E-textile Based Imaging Array: Phase Four
Introduction
This web page describes the work done on the fourth phase of the
project and the results obtained. We have successfully implemented
the fixed routing algorithm for the sending of sensor data. The
node in the center of the array, Node 13, has been designated as
the single collection point of all the sensor data and to form the
image. We have also collected some preliminary power statistics.
Work Done
The Fixed Routing Algorithm
Figure 1 shows how the fixed routing algorithm works. Node 13 is
designated to be the node to form the sensor image. The array will
be divided into four quadrants. Nodes in the quadrants will send
packets to the two axes, which separate the quadrants. The nodes
on axes will then relay the packets to the center node in the array.
The fixed routing algorithm was successfully implemented.
|
Figure 1
The Fixed routing algorithm |
Tuning of system parameters
After the entire imaging algorithm is completed, we attempt to tune
the system for performance. The following parameters are optimized:
Network segment frame size
The network segment frame size determines the largest size of
a data frame that can be sent across a segment at any one point.
However, if the data packet is smaller than the network segment
frame size, the processor will still read in this maximum frame
size worth of data, resulting in unnecessary delays and wasting
processing power. By adjusting the simulator configuration, we
reduced this frame size to 64 bytes from the previous 128.
Application data packet size
The network segment frame size represents the physical layer constraints.
On the application layer, we reduced the application data packet
size to fit into the maximum frame size. We convert some of the
integers into unsigned characters, downsizing from 4 bytes to
1 byte each. The impact of such a downsizing meant that there
should not be more than 256 nodes in a single array (an unsigned
character is used to represent node IDs). We do not foresee an
e-textile array to be greater than 256 nodes, due to latency issues.
To have more than 256 nodes, it is probably necessary to have
subnets of smaller arrays connected by high-speed links.
Interval between ID Map formation and sending of sensor
data
Since we use two separate algorithms to form the ID map and to
send sensor data, it will be good to separate these two phases
by a time interval. During this interval, nodes that have completed
ID map formation will not send out sensor data but continue passing
neighbor-ID information. This will allow nodes that are still
forming the ID map to have the maximum bandwidth.
From the time taken to form the ID map (Figure 2), we determined
that a time interval of 0.7 sec should be sufficient.
Transmit buffer on each node
The employment of the fixed routing algorithm means that the nodes
surrounding the collection node (on the two axes) will to route
a large volume of data. The transmit buffers on these nodes have
to be increased to accommodate this higher data volume.
Sensor data rate
We attempt to determine a suitable sensor data rate that could
be used for the imaging array. We are looking at a value of around
10Hz. Too high a data rate would “choke up” the network
and prevent the image from forming up properly. On the other hand,
too low a data rate would not produce a useful image since it
will not be updated frequently enough.
However, we encounter a “Segmentation fault” error
when the simulation runs for some time. Although we are able to
see the image being formed, we have problem with the printout
such that it is difficult to validate the image.
Results and Statistics
The results are shown in the Figures 2 to 5. Figure 2 shows
the time taken for each node to form the ID map. We can observe
that the nodes at the corners typically take the longest time to
form the ID map, whereas the center nodes takes the shortest time.
Figure 3 to 5 show the power consumption charts, i.e. the energy
the CPU uses, the battery drainage and the maximum current. These
are preliminary power statistics.
|
Figure 2
Time taken to form ID map |
|
Figure 3
Total CPU Energy for Node 13 |
|
Figure 4
Max current for Node 13 |
|
Figure 5
Battery energy remaining for Node 13 |
Next phase
In the final phase, we will continue to tune the parameters if we
manage to get the simulator to run properly. Following that, we will
obtain power statistics to compare different routing schemes. We will
compare the Ring algorithm against the flooding algorithm in the ID
map formation step. We will also compare the Fixed routing algorithm
against the flooding algorithm in the sending of sensor data step.
If time permits, we will also look at fault tolerance. For example,
we could disable one node and see how the image will form. Currently,
a node will not start sending sensor data until it has form the
ID map completely. We can change this criterion such that it can
start sending data even when its ID map has not formed completely,
due to faulty nodes.
Conclusion
In conclusion, we have successfully implemented the fixed routing
algorithm for the sending of sensor data. We have tried to tune some
of the system parameters to make the image array runs better. We also
tried to determine a suitable sensor data rate. However, we have not
been able to conclude on a data rate due to some problem we had with
the simulator. We have also collected some preliminary power statistics.
References
[1] “A Survey of Technologies for Smart Fabrics(Computational
Textiles), DRAFT, Summer 2001”, Phillip Stanley-Marbell
[2] “Project proposal, E-Textile-based Ultra-sound Imaging Array”,
Seng Teck, Sing & Chee Wan, Teng
[3] “Project Report Phase 1, E-Textile-based Ultra-sound Imaging
Array”, Seng Teck, Sing & Chee Wan, Teng
[4] “Project Report Phase 2, E-Textile-based Ultra-sound Imaging
Array”, Seng Teck, Sing & Chee Wan, Teng
[5] “Project Report Phase 3, E-Textile-based Ultra-sound Imaging
Array”
[6] “Myrmigki Simulator Manual, Release 0.1.ece743”, Philip
Stanley-Marbell. |