A recent study conducted by Tel Aviv University showed that stressed plants emit airborne ultrasonic sounds that can be heard from a distance. Thanks to their discoveries and our experiments we were able to identify features common to acoustic signals.
Frequency range
Typically between 20 kHz and 80 kHz
Most observed peaks occur between 40 kHz and 80 kHz
Some species show dominant spectral peaks around 50–60 kHz
Click duration
Extremely short impulsive events
Typical duration between 0.1 ms and 0.5 ms
Sound pressure level
Approximately 60–70 dB SPL at 10 cm
Can still be detected at distances of 3–5 meters under controlled conditions
Event rate
Healthy plants produce very few clicks
Plants under stress can produce tens of acoustic events per hour
These emissions appear as isolated ultrasonic impulses rather than continuous signals
Biological origin of the signals
Biological origin of the signals
The emissions are thought to originate from xylem cavitation. During water transport, plants pull water upward through the xylem. Under stress (especially drought), increased tension can form and collapse air bubbles, generating microscopic mechanical vibrations. These vibrations propagate through plant tissue and into the air as ultrasonic clicks. While widely supported, the exact link between cavitation and airborne emissions is still under study.
What we know
What we know
What is clear from these studies and our experiments is that
Stressed plants emit more ultrasonic clicks than healthy ones.
Signals are detectable several meters away.
Algorithms can classify plant conditions from the acoustic data.
Our study does have some limitations. Only a few plant species were tested, the experiments were conducted under semi-silent rooms and more data still needs to be collected. However, the results are promising and show that our instrumentation can effectively capture and analyze plant bioacoustic signals.
System overview
System overview
The ASEB board is the analog interface between the ultrasonic sensor and the digital acquisition system.
Main objectives
Low-noise amplification of ultrasonic signals
Adjustable gain for different signal amplitudes
Bandwidth limitation to reduce noise
Stable analog output for ADC acquisition
Compact and low-power design
The architecture includes a two-stage analog processing chain, followed by digital acquisition through an STM32F411 microcontroller.
Hardware architecture
Ultrasonic Microphone
The ASEB uses the SPU0410LR5H-QB MEMS microphone by Knowles as the primary sensing element. The sensor converts acoustic pressure variations into an analog electrical signal, allowing the detection of ultrasonic emissions potentially produced by plants.
The ASEB analog front-end is built around three main components which provide low-noise amplification, configurable gain, and signal conditioning before the signal reaches the digital acquisition stage.
Low-Noise Amplification — OPA1652
The first stage of the ASEB uses the OPA1652 low-noise operational amplifier.
Very low noise density
High bandwidth
Excellent performance for audio and ultrasonic signals
Low distortion
Configurable Gain — ADG704
The ASEB includes an ADG704 analog switch to allow dynamic selection of different gain configurations.
It can connect different resistor networks into the amplifier feedback path.
Advantages of this approach:
Adjustable sensitivity
Improved measurement flexibility
Optimized signal-to-noise ratio
The switches are controlled digitally by the STM32 microcontroller.
Analog Filtering
To improve signal quality, the ASEB uses RC low-pass filters.
These filters are implemented using resistor–capacitor networks that limit the signal bandwidth and suppress unwanted noise.
Filtering helps remove:
High-frequency electronic noise
Switching artifacts
Environmental electromagnetic interference
Digital Acquisition & Firmware
Digital Acquisition & Firmware
The analog signal is acquired and digitized by the STM32F411CEU6, which provides:
12-bit ADC conversion
High-speed sampling
USB CDC communication
Real-time data streaming
The digitized data is then transmitted to the PlantLeaf Desktop Application, where ultrasonic activity can be visualized and analyzed.
From an applied perspective, this work aims to lay the foundations for a non-invasive, low-cost system capable of monitoring plant stress in real-world conditions. Such a tool could find applications in plant research, education, and potentially in precision agriculture, where early detection of stress could support more sustainable and targeted interventions. The emphasis on accessibility and scalability remains central, with the goal of making advanced plant monitoring techniques available beyond specialized laboratories, and all of this by starting from our own vegetable garden.
Main characteristics
Type: Analog MEMS microphone
Frequency response: up to ~100 kHz
Output: analog voltage
Package: Surface-mount (SMD)
Advantages of MEMS microphones
Very small size
High manufacturing consistency
Low cost
Good sensitivity and stability
SPU0410LR5H-QB
The ASEB uses the SPU0410LR5H-QB MEMS microphone by Knowles as the primary sensing element. The sensor converts acoustic pressure variations into an analog electrical signal, allowing the detection of ultrasonic emissions potentially produced by plants.
Experiments
Audio Experiments
We have now accumulated roughly 160 hours of recordings across more than 150 sessions, grouped into three categories: ambient noise only, unstressed plants and mechanically stressed plants. We progressively moved from silent controlled rooms to noisier indoor environments and outdoor settings, annotating every session with the exact stress timing and any contaminating noise sources.
Click rates are computed with our automatic click-detection algorithm (v5, SVM-based). The contrast between resting and post-stimulus activity is striking.
Against an indoor baseline of 0.45 clicks/hour, Aloe vera rose to 16.0 clicks/hour in the 15 minutes after mechanical stress — about a 35× increase — while resting at 0.47 clicks/hour.
Ferocactus reached 40.7 clicks/hour post-stimulus (roughly 90× baseline), resting at 1.40 clicks/hour. Emissions concentrate in the first ~2 minutes after each stimulus, and rates stay elevated even after watering.
Further exploratory tests — including carnivorous plants feeding on insects (Dionaea muscipula at 12.41 clicks/hour) — are ongoing, but these results already demonstrate our system's ability to detect stress-related ultrasonic clicks and contribute new data on species never studied before.
Examples of confirmed clicks detected by the app — each shows the FFT spectrum (left) and the inverse-FFT reconstruction with its exponential-decay fit (right), together with the SVM confidence assigned by the v5 algorithm.
Aloe vera — mechanical stress (SVM confidence 0.74)Ferocactus — mechanical stress (SVM confidence 0.84)Aloe vera — after watering (SVM confidence 0.90)
Method for Recording and Analysis
Finding a method to reliably spot ultrasonic clicks generated by plants played a crucial role. We collected three categories of recordings: only ambient noise, unstressed plants and stressed plants. Starting from silent, controlled rooms, we progressively shifted toward noisier indoor environments and eventually outdoor settings. In total, we accumulated approximately 160 hours of recordings across over 150 sessions, each annotated with the exact timing of stress events and any noise sources that could have contaminated the recording. Careful analysis initially consisted of comparing potential clicks with those of Khait et al. and examining their most consistent characteristics: broadband FFT spectra and iFFTs resembling damped sine waves lasting 0.1–0.5 ms. As confirmed clicks accumulated, we began extracting physical features to describe them quantitatively. This process, eventually, led to the development of the automatic Click Detection Algorithm.
