Every Piston Tells a Story: Designing a Vehicle Noise Simulator

By Carol Thomas Christou

A netted sensor network embedded in a mountain road along the Afghan-Pakistani border detects a rumble. Is it a herd of goats? A rockfall? A Russian-made BTR-60 armored personnel carrier transporting Al-Qaeda operatives? Thanks to MITRE's simulation expertise and innovation, netted sensor networks may one day be able to identify not only the passing of a vehicle, but its exact make—just by the sounds it creates.

Vehicles emit a complex spectrum of sounds that reveal more than you might think: the vehicle's number of cylinders, muffler type, exhaust pipe radius, tire tread length, and frame size. In order to develop a reliable way of detecting and classifying ground vehicles, however, we have to first understand their intricate acoustical properties. During its first year, the MITRE Netted Sensors Initiative set out to design a high-fidelity vehicle noise simulator to achieve this understanding.

Not a Simple Sound Source

There are many sources of noise in any vehicle; an exhaustive simulation of all of them is not required. For identification purposes, the engine cylinder harmonics, muffler resonances and tire noise are the most important. MITRE engineers had to distill the distinguishing elements of these noise sources and then design methods to simulate those elements.

Engine and muffler sounds arise from the motion of the cylinders within the engine block. Since cylinder motion is periodic, engine noise can be recreated through simple (or not so simple) mathematics. Assuming half the cylinders fire in every revolution, we can derive the engine time-series acoustic waveform as a function of the RPM rates and the mechanical properties of the piston. These properties include transient settling times, damping factors, and the relative ping amplitudes. To find the engine noise acoustic spectrum, the time series is Fourier-transformed into the frequency domain, and the firing rates for all the cylinders are time-aligned so they sum coherently.

To recreate pipe noise, MITRE simulation designers again need to dabble in acoustics. Each cylinder in a vehicle engine is connected to the muffler housing through a pipe that can be thought of as an open-ended tube with a particular set of natural frequencies (like an organ pipe). The muffler is an acoustic bandpass-filter that amplifies some frequencies while suppressing others. This creates the characteristic pipe note valued by car aficionados. In our model, pipe length differences are also taken into account. The waveform at the output of each pipe is then expressed as a harmonic series (sums of sine and cosine waves).

Tire noise is a more complicated phenomenon. It depends on the vehicle speed, tire tread pattern, and road conditions. To simulate the noise made by a rotating tire, the waveforms were filtered so that only certain frequencies close to the special tread impact frequencies (harmonics) were kept. This generates the "tread impact pattern." The tire resonance pattern is formed by passing the tread impact waveform through a filter tuned to the range of 800-1200 Hz. A resonance peak emerges only when the tread impact waveform has components within this frequency range. This typically happens at speeds in excess of 30 MPH and accounts for the characteristic "tire hum" of vehicles moving down a highway. The tread impact noise can be heard equally well in all directions, while the tire hum is primarily audible in certain relative positions.

Hearing Is Relative

The quality and intensity of car noise varies depending on the position of the listener. In the simulator, these directional acoustic patterns were modeled for each car part. The engine directivity pattern is modeled by a randomly vibrating plate that simulates the coupling of the engine block to the vehicle frame. It radiates more or less uniformly in the forward direction (-90 to 90 degrees, with 0 degrees straight ahead) where the coupling is strong, and minimally in the backward direction where the coupling is weaker.

The exhaust directivity pattern is pronounced in the backward direction (like a mirror image of the engine pattern). The tire whine directivity pattern is modeled by a vibrating disk (when the tire is viewed from the side). The resulting radiation pattern is uniform in all directions at low frequencies, with the pattern becoming more "elongated" perpendicular to the car axis as the frequency increases. Except at low frequencies, sound waves do not diffract well around corners. This means that if part of the vehicle frame blocks a line-of-sight path from a noise source (like the engine) to the observer, then that component is not heard. In order to account for small diffraction effects at lower frequencies, an empirical model was developed to account for sounds that would otherwise not be heard because of vehicle frame occlusions.

It Depends Where You Are

Sound does not arrive instantaneously from source to receiver. Sound can also reach a receiver along many different paths. As a wave, it can be reflected off surfaces, diffracted through openings, and absorbed by materials along the way. Sound is also attenuated, getting weaker and weaker as it travels away from its source. To capture some of these effects, a reflecting wall was positioned arbitrarily between the car and the sensors, so that in addition to a direct sound path, there was an option for a second reflected path. From the vehicle displacements and sensor positions, the car-sensor time delays and course angles (between the velocity vector and the direct path vector) were calculated using simple geometry. The course angles figure in the estimation of the spatial directivity patterns of the sound.

The direct and reflected path-spreading losses were then computed, and the three waveforms for the engine, muffler, and tire noises combined and adjusted for time delays. The composite waveforms were then multiplied by the spreading factors and a reflectivity loss parameter. Combining the direct and reflected paths, we finally had the received waveforms at the sensor locations. Happily, the final product from the simulator sounded like a car, and with a little tweaking of the parameters, like a truck.

Yes, But What Kind of Car?

The proof of a successful simulation is in its agreement with the real world. Fortunately, the results of the Vehicle Acoustics Simulator agreed with real data that was collected around the MITRE Washington campus last year. Its performance was encouraging, and with some more hard work it will provide the necessary tools to help us build a classifier that will identify a vehicle by size and, eventually, perhaps by its make.

For more information, please contact Carol Thomas Christou using the employee directory.

Solutions That Make a Difference.®
Copyright © 1997-2013, The MITRE Corporation. All rights reserved.
MITRE is a registered trademark of The MITRE Corporation.
Material on this site may be copied and distributed with permission only.

Privacy Policy | Contact Us