RealWSN 2015 Trace File Description

Here we present and briefly outline the traces analysed and presented in our RealWSN 2015 paper 'IEEE 802.15.4 Channel Diversity in an Outdoor Environment'

We used a 16-node test network permanently deployed in a field in Marsta (near Uppsala, Sweden). During June and July of 2015 we collected link quality data as well as temperature and humidity measurements. The nodes were mounted on four 1.5 m high poles; for each pole, two nodes are placed near the top of it, and two placed approximately 0.5 m above ground. All nodes were connected (though serial cables) to Linux routers, which in turn were linked together in a single LAN with Ethernet cables. The LAN also included test server which controlled the experiments and stored the output.

Trace files can be downloaded using the following links

Preprocessed traces: download (203 MB)

Raw traces: download (270 MB)

The test application used a custom TDMA protocol to achieve collision free operation and network-wide synchronized channel switching.

The nodes in the network had three roles:
1) The test initiator (1 node)
2) The test initiator helper's (3 nodes)
3) Other nodes

The experiment consisted of doing centrally-triggered test runs repeatedly. Each run had the following schedule:

1) The central server initiated the test run by sending "start" command to the test initiator node.
2) The test initiator node broadcasted out a test preamble message, which included the test ID and the test's channel number. The preamble message was repeated 100 times (with 512 Hz frequency) to increase probability some nodes hear it. It was sent out in the common control channel.
3) The initiator's helpers each repeated the preamble message 100 times. As each of the helpers were located on a different pole, with high probability each nodes in the network heard at least one of these preamble messages.
4) After receiving a preamble message, a node synchronized its time to the test initiator and switched its channel to the designated test channel.
5) A fixed time after the first preamble message, the test run was started on the designated test channel.
6) A test run consisted of each of the nodes broadcasting a 100 packet burst on the designated test channel. Packets were sent with 200 Hz frequency. All non-sending nodes continuously listened for packets, and at the end of a burst logged cumulative stats: PDR, RSSI, and LQI. The sending sequence of the nodes was predefined, and was never changed. Each packet in the burst was 66 bytes long.
7) At the end of packet sending, the sending node did RSSI sampling to detect presence of external interference.
8) At the end of a run, nodes switched back to the control channel.

Despite repeating the preamble message on 4 nodes for 100 times on each, in some tests not all nodes had heard it. These tests were excluded from the results in data analysis stage (less than 1% of all).

At the end of the test run, the next channel was selected according to a pre-computed pseudorandom schedule (16 elements long). After all 16 channels had been tested, a new schedule was computed (using Fisher-Yates random shuffling algorithm) and the whole cycle restarted. We added randomness to avoid unwanted synchronization with unknown periodic environemtntal processes.

All nodes also periodically collected temperature measurements. Sky nodes also periodically collected humidity measurements.

During June we used Tmote Sky nodes; for the experiments in July we replaced every other Sky node with Zolertia Z1 (nodes 2, 4, 5, 7, 9, 11, 14, 16); both have TI CC2420 radio chip.

We selected data from consecutive 7 days in June and consecutive 6 days in July for further analysis in the paper, as during these periods the experiment was running without interruptions. In the first period, around 57 million packets were sent; in the second, around 59 million.

The effective frequency was approximately one 100-packet test each 5 minutes on each of the 16 x 240 = 3840 different channel-links in the network.

The approximate average data rate was around 100 packets per second.

The site is located here, outside the city of Uppsala, Sweden. The deployment sits on an open grass field surrounded by farmland and a meteorological measurement station. The site is not open to the general public and only occasional visits by researchers.

Pictures and Figures from the site:

Node layout and site pictures here

Additional picture of the layout of the nodes here

Weather and Climate

The site is also co-located with a research grade weather station run by Uppsala University, link to data.

Preprocessed data:
=================
All data is sorted in many files. Files are named by using a simple convention:

<sender_id>-<receiver_id>.csv - all channel data for a single link

File format:
timestamp goodPDR allPDR RSSI LQI channel temperatureSender humiditySender temperatureReceiver humidityReceiver noiseFloor channelAveragePDR channelAverageRSSI

"allPDR" includes packets with bad CRC and invalid length.

<sender_id>-<receiver_id>_<channel_number>.csv - 5 min average data for a single channel on a single link
<sender_id>-<receiver_id>_<channel_number>_avg.csv - hourly average data
<sender_id>-<receiver_id>_<channel_number>_all.csv - daily average data

File format:
timestamp PDR RSSI averageTemperatureSender averageTemperatureReceiver

<sender_id>-<receiver_id>_<channel_number>_temp.csv - data averaged by temperature ranges (5 degree ranges)

File format:
PDR RSSI temperature

Raw data:
=========
These are the log files from which the preprocessed data was extracted.

For more information contact:

Atis Elsts, atis.elsts at gmail dot com
Hjalmar Wennerström
Christian Rohner