The performance of the different tag positions was evaluated on different aspects. Accuracy, received power level, number of line-of-sight anchors, comfort, maximum communication range, and energy consumption were investigated. The used UWB radio settings are shown in
Table 1. The low data rate and high preamble length resulted in a long range. The highest possible pulse repetition frequency was chosen to obtain higher accuracy on the first path time stamp. The downside of this choice was that this required greater power consumption. All raw data of the tests, as well as all constructed graphs to visualize the measured signals and the calculated values are available in the related dataset [
8].
5.1. Accuracy
The range between the tag and an anchor was calculated based on symmetric double-sided-two way ranging (SDS-TWR). Three UWB messages were exchanged to calculate a range between the tag and anchor: poll, reply, final. The message exchange is shown in
Figure 6a. This way of ranging has the advantage of eliminating clock drift. The range can be calculated using:
, with
c equal to the speed of light. The time of flight (ToF) was calculated with Formula (
1), where the variables are related to the timestamps used in
Figure 6a.
The ranges between the tag and the eight anchors were read from the serial output of the tag. Ranging with multiple anchors was done sequentially, starting with Anchor 0 till Anchor 7. In total, 100 samples per anchor per test were collected. The communication procedure for the ranging with three anchors is shown in
Figure 6b. It started with the tag broadcasting a beacon using sub-GHz (blue) containing the set of anchors with which to determine the range, followed by the tag broadcasting a poll to the set of anchors. Each anchor replied sequentially and received the final message from the tag. In the end, the ranges were broadcast from the anchors using sub-GHz. The tag received and printed this range.
5.1.1. Absolute Ranging Error
To be able to calculate the absolute ranging error, the ground-truth is required. The ground-truth was determined with a mm-accuracy laser meter. The absolute error to each anchor for six different tag positions is plotted in
Figure 7a. The cyclist was placed on the track bike, and the anchors were mounted at a height of
m. From this plot, it was clear that the ranging for the tag positions underneath the saddle and seat post to the anchors in front (Anchors A1, A0, and A7) were the least accurate. When the tag was placed on the chest of the cyclist, the ranging to the anchor behind the cyclist (anchor four) was less accurate. For the upper arm tag position, Anchors 3 and 4 had the greatest absolute ranging error. Except for Anchor 5 and 7, the error was smaller than 200 mm when the tag was placed at the upper back. The greatest error for a lower back-mounted tag was for Anchor 0 and Anchor 5. The median values of the absolute ranging error to each anchor for all tests can be found in the related dataset [
8].
5.1.2. Cumulative Distribution Function of the Ranging Error
The cumulative distribution of the ranging error per anchor was plotted on the same figure, and an example is shown in
Figure 8a. In this example, the tag was placed on the lower back. Anchor 4 had the smallest error, and this was the closest anchor.
To be able to compare different tag positions, the absolute errors to each anchor were aggregated per test.
Figure 8b gives the cumulative distribution of the ranging error for six different tag positions. This plot was made for both bikes and both anchor heights and can be found in the related dataset [
8]. For the setup used in
Figure 8b, the upper arm tag position had the highest probability to have a small ranging error.
5.2. Received Power Level
Figure 7c shows the measured received power (RP) level of each anchor for six different tag positions, again for the track bike and an anchor height of
m. We remark that the center of this plot had the smallest received power level, so the ideal tag position had a received power level plot that was a wide circle.
The worst tag position was the chest, as the signal was blocked by the cyclist and the RP level for Anchors A3, A4, and A5 was very low. The human body influences propagation behavior by reflecting and absorbing (for the most part) the radiated waves [
29]. The same blocking behavior was noticeable when the tag was mounted on the bike, tag positions being underneath the saddle and seat post. A seat post-mounted tag on the pursuit bike is shown in
Figure 9b. Now, the RP level of the anchors in front of the cyclist was low.
Figure 7d shows the RP level for an anchor height of
m. For the chest tag position, the RP level of anchor 0 decreased if the anchor height increased, and this was due to bad antenna orientation. The RP level of Anchor 0 was even lower if the cyclist was on the pursuit bike at this anchor height. This setup is shown in
Figure 9a. It can be seen that the antenna orientation was bad, and the antenna radiated to the body and to the ground. The chest tag was also enclosed by the body of the cyclist.
Increasing the anchor height did not always have a bad influence on the received power level. When the tag was mounted on the lower back of the cyclist, the performance actually improved and when anchors were mounted higher. The received power level of Anchors 2 and 6 increased. The RP level of this tag position for the pursuit bike was similar to the track bike, which was good. This means that this tag position was suitable for races with the track and the pursuit bike.
We now evaluate the upper arm tag position; the tag was mounted on the right upper arm of the cyclist, shown in
Figure 9c. Anchors A0, A7, and A6 have the highest RP level when anchors are mounted at
m, which is the logical result of the chosen setup. For an anchor height of 2.3 m, the behavior changed: the RP level of the anchors on the left of the test person increased. This position also had a similar performance for the pursuit bike.
The last tag position that is discussed is the upper back. For the track bike and an anchor height of m, the RP level of Anchor 3 was the smallest power level of all considered tag positions. This is not to be expected, and probably, something went wrong with this measurement. In the next section, the criteria for an anchor to be in the line-of-sight is discussed. It will become clear that Anchor 3 is non-line-of-sight. This behavior was not observed for the pursuit bike and at a higher anchor height. Because of above reasons, it was assumed that this RP level was the consequence of a software problem. Neglecting Anchor 3, the behavior of this tag position with anchors mounted at m was similar to the lower mounted anchors.
5.3. Line-of-Sight Detection
We want to determine the number of anchor nodes that are in the line-of-sight and the number of anchors that are not in the line-of-sight during a test. This gives an indication of the level of blocking by the cyclist or by the bike. The following rule of thumb was used [
30]:
If the absolute value of the difference between the first path power (FP) level and the received power (RP) level was smaller than or equal to 6 dBm, the anchor was likely to be in the line-of-sight (LOS). If this value was greater or equal to 10 dBm, the anchor node was likely to be in the non-line-of-sight (NLOS). If the difference lied between 6 and 10 dBm, we did not know if the anchor was LOS or NLOS.
The RP and FP values were captured during the static measurement described in
Section 4, therefore, this difference can easily be calculated. An overview of the number of LOS, NLOS, and undetermined anchors of all measurements can be found in the related dataset [
8]. The value of the absolute difference between the first path and received power level for the setup with the track bike and anchors mounted at
m is depicted in
Figure 7b. The green zone on this plot indicates a value smaller than or equal to 6 dBm.
A table that gives an overview of this absolute difference for all measurements can be found in the related dataset [
8]. From this table, it can be deduced that only three configurations had all eight anchors’ LOS, and these are shown in
Table 2. When the tag was mounted on the lower back of the cyclist and anchors were mounted at a height of
m, all anchors were in the line-of-sight, independent of the type of bike that was used.
Figure 7b shows the result when anchors were mounted at a height of
m and the track bike was considered. The tag position that had the most LOS anchors was the upper back; seven anchors were line-of-sight, and one anchor was non-line-of-sight. This was already mentioned in the section about the received power level.
The blocking behavior of the bike and cyclist for tag positions underneath the saddle and seat post was confirmed by this absolute value. For all considered anchor heights and bikes, Anchors A1, A0, and A7 were never LOS. These anchors were even always NLOS for the seat post tag position. The blocking of the signal by the body of the cyclist for the chest tag position was also confirmed: Anchor 4 was always NLOS for all measurements. The worst situation was a chest-mounted tag, track bike, and anchors at m; in this case, there were no LOS anchors.
5.4. Comfort
The comfort of the cyclist wearing the UWB hardware was taken into account. It is possible that a tag position has excellent performance on received power level, the number of LOS anchors, and accuracy, but that this position is not suitable to use in practice because the tag hinders the athlete. The tracking system is to be used during training and competition. If the tag has a negative impact on the performance of the cyclist, it is very likely that the cyclist will refuse to use the system. The measurements happened with a semi-professional cyclist, who is currently the performance analyst of the Belgian national indoor track cycling team. With his advice, the different tag positions were given a comfort rating, shown in
Table 3. A score of five equals very comfortable, and zero is very uncomfortable.
From the ratings above, it is clear that the upper back position was the most uncomfortable. This is because of the setup with the pursuit bike. In this case, the cyclist wore a special aerodynamic helmet, shown in
Figure 5b. This helmet had a long tail, and this tail hindered the tag. Because this aerodynamic helmet was always used when the cyclist rode the pursuit bike, this tag position was not suitable.
The most comfortable tag positions were the ones where the tag was mounted on the bike. The second most comfortable position was the lower back. The tag can easily be mounted here because there is a small pocket in the skin suit that riders wear during training and competition.
5.5. Maximum Communication Range
The possible communication range of a system is an important factor when analyzing the feasibility. Based on this, the number of necessary anchors for the enrollment of the tracking system can be determined. We remark that an indoor cycling track is an open area with no obstacles between tags and anchors. In this section, the maximum communication range, assuming an open area between tags and anchors, is discussed. The sensitivity power level determines the power level after which the packet cannot be detected or decoded anymore. By adapting the modulation setting, the sensitivity level can be changed [
31]. The maximum communication range can be increased by increasing the transmission power.
We now consider the two-path propagation model shown in
Figure 10 [
20]. This model assumes a direct wave between the tag and anchor and a reflected wave from the planar ground. It takes interference between these two waves into account, and the total received energy is the vector sum of the direct and the reflected wave. The difference in path length will result in distances where both signals will interfere destructively and distances with constructive interference.
The difference between transmitted and received power is called path loss. The path loss fluctuates significantly when the distance between tag and anchor is short, and the envelope of the path loss is then proportional to the square of the distance (
). When the distance between tag and anchor is much greater than the height of the tag and anchor, the path loss increases in proportion to the fourth power of the distance (
). The boundary between these two regions is called the breakpoint
and can be calculated with Formula (
2).
where
is the wavelength,
. The center frequency
f for Channel 1 was equal to 3494.4 MHz and
c equal to the speed of light. If we calculate the breakpoint distance for anchor and tag height
m, we get a breakpoint distance
m.
The path loss (dB) at distance
d can be calculated with Equation (
3). It consists of the path loss measured at a reference distance
m and a logarithmic term on the relative increase in distance between transmitter and receiver. Further, it also takes shadowing into account by the term
.
where:
transmitted power in dBm
received power in dBm
path loss at reference distance in dB
path loss exponent
distance in m,
reference distance in m
shadowing term ∼
The transmitted power by the Decawave DW1000 chip Channel 1 was
dBm [
32]. The path loss exponent
n was equal to two because the considered distances were smaller than the calculated breakpoint distance
m.
The maximum possible communication range from the tag to each anchor can be calculated using the measured RP levels. The minimum received power that was required for reliable ranging was
dBm, and the transmitted power was equal to
dBm, which results in a maximum allowable path loss of 91.68 dB. The path loss at the reference distance was equal to the difference between transmitted power and the measured received power. The maximum communication range can now be found using Formula (
3) (remember,
and
m). The results of these calculations for each anchor and for each possible configuration of tag, bike, and anchor heights were appended to the related dataset [
8]. These results were plotted on polar plots and can also be found in the related dataset [
8].
Figure 11a,b show the maximum communication range to each anchor for the track bike and both anchor heights.
When the tag was mounted on the bike, the maximum communication range to the anchors in front of the bike was the lowest for both anchor heights. A chest-mounted tag achieved the greatest communication range to the anchors in front. The maximum range of tag positions lower back, upper back, and upper arm approximated a circular pattern for an anchor height of 2.3 m. Considering these three tag positions, the upper arm tag had the greatest range to the anchor in front, and the lower back tag had the greatest range to the anchor behind the cyclist.
For each combination of tag position, bike, and anchor height, the minimum and maximum value of the calculated maximum communication ranges to the eight anchors were determined. These results can be found in the table: Overview of the maximum ranges, which is available in the related dataset [
8]. These values were aggregated per tag position, and the result is shown in
Table 4. A similar conclusion as the one that followed from the received power level can be drawn. For tag positions underneath the saddle and seat post, the maximum communication range achieved a minimal value for an anchor in front of the rider. For the chest position, this value was the smallest for the anchor behind the rider. This is also noticeable in
Figure 11a,b. A second conclusion is that the difference between the smallest and largest value of the maximum communication range was the smallest for the lower back tag position.
5.6. Energy Consumption
The energy consumption of the solution is another aspect that needs to be taken into account. Because the tag is mounted on the cyclist or on the bike, it needs to be battery powered. For indoor track cycling, a tag that can work for four consecutive hours is sufficient. This is because training and competition almost never exceed four hours. An anchor has a fixed location, for example mounted on a tripod, and can be connected to the electrical grid. Battery-powered anchors have the advantage of an easy and quick setup because no cables are required. In general, the power consumption of an anchor can be much lower than the power consumption of a tag, and this will become clear in the following paragraphs.
In [
33], the energy consumption of an anchor of our hardware and software solution was already discussed. The hardware consisted of the Decawave DW1000 UWB transceiver [
31] and a CC1200 sub-GHz radio [
34]. The difference between our solution and a traditional UWB system is that a traditional approach always listens to the UWB radio. In our solution, the anchors can set their UWB radio in sleep-mode while they are not selected by the tag. They only listen to the beginning of the superframe for the synchronization message using the sub-GHz radio. The current consumption of a stand-by anchor was 3.4 mA. When an anchor is selected to determine the range, it will only power on the UWB radio on the slots on which they are supposed to range. The current consumption of an active anchor was 26.6 mA. In comparison, a traditional approach consumes on average of 130 mA. This means that our solution had a significant current reduction on the anchor nodes. The current consumption of the tag will be much higher than the current consumption of an anchor as the tag is constantly ranging with different anchors.
In [
24], the current consumption of the tag was measured. Furthermore, the average current consumption of a tag and anchor node was calculated. When the tag is powered by a 6000-mAh battery pack, the tag can be powered for 74 h. A powered anchor with this battery pack can last for 99 h, which again shows that a tag consumes more energy than an anchor.
5.7. Summary of the Main Results
Six different tag positions were evaluated on two different bikes. Different anchor heights were considered, namely
and
m.
Table 5 gives the results for an anchor height of 1.5 m. The tag positions are ordered on the number of LOS anchors and received power level. The column “Comfort” tries to rate how comfortable the tag position was. A score of five means very comfortable, zero very uncomfortable. For each measured value, the
and
percentiles are given. The column “Max.” range gives an overview of the lowest and highest value for the calculated maximum communication range to the respectively worst and best anchor per tag position.
Table 6 gives the same results, but for an anchor height of 2.3 m.
From these two tables, the following can be concluded. The tag position upper back had the most LOS anchors for both anchor heights. The chest position performed well for an anchor height of 1.5 m, but had a dramatic performance for an anchor height of 2.3 m with zero LOS anchors. The value of the maximum communication range to the best possible anchor decreased from 43.599 m to only 34.766 m. For an anchor height of 1.5 m, three tag positions had five LOS anchors, and based on the RP level and ranging error, we concluded that the tag positions chest and upper arm had a similar performance. The lower back performances increased if the anchor height increased. The ranging error decreased, and the number of LOS anchors and RP level increased. However, the maximum communication range to the best and worst possible anchor stayed approximately the same. Tag positions on the bike, seat post, and underneath the saddle performed poorly for both anchor heights.
Three tag positions were uncomfortable for the cyclist; therefore, the tag cannot be mounted on the upper back, the upper arm, and chest. The signal from the tag positions on the bike, underneath the saddle, and seat post was disturbed by the bike and cyclist. The signal from the tag mounted on the chest was also disturbed by the cyclist. Therefore, these three tag positions are not suitable. The tag position that had an acceptable performance level (especially at higher anchor heights) and was comfortable for the cyclist was the lower back. For our use case, this tag position outperformed the other positions.
From the above results, it is clear that a larger anchor height improves the performance of the lower back-mounted tag. Ceiling-mounted anchors are preferred. An important remark is that the vertical and horizontal dilution of precision (VDOP and HDOP) values of the setup should be taken into account. A low value for the VDOP and HDOP is of great importance for the performance of the localization system. If all anchors are mounted at the same height above the velodrome, the VDOP value will be very large. Therefore, a combination of ceiling-mounted anchors and anchors on the central square mounted on tripods should be installed. This will prevent all anchors being in the same plane and will improve the localization accuracy.