Our Mark 1 Bioacoustic Monitoring Systems were designed and built using Australian government funding administered by the CSIRO.

The project was conceived by Gordon Grigg. Gordon's account of the project's conception and development can be found here. Other people involved in their design and installation included Les Fletcher (then at Queensland Uni.), Hamish McCallum, Andrew Taylor and Graeme Watson (then at Melbourne Uni.)

Between January 1996 and November 1998 14 Mark 1 systems were deployed, 10 in the Roper valley and 4 in Kakadu National Park.

These systems were deployed in locations subject to human interference, dry season fires, wet season storms & flooding and occasional cyclones. Les Fletcher and Bill Stablum obtained (for free) street lights damaged by vehicle collisions and modified the 5-6 meter high steel poles for our purposes. Microphone, solar panel, rain gauge and temperature sensor were mounted at the top of the pole. Batteries and electronics were raised by a pulley system into the centre of the pole.

These systems are based on a AT304 single board computer with a 25 Mhz Cyrix 486SLC processor, 1 Mb of RAM and 2 Mb of flash.

This provides sufficient computational power to process approximately 25% of incoming sound. Approximately every 5 minutes, details of the frog species identified in that interval are logged to flash memory. Rainfall, temperature and humidity data are logged hourly. The 10 watt power consumption of the AT304 can only be sustained for a few hours per day. Frog monitoring is limited to 4 hours (8pm to midnight), less if battery is low.

A Miniboard, designed at MIT for robotics, is used to power the AT304 on when needed for frog monitoring. The Miniboard has a Motorola 68HC11 processor with 2K bytes of EEPROM and 256 bytes of RAM. Its 1 watt power consumption can be sustained 24 hours/day. An LCD display and temperature & rainfall sensors are connected to the Miniboard. It samples these hourly and stores the data in RAM for transfer to the 304 when it is woken for frog monitoring.

The frog species identification is based around classification of local peaks in the spectrogram of the audio signal the C4.5 machine learning system. Unreliable identifications of peaks are aggregated together using a hierarchical structure of segments based on the typical temporal vocalisation species' patterns. More information can be found here.

A single balanced unidirectional dynamic microphone (Shure SM58 clone from Jaycar) is used. Microphone were wrapped in 2 layers of plastic film (glad-wrap) and deployed in a steel tube with speaker cloth end-covering preventing insect intrusion. We have experienced several failures where insects have penetrated the speaker cloth, likely assisted by UV damage, and filled the microphone with frass.

A balanced preamp using an SSM2017 IC on a small PCB connects the microphone to a Sound Blaster16 card mounted with the AT304 on a passive backplane. Sampling is at 16khz and 16 bit.

System electronics are enclosed in approximately 1 metre of 6" PVC (sewer) pipe below a carrier containing a Panasonic 12V 7Ahr seal lead acid batteries . The batteries are charged from the 30 watt solar panel via a SunSaver 6A-LVD regulator. An ETA-USA VTA01C12 DC-DC Converter produces the +12,+5 and -12 voltages required by sound and computer boards.

The Mark 1 systems were generally deployed for the length of the wet season (November-April). Over the course of the wet season system failure rate was generally in the region of 20-30% but higher in some early seasons.