u-Bat (earlier ULP-ACE)

u-Bat aims to design a deployable system on edge to detect and classify the audio signatures in real-time with an ultra-low power, ultra-light acoustic classifier.

How to use?

System requirements:

1. Operating system:
   • Ubuntu 18.04 LTS or any linux system^[1].
2. Software setup:
   • Download ULP-ACE repository.
   • Python 2.7
     • run sudo apt update
     • install pip sudo apt install python-pip
     • pip install python
     • python 2 dependencies pip install numpy scipy matplotlib
   • Python 3
     • install pip3 sudo apt install python-pip3
     • pip3 install python
     • python 3 dependencies pip install pydub pylab-sdk Wave pytest-shutil playsound numpy scipy matplotlib tqdm
     • other dependencies sudo apt install ffmpeg
     • to change the permissions of the clean.sh: sudo chmod 777 clean.sh
   • Code Composer Studio (CCS) v10.2.0
     • download on demand installer for CCS IDE
     • update system: sudo apt update
     • install dependent libraries sudo apt install libc6:i386 libusb-0.1-4 libgconf-2-4 build-essential
     • install CCS
       • sudo chmod 777 *.run
       • sudo ./*.run
         • select SimpleLink SDK for CC13xx, spectrum digital debug probes and boards while installing.
     • launch CCS and Install the following packages/SDK through resource explorer.
       • go to Software -> SimpleLink SDK Plugins -> SimpleLink Audio Plugin 3.30.0006
       • manage SimpleLink CC13x2 26x2 SDK version to 3.30.00.03
     • import empty project and build the project
       • go to Software -> SimpleLink CC13x2 26x2 SDK 3.30.00.03 -> Examples -> Development Tools -> CC13152R LaunchPad -> TI Drivers -> empty project -> TI-RTOS-> CCS Complier -> empty
       • Build the empty project to generate TI-RTOS kernal.
   • MATLAB
     • download and install
       • sudo ./install
       • select the tools to be installed:
         • MATLAB
         • Audio toolbox
         • DSP system toolbox
         • Signal Processing toolbox
3. Hardware setup:
   • TI's CC1352R launchpad
   • Adafruit PDM MEMS Microphone.
   • Microphone connections:

     PDM microphone	CC1352R launchpad
     3.3V	3.3V
     GND	GND
     SEL	GND
     DAT	DIO26
     CLK	DIO27

   • serial port connection: Identify the port name of the serial connection using dmesg command. Note the port name(ex /dev/ttyACM0).

Data acquisition:

In this framework, get the audio dataset for 1 class or 4 classes.

1. Import the data_dump project into CCS^[2]
2. Build and flash the project in CSS.
3. Navigate to data_acquisition folder and run python audio_frame_serial_print.py /dev/ttyACM0 68 NONE -b 460800 replace /dev/ttyACMO with the port name noted earlier.
4. Play the recordings for the microphone dataset.
5. Press button 1 (BIN-1) on the board to start acquiring the data and button 2 (BIN-2) can be used to stop recording.
6. The microphone recordings are dumped in the folder samples/
7. Follow steps 4-6 for all the recordings of every class + background noise.

Training framework:

1. Follow 'Steps to execute' in the documentation

Demo and deployment:

1. Import it to any of the example demos given in the CCS_projects/ either single-class or multi-class.
2. Paste the header files created by the framework - header_files into the CCS project.
3. If it's multi-class (4-classes), connect four LEDs with 500-ohm resistor in series and refer to the below table for launchpad connections:

   LEDs	CC1352R launchpad
   LED1 (Dingo)	Board_DIO21
   LED2 (Kangaroo)	Board_DIO22
   LED3 (Fox)	Board_DIO24
   LED4 (Koala)	Board_DIO29

4. Build and flash the binary to the board using CCS.
5. Use button 1 (BIN-1) and button 2 (BIN-2) to start and stop detecting respectively

Happy Detecting!

Tasks list:

• Add the UART code into the deployment folder for integration of LoRa @Nadir
• Analyse the neuron outputs and filter outputs for robustness of the algorithm. @Vishnu and Hemanth
• Convert MATLAB scripts to python @Abhishek -- By next week.
• A simple script that can install all the softwares and their dependencies. Bash script for it @Hemanth
• Try different dataset and their applications. @Vishnu and Hemanth
• Videos for the website. @Vishnu and Hemanth
• Change the python 3 script to python 2, take guidance from Sathya or Abhishek.

Footnotes:

1. Windows support is not available yet, stay stunned to GitHub repository for updates
2. Refer to the CCS documentation in case of any issue with importing projects.