Diary
50
Site
Survey, Lansing Michigan
For the Sound Digitizing Device
July
19, 2002
Professor Stuart Gage, Director of the Computational Ecology and
Visualization Laboratory within Michigan State University (MSU),
met with me to discuss the implementation of a Sound Digitizing
Device. I had several questions regarding the use of sound for
ecological studies, and wanted to form an opinion on how we might
proceed to design a system that would meet the needs of using
sound as an indicator of the environment. Stuart was very enthusiastic
about this approach and open about the entire concept. I soon
felt if I were being immersed in "The Ecological Study Through
Sound -101", a study of the environment with emphasis on
the relationship of scientific data mixed with human sensory input,
and I could not get enough. Immediately ideas of virtual reality
tours, allowing a person to "experience a variety of environments"
crossed my mind. But, at this point, the study is not so extreme
as my tangential imagination. Indeed, it possesses a solid basis
of scientific quantification, which includes several concurrent
attributes of the environment along with the ability to recognize
differences between environments simply by listening.
How,
why where, when:
Several
questions and objectives were noted for this meeting:
*
How does recording and categorizing sounds aid to the study of
Long Term Ecological Research (LTER)?
*
What tools are currently used to collect the data of interest?
*
If the world was perfect, and a person only had to snap their
fingers to develop the perfect solution for this type of study,
what would it be?
*
When and where would a pilot project be implemented to prove the
concept of a direct-to-internet sound recording device?
*
How do we quantify and qualify the data into values easily imported
into tools we may use for statistical analysis?
Overview
of the day:
The
meeting opened with coffee at the home of Stuart Gage in Okemos,
a small township east of Lansing, Michigan. The home is located
in an area with a hint of being removed from the metropolis of
industrial Lansing, yet only 15 minutes from the MSU campus. This
location is a fine setting for controlled experiments away from
the main campus, and Stuart capitalizes on this location for his
studies to a great extent. Follow this link to his home:
http://www.cevl.msu.edu/envirosonics/locations/gagehome_sum.htm
External
to the home you will find microphones http://www.natural-technology.com/birdbug2a.htm
used for recording the sounds that exist in the general area.
A computer system automatically generates a 30-second recording
every hour. To demonstrate the results of the event, Stuart played
back samples from the computer and I listened to the sounds of
earlier that morning. Our conversations on the back deck became
a part of this record, and it became clear our presence and conversation
left a fingerprint on the environment for a short period of time.
Also recorded were the sounds of Wrens hustling about in the tree
next to the back patio along with many other sounds, some natural,
but at that time of day, the man made sounds were becoming an
overwhelming component of the audio composition. However, in contrast,
the sounds recorded the night before and very early that morning
revealed a composition more oriented toward what you would expect
in an area located away from the city; sounds of crickets, an
occasional amphibious calls and a great many insect sounds I could
not identify.
Internal
to the home you find interesting tools for creating recordings
of the environment. This includes a computer with a sound card
connected to a DirecPC (Hughes Satellite System), providing the
connectivity of the data collected (audio, video and meteorological)
to the main web server located at MSU. The audio samples are recorded
on the local PC in a wav file format. The sample is then transferred
via File Transfer Protocol (FTP) to the server at the MSU campus
via the satellite system. In essence, other than the power source
for the system being supplied by the local utilities, all the
basic conceptual components are represented for a remote data
collection system: a sound capturing system (the PC), communications
(a satellite system) and a place and method to move and store
the data.
Quantifying
sound seems subjective. What one person may hear, another may
not, or may interpret differently. In this pilot project, Stuart
uses a small number of tools to gather the samples of sound. Some
sound data is transferred to servers at MSU. Some sound is recorded
on audio CDs. To analyze the data, Stuart employed a bright student
to write code to break the digital representation of sounds into
its various frequency and amplitude components. Stuart demonstrated
the visual equivalent of the sounds I heard, which provided another
form of sensory input. Along with audio for my senses, I could
see the mid range frequencies, perhaps the call of the amphibians
or a crow, the croak of a frog, and high frequencies of other
animals along with the mechanically generated frequencies most
common to man. See below:
Taking this one step further, the components were also broken
down into primary frequency bands and represented in the true
digital, numerical form with fixed, objective measurements, correlated
to time and aural events. This data can now be compared to data
of a different time for extrapolation of a data subset reflecting
a possible change in the environment, and furthermore linked to
trend analysis as the environment evolves.
A
few additional sound gathering tools include devices such as MP3
and DAT recorders, which may be deployed by one individual or
a team of individuals in an effort to easily record samples of
sound at a remote location. However, this requires human intervention
to penetrate the site of interest, record a sample, and then leave
the area of interest. It is clear that the human impact on the
environment changes the quality or natural characteristics of
the sample. For example, there are human limitations to remain
at the recording site. The human impact changes the natural state
of the environment: - birds take flight to other locations, amphibians
and insects quit singing, the human scent creates an immediate
or perhaps a fairly long-term barrier many animals may avoid.
The data (sounds) collected, during or near the period of time
of the human presence, probably does not represent the natural
state of the environment before the addition of the human component,
other than that occurrence of the incidental influence of a human
in the area. Also, the area covered is limited, since sound fades
over distance. From this, we can see the need for remote, automatic
sample requirements via a method acceptable and non-intrusive
to the environment.
Recognizing
this need, Stuart has piloted a few wireless systems for collecting
data. These include wireless video monitoring via 2.4GHz low-power,
narrow band transmitter / receiver and wireless microphones typically
utilized for public address / mobile speaking. A very successful
experiment used wireless microphones to capture the beginning
life cycle of bluebirds nested within a birdhouse located in the
middle of the back yard of Stuarts' home. A direct correlation
could be made from the audio clips to the the family of birds
progressing through a new generation life cycle.
Different
than other LTER data collection / monitoring efforts observed
during this project, Stuart has gone beyond the typical Campbell-Scientific
based data collection system used at most LTER sites, and has
implemented a wide variety of tools and methodologies in an effort
to provide a larger picture of the summation of elements creating
the environment at the time of the sample. For example, sounds
recorded are tied to meteorological data, as well as visual data,
such as a view of the observation area in a captured JPEG image.
From my novice viewpoint, this audio / visual / meteorological
data composition provides a more complete picture of the environment
at the time of the sample, which helps all of us understand and
quantify the data presented.
During
the time spent with Stuart it became clear that sound is another
valid indicator of the environment, and the environment is far
more than plant growth rates, PH levels, temperature and wind
direction. The environment consists of possibly an infinite number
of ingredients mixed together and co-existing to create a both
a perception and reality of life in a limited area at a given
time. To capture only a few attributes of the environment will
not provide a valid picture of the environment, it is only an
indication of those attributes and other attributes known to be
directly related, but many of the components of the environment
have to be assumed to be relatively constant.
"The environment consists of possibly an infinite number
of ingredients mixed together and co-existing to create a both
a perception and reality of life in a limited area at a given
time."
To
capture all of the individual attributes of an environment would
be nearly impossible by man, but to capture many of the attributes
humans easily relate to (audio / visual / meteorological) can
provide a greater understanding of the data collected, help with
analyzing trends and possibly provide a much greater confidence
of the conclusions we draw from observations. To this end, Stuart
has made great strides, not only capturing meteorological data,
but also including environmental indicators such as what we humans
sense (audio / visual) and upon which we base our understanding
of the environment.
I
came away from the brief meeting at Stuart's home with a strong
feeling that this unique approach to long term ecological study,
using audio as another indicator of the environment, is a very
under-utilized approach. I feel it provides an important contribution
to standard environmental monitoring techniques, but reaches beyond
the standard data collection processes to provide a more immediate
comparison of the environment in a number of ways. The data can
be analyzed by numerical values, visual displays or by simply
listening to sounds recorded and compared from different times
or places.
This
opened up questions related to standardized data capture once
the sound data collection pilot project becomes a more widely
deployed tool. If sound recording / capturing devices are used
as a status indication of the environment, the measurement must
be repeatable and relative to a known reference point over long
periods of time. The acoustical sensors should be calibrated to
a known standard, just like temperature and humidity sensors.
Otherwise, comparative data from two or more separate audio sensors
may yield different data under the same acoustical / environmental
conditions. This appeared to be a fairly major problem from my
viewpoint, since we are developing a sound digitizing device that
may be constructed by scientists all over the world. I started
digging into audio standards wherever I could find information.
In my opinion, the microphone being the actual sensor, was most
important.
There
does appear to be fairly inexpensive solutions for standardizing
frequency response from microphone-to microphone. The following
graph represents a frequency response for a calibrated microphone,
or a laboratory grade microphone, calibrated to ANSI standard
S1.4-1983, Type 1 - "a precision microphone for accurate
measurements in the field and the laboratory."
From
the graph above, it is clear that the grade of the microphone
solidifies the correlation of acoustical sensing if the user is
interested in frequencies some humans can hear (20Hz to 20KHz).
The input to the sampling device, or recorder is flat within approximately
½ db through that range. Great! However, some microphone
capabilities go even further:
The
chart above from ACO http://www.acopacific.com
depicts frequency response choices available for microphones of
high accuracy and for calibration and characterization of less
expensive microphone systems.
As
you can see, some microphones may be able to reach beyond human
audio levels into the high ultrasonic range. This range may be
interesting to characterize acoustical attributes of insects,
marine animals, or perhaps acoustical events that happen we have
yet to discover.
Other
attributes also affect audio fidelity of the sensing device. These
include phase response over a given set of frequencies, impulse
response to transient events, signal to noise ratios, amplitude
output for a given input, and ability to accurately reproduce
complex composition signals without distortion of the fundamental
signals. Add to this the acoustical diversity due to the location
of the microphone within a weather shelter or placement of the
microphone from the object or area of interest and there still
remains quite an opportunity for variation in the samples gathered.
However, if the sampling characteristics remain fairly constant,
such as the microphone, and the samples are recorded in a fairly
consistent format, the attributes of the sampling system can be
offset and no longer appear to be a major contributor to sample
variation when the interest is to monitor local change.
The
Prototype Sound Digitizing Device (SDD):
Utilizing
a calibrated microphone, we now have to digitize the analog sample
and store it, or provide it in a near real time stream. The digitizing
device must provide several things: A dynamic range capacity to
prevent distortion or clipping of the input signal, a signal to
noise ratio (SNR) great enough to not add machine induced information,
and a sampling rate that will allow a fair reproduction of the
highest frequency of interest. The analog to digital converters
would require either enough bits to alleviate quantizing error,
or an adaptive algorithm similar to companding the signal of interest
at its lowest level. Some of the better sound cards utilize 16bit
A/Ds minimizing the digital steps or quantizing errors. The Nyquist
Theorem states the sampling rate must be at least twice the highest
frequency of interest. If the SDD accurately captures the samples
of interest, and the highest frequency is 20KHz, the SDD requires
a sample rate greater than 40 KHz. In practice, the input to the
SDD must be limited to approximately 20KHz per channel to alleviate
alias components being introduced to the analog to digital (A/D)
section of the SDD. A lower frequency response microphone, perhaps
limited to 20KHz, would be an obvious solution to high frequency
roll-off or defacto front-end filtering. By sampling at a higher
rate, such as 44KHz, (CD quality) a 20KHz signal can be reproduced
in its most abstract form.
To
implement a device that can meet the needs of aural data gathering
other criteria would need to be met, such as low power requirements,
portability, reliability, and a device that is fairly standard
and reproducible. To meet these requirements the PC-104 platform
was the choice. First a fairly high-speed embedded processor was
needed to provide enough processing performance to process the
incoming audio into a standard file format such as wav, or mp3.
Second a fairly standard audio system, such as found on many computers
today would have to be used so the operating system would be able
to communicate it without special drivers or code generation.
Third, a simple power supply requirement is required to allow
deployment in remote areas, such as being powered from solar or
wind. Finally, a severely stable operating system is required,
with a proven track record of being able to run un-attended for
months or even years at a time. For the OS, Linux was chosen,
although FreeBSD, BSDI, and other Unix-like operating systems
would probably work very well, having a very proven track record
for long-term stability and flexibility to allow adaptation to
almost any environment. Our particular choice would be either
Red-Hat Linux or FreeBSD. Linux was chosen primarily because we
had considerable experience with the latest releases of Red-Hat
Linux.
To
facilitate the hardware requirements, the following PC-104 boards
were selected and implemented, stated in order from the bottom
of the stack to the top:
1:
266MHz CPU, capable of running Linux, 256 Megabytes of SDRAM,
a variety of standard I/O, such as PS2 mouse and keyboard, USB,
serial, parallel, Ethernet and IDE drive interface cable.
2:
A Sound-blaster™ compatible audio card, since there is fairly
good support for the devices by several operating systems. In
hind-sight, we would have picked a card that was 5V only, but
at the time this board, with 5V, +12V and -12V requirements would
work fine for the prototype.
3:
A Flash card adapter, which allows us to mount an IBM Microdrive™
to the system providing 1 Gigabyte of storage for the operating
system and data sample storage.
4:
A DC to DC converter, which allows us to adapt the system to a
wide variety of power supply voltages. This particular DC to DC
converter allows 12-60V input and provides +5V, +12V and -12V.
Again in retrospect, a single 5V output DC-DC converter would
be preferred, but the remaining boards (above) would need to facilitate
this requirement.
5:
For development purposes only, a video card is preferred to allow
a standard PC video monitor to be connected to allow development
right on the platform itself.
Now
that the hardware and operating system pieces are chosen, Ice-Cast
and Live-Ice are utilized for the streaming and encoding functions,
respectively. Live-Ice performs the actual compressing / encoding
function, while Ice-Cast provides the streaming function. These
two programs can run simultaneously, providing a single platform
solution for data encoding and streaming. The two programs were
found to be very versatile in their implementation, capable of
adapting to a number of client programs, such as Win-Amp, Real-Audio,
etc.
To
provide some of the desired output streams, several shell scripts
are used to provide the automagic sampling of data, house-keeping
and other functions, showing in my opinion where Unix-like operating
systems shine. In addition, the platform also provides several
Web-based interfaces, one for administration of the Linux system
via a GUI interface, and another for presenting audio streams
for a point-and-click interface.
Finally
the whole system is housed in Stahlin (http://www.stahlin.com)
RJ1412HPL enclosure. This particular enclosure measures 12"
wide, 14" high and 6" deep. It is UV protected from
degradation and is submersion capable rated. A 1.5" PVC pipe
fitting was used to facilitate wiring to the power system of choice
and to provide a feed-through for microphones.
Communications:
Unlike
many other challenges in projects previous to this project, the
wireless implementation could not have been an easier task. This
prototype audio system would be located in the back yard of the
Gages' home, and a simple LAN based wireless system could be used
for this communication link. For this link, we used the Aironet
(now owned by Cisco Systems) 4800 series, IEEE 802.11 standards
based system. However, for this short link, just about any LAN
WI-FI radio would work. The Aironet radio will be connected to
the Ethernet port of the processor of the sound digitizing device
at the encoding side, and the host side will be connected to the
dual-homed Windows 2000 based system that is in turn connected
to the DirecPC satellite communications system. The dual-homed
PC presents the biggest challenge for the networking portion of
the implementation.
In
a very remote deployment or greater distance, radios aimed toward
the Wide
Area Network (WAN) market would
be used with higher gain antennas and appropriate site survey
to determine exactly what will be needed to successfully deploy
the sound system at that particular location.
Conclusion:
This
project requires more research into the process and implementation
to provide a standardized platform for repeatable and consistent
envirosonics research. The days coming will be focused on problems
more related to the variables in the sound capturing device, such
as microphone characteristics, a standardized recording amplitude
for a given audio input level, characterization of the encoding
system, possibly equalization and determining what characteristics
are important to this research.
Data
Sources:
ANSI
standard S1.4-1983, Type 1 "a precision microphone for accurate
measurements in the field and the laboratory."
http://www.britannica.com/seo/h/harry-nyquist/
http://www.und.nodak.edu/news/messages/386.html
http://www.geocities.com/bioelectrochemistry/nyquist.htm
