Audio - PlantLeaf

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

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 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

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.

More Information

System architecture

The ASEB

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

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

Experiments

Audio Experiments

After having recorded for 50 hours, 34 of them were analysed for 3 different conditions: empty room, the plant alone and the plant mechanically stressed. The plants we recorded were the Ferrocactus (14 hours) and the Aloe (15 hours).

Using the automatic clicks detector algorithm we developed, we calculated the click rate during each condition. The difference is highly noticeable.

The click rate of a non-stimulated Ferrocactus was 0.71, while when stimulated the rate was 6.7 clicks/hour.
The click rate of a non-stimulated Aloe was 0.37, while when stimulated the rate was 4.57 clicks/hour.

Even though further research and more recordings can be done, these results demonstrate the capability of our system to detect the clicks and already contribute to the existing research, since experiments were never conducted on these plants.

Algorithm

Method for Recording and Analysis

Finding a method to reliably spot ultrasonic clicks generated by plants plays a crucial role in the research, since it guarantees to spot the actual ultrasonic clicks emitted by plants.

Here is the method we used and suggest:

Recording:

1. Place the instrumentation and plant in a silent room;
2. Turn off electrical devices that could create interferences;
3. After starting the recording, take note of the exact time the plant is stressed;
4. Leave the room to reduce any kind of noise and record for about an hour.

Analysis:

5. Open and navigate the file to cut the initial and final parts, since noise can be generated while inside the room;
6. Select a threshold (the one we used was equal to the mean energy of FFTs plus 5 times standard deviation) and use the app's integrated function to export the FFT, iFFT and important values of the peaks detected;
7. Analyse them by confronting to existing data regarding both ultrasonic clicks of plants and possible interferences;
8. Once excluded the interferences, clicks can reliably be spotted.

Once you have enough data, you can use the integrated algorithm to find the ultrasonic clicks in other recordings.

Algorithm