Glass-Break Sensors Using Silicon Labs AI/ML-Enabled SoCs
In an age where smart, IoT-enabled homes and advanced security systems are becoming the norm, the need for precise and reliable sensors has never been greater. Among the various types of security devices, glass break sensors are one of the first lines of defense. When designed to correctly detect the sound of breaking glass - a common method of forced entry - it can accurately alert homeowners or security systems to potentially dangerous intrusions. The advent of AI/ML has revolutionized smart sensor and glass break detector technology by introducing higher accuracy powered by adaptive learning, and also smaller device sizes at a lower price. Silicon Labs’ AI/ML-enabled wireless SoCs stand at the forefront of this pathbreaking innovation.
Advantages of AI/ML-Enabled SoCs in Glass Break Sensing
Existing acoustic glass break sensors often rely on simple traditional DSP algorithms running a microphone’s level thresholding and sound frequency analysis to detect whether the glass on a window is being broken. While effective to some extent, these sensors are prone to false alarms due to other common high-frequency and sharp loud noises such as doorbells, the sound of dishes clattering, voices speaking or even loud music. Another category of glass break sensors called shock sensors, utilize shock impact to initiate an alarm. They detect vibrations on the panes and activate the alarm when the amplitude of the vibration exceeds a preset threshold. However, even a loud knock or a tap on a window can sometimes be misinterpreted as a break-in attempt.
The drawbacks with both these types of sensors include the false positive sensing and that the sensitivity levels need to be adjusted, and the accuracy of these sensors can vary, often requiring manual calibration and adjustments to function optimally. Additionally, in the case of shock sensors, a separate device is required for every pane of glass, increasing the overall installation cost.
This is where Silicon Labs' AI/ML-enabled SoCs offer a solution to the limitations of traditional glass break sensors. By integrating AI and ML capabilities into the sensor design, these SoCs allow fast-track development and greater flexibility, significantly enhancing the accuracy and reliability of glass break detection by reducing the rate of false triggers.
This is achieved by training the ML based pattern matching algorithms on large datasets of various ambient environment sounds and glass break audio samples, enabling the sensor to learn and recognize the unique sound signature of glass shattering and differentiate it from ambient sounds. As a result, the sensors can effectively minimize false alarms and only trigger alerts when a genuine glass break event is detected.
This video demonstrates the superior efficiency of Silicon Labs’ ML-enabled Glass Break Sensor as compared to a traditional algorithm operated glass break sensor available on the market.
Silicon Labs’ Glass Break Sensing Solutions
Silicon Labs SoCs being designed with power efficiency in mind means that the integration of ML capabilities reduces the need for external components and minimizes power consumption and design cost. Silicon Labs’ xG24, xG26 and xG28 SoCs are equipped with a Matrix Vector Processor (MVP) that speeds up ML inferencing by up to 8x while consuming 6x less power, thereby optimizing performance. This is particularly important for battery-powered sensors, as it extends their operational life and reduces the need for frequent maintenance. In turn, the reduced need for maintenance and replacement reduces overall costs as well in the long run.
Silicon Labs has partnered with market leading machine learning solution provider AIZIP to create an audio-based glass break sensing solution using the machine learning capabilities of EFR32 SoCs. This joint solution accurately detects genuine glass shattering sounds in a given room by making use of AIZIP’s ML applications while running efficiently on low power, enabling many years of battery life. Some of the test results of the performance of this jointly developed solution is given in the following table, showing its market-leading capabilities:
Detection Rate By Distance
| Difficult Glass Breaking Test Sounds (105 dB), Played 5 Times | Commercially Available Glass-Break Sensor | Silicon Labs + AIZIP |
|---|---|---|
| 0.5 m | 3/5 | 5/5 |
| 3 m | 0/5 | 5/5 |
| 5 m | 0/5 | 5/5 |
| 7 m | 0/5 | 5/5 |
False Positive Testing
| False Positive Test Sounds | Commercially Available Glass-Break Sensor | Silicon Labs + AIZIP |
|---|---|---|
| Urban sound (1 hour) | No trigger | No trigger |
| Factory noise | No trigger | No trigger |
| Forging sound | 1 trigger | No trigger |
| Glass clinking | 1 trigger | No trigger |
| Office noise (real-time) 3 hour | 1 tigger | No trigger |
| High impulse noise (like close-distance claps) | 1 trigger | No trigger |
Our reference design hardware is a combination of a EFR32xG28 Explorer Kit (xG28-EK2705A) and a custom designed glass-break sensor board consisting of an analog and digital microphone along with IMU sensors.
The board is powered by a CR-123 battery, providing 3 to 6 years of operation, depending on the usage pattern. The reference design is ready to test and comes with Silicon Labs’ software application, including features like Bluetooth connectivity, adaptive thresholding, audio front end and AIZIP’s ML algorithm.
For more information and availability of our glass break sensor reference design, please contact your sales representative.
