Reproducibility

In this section the steps required to reproduce the results presented in the paper are explained.

Applications

Bike speedometer

  1. Replace Botoks’s solar cell with a magnetic energy harvester (one commonly found in inductive bike lights), and attach it to the fork of a bike, behind the wheel.

  2. Connect one GPIO of Botoks to a logic analyzer to measure reboot periods and have a ground truth for the RPM of the wheel.

  3. Connect Botoks to a debugger capable of debugging the MSP430 line of products. Set the target voltage to 2.2V to avoid back-feeding. Then power Botox using the auxiliary power connector on the PCB. The voltage provided needs to be between 3.4 and 5V.

  4. Program the bike-rpm-tx application using UniFlash:

    $ ./flash.sh bin/bike-rpm-tx.out

  5. Program another Botoks node to receive radio packets from the other node on the bike. Have this latter node continuously powered, and record reboot periods contained in radio packets.

To produce the graph in the paper (Figure 9) convert reboot periods to RPM and then to km/h (that depends on the radius of the wheel) of both the ground truth, extracted with the logic analyzer, and the CHRT-measured results, collected by the continuously-powered Botoks receiver.

Network synchronization

We have established a point-to-point link with two nodes powered by solar energy, in one experiment, and radio-frequency energy, in another experiment. The solar energy was generated by two independent light bulbs placed in two closed boxes and collected by Botoks’s solar panel. The radio-frequency energy was provided by a Powercast transmitter Powercast transmitter and harvested by a Powercast receiver connected to Botoks’s power supply (in place of the solar cell). In the application, the transmitting node would wake up to send one packet of data, using its buffered energy, eventually incurring a power failure, proceeding then to harvest energy and finally recharge and wake up again. Similarly, the receiving node would wake up and start listening until receiving a packet, or until a power outage.

  1. Connect one Botoks to a debugger capable of debugging the MSP430 line of products. Set the target voltage to 2.2V to avoid back-feeding. Then power Botox using the auxiliary power connector on the PCB. The voltage provided needs to be between 3.4 and 5V.

  2. Program the intermittent-tx application using UniFlash:

    $ ./flash.sh bin/intermittent-tx.out

  3. Program the second Botoks with the intermittent-rx application:

    $ ./flash.sh bin/intermittent-rx.out

  4. Connect both nodes to a logic analyzer to record sent and received packets.

  5. For the solar energy experiment, place each node in a box that can filter out external light, and place a controllable light bulb in each of the box. Then turn on the light bulbs to start the experiment.

  6. After 10 minutes, turn off the lights to conclude the experiment.

Use the logic analyzer data to compute the ratio between received and sent packets and produce the graphs as in the paper (Figure 10).

CHRT Accuracy

In this procedure we assume the CHRT already has been calibrated. Please refer to the CHRT section for instructions on how to calibrate the CHRT.

In order to verify the results presented about the accuracy of the CHRT the following steps are executed:

  1. Connect Botoks to a debugger capable of debugging the MSP430 line of products. Set the target voltage to 2.2V to avoid back-feeding. Then power Botox using the auxiliary power connector on the PCB. The voltage provided needs to be between 3.4 and 5V.

  2. Program the calib-accuracy application using Uniflash.

$ ./flash.sh  bin/calib-accuracy-t0.out

  1. Monitor the output.

$ picocom /dev/ttyACM1 -b 19200 --imap lfcrlf

  1. Repeat this process 10 times (go back to step 2).

  2. In previous steps the accuracy of tier 0 was measured. Now this process is repeated for every other tier. Change t0 to t1, t2 or t3 in the bash call to program the device.

With the measurements completed, we processed the data as follows:

  • Average the reported time for each time-step of the 10 runs for tier 0 and tier 3. These results are presented in figure 7a and 7b.

  • For figure 7c and 7d, first the error is computed by deducting the expected time from the measurement. Next all of the computed errors for a tier (so over the whole timing range and 10 executions) are binned into -2, -1, 0, 1, 2 bins and from this the percentages are computed that are shown in figure 7c and 7d.