What a Passing Car Really Sounds Like: New Method Could Improve Traffic Noise Maps

A vehicle may pass a roadside microphone in mere seconds, yet how that brief moment is measured can affect its estimated noise level. The study 'Improving Vehicle Sound Power Estimation for Road Traffic Noise Models,' published in npj Sustainable Mobility and Transport, examines this problem by comparing several acoustic measurements and proposing a new way to describe the sound produced across a complete pass-by.

Researchers Antonio Pascale, Claudio Guarnaccia and Margarida C. Coelho found that the proposed method closely matches conventional estimates during steady driving while promising a foundation for studying acceleration, deceleration and other conditions that better resemble everyday traffic.

Traffic noise needs measurements that reflect how vehicles actually move

Road traffic is Europe's largest source of environmental noise, exposing more than 110 million people to day-evening-night levels above 55 dB(A). The European Environment Agency estimates that environmental noise contributes to more than 12,000 premature deaths and nearly 50,000 new cases of ischemic heart disease each year, and millions of people experience chronic annoyance or disturbed sleep. These health effects make reliable traffic-noise models essential for city planning, road design and public-health protection.

Many prediction systems separate the noise produced by a vehicle from the way that sound travels through the surrounding environment. Their usefulness depends heavily on the quality of the original vehicle-emission data. This becomes challenging as traffic fleets gain hybrid and electric vehicles whose acoustic behaviour can differ from that of conventional petrol and diesel cars.

Controlled Pass-By testing is one method used to collect these measurements as a vehicle travels along a test track at a fixed speed, and a precision sound-level meter records the event from 7.5 metres away. The traditional indicator, known as LAFmax, captures the maximum A-weighted sound pressure level during the pass. It is simple and widely used, though a single peak cannot describe the complete acoustic history of a moving vehicle or show how its sound changes before and after reaching the microphone.

Seven vehicles reveal the strengths and weaknesses of three noise metrics

The researchers analysed Controlled Pass-By measurements from seven light-duty vehicles: two diesel cars, two gasoline cars, two hybrids and one LPG vehicle. Tests took place on open, nearly flat sites in Aveiro, Portugal, and Battipaglia, Italy, using calibrated Class 1 sound-level meters. Vehicle speed was recorded through onboard diagnostic systems, while acoustic readings were collected every 0.1 seconds.

Each vehicle completed multiple runs involving different speeds and, for manual cars, several gear combinations. The final analysis covered 30 to 90 kilometres per hour because measurements below 30 kilometres per hour were too easily affected by background noise. Between 24 and 32 valid tests were available for each vehicle.

Three acoustic metrics were compared: LAFmax represented the loudest instant of the event. SEL10dB measured the accumulated sound energy within a window extending from the point where noise rose to 10 dB below the peak until it fell to the same level after the vehicle passed. SELspeed used a speed-dependent window covering a fixed 15-metre section of road, beginning 7.5 metres before the microphone and ending 7.5 metres after it.

The team converted each metric into an estimated sound power level through backward sound propagation, treating the vehicle as a point-like source. Vehicle Noise Specific Power regression was then used to produce emission curves linking sound power with speed. Every fitted model performed strongly, with coefficients of determination above 0.95, while residuals generally remained within ±1 dB(A).

Background noise and short measurement windows can distort the result

SEL10dB usually produced higher sound-power estimates between 30 and roughly 60–70 kilometres per hour. A slower vehicle remains near the microphone for longer, allowing the calculation to accumulate more acoustic energy, yet the extended window also collects more background sound. The acoustically relevant pass-by space commonly stretched across 50 to 60 metres, with quieter vehicles producing longer apparent windows because their sound peaks sat closer to the background level. The loudest car, the LPG vehicle, had a shorter estimated acoustic space of about 39 metres.

At higher speeds, SEL10dB moved closer to LAFmax because the pass-by lasted less time. SELspeed reduced some low-speed background interference by limiting the calculation to a defined section of road. A different problem emerged at 80–90 kilometres per hour: the vehicle could cross the 15-metre window in less than one second, causing sound power to be underestimated by around 3 dB(A).

Differences between LAFmax and SEL10dB also reached about 3 dB(A) at 30–50 kilometres per hour. Hybrid vehicles showed the largest disagreements among metrics because their lower sound levels made them especially vulnerable to background contamination. Their emission curves were generally lower than those of combustion-engine vehicles despite the hybrids having greater power and weight, a pattern linked to assistance from the electric motor and reduced engine load. Across much of the tested range, tyre-road noise dominated the acoustic profile and produced an almost logarithmic relationship between speed and sound.

A 15-metre average offers a path toward more realistic traffic models

The researchers proposed an average sound power level, calculated from instantaneous sound-power estimates taken as each vehicle travelled through the 15-metre reference space. Instead of relying on the single loudest moment or integrating raw sound pressure across a potentially problematic time window, this approach reconstructs the vehicle's sound power at several positions and averages the values.

During steady-speed tests, emission curves from the new average and LAFmax overlapped almost perfectly, with differences below 1 dB(A) for every vehicle. The average typically varied by only 0.5–1 dB(A) within individual pass-bys and reached 2 dB(A) only in rare cases. No meaningful relationship appeared between this variation and the small differences from LAFmax, suggesting that measurement geometry or sound-propagation effects were more likely causes than unstable vehicle speed.

This close agreement confirms that LAFmax remains suitable for constant-speed testing, and the proposed spatial average may become more valuable when a vehicle is accelerating or slowing down because it can preserve acoustic information from several points along the trajectory. It could also provide a safer and more realistic alternative to wide-open-throttle testing, where drivers fully depress the accelerator in a controlled measurement zone despite that manoeuvre poorly representing normal urban driving.

The experiments did not include acceleration or deceleration, which the authors identify as the study's main limitation. The arithmetic averaging method may also need to be replaced by energy-based logarithmic averaging when sound power changes sharply along the route. Future work must test the approach with real dynamic driving, particularly electric and hybrid vehicles operating at low speeds, while examining the combined effects of speed, acceleration, gear selection, engine size and powertrain.

The findings show that no single acoustic metric works best in every situation. Peak measurements are dependable during steady travel, energy-based indicators preserve more of the pass-by but can be distorted by background noise or short windows, and the new spatially averaged method offers a flexible route toward traffic-noise models that better represent the movement heard on real streets.