a number of reasons; either they have no sensor or communications IO for interfacing with the UAV controller, or they are too heavy for a UAV to lift, or they are designed to give positional up- dates too infrequently. However, that being said, the comparison made between the Durham tracker and other alternatives does allow for some degree of benchmarking — allowing us to see whether the tracker behaves notably worse or better than the alternatives in situations where they are on a relatively equal footing.

Two alternative tracker options were extensively compared against the Durham Tracker, in a variety of use cases:

• Stationary outside • Stationary in a car • Stationary inside

146 Chapter 5. GPS tracking • Out of range of GPS

• Out of range of GPRS

• Walking in an urban environment • Driving in a car

The two alternatives used were the Globalsat TR-102 (see Section 5.5.1) and a smartphone18 running OsmAnd (see Section 5.5.2.2). Further trackers not tested are also included below, using data obtained mainly from the manufacturers’ speci- fications.

The trackers were compared on a number of categories: • Frequency of samples

• Frequency of uploads

• How well it copes when out of range of network • How well it copes when out of range of GPS • How well it copes when stationary

• Size • Cost • Weight • Map quality

While some of these are more qualitative than quantitative, an attempt was made in all cases to put the tracker options in order of preference.

5.5.12.1 Experiments

Here follows a more in depth description of the use cases tested.

5.5. Comparison with commercial trackers 147 Stationary outside The three trackers were placed on a table in a garden, 5.25m from any obstacles, with a clear view of the sky. They remained there for over three hours.

Stationary in a car The three trackers were placed on the dashboard of a car, parked on a suburban street on the outskirts of Durham. They remained there overnight.

Stationary inside The three trackers were placed on a surface one metre away from a window, remaining there for over three hours.

Out of GPS range The trackers were moved to a position from where they had no GPS signal but could still access GPRS, remaining there for a short period to ascertain their behaviour.

Out of GPRS range The trackers had their GPRS capabilities removed, and were placed indoors by a window where a GPS signal was available. They remained there for a short period to ascertain their behaviour, and then had their GPRS capabilities restored to them.

Walking in an urban environment The three trackers were placed into a small rucksack and carried across Durham. The walk included travelling through relatively built up areas, although Durham does not have any concentrations of high rise buildings. It took 40 minutes, and was 3km long.

Driving in a car The three trackers were placed inside a car and taken on a four hour drive, a journey of approximately 200 miles.

5.5.12.2 Frequency of samples

If tracking a fast-moving object such as a UAV, it is important to be able to take rapid readings. It is also important to be able to upload those readings in such a way as to reduce any down time where readings cannot be taken.

148 Chapter 5. GPS tracking Durham Tracker TR-102 OsmAnd Experiment No. Avg

Total

avg No. Avg No. Avg Stationary outside 1122 8.92 9.63 259 41.79 1893 5.69 Stationary in a car 1423 9.31 10.12 - - 2642 5.45 Stationary inside 1100 8.89 9.82 58 187.56 1870 5.78 774 12.71 Walking in an urban environment 229 9.42 10.11 32 75.29 391 6.14 Driving in a car 1301 9.70 10.61 24 60.26 2175 6.39 Totals 5175 9.24 10.07 1147 30.86 8971 5.83 Table 5.3: Number of readings, average time between readings, and (in the case of the Durham tracker) average time between readings if upload time is not taken into account. All times are in seconds.

The Durham tracker was set to take readings as often as possible, with uploads to the server once every half hour. The TR-102 was set to take readings at its minimum time of 5s, and upload every point when taken. OsmAnd, on the smartphone, was set to its minimum time of 5s, and also uploads every point when taken — although it is able to do so while simultaneously continuing to track, which the other two trackers are unable to do.

Looking at the results from all experiments (except those where GPS and GPRS were artificially made unavailable), and discounting time spent bulk uploading for the Durham tracker, the three trackers had average reading times of 9.24s (Durham Tracker), 30.86s (TR-102), and 5.83s (OsmAnd) (see Table 5.3). Including the bulk upload time changes the Durham tracker’s average upload time to 10.07s. Interest- ingly the TR-102 reports every 5 seconds even when it has not achieved a new GPS reading, giving multiple reports for each timestamp. These have been discarded for calculating the above statistics.

Table 5.3 also demonstrates the reliability of the three trackers. The Durham Tracker and the smartphone both show a high level of reliability in the tests taken; the TR-102, however, has regularly stopped responding. For two of the stationary experiments it only reported for limited periods, and for the third it failed to report a valid position at all. In the driving experiment the TR-102 only reported 24 positions before cutting out, a very short way into the four hour journey.

5.5. Comparison with commercial trackers 149 Experiment Number of uploads Average time per upload Stationary outside 5 167.80

Stationary in a car 7 174.43 Stationary inside 5 214.60 Walking in an urban environment 1 167.00 Driving in a car 7 179.00

Totals 25 182.12

Table 5.4: Number of uploads and average time taken per upload for the Durham tracker. All times are in seconds.

5.5.12.3 Frequency of uploads

The remote user will want to be able to know where the tracker is. In many cases it is not worthwhile tracking something if you can only find out where it was a long time ago, rather than where it is at the moment. It is therefore important for the tracker to upload its position regularly and often.

This is one category in which the Durham tracker is somewhat behind, in that it is set to bulk upload readings every half hour. The other two trackers tested upload instantly.

The main advantage of bulk uploading is that it can continue to take and upload readings even when no mobile phone reception is available — storing them up for upload later. It also greatly speeds up the time to take readings, although against the smartphone particularly this improvement is of a lesser importance.

The disadvantage is that, from the user’s perspective, it is invisible for the length of time between bulk uploads. This can be improved by increasing the upload frequency.

Of the three trackers tested only the Durham tracker has a bulk upload. Over the course of the experiments, with it set to upload half hourly, the average time taken to bulk upload (during which time no measurements could be taken) was a little over 3 minutes (see table 5.4).

5.5.12.4 Out of network range

In many cases — particularly in situations where there is an element of risk or danger such as when searching for a missing person in a remote part of the country,

150 Chapter 5. GPS tracking but also in everyday use such as when inside a large building — the GPRS network will not be always available. The tracker therefore needs to be able to cope with situations where it is unable to upload to the server immediately. In particular, it needs to be able to deal with situations where GPRS is unavailable but GPS (or other sensor data) is.

The TR-102 tracker, when out of GPRS range, was unable to store readings for later upload — if it does not upload the data immediately it is lost when the next reading is taken. OsmAnd, similarly, does not backfill readings — although it does have the capability of storing all data into the phone’s memory, which could be used for later retrieval. Backfilling would be easy to implement, but is not included in this software. The Durham tracker stores all data that has not been successfully uploaded, aiming for perfect retention. As such, when out of GPRS range, it continues to take readings and uploads them when next back in range.

5.5.12.5 Out of GPS range

Similarly to the section above, it is not an uncommon occurrence to be out of GPS range. This is often the case when indoors, or in an urban canyon, or under heavy tree cover. The GPS tracker must be able to respond appropriately when unable to track via GPS.

When inside and out of GPS range (but still in GPRS range), the Durham tracker reported positions with no latitude or longitude, and reported that the positions were invalid. It still reported other information such as sensor readings, however.

The TR-102 continued to report positions every 5s, but with the timestamp given being the time of the last reading it had been able to take. The readings were reported as invalid.

The smartphone stopped reporting positions entirely. The software used has no means for reporting that a position is invalid, and this being the case it seems the best solution — although it does leave the user unknowing as to whether the tracker is out of GPS range, out of mobile phone reception, or turned off entirely.

5.5. Comparison with commercial trackers 151 5.5.12.6 Stationary

GPS trackers tend to give more accurate positions when moving than when station- ary. When the Durham tracker is still, for example, it will often report positions over a large area; as soon as it is carried away it will give readings which are very close to the actual positions of the tracker (other effects such as multipath aside).

The reason for this is that the SiRFstarIII chip used in the Telit module takes the Doppler effect into account when the unit is moving. The GPS satellites are all moving at a high speed in their orbits, and there is a discernable Doppler shift in the signals being received from them. This shift is caused by the movement of the satellite (known), the rotation of the Earth (known), and the movement of the tracker. The Doppler shift can be measured with a far greater accuracy than the tracker position, as it will not be susceptible to things like multipath effects. This means that the positions of GPS readings — which would normally give a much higher error — can be smoothed by taking into account the more accurately known velocity when the tracker is moving at higher speeds.

The SiRFstarIII chip contains two further, optional filters that are not used by the Durham tracker but which may be used by other trackers. The first — ‘Track Smoothing’ — extrapolates future positions from previous ones, which greatly reduces the error when travelling in a relatively straight line (such as in a car), but causes the recorded position to continue moving for a short while after the tracker has come to a halt. The second — ‘Static Navigation’ — will freeze the tracker’s position when the measured velocity drops below 1m/s. The recorded position will remain static until either the velocity rises approx. 10% above this, or the position reports an approx. 50m change. Both of these filters work better in vehicles such as cars than in situations with slower movement or movement in often changing directions.

Three experiments involved the trackers being stationary — outside, in a car, and inside near a window.

Outside For the outside experiment all three trackers had a good view of the sky, although they did not have a view to the horizon. The smartphone software does

152 Chapter 5. GPS tracking not report the number of satellites used to calculate its positions; of the other two, the Durham tracker reported 6–9 satellites, mainly 8, and the TR-102 reported 3–5, mainly 5. All three trackers gave a range of points which were centred about the actual location. The Durham tracker and the smartphone both behaved similarly; their readings occurred regularly throughout the three hour period, and the positions obtained were both within a similar area (see Fig 5.12). The TR-102 was less reliable, however. It didn’t respond consistently (spending long periods where it was unable to get a valid GPS reading), and the positions it did obtain covered a far greater area.

Ignoring the first few minutes of the experiment, where the errors were unusually high, the TR-102 showed errors of up to 270m. It also did not report consistently; there is a large period in the middle of this experiment where only a handful of readings were taken. The errors on this tracker are extremely high; for the latter part of the experiment the majority of errors were under 20m, but there were still a great number which were above this.

Both the smartphone and the Durham tracker have a much more consistent position. The error — while not static throughout — remains relatively constant for long periods. Oddly, for both these trackers, the error changes at about the same point. At around 1 hour 45 minutes the error increases from around 1.5m (Durham tracker) / 1.5–3m (OsmAnd) to approx. 5.5m.

Inside For the inside experiment the trackers had a much more limited view of the sky. The TR-102 reported 3–4 satellites, while the Durham tracker reported 3–9, averaging around 7. As above, the points measured by the trackers were centred around the correct position, but again the positions from the TR-102 covered a much larger area. The errors can be seen in Figure 5.13.

Again ignoring the first few minutes, the TR-102 gives much larger errors than the other two trackers. The errors go up to almost 50m, and for the most part are reasonably evenly spaced between 10 and 40m. It is also notable that the TR-102 reports extremely infrequently, with large gaps in reporting. Its time to first fix was also extremely high, and has been for all experiments run.

5.5. Comparison with commercial trackers 153

Figure 5.12: Positional error from three trackers lying stationary outdoors. The graph is cropped to errors below 50m. Errors are based on a central position taken from Google Maps, which may introduce a small error of a couple of metres. The three trackers are the Durham Tracker (red), the TR-102 (blue), and the smartphone running OsmAnd (black).

154 Chapter 5. GPS tracking

Figure 5.13: Positional error from three stationary trackers indoors, lying 1m away from a window. The graph is cropped to errors below 50m. Errors are based on a central position taken from Google Maps, which may introduce a small error of a couple of metres. The three trackers are the Durham Tracker (red), the TR-102 (blue), and the smartphone running OsmAnd (black).

due to it locking into a position and continuing to report this position while speed is low. However, after a short period while the error was extremely low, the reported position error rises to 10m.

The error on the positions reported by the Durham Tracker varies between 0 and 30m, with the majority of points between 0 and 10m.

To ascertain whether the poor behaviour of the TR-102 was a one off problem this experiment was rerun on a second occasion, with just the TR-102. On this occasion it gave a considerably improved response for the first hour of operation — reporting positions regularly and accurately, with an average time between readings of 12.71s. However, after this first hour it stopped responding entirely for the remainder of the three hour period.

This shows that the TR-102 has the capability to perform to a similar standard as the other two trackers — but on this and many of the other experiments it has failed to do so. It appears to be somewhat unreliable.

5.5. Comparison with commercial trackers 155

Figure 5.14: Positional error from two trackers in a stationary car. Errors are based on a central position taken from Google Maps, which may introduce a small error of a couple of metres. The two trackers are the Durham Tracker (red) and the smartphone running OsmAnd (black). The TR-102 failed to report a valid position during this experiment.

In a car The three trackers were left in a car overnight. Unfortunately the TR-102 failed to report any valid positions during this experiment, so the only data obtained is for the Durham Tracker and the smartphone running OsmAnd. The experiment lasted four hours, after which time the smartphone ran out of battery. The results can be seen in Figure 5.14.

The results for this experiment are very similar to the two previous stationary experiments. The errors for both trackers range from 0–25m, with the majority of points under 10m. The errors for both drift over time, rather than jumping around. Of the two, the smartphone gives slightly better results. Both give a regular set of readings, without any breaks.

156 Chapter 5. GPS tracking 5.5.12.7 Size, Weight and Cost

A tracker that is too large or too heavy to use, or which is too costly to buy, will not be of practical use. It therefore needs to be sufficiently small and light to be — depending on the context — carried around by a person, attached to a UAV, attached to an item, or placed in a car.

Table 5.5 lists the size, weight and cost of the various trackers described above. The first three are the trackers being directly compared; the remainder were de- scribed in Section 5.5. The last two sections are model aircraft trackers and animal trackers respectively, and may not be directly comparable — see Sections 5.5.9 and 5.5.10. For several of the trackers listed some information was unavailable.

5.5. Comparison with commercial trackers 157 Tracker Width / mm Height / mm Depth / mm Volume / cm3 Weight / g Cost / £19 Durham Tracker (boxed)20 130 100 30 390 18521 - Durham Tracker (unboxed)22 85 67 14 79.7 65 - TR-102 115 45 22.5 116.4 102 167.99 OsmAnd 83 50 16 66.4 8823 10/mo24 OsmAnd — HTC Sensation 126.1 65.4 11.3 93.2 14825 -

SPOT Personal Tracker 111 69 44 337.0 209 >63.0726 Globalstar SmartONE 165 82.5 25.5 347.1 385 - MeiTrack MVT-600 103 98 32 323.0 220 - MeiTrack VT-400 123 83 37 377.7 350 - CelloTrack Solar 585 130 30 2,281.5 - - ATrack AY5i 100 65 26 169.0 161 - TinyTelemetry 55 15 10 8.25 12 85.12 EagleEyes 100 70 40 280 78 63.07 TinyTrak SMT 25.5 23.5 4.1 2.527 - 30.27 Vectronic’s GPS Plus Telemetry

Collar

- - - - 100 - Quantum 4000 Advanced GPS

Collars

- - - - 30 - Table 5.5: Comparison of sizes and weights for a variety of trackers. Not all data is available for all trackers.

19_{All currencies have been converted to GBP.} 20

The current box design could be reduced in size for a future version.

21_{185g without 4xAA batteries; 341g with.} 22

This does not include battery or aerials.

23_{Smartphone size and weight based on Sony Ericsson X10 Mini.} 24

The app is free; smartphones can cost under £10/month, but prices vary enormously.

25

The experiments were taken using an HTC Sensation.

26_{£63.07 plus subscription.} 27

A separate power supply, radio and serial based GPS receiver are also required, which will take additional space.

158 Chapter 5. GPS tracking 5.5.12.8 Map quality

The user will want to be able to see where the tracker is, and may want to take in other information (such as historical tracks). This is a comparison between the Durham Tracker website and some of the other options — either as supplied with one of the trackers tested, or a generic tracking website as described in Section 5.5.11.

The servers compared below are:

EZFinder is a web based tracking server produced by Globalsat (see Section 5.5.11)