Every year in the U.S., active shooter incidents tragically cause hundreds of casualties, often because alerts reach responders too late. To solve this, I propose SonicSentinel, an ultra-low-cost, battery-powered acoustic detection node built around the PIC16F13145 Curiosity Nano.
Introduction
Active shooter incidents often go undetected until after multiple rounds have been fired, leading to delayed responses and tragic outcomes. SonicSentinel addresses this challenge by providing a low-cost, battery-powered acoustic detection node built around the PIC16F13145 Curiosity Nano. The system continuously monitors ambient sound through a Max4466 microphone amplifier, processes the incoming audio with a real-time FFT on the microcontroller, and uses simple visual feedback to indicate safety (green) or potential danger (red). By leveraging optimized signal processing and the CLB, SonicSentinel can detect the characteristic frequency signature of gunshots within milliseconds and notify nearby responders or trigger additional alerts.
Methodology
Four PIC16F13145 peripherals and their drivers were used:
- EUSART1
- Interrups
- Configurable Logic Block clocked on Timer2
- ADC clocked on Timer1.
Figure 2.1: PIC16F13145 Design Tree.
CLB Design
Figure 2.2: CLB Design Flow
The CLB takes input from the software-driven CLBSWIN0. When CLBSWIN0 is true, the Safety LED Indicators are turned on, and the simulation for the Danger Alert LED Indicator Array remains off. When CLBWIN0 is false, a Danger Alert is triggered and the Verilog simulation runs until CLBWIN0 returns to true by triggering the reset. Below is a State Transition Table for the State Machine Architecture that the Verilog simulation code is built upon.
Hardware Setup:
- Microphone: A Max4466 amplifier module is connected to the PIC’s analog input (RA0) to capture sound pressure levels.
- Analog-to-Digital Conversion: The built-in ADC on the PIC16F13145 samples the audio signal.
- LED Indicator Arrays: 10 GPIO pins drive 6 red and 4 green LEDs. Under normal conditions, the green LED remains lit. A detected gunshot or emergency button press triggers the red LED Array simulation.
Figure 2.3: Schematic Diagram of the Hardware Setup Connection.
FFT Implementation
Blocks of 64 audio samples were collected, and a standard Cooley‑Tukey FFT run was performed to turn each time‑domain frame into its frequency‑domain representation. Because the input is real, only the first 32 bins actually carry unique information (the remaining bins mirror them), so the redundant half was discarded. These 32 bins gave an evenly‑spaced spectrum, from DC up to Nyquist. Lastly, a simple magnitude for each bin was computed, which became the basis for the detection logic.
Detection Algorithm
To reliably spot gunshots "brief, broadband impulses," a two‑metric test was used, tuned, and developed against test audio samples conducted.
Spectral Flux
Measures how much the spectrum jumps from one FFT frame to the next. Gunshots show a large, sudden spike; ordinary sounds (speech, music, ambient noise) tend to change more gradually. Using this formula.
The spectrum flux formula compares consecutive audio sample frames “X” to compute the sum of the difference between individual frequency bins, given that a jump is detected.
Active‑Bin Count
Counts how many of the 32 bins exceed their “ambient‑only” noise floor. Utilizing collected background data, a per‑bin threshold was set so that everyday noises rarely light up more than a handful of bins, whereas a gunshot lights up many at once.
Results
Figure 3.1: Example Collected Ambient Noise Frequency Bins Data Samples
Figure 3.2: Example Collected Gun Shot Detected Frequency Bins Data Samples
Figure 3.3: Example Collected Loud Music Frequency Bins Data Samples
Resource Utilization:
- Program Memory: 5,715 of 8,192 words used (~70%)
- Data Memory: 503 of 1,024 bytes used (~49%)
Discussion
The decision for the detection algorithm used was mainly data-driven. For the purpose of demonstrating in this project, A pistol sound from the upbeat.io website was used for gunshot sounds. No actual gun was used in the calibration and the interpretation of the sampling data. However, the code structure was written so well, allowing easier calibration on any alternative scope of variables. One of the major challenges during this project was finding the right documentation and learning resources. This slightly slowed down progress. To solve this, I mainly used the documentation provided, and also read other project examples available and related to my project.
Conclusion
SonicSentinel showcases how dynamic, real-time signal processing on a low-cost microcontroller can deliver rapid, reliable gunshot detection with minimal hardware. The integration of a Max4466 microphone, optimized 64-point FFT, clear LED indicators, and a UART peripheral provides an effective proof of that concept.
