CRAWDAD cu/antenna

Dataset of signal strength collected from 2.4 GHz directional antenna.

We collected signal strength data to derive a parametric model for 2.4 GHz directional antennas.

date/time of measurement start: 2007-08-16

date/time of measurement end: 2008-03-07

collection environment: As the demand for wireless networks grows, the research community continues to seek methods for improving network performance. One of the method for improving network throughput involves using directional antennas to increase signal gain and/or decrease interference. We collected signal strength data to derive a parametric model for (2.4 GHz) directional antennas.

network configuration: We use two laptops, one receiver and one transmitter. Each is equipped with an Atheros-based MiniPCI-Express radio which is connected to an external antenna using a U.Fl to N pigtail adapter and a length of LMR-400 low-loss antenna cable. The receiver laptop is connected to a 7 dBi omnidirectional antenna on a tripod approximately two meters off the ground. The transmitter laptop is connected to the antenna we intend to model on a tripod 100 feet from the receiver and also two meters off the ground. The transmitter tripod features a geared triaxial head which allows precise rotation.

data collection methodology: The transmitter radio is put in 802.11x ad hoc mode on the least congested channel. The transmitter’s ARP table is manually hacked to allow it to send UDP packets to a non-existent receiver. The receiver is put in monitor mode on the same channel and logs packets with tcpdump. Finally, both the receiver and transmitter must have antenna diversity disabled. With the equipment in place, the procedure is as follows: For each 5 degree position about the azimuth, send 500 un-acknowledged UDP packets. Without intervention otherwise, due to MAC-layer retransmits, each will be retried k times (where k is radio-vendor and/or driver implementation specific), resulting in k ∗ 500 measurements. During the experiment, the researchers themselves must be careful to stay well out of the near-field of the antennas and to move to the same location during runs (so that they, in effect, become a static part of the environment). If additional data is desired for a given location, multiple receivers can be used, provided the data from them is treated separately (as each unique path describes a unique environment).

limitation: We were unable to aquire access to an anechoic chamber in time for this study, but would like to make use of one in future work, for even cleaner reference measurements. 

Tracesets

rss

Traceset of signal strength collected from 2.4 GHz directional antenna.

• file:cu-antenna-data-200905.tar.gz
• description: We collected signal strength data to derive a parametric model for 2.4 GHz directional antennas.
• measurement purpose: Network Performance Analysis
• methodology:

1. Testing Commodity Hardware

To ensure that it is safe to use commodity 802.11x-based hardware to measure antenna patterns, we calibrate the sensitivity of our radios and analyze losses in the packet-based measurement platform. 

In the process of collection, some packets will be dropped due to interference or poor signal. In our experience, the percentage of dropped frames per angle is very small: the maximum lost frames per-angle in our datasets is on the order of 5%, with less than 1% lost being more common (the mean is 0.01675%).

Moreover, the correlation coeffient between angle and loss percentage is -0.0451, suggesting that losses are uniformly distributed across angles. Given that we have taken 4000 samples in each direction (k=8 for our configuration), noise in our measurements due to packet loss is negligible.

2. Experiment Setting

We collected data in several disparate environments with three different antennas.

With the exception of the reference patterns, all of the measurements were made with commodity hardware by sending many measurement packets between two antennas and logging received signal strength (RSS) at the receiver. The three antenna configurations used include:

- a HyperLink 24dBi parabolic dish with an 8-degree horizontal beam-width,

- a HyperLink 14dBi patch with a 30 degree horizontal beam-width, and

- a Fidelity Comtech Phocus 3000 8-element uniform circular phased array with a main-lobe beam-width of approximately 52 degrees.

This phased array functions as a switched-beam antenna and can form this beam in one of 16 directions (on 22.5 degree increments around the azimuth). For the HyperLink antennas, we used the same antenna in all experiments to avoid intra-antenna variation due to manufacturing differences.

In addition to the in-situ experiments, we have a “reference” data set for each configuration. The Array-Reference data set was provided to us by the antenna manufacturer. Because HyperLink could not provide us with data on their antennas, Parabolic-Reference and Patch-Reference were derived using an Agilent 89600S VSA and an Agilent E4438C VSG in a remote floodplain.

3. Experiments

Following is a brief description of each of the experiments:

- Parabolic-Outdoor-A, Patch-Outdoor-A: A large open field on the University of X campus was used for these experiments. The field is roughly 500-feet on a side and is surrounded by brick buildings on two of the four sides. Although there is line of sight and little obstruction, the surrounding infrastructure makes this location most representative of an urban outdoor deployment.

- Parabolic-Outdoor-B, Patch-Outdoor-B: A large University-owned floodplain on the edge of town was used for our most isolated data sets. The floodplain is flat, recessed, and is free from obstruction for nearly a quarter mile in all directions. This location is most representative of a rural backhaul link.

- Array-Outdoor-A: The same open field is used as in the Parabolic-Outdoor-A and Patch-Outdoor-A data sets. The collection method here differs from that described in section 3. A single phased array antenna is placed approximately 100 feet away from an omni-directional transmitter. The transmitter sends a volley of packets from its fixed position as the phased array antenna electronically steers its antenna across each of its 16 states, spending 20 ms in each state. Several packets are received in each directional state. The phased array antenna is then manually rotated in 10 degree increments while the omni-directional emitter remains fixed. The same procedure is repeated for each of 36 increments. Moving the emitter changes not only the angle relative to the antenna but also the nodes’ positions relative to their environment.

To address this confound, each physical position is treated as a separate experiment. This means that the number of angles relative to the steered antenna pattern is limited to the number of distinct antenna states (16). The tx-power of the radio attached to the directional antenna was turned down to 10dBm to produce more tractable noise effects (for the purpose of modeling small-scale behavior the default EIRP is much too high).

- Parabolic-Indoor-A and Patch-Indoor-A: For this data set we used the University of X Systems Lab. The directional transmitter was positioned approximately 20 feet from the receiver in a walkway between cubicles and desks. This is our most cluttered environment.

- Parabolic-Indoor-B, Parabolic-Indoor-C, Patch-Indoor-B, and Patch-Indoor-C: An indoor offce space was used for this set of tests. See figure 11 for the floor-floorplan of this office space. Two receivers were used here: one with line of sight and one without line of sight, placed amidst desks and offices.

- Array-Indoor-A and Array-Indoor-B: Seven phased array antennas are deployed in the same 25x30m indoor office space used for Parabolic-Indoor-B, Parabolic-Indoor-C, Patch-Indoor-B, and Patch-Indoor-C. Data from two of the seven antennas are analyzed here. Each antenna electronically steers through its 16 directional states, spending 20 ms at each state. Two mobile omni-directional transmitters move through the space and transmit 500 packets at 44 distinct positions. For each packet received by a phased array, the packet’s transmission location and orientation is recorded (i.e., which of the four cardinal directions was the transmitter facing) along with the directional state in which the packet arrived and the RSSI value.

- Parabolic-Reference and Patch-Reference: The large flood-plain is used here. An Agilent VSA is connected to the omni-directional receiver and makes a 10-second running average of power samples on a specific frequency (2.412 GHz was used). Three consecutive averages of both peak and band power are recorded for each direction. The directional transmitter is rotated in five degree increments and is connected to a VSG outputting a constant sinusoidal tone at 25 dBm on a specific frequency. Before, after, and between experiments we made noise floor measurements and as a post-processing step, we have subtracted the mean of this value (-59.61811 dBm or 0.0011 µW) from the measurements.

4. Normalization

Our first task in comparing data sets is to come up with a scheme for normalization so that they can be compared to one another directly. For each data set, we find the mean peak value which is the maximum of the mean of samples for each discrete angle. This value is then subtracted from every value in the data set. The net effect is that the peak of the measurements in each data set will be shifted to zero.

rss Traces

• rss/reference: Trace of signal strength collected from 2.4 GHz directional antenna.
• configuration: Clean "reference" antenna measurements supplied to us by the manufacturer or taken with a VSG and VSA in remote floodplain. 
• format: ::: baseline.txt.bz2 ::: 

Contains a header line followed by newline-delimited records of whitespace-delimited fields. The first column is just the record number and doesn't correspond to a header label. This is kind of confusing, but it's the native format for R's read.table() and write.table(), so if you use R, your life is especially easy. If you don't, a command like this will put the data in a more ammenable format:

bzip2 -cd baseline.txt.bz2 | tail -n $((`bzip2 -d -c baseline.txt.bz2 | wc -l`-1)) \

| cut -f 2- -d ' ' > baseline.txt

Some sample lines are as follows:

cphillips@shannon:~/data$ bzip2 -dc baseline.txt.bz2 | head -n 2 "position" "ctr" "batch" "tag" "norm.rss" "1" 0 -34.0841277279277 "patch" "default" -0.313844191808094

The fields are (i.e. as appear L to R in the 2-Nth line):

- id: quoted record id as produced by R's write.table()

- position: angle about the azimuth

- ctr: measured power value at center frequency with noise floor subtracted (i.e. RSS)

- batch: experiment/antenna label: patch, parabolic, or patty (patty is an 8-element uniform circular phased-array antenna)

- tag: always default. used as a sub-batch identifier

- norm.rss - Within a given trace (i.e., unique batch/tag), the normalized RSS of each packet is defined as the absolute RSS less the "reference maximum" for that trace. The reference maximum is the greatest mean value for any angle within the trace.

• rss/in-situ: Trace of signal strength collected from 2.4 GHz directional antenna.
• configuration: In-situ antenna measurements were taken using (calibrated) commodity 802.11 hardware. 
• format:  ::: packets.txt.bz2 ::: 

Contains a header line followed by newline-delimited records of whitespace-delimited fields. The first column is just the record number and doesn't correspond to a header label. 

This is kind of confusing, but it's the native format for R's read.table() and write.table(), so if you use R, your life is especially easy. If you don't, a command like this will put the data in a more ammenable format:

bzip2 -cd baseline.txt.bz2 | tail -n $((`bzip2 -d -c baseline.txt.bz2 | wc -l`-1)) \

| cut -f 2- -d ' ' > baseline.txt

Some sample lines are as follows:

cphillips@shannon:~/data$ bzip2 -dc packets.txt.bz2 | head -n 2

"rss" "batch" "position" "tag" "norm.rss" "norm.diff"

"1" 48 "parabolic-field2" 0 "default" -2.66053226698007 2.66053226698007

The fields are (i.e. as appear L to R in the 2-Nth line)

- id: quoted record id as produced by R's write.table()

- rss: packet received signal strength as reported by the MadWiFi driver

- batch: same as above

- tag: near, far, or default. used as a sub-batch identifier for the indoor experiments.

- norm.rss: same as above

- norm.diff: normalized difference from the corresponding baseline pattern

The batch/tag names are related to the "pretty names" used in our 2008 tech. report by the following space-delimited mapping:

pretty-name batch tag

Parabolic-Outdoor-A parabolic-field2 default

Parabolic-Outdoor-B para-floodplain default

Parabolic-Indoor-A indoor-lab default

Parabolic-Indoor-B parabolic-cinc far

Parabolic-Indoor-C parabolic-cinc near

Parabolic-Reference parabolic default

Patch-Outdoor-A patch-field default

Patch-Outdoor-B patch-floodplain default

Patch-Indoor-A patch-indoor-lab2 default

Patch-Indoor-B patch-cinc far

Patch-Indoor-C patch-cinc near

Patch-Reference patch default

Array-Outdoor-A patty-field default

Array-Indoor-A patty-cinc-1 default

Array-Indoor-B patty-cinc-7 default

Array-Reference patty default.