This month, Sofar Ocean and Applied Ocean Sciences (AOS) released a new Bristlemouth enabled sensor for underwater acoustics monitoring. This is the second iteration of Spotter Sound. The first iteration was released around this time last year. There are some awesome features that this product has to offer including[1]:
-
AI powered detection and classification of vessels and dolphins
-
Real time Sound Pressure Level (SPL) spectrogram
-
Improved audio quality and sensitivity over the first iteration
The attached images are from Sofar Ocean’s dashboard[2].
But the most important feature of this new generation acoustic system is the power savings gained. Within the next few sections the following information will be covered:
- Hardware updates between the first and second iterations
- Software improvements
- Power statistics of the next generation
Updated Hardware
The next generation of Spotter Sound has three processing elements within the sensor:
- SCARI DAQ: Hard real-time data acquisition
a. Responsible for data acquisition from the hydrophone element, saving that data to an onboard SD card and forwarding that data to the other 2 processors - Bristlemouth Mote: Network protocol plus aggregate analysis
a. Responsible for taking the readings collected by the DAQ, transforming the data so that it may be compatible with Bristlemouth and forwarding the data onto the network, as well as performing aggregated analysis of the data - Raspberry Pi Zero 2W Single Board Computer (or SBC): Heavyweight inference
a. Responsible for running detection models (custom user detectors, or Blue Oasis’s Hydrotwin[5]) via ingested data from the DAQ
Significant updates were made to the electrical hardware from the previous iteration which include:
- A new Microcontroller for the aforementioned DAQ processing element
- Improved power supply design
The Microcontroller chosen for this new iteration is the Raspberry Pi RP2350.
It is a dual core ARM Cortex-M33 processor, which provides significantly better power improvements over the previous iterations Cortex-M4 Microcontroller, the ATSAMD51 by Microchip [3][4]. Tighter control of the power domain can be achieved in the RP2350, giving the real advantage. There are also massive improvements to the onboard RAM, 520kB compared to 256kB, allowing more data to be buffered before offloading to the Bristlemouth mote or SBC.
The simplified power design also helped improve efficiencies on board. On the PCB, power is tapped directly from the 3v3 output of the Mote to power the SCARI DAQ element, rather than stepping down the 24V Bristlemouth network.
With these changes, testing has shown a significant decrease in power of 40mW.
Updated Software
On the first iteration of Spotter Sound, if the SBC was enabled, it would be on 100% of the time. Being that it is the most power hungry element on the system, this is less than ideal. Spotter Sound 's new generations adds the ability to duty cycle the power of the SBC. The following figure depicts the differences, the red shaded sections capture when the SBC is powered:
The DAQ now has the ability to replay the data while the SBC was powered off faster than real time, buffering the data on the SD card allows 100% coverage of audio data to feed into the detection model.
Some other improvements were made to the Linux image of the SBC to improve boot time and save power that can be viewed in the borealis_sbc Git repository[6]
The amount of power saved due to the optimizations mentioned above is ~540mW.
Power Statistics
The goal of the next generation Spotter Sound was to be able to run the system year round without needing to power down the device, and that is now possible with the aforementioned improvements. Power draw for the new Spotter Sound in two common configurations provided for customers is shown below:
| Configuration | Gen 1 | Gen 2 |
|---|---|---|
| SBC enabled, Hydrotwin | 960 mW | 360 mW |
| SBC disabled | 180 mW | 140 mW |
What This Unlocks
Because the Spotter’s energy budget is set by solar harvest, and harvest falls off with latitude, power draw translates directly into how far from the equator a sensor can sit and still run through the winter. At 360 mW with the SBC enabled and running Hydrotwin, Spotter Sound can be deployed year round within 20° of the equator. With the SBC disabled, the 140 mW draw extends that to 40° N/S, covering most of the world’s continental shelves and busiest shipping lanes.
Both figures assume the sensor runs continuously. Duty-cycling the sensor itself pushes the operational envelope higher still, trading temporal coverage for latitude.
Resources
[1] Sofar Ocean Spotter Sound
[2] Sofar Ocean Dashboard Demo
[3] Raspberry Pi RP2350 Power Consumption
[4] Microchip ATSAMD51 Power Consumption
[5] blueOASIS - Hydrotwin
[6] Borealis SBC Git Repository