Self-Contained Earthquake Early Warning System Based on Characteristic Period Computed in the Frequency Domain

Abstract

This study presents the design, implementation, and experimental validation of a self-contained earthquake early warning system (EEWS) based on real-time frequency-domain analysis of ground motion. The proposed system integrates a low-noise triaxial micro-electro-mechanical system (MEMS) accelerometer with a high-performance microcontroller, enabling autonomous seismic event detection without dependence on external communications or centralized infrastructure. The characteristic period of ground motion (τ_c) is estimated using a spectral moment method applied to the first three seconds of vertical acceleration following P-wave arrival. Event triggering is based on a short-term average/long-term average (STA/LTA) algorithm, with alarm logic incorporating both spectral and amplitude thresholds to reduce false positives from low-intensity or distant events. Experimental validation was conducted using a custom-built uniaxial shaking table, replaying 10 real earthquake records (Mw 4.1–7.7) in 20 repeated trials each. Results show high repeatability in τ_c estimation and strong correlation with event magnitude, demonstrating the system’s reliability. The findings confirm that modern embedded platforms can deliver rapid, robust, and cost-effective seismic warning capabilities. The proposed EEW solution is well-suited for deployment in critical infrastructure and resource-limited seismic regions, supporting scalable and decentralized early warning applications.

1. Introduction

Earthquakes remain one of the most devastating natural hazards, responsible for significant loss of life, economic disruption, and long-term infrastructure damage. Unlike weather-related phenomena, earthquakes strike without visible warning, leaving little to no time for preparation or evacuation. Despite decades of research, no scientifically validated method currently exists that can accurately predict the time, location, or magnitude of an impending earthquake [1]. In this context, the most practical strategy available today to mitigate seismic risk is the earthquake early warning system (EEWS), a technology that aims to provide real-time alerts immediately after an earthquake begins, before the most damaging ground shaking arrives.

EEWSs exploit a fundamental property of seismic wave propagation: P-waves (primary waves) travel faster than S-waves (secondary waves) and surface waves. P-waves are typically nondestructive and arrive first, while S-waves and surface waves carry most of the destructive energy. By detecting P-waves as soon as they reach a sensor, it becomes possible to estimate the earthquake’s location and magnitude and issue a warning seconds to tens of seconds before significant shaking occurs [2]. This brief but critical time window enables individuals and automated systems to take protective actions that can dramatically reduce casualties and economic losses [3].

EEWSs have evolved into a powerful tool for risk mitigation and resilience enhancement, particularly in densely populated urban areas. For example, Japan’s Kinkyu Jishin Sokuho system [4] and Mexico’s SASMEX [5] have demonstrated the potential of EEWSs to save lives and prevent infrastructure failures. When properly implemented, EEWSs can trigger safety protocols such as

• stopping elevators at the nearest floor and opening their doors to avoid entrapment;
• slowing or stopping high-speed trains to reduce derailment risk;
• shutting down gas pipelines to prevent fires and explosions;
• halting vehicular access to bridges, tunnels, or hazardous zones;
• activating emergency broadcasting systems or mobile alerts [6].

These applications highlight how even a few seconds of advance notice can make a substantial difference in reducing disaster impacts.

EEWSs are generally divided into two categories, based on their deployment strategy and data-processing architecture:

• Regional EEWSs operate based on a dense network of seismometers deployed over wide geographic areas with known seismic activity. Data from multiple sensors are collected in real time and processed in centralized facilities, where algorithms rapidly compute the epicenter, depth, and magnitude of the earthquake and disseminate alerts to target regions [7]. These systems are highly effective for providing warnings to locations at significant distances from the epicenter but are often costly, infrastructure-dependent, and subject to communication delays.
• On-site EEWSs, in contrast, rely on local sensors placed near the point of interest (e.g., a building, industrial facility, or bridge). These systems detect incoming P-waves and estimate local shaking intensity in real time, without the need for centralized processing or external communication infrastructure [8]. While on-site systems may not provide warnings as early as regional ones (especially for distant events), they offer extremely fast local response—typically within milliseconds—making them ideal for autonomous safety triggers [9].

As global interest in earthquake resilience grows, there is an increasing demand for scalable, cost-effective, and decentralized EEW solutions, particularly in regions without access to centralized seismic networks. Much critical infrastructure—such as hospitals, schools, historical buildings, or rural communities—remains unprotected due to the prohibitive cost and complexity of regional EEWSs. Furthermore, developing countries often lack the resources or infrastructure to maintain dense seismic sensor networks and real-time communication frameworks [10].

Recent technological advances have opened the door for stand-alone EEW devices: compact systems composed of microprocessors and low-cost MEMS (micro-electro-mechanical system) accelerometers that can operate independently, without reliance on cloud services or external data centers. These devices can be installed directly in buildings or assets of interest and configured to trigger localized safety actions such as closing valves, sounding alarms, or initiating shutdown protocols [11,12]. Their low power consumption [13,14], small form factor, and affordability make them ideal candidates for democratizing seismic protection, particularly in vulnerable or under-resourced regions.

Wu and Kanamori [15] were among the first to propose an on-site warning system using local acceleration measurements to issue rapid alerts. Their work laid the foundation for compact devices that could be installed directly in buildings or infrastructure. Later studies explored the use of MEMS accelerometers coupled with microcontrollers, such as Arduino or Raspberry Pi, to create low-cost EEWS nodes capable of triggering alarms or protective actions locally [16,17]. These systems typically include embedded signal-processing algorithms for event detection, classification, and intensity estimation, and have proven effective in experimental settings.

Another growing area of interest is the use of smartphones as seismic sensors. Modern mobile phones are equipped with triaxial accelerometers, GPS, and network connectivity, making them suitable for opportunistic EEWS applications. The MyShake project developed at the University of California, Berkeley, is a notable example, demonstrating that crowdsourced phone data can be used to detect earthquakes and issue warnings in real time [18]. Similar efforts, such as the Earthquake Network project, have implemented early warning logic directly on Android devices, allowing alerts to be issued even in the absence of centralized infrastructure [19]. These mobile-based systems have shown promise for large-scale deployments, particularly in regions where traditional seismic networks are sparse or nonexistent.

While smartphone-based EEWSs offer scalability and ease of distribution, they are also limited by factors such as variable sensor quality, user behavior, and network latency [20]. In contrast, dedicated microcontroller-based EEW devices offer improved control, reliability, and deterministic behavior, making them more suitable for critical infrastructure applications [21,22]. Both approaches contribute to a growing paradigm shift in seismology, from centralized, high-cost networks toward distributed, autonomous, and low-cost warning systems.

This paper presents the development of a self-contained earthquake early warning system, based on a triaxial MEMS accelerometer and a microcontroller platform capable of real-time signal processing. The system is designed to detect the arrival of P-waves, distinguish seismic events from nonseismic vibrations, and trigger predefined actions before the destructive S-waves arrive. Its design emphasizes

• real-time performance, using threshold-based signal analysis;
• portability, allowing deployment in isolated or mobile environments;
• autonomous operation, independent of cloud or internet connectivity;
• energy efficiency, enabling battery or solar-powered operation.

Unlike smartphone accelerometers, which are affected by handling and placement variability leading to nonseismic noise and false triggers, the fixed MEMS sensor used in the developed system is mounted on a stable surface, significantly reducing unrelated vibration noise and improving the signal-to-noise ratio for low-frequency signals, thereby enhancing the reliability of P-wave detection and characteristic period estimation.

By integrating sensing, processing, and response into a single unit, this system contributes to the emerging field of embedded seismology and demonstrates how modern electronics can be leveraged to enhance safety in earthquake-prone areas. The proposed device also has potential applications in smart buildings, critical infrastructure monitoring, and early-stage deployment in regions that lack broader EEWS coverage.

2. Methods and Materials

In an earthquake early warning system (EEWS), the characteristic period τ_c of the P-wave is a key parameter used to estimate the earthquake’s moment magnitude (Mw) within the first few seconds after rupture. Traditionally, this period is calculated in the time domain based on velocity and displacement measurements. However, recent work [23] proposes a novel estimation technique in the frequency domain, which improves robustness against noise and signal variability across seismic stations.

The classical formulation of the characteristic period relies on the energy ratio between the velocity and displacement of the vertical ground motion over the first few seconds (typically 3 s) after P-wave arrival. Let u(t) be the ground displacement and $\dot{u}$(t) the corresponding velocity. The energy ratio r is computed as

$$r={\displaystyle \frac{{\int }_0^{t_0}{\dot{u}}^2\left(t\right)dt}{{\int }_0^{t_0}{u}^2\left(t\right)dt}}$$

(1)

where t₀ = 3 s in most EEWSs. The characteristic period τ_c is then given by

$${\mathsf{\tau }}_{\mathrm{c}}={\displaystyle \frac{2\mathsf{\pi }}{\sqrt{\mathrm{r}}}}$$

(2)

This method is widely accepted in EEWSs due to its simplicity and physical interpretability, and is referenced in studies [24,25]. In [23], the method is reformulated using Parseval’s theorem, which equates the total energy of a signal in the time and frequency domains. Parseval’s identity allows Equation (1) to be expressed in terms of the signal’s frequency content:

$$\mathrm{r}={\displaystyle \frac{4{\mathsf{\pi }}^2{\int }_0^{\mathrm{\infty }}{\mathrm{f}}^2{\left|\widehat{\mathrm{u}}\left(\mathrm{f}\right)\right|}^2\mathrm{d}\mathrm{f}}{{\int }_0^{\mathrm{\infty }}{\left|\widehat{\mathrm{u}}\left(\mathrm{f}\right)\right|}^2\mathrm{d}\mathrm{f}}}=4{\mathsf{\pi }}^2〈{\mathrm{f}}^2〉$$

(3)

where f is the frequency, û(f) is the frequency spectrum of the displacement wave u(t), and $〈{\mathrm{f}}^2〉$ is the average of f² weighted by |û(f)|², hereafter denoted by ${\mathrm{f}}_{\mathrm{c}}^2$.

The characteristic period is defined as

$${\mathsf{\tau }}_{\mathrm{c}}={\displaystyle \frac{2\mathsf{\pi }}{\sqrt{\mathrm{r}}}}={\displaystyle \frac{1}{\sqrt{{\mathrm{f}}_{\mathrm{c}}^2}}}$$

(4)

Although the characteristic period of the P-wave is calculated based on the ratio r, the characteristic period can also be expressed based on the frequency spectrum of the displacement. More precisely, the squared characteristic frequency is the average of the squared frequencies weighted by the square of the amplitudes in the frequency spectrum.

$${\mathrm{f}}_{\mathrm{c}}^2={\displaystyle \frac{{\mathrm{A}}_1^2·{\mathrm{f}}_1^2+{\mathrm{A}}_2^2·{\mathrm{f}}_2^2+\dots +{\mathrm{A}}_{\mathrm{n}}^2·{\mathrm{f}}_{\mathrm{n}}^2}{{\mathrm{A}}_1^2+{\mathrm{A}}_2^2+\dots +{\mathrm{A}}_{\mathrm{n}}^2}}$$

(5)

where A₁…A_n are the amplitudes of the spectral components, and f₁…f_n are the frequencies of the spectral components. These values of the spectral components are obtained by applying a fast Fourier transform (FFT) to the displacement sequence in the time domain.

To apply this method practically, in [23] is developed a step-by-step procedure that enhances spectral resolution and minimizes numerical artifacts:

• Extract the first 3 s of the vertical displacement signal after P-wave arrival;
• Zero pad the signal to increase the FFT resolution and compute the FFT to obtain the frequency-domain representation;
• Identify the local maxima in the amplitude spectrum;
• Calculate the weighted average of squared frequencies over the identified peaks.

  $$f_c^2={\displaystyle \frac{A_1^2·f_1^2+A_2^2·f_2^2+\dots +A_k^2·f_k^2}{A_1^2+A_2^2+\dots +A_k^2}}$$

  (6)

  where A_k and f_k are the amplitude and frequency of the k-th local spectral peak;
• Compute the characteristic period.

The hardware architecture of the proposed embedded seismic early warning module has been meticulously engineered to achieve a balance between high-fidelity ground motion sensing, computational efficiency, and low power consumption. Emphasizing compactness and deployability, the system integrates advanced sensing and processing capabilities into a self-contained unit, suitable for installation in critical infrastructure or remote environments.

The design is centered around two principal components: a high-performance 32-bit ARM Cortex-M4 microcontroller (STM32F466RET6) and a triaxial digital MEMS accelerometer (ADXL355), as rendered in Figure 1. These components were selected based on their demonstrated reliability in embedded systems and precision sensing applications. The STM32F466RET6 has been used in a variety of industrial and real-time control applications due to its high-speed processing and rich peripheral set [26], while the ADXL355 has been validated for structural health monitoring [27,28] and seismic signal acquisition [29], where its low noise and long-term stability are essential. The triaxial MEMS accelerometer ADXL355 was selected to enable orientation-independent deployment, with vertical-axis alignment achieved through software axis remapping, and to exploit its superior noise density, temperature stability, and digital output integrity relative to most uniaxial MEMS devices. Although τ_c computation utilizes a single axis, the triaxial configuration ensures mechanical installation robustness and maintains high signal fidelity in the low-frequency range critical for seismic P-wave detection.

At the core of the system lies the STM32F466RET6, which is optimized for signal-intensive, real-time applications. The microcontroller unit (MCU) operates at a clock frequency of 180 MHz and includes an integrated hardware floating point unit (FPU), which significantly enhances the performance of filtering algorithms and feature extraction routines applied to seismic signals.

Key features include

• 512 KB Flash memory and 128 KB SRAM, enabling storage of real-time data buffers, firmware logic, and logging structures;
• A rich peripheral set, including a Serial Peripheral Interface (SPI), Inter-Integrated Circuit (I²C), Universal Synchronous/Asynchronous Receiver/Transmitter (USART), Universal Serial Bus On-The-Go Full Speed (USB OTG FS), and Controller Area Network (CAN), facilitating flexible integration with both sensors and communication modules;
• Multiple low-power modes, supporting energy-efficient operation for deployments in resource-constrained or solar-powered environments.

To ensure power stability, the MCU is powered through a regulated 3.3 V supply, with localized power decoupling achieved using a constellation of multilayer ceramic capacitors (MLCCs) placed near the VDD and VCAP pins. This configuration minimizes voltage ripple and sustains reliable digital operation, even during high-frequency SPI transactions.

Ground motion detection is handled by the ADXL355, a high-resolution, 3-axis digital MEMS accelerometer from Analog Devices. This sensor is specifically tailored for low-frequency, low-amplitude applications such as seismic monitoring. Its typical noise density of 22.5 μg/√Hz, combined with 20-bit digital resolution, allows it to detect minute accelerations consistent with early-stage P-wave activity.

The ADXL355 offers a programmable full-scale range of ±2 g, ±4 g, and ±8 g, ensuring both sensitivity and resilience across various seismic intensities. It communicates via a high-speed SPI interface, with hardware interrupt capabilities for efficient data synchronization.

Sensor integration details:

• SPI1, the first Serial Peripheral Interface module of the STM32 microcontroller is utilized, with the following pin assignments: NSS: PA4 (Negative Select Slave, a chip-select signal, assigned to pin PA4, which enables communication with ADXL355), SCK: PA5 (Serial Clock—clock signal for synchronizing data transfer, assigned to pin PA5), MISO: PA6 (Master In Slave Out—data line for communication from the sensor (slave) to the microcontroller (master), assigned to pin PA6), MOSI: PA7 (Master Out Slave In—data line for communication from the microcontroller (master) to the sensor (slave), assigned to pin PA7).
• The INT1 (Interrupt 1) output of the ADXL355 accelerometer is connected to PA0 (Port A, Pin 0) on the STM32 microcontroller, which corresponds to the EXTI0 (External Interrupt Line 0). This setup enables an interrupt-driven data acquisition routine, triggered directly by the sensor’s hardware signal, thereby minimizing latency and maximizing system responsiveness.

To ensure clean power delivery to the sensor, the analog and digital supply lines are decoupled using a combination of three 100 nF MLCCs (C1–C3) and a bulk capacitor (e.g., 1–2.2 µF), placed in close proximity to the power input pins of the accelerometer. This power conditioning strategy reduces supply noise and protects the sensor from transient disturbances during switching events.

While the STM32F466RET6 includes an internal RC oscillator, applications requiring precise event timestamping or synchronization across distributed nodes may benefit from an external 8 MHz high-stability crystal oscillator (HSE). This oscillator can be optionally enabled via hardware configuration, allowing trade-offs between timing accuracy and system cost.

A manual reset mechanism is implemented using a push-button switch in conjunction with an RC delay filter (10 kΩ, 100 nF) tied to the NRST (Negative Reset) pin. This guarantees consistent behavior during startup or in response to brownout conditions. A 10 kΩ pull-down resistor on BOOT0 (the boot mode selection pin determining startup source) ensures that the microcontroller enters user Flash mode on boot, while still supporting optional bootloader entry for firmware upgrades via USART or USB.

To facilitate user interaction and debugging, the system includes two general-purpose LEDs:

• LED1 (PA8): Illuminates during valid seismic event detection or data logging;
• LED2 (PA1): Signals error conditions, including sensor faults, communication errors, or memory overflows.

3. Experimental Setup and Results

To evaluate the performance of the embedded EEWS, a series of controlled experiments were conducted using a custom-built uniaxial shaking table (Figure 2). This platform is designed to replicate realistic seismic events by generating programmable displacement profiles derived from both synthetic and real earthquake data.

The mechanical structure comprises a low-friction linear guide rail supporting a precision-machined aluminum cart mounted on a steel baseplate (1200 mm × 800 mm). The cart is constrained to move along a single translational axis to minimize lateral displacement artifacts. Actuation is performed by a high-torque brushed DC servomotor coupled via a belt or lead-screw mechanism to the cart. The actuator incorporates an optical rotary encoder, enabling high-resolution position feedback.

The motion system operates under a closed-loop proportional–integral–derivative (PID) control scheme implemented on a host PC. This configuration allows real-time tracking of the reference trajectory with minimal steady-state error, supporting seismic waveform reproduction in the 0.1–10 Hz frequency range—characteristic of structural vibration modes in civil engineering.

Motion control, data acquisition, and system monitoring are coordinated via a National Instruments (NI) data acquisition (DAQ) system interfaced with a LabVIEW-based graphical user interface. Earthquake input signals are sourced from the Engineering Strong Motion Database (ESM), integrated into displacement time series, and uploaded to the host controller.

The shaking table also includes the following instrumentation:

• Actuator: Brushed DC motor (4160 rpm, 7.67 mNm/A torque constant);
• Gearbox: Single-stage planetary gear with 3.71:1 reduction ratio;
• Encoder: 4096 counts/rev optical encoder mounted on a 56-tooth pinion; resolution ≈ 0.0235 mm;
• Power amplifier: Linear type, delivering ±10 V analog control with a 24 V DC output.

The PID controller continuously adjusts the actuator input based on encoder feedback, ensuring accurate reproduction of seismic waveforms. Motion is unidirectional and monotonic, allowing it to be approximated as a unit-step response system with a measurable time constant and delay.

Ten earthquake records were selected from the ESM database, covering a wide range of magnitudes (Mw 4.1 to 7.7), distances (39–264 km), and spectral characteristics (

Table 1. The average characteristic periods and standard deviations for each tested event.

Date_Event	Magnitude [Mw]	Epicentral Distance [km]	Station	Mean [s] $\overline{{\mathit{\tau }}_{\mathit{c}}}$	Standard Deviation [s] $\mathit{s}\left({\mathit{\tau }}_{\mathit{c}}\right)$
2023.06.21_0000149	4.1	39	LSN	0.29	±0.03
2023.06.06_0000145	4.8	48	TIM	0.54	±0.6
2023.02.14_0000139	5.5	72	LOT	1.09	±0.11
2021.08.01_0000049	5.7	90	KARP	1.18	±0.13
2022.11.23_0000008	6.1	139	IZI	1.23	±0.17
2023.02.20_0000197	6.3	205	2713	1.31	±0.20
2023.02.06_0000011	6.7	255	ARPRA	1.55	±0.21
2020.10.30_0000082	7.0	249	DST	2.15	±0.23
2023.02.06_0000222	7.5	239	KOZK	3.46	±0.34
2023.02.06_0000008	7.7	264	6203	4.25	±0.38

). Each event was replayed on the shaking table 20 times to assess the repeatability of the system’s response. For each trial, the characteristic period (τ_c) was estimated locally by the embedded system using a frequency-domain approach based on spectral moment analysis.

The embedded seismic node—composed of an STM32F4-series microcontroller and an ADXL355 MEMS accelerometer—performed real-time detection and classification of seismic events. The signal-processing chain included

• Event Detection: Based on the short-term average/long-term average (STA/LTA) algorithm: STA window: 2 s; LTA window: 20 s; trigger threshold: 4.0; detrigger threshold: 1.5. The STA/LTA trigger and detrigger thresholds were selected based on prior EEWS studies [15,16] and empirical calibration specific to the ADXL355 MEMS accelerometer’s noise characteristics (noise density ~22.5 µg/√Hz), optimizing the balance between sensitivity to weak P-wave arrivals and suppression of false triggers caused by ambient microvibrations and sensor drift.
• Characteristic Period Estimation:
  • The first 3 s of the vertical acceleration signal were recorded following event onset. This FFT input is sampled at 200 Hz (fully compatible with the hardware configuration), yielding 600 samples. The sampling rate provides a Nyquist frequency of 100 Hz, far above the 0.1–10 Hz band relevant for P-wave analysis, allowing oversampling that improves signal-to-noise ratio before numerical integration.
  • The signal was zero-padded (to extend the sequence to 2048 points, resulting in a frequency resolution of approximately 0.098 Hz (i.e., 200 Hz/2048)) and converted to the frequency domain using FFT.
  • Local spectral maxima were extracted.
  • A weighted average of squared frequencies was computed using peak amplitudes as weights.
  • The characteristic period was derived as

    $${\mathsf{\tau }}_{\mathrm{c}}={\displaystyle \frac{1}{\sqrt{\frac{A_1^2·f_1^2+A_2^2·f_2^2+\dots +A_k^2·f_k^2}{A_1^2+A_2^2+\dots +A_k^2}}}}$$

    (7)

    where A_k and f_k are the the amplitude and frequency of the kth local spectral peak.

For testing, the single-axis shaking table was operated in a horizontal translational mode, while the sensor board was mounted vertically so that the actuator’s motion aligned with the internal vertical (Z) sensing axis of the triaxial MEMS accelerometer (ADXL355). Only the vertical component of each ground motion record from the ESM database was used, converted to displacement and reproduced on the table in accordance with the τ_c estimation methodology above.

4. Discussions

The experimental validation employed a uniaxial shaking table simulating vertical (Z-axis) motion exclusively, consistent with standard EEW practices that utilize the vertical ground motion component to detect P-wave onset and compute the characteristic period (τ_c). This approach is based on the predominance of vertical propagation in P-waves, while horizontal components—dominated by later-arriving S-waves and surface waves—are not used in the initial warning computation. In the present study, the first 3 s of vertical ground motion following the STA/LTA-triggered P-wave arrival were analyzed using the τ_c–Mw method to characterize the spectral content within this window.

The frequency-based method offers improved noise robustness compared to traditional time-domain energy ratios, making it particularly suitable for MEMS-based instrumentation.

Alarm activation was governed by a dual-parameter decision rule:

• The estimated characteristic period τ_c must exceed a predefined threshold (e.g., 0.8 s), indicating a potentially damaging event.
• Simultaneously, the P-wave amplitude must exceed 0.5 cm to ensure that weak or distant earthquakes do not cause false positives.

These alarm activation criteria were established through a combination of prior studies [15,25] and empirical calibration with the experimental setup. This dual-threshold approach effectively filters out low-magnitude or distant events and was validated against ten shaking table ground motion records, consistently distinguishing potentially damaging seismic events from noncritical ones. The hybrid criterion increases the specificity of warnings by incorporating both spectral and amplitude-based discrimination.

Since the system computes the characteristic period (τ_c) using the first 3 s of vertical acceleration following the STA/LTA trigger, which itself activates within 0.5 to 1.5 s after P-wave onset, and executes τ_c estimation and alarm logic in under 0.3 s, the total latency from P-wave arrival to alarm issuance is approximately 4 to 5 s; this delay still allows meaningful early warning, as S-waves typically arrive 10 to 30 s later depending on epicentral distance, providing sufficient lead time for automated safety measures.

During experimental validation, the system was tested across 200 shaking table trials replaying 10 earthquake records, successfully triggering alarms for all events exceeding the dual criteria of τ_c > 0.8 s and peak amplitude > 0.5 cm with no false negatives, while no false positives occurred when these thresholds were not met; this dual-parameter approach benefits from the weak correlation between τ_c and peak amplitude, reducing false alarms by requiring simultaneous exceedance, though real-world deployments are necessary to assess performance under environmental noise and may prompt threshold adjustments or enhanced filtering.

Across the entire test series, the system demonstrated remarkable repeatability. For each of the ten earthquake records, the 20 repeated trials yielded nearly identical characteristic period values, with standard deviations consistently within a narrow range. This level of reproducibility highlights both the mechanical stability of the shaking table and the signal-processing consistency of the embedded system.

Furthermore, the observed correlation between the event magnitude and characteristic period aligns well with the theoretical expectations and previously published literature. Low-magnitude, near-field events resulted in short-period estimates (e.g., <0.5 s), while large-magnitude, distant events produced longer characteristic periods, sometimes exceeding 4 s. This confirms the method’s ability to capture fundamental seismic parameters using compact, low-cost hardware and simplified frequency-domain analytics.

5. Conclusions

The experimental results presented in this study demonstrate the feasibility, consistency, and practical relevance of using a low-cost, embedded system for on-site earthquake early warning (EEW). The implemented spectral method for estimating the characteristic period (τ_c) showed strong agreement with seismic theory and the literature, confirming its potential as a reliable proxy for rapid magnitude assessment.

One of the most significant observations is the high repeatability of the system across multiple trials for the same input signal. The standard deviations of the estimated characteristic periods were consistently low, indicating that the system’s performance is deterministic and resilient to noise, sensor drift, or signal-processing artifacts. This consistency is crucial in early warning applications, where false positives or missed detections can have serious safety and operational consequences.

The hybrid decision logic, which combines both spectral (period-based) and amplitude thresholds, has proven effective in filtering out low-intensity or distant events that are unlikely to cause damage at the monitored location. This approach minimizes the risk of false alarms, thereby increasing the system’s reliability and acceptability in real-world deployments. Such dual-parameter validation is particularly important for autonomous or resource-constrained environments where human verification is not always feasible. Future developments will explore the integration of machine learning techniques for event classification and false alarm reduction, complementing the current threshold-based detection approach.

Moreover, while the system is fully self-contained, integration with wireless communication protocols (e.g., LoRa, LTE-M, or Wi-Fi) could enable event reporting and networked early warning systems across larger geographical areas. Combined deployments of these nodes could support distributed seismic sensing with redundancy and spatial correlation.

Finally, field validation in operational environments remains a critical next step. Installing and monitoring the device in buildings, bridges, or industrial facilities in seismically active regions would provide valuable insights into its long-term stability, environmental resilience, and user acceptance.

In conclusion, the proposed embedded EEW platform represents a robust and cost-effective solution for real-time seismic risk mitigation. Its compact design, autonomous operation and fast local response capabilities make it well-suited for decentralized applications, particularly in regions lacking access to traditional early warning infrastructure. Also, its dual-threshold alarm logic (τ_c and amplitude) helps discriminate between low-magnitude/nearby and high-magnitude/distant events. The findings of this work contribute to the growing field of embedded seismology and open promising avenues for scalable, community-level seismic resilience.