wireless ad hoc networking

102 Wireless Ad Hoc Networking
3.5.2 Voice Streaming
In our second deployment, we wanted to add the ability for the network to
switch into a high rate of operation capable of streaming interactive voice
from a gateway to a mobile node. Under normal circumstances, sensor data
is collected once every 20 s from all nodes. This includes light, temperature,
average energy level, battery voltage, and the SNR values associated
with any nearby mobile nodes. During the audio streaming mode of operation,
the network sends compressed voice data at 13 kBps. Our primary
focus was on the networking aspects and evaluating the feasibility of such
a system. For our tests, the mobile node was able to sample audio from the
onboard microphone and compress the data while running the networking
task. Our current mobile nodes do not have an onboard digital-to-analog
converter (DAC) and speaker output, so we used a laptop connected to
the node with a serial port to playback the received audio. To simplify
tests, we transferred the raw packet data over the universal asynchronous
reciever/transmitter (UART) and performed the decompression and playback
live on the PC. In a next-generation device, the handheld mobile
node would have a high-end microcontroller capable of doing both the
compression and decompression with a built-in DAC and speaker system.
In controlled environments outside of the mine, we found that the system
performed with below 3% packet loss per hop. Sending redundant
data in separate packets allowed for easily understandable end-to-end voice
transfers. Figure 3.23 shows the distribution of packet loss clustering at four
different hops along an eight-hop series of nodes inside the coal mine. The
end-to-end latency across the eight hops between when audio was sampled
and when the playback occurred was just under 200 ms. Each hop along
a prescheduled path toward the gateway maintained an average latency
of 24 ms. We found that while the mine corridor is clear of obstructions
the wireless channel shows few packet drops. In some situations when
a machine blocks the narrow corridor we see packet loss rates as high
as 50%. Under these circumstances, packet drops are heavily clustering
making error concealment or recovery difficult. Since occupancy inside a
coal mine is relatively sparse (usually less than five groups) compared to
the mine’s size, clear paths are quite common. Future work will investigate
protocols that use the mesh nature of sensor networks to ameliorate broken
links by using redundant paths.
3.6 Summary and Concluding Remarks
We presented in this chapter a detailed description of the FireFly WSN
platform, developed by the Real-Time and Multimedia Systems Laboratory
at Carnegie Mellon University. This platform consists of extensible FireFly
sensor nodes, a time-driven link layer multihop networking protocol called