Source-linked AI summary
The Implementation of Low-cost Urban Acoustic Monitoring Devices
Charlie Mydlarz, Justin Salamon, Juan Pablo Bello
TL;DR
Urban noise in NYC is widespread, but comprehensive long-term monitoring is limited by expensive equipment and challenging calibration requirements. The paper develops a low-cost, static acoustic sensor network using consumer hardware and calibrates its MEMS microphone against acoustic standards. Preliminary tests show type 2-level acoustic performance in relevant urban sound conditions, while current IEC certification procedures and low-level noise performance remain boundaries.
Problem
NYC lacks comprehensive validated urban-sound evidence, while conventional monitoring relies on expensive short-term measurements and MEMS microphones face IEC type-rating constraints.
Method
The paper develops low-cost autonomous sensing nodes using consumer computing hardware and a MEMS microphone, then evaluates the microphone against type 2 measurement requirements.
Results
The analog MEMS microphone produces accurate sound-pressure-level data at or above type 2 performance in the undertaken tests, operating within type 2 linearity tolerances above 40dBA on average across 31.5Hz–8kHz.
Takeaways & Limitations
The sensor solution supports prospective low-cost, scalable, long-term urban acoustic monitoring and real-time acoustic data collection in NYC-like environments.
Takeaways & Limitations
Further environmental testing is needed for temperature, humidity, deployment location, and the extended IEC 61672-1 requirements; current noise-floor limits prevent effective operation below 30dBA.
Abstract
from arXiv · showhide
The urban sound environment of New York City (NYC) can be, amongst other things: loud, intrusive, exciting and dynamic. As indicated by the large majority of noise complaints registered with the NYC 311 information/complaints line, the urban sound environment has a profound effect on the quality of life of the city's inhabitants. To monitor and ultimately understand these sonic environments, a process of long-term acoustic measurement and analysis is required. The traditional method of environmental acoustic monitoring utilizes short term measurement periods using expensive equipment, setup and operated by experienced and costly personnel. In this paper a different approach is proposed to this application which implements a smart, low-cost, static, acoustic sensing device based around consumer hardware. These devices can be deployed in numerous and varied urban locations for long periods of time, allowing for the collection of longitudinal urban acoustic data. The varied environmental conditions of urban settings make for a challenge in gathering calibrated sound pressure level data for prospective stakeholders. This paper details the sensors' design, development and potential future applications, with a focus on the calibration of the devices' Microelectromechanical systems (MEMS) microphone in order to generate reliable decibel levels at the type/class 2 level.
1. Introduction
NYC faces substantial noise exposure and lacks comprehensive, validated city-wide understanding of its urban sound environment. The paper addresses this gap with a low-cost, scalable sensing approach while noting constraints in current IEC type certification for MEMS microphones.
- Around 90% of NYC residents are exposed to noise levels exceeding EPA guidelines considered harmful.
- NYC lacks a comprehensive city-wide study providing a validated urban sound model for long-lasting operational or policy interventions.
- The proposed network uses low-cost MEMS-based sensing devices to capture long-term audio and objective acoustic measurements from strategic urban locations.
- NYC lacks resources to systematically monitor noise pollution, enforce mitigation, and validate the effectiveness of noise-control actions.
- Traditional short-term studies and logging-meter deployments provide limited evidence about longer-duration urban noise patterns.
- Under the 2013 IEC 61672-1 specifications, MEMS microphones cannot currently receive a type rating because their internal pre-amplifier prevents all required test procedures.
2. A high quality & scalable acoustic sensor network
Existing acoustic-monitoring networks trade off accuracy, scalability, processing capability, autonomy, and cost. The presented solution targets large-scale deployment by combining consumer computing hardware with low-cost sensing and type 2 acoustic performance.
- 2.1. Category 1 - Dedicated monitoring stations: Dedicated commercial monitoring systems can cost upwards of $10,000USD per unit while providing accurate, reliable, networked acoustic measurement.
- 2.2. Category 2 - Moderately scalable sensor network: Moderately scalable networks typically cost about $600USD per node, with varied accuracy and limitations in processing or measurement reliability.
- 2.3. Category 3 - Low-cost scalable sensor network: Custom low-cost networks use inexpensive, low-power, autonomous nodes priced at approximately $150 per sensor.
- 2.4. What makes a high quality & truly scalable acoustic sensor network?: A scalable high-quality network should combine agency-comparable accuracy, in-situ processing, wireless audio transmission, autonomous operation, and approximately $100USD per-node cost.
- 2.4. What makes a high quality & truly scalable acoustic sensor network?: The presented solution uses components costing less than $100USD and aims to produce type 2 acoustic data while operating autonomously with substantial processing power.
3. Applications
The sensor network is intended to turn continuous acoustic measurements into actionable urban-noise intelligence. Its applications include identifying patterns, guiding inspection resources, evaluating interventions, and supporting high-resolution noise mapping.
- Network data can identify important patterns of noise pollution across urban settings.
- City agencies could use automatically identified offending locations to deploy costly noise inspectors more strategically.
- Continuous monitoring can help validate mitigation effects across time and space and inform future intervention decisions.
- Expanded deployment could enable noise maps with high spatial and temporal resolution and support studies linking sound to health, crime, education, and real-estate values.
4. Summary of contributions
The paper presents a low-cost MEMS-based acoustic sensing solution for smart-city noise monitoring, demonstrating type 2 measurement suitability and supporting scalable, in-situ processing.
- IEC 61672-1 measurements show that an analog MEMS microphone solution is suitable for accurate urban acoustic monitoring at the type 2 level.
- Consumer mini PCs enable advanced signal processing directly within acoustic sensing devices, including automatic sound source classification.
- Low-cost core components support an advanced and scalable acoustic sensing system for smart cities.
- The paper focuses on hardware development and MEMS microphone testing, while omitting the sensor network’s software and networking elements.
5. Hardware
The hardware uses inexpensive consumer computing, audio, MEMS microphone, mounting, and power components to support distributed urban acoustic sensing. Testing addresses frequency response, environmental stability, mechanical effects, and power-supply noise.
- Computing core: The Tronsmart MK908ii costs $50USD and combines a 1.6GHz quad-core processor, 2GB RAM, 8GB storage, USB I/O, and Wi-Fi.Its processing capacity supports on-device digital signal processing, reducing the need to transmit large audio volumes.
- Audio interface: The eForCity USB audio interface costs $5USD, provides one microphone input with adjustable gain, and has a measured noise floor of -90.1dBV(A).Its response is relatively unaffected across the audible range, despite steep roll-offs below 20Hz and above 20kHz.
- MEMS microphone: Temperature changes below 0.017dB/°C and humidity changes below 0.1dB between 40% and 90% RH indicate limited sensitivity variation under tested conditions.These characteristics support consideration of MEMS microphones for long-term monitoring in varying urban environments.
- MEMS microphone: The Knowles SPU0410LR5H-QB MEMS microphone is specified at -38dB re. 1V/Pa sensitivity, 63dBA signal-to-noise ratio, and 120µA current draw.The microphone is quoted as having a flat response from 100Hz to 10kHz and requires a maximum 3.6V supply.
- Mounting: The custom mount protects the microphone port, permits a windshield, avoids closed-cavity Helmholtz resonances, and may introduce diffraction effects above 8.5kHz.PCB dimensions may also affect response above 13.5kHz; these effects were deferred to further testing.
- Power supply: 350mV average peak-to-peak noise from the unregulated PSU fell to 17mV with regulation, while harmonic noise was reduced by up to 26dBu at certain frequencies.The unregulated supply had a fundamental switching peak around 750Hz; an LT1086 regulator was introduced to provide cleaner DC power.
6. Software & network
The prototype software configuration continuously captures high-quality audio, encrypts compressed segments for remote collection, and supports periodic uploads with remote control commands.
- Audio capture: Each sensor node continuously samples 16-bit audio at 44.1kHz and can compress contiguous one-minute segments using lossless FLAC.The compressed files are encrypted with AES, while the AES password is protected using RSA public/private-key encryption.
- Networking: Nodes upload audio at one-minute intervals through an internet-connected Wi-Fi router, which also supports sensor communication and control.Commands include data flushing, rebooting, manual microphone-gain adjustment, and software updates.
7. Signal pre-processing
The device’s MEMS microphone response was characterized and compensated using reference measurements, then calibrated to produce A-weighted SPL output. The setup positioned the device under test beside a calibrated sound-level-meter microphone and applied a 1 kHz, 94 dBA offset.
- Frequency-response characterization: Swept-sine impulse responses from the reference and MEMS microphones were measured after removing room and speaker coloration.A calibrated reference microphone and pre-amplifier were used with a studio speaker to characterize the device under test.
- Frequency-response characterization: The 10 MEMS microphones showed negligible frequency-response differences, indicating strong part-to-part consistency.Observed peaks and troughs from 2–20 kHz were partly attributed to mounting conditions, while the rise above 10 kHz was attributed to Helmholtz resonance.
- Frequency-response compensation: An averaged response was used to design a regularized inverse linear-phase FIR filter for time-domain compensation.The filter used 8192 coefficients and was regularized to avoid extreme attenuation or amplification at frequency extremes.
- Calibration setup: The DUT was mounted beside the calibrated reference SLM microphone, 1.3 m high and 1 m on-axis from the speaker.The microphone capsules were separated by 20 mm, producing negligible (<0.1 dBA) level variation when positions were matched.
- Calibration setup: A-weighted sample values were converted to dBA using a 1 kHz, 94 dBA calibration signal and an offset adjustment.The processing steps for generating calibrated SPL output from the DUT are shown in the sensor’s SLM functionality block diagram.
8. Measurements
The DUT was evaluated against IEC 61672 procedures using a calibrated type 1 SLM as the comparison reference. It met the reported type 2 criteria across frequency weighting, stability, linearity, toneburst response, and urban-audio time-history tests, within stated level boundaries.
- Test framework: The DUT assessment used IEC 61672 procedures and adjusted type 2 tolerances that account for the type 1 reference SLM’s own tolerance bounds.For example, ±2.0 dB type 2 bounds and ±1.0 dB type 1 bounds produce adjusted bounds of ±1.0 dB.
- Self-generated noise: The DUT’s self-generated noise was 29.9 dBA, with a 0.1 dBA standard deviation, while the reference SLM averaged 22.5 dBA.The authors report that the resulting dynamic range was adequate for urban sound environments.
- Frequency response: The DUT met all adjusted type 2 dBA frequency-weighting criteria and differed from the SLM by at most 0.5 dBA for pink and white noise.Measurements used octave-frequency sine waves from 31.5 Hz to 8 kHz and continuous broadband noise signals.
- Long-term stability: 0.07 dBA was the observed beginning-to-end difference during the 30-minute, 94 dBA stability test, within the ±0.2 dBA type 2 tolerance.The mean and standard deviation throughout the test were both reported as <0.1 dBA.
- Level linearity: Above 40 dBA on average, the DUT operated within ±0.6 dB adjusted type 2 linearity tolerances from 31.5 Hz to 8 kHz.The type 2 lower limits were 37.2 dBA for pink noise and 36.6 dBA for white noise.
- Toneburst response: The DUT met all IEC 61672-1 toneburst criteria for 4 kHz signals with durations from 1000 ms to 0.25 ms.These measurements used the standard’s relative type 2 tolerance limits rather than the SLM as a reference.
- Urban-audio time history: R^2 was 0.9723 (p ≤0.0001) between DUT and SLM 15-minute urban-audio time histories, with a mean difference of 0.4 dB and standard deviation of 0.1 dB.The recording included impulsive events such as door closures and banging sounds, and the DUT closely followed the type 1 SLM.
- Urban-audio time history: The MEMS system slightly overestimated rising transient levels and underestimated falling-edge levels relative to the SLM.The authors suggest that the difference may result from the DUT sampling more frequently than the SLM.
9. Future work
Future work extends validation to full-housing IEC requirements, environmental conditions, hardware robustness, calibration form factor, and energy management. The paper also proposes on-device sound-source classification while recognizing the computational challenge of deploying complex models.
- Further measurements: The final prototype will undergo additional testing for directivity, high-level thresholds, environmental variation, housing effects, and long-term outdoor comparison with a type 1 SLM.Planned environmental tests span −20°C to +50°C and 25%–100% relative humidity.
- Hardware development: Future hardware work includes digital MEMS microphones for improved immunity to radio-frequency interference and power-supply noise.The proposed digital solution addresses external influences affecting the analog MEMS board’s output.
- Hardware development: A circular 1/2-inch PCB form factor could enable standard acoustic-calibrator use and simplify calibration across sensor nodes.The authors suggest this could potentially improve calibration accuracy and ease.
- Power management: Battery-powered nodes will be investigated through power-mode cycling and adaptation to periods of low acoustic activity.
- Automatic sound-source identification: On-device sound-source classification could avoid transmitting audio to a centralized server, but model computational complexity motivates research into model compression.The proposed application targets automatic identification of urban sound sources while addressing privacy and deployment constraints.
10. Conclusion
The paper presents a low-cost MEMS sensor network that produced accurate SPL data and met type 2 specifications for the tests undertaken. Its noise floor limits operation below 30 dBA and type 2 accuracy below 40 dBA, but the authors state these levels are uncommon in NYC urban environments.
- Conclusion: The presented low-cost sensor network generated accurate, real-time acoustic data at or above the type 2 level in the preliminary tests.The conclusion attributes this performance to the sensor network and frequency-compensation procedures.
- Conclusion: The system cannot effectively operate below 30 dBA or provide type 2 accuracy below 40 dBA because of its noise floor.The authors state that these levels would rarely occur in NYC’s urban sound environment.
- Conclusion: The authors position reliable low-cost acoustic data as a foundation for cyber-physical noise-monitoring systems and noise-enforcement prioritization.