With the advancement of digital signal processing technologies, consumers are more concerned with the quality of multimedia entertainment in automobiles. In order to meet this demand, an audio enhancement system is needed to improve bass reproduction and cancel engine noise in the cabins. This paper presents an integrated active noise control system that is based on frequency-sampling filters to track and extract the bass information from the audio signal, and a multifrequency active noise equalizer to tune the low-frequency engine harmonics to enhance the bass reproduction. In the noise cancellation mode, a maximum of 3 dB bass enhancement can be achieved with significant noise suppression, while higher bass enhancement can be achieved in the bass enhance mode. The results show that the proposed system is effective for solving both the bass audio reproduction and the noise control problems in automobile cabins.
1. Introduction
Noise control and the
high-quality bass reproduction in automobile cabins are two interrelated
problems. The later can be difficult due to the high-level noise present and
the size of the loudspeakers that can be installed inside the cars. Traditional
passive noise control techniques are only efficient at high frequencies. For
the low-frequency engine noises, passive techniques become costly and bulky,
which are not suitable for the use in automobile cabins. Due to its
effectiveness in reducing low-frequency noise, the active noise control (ANC)
[1] technique has received much attention since 1980s [2, 3].
On the other hand,
with the advancement of multimedia digital signal processing (DSP)
technologies, high-quality audio reproduction is becoming possible for the
automobiles. However, there are many challenges in reproducing high-quality
bass in cars due to the limited space and acoustic properties, and the
low-frequency noise present in the cabins.
The ANC techniques generally produce good performance in canceling the
narrowband engine noise. However, it
does not offer complete control over the engine noise in cabins. In some
practical applications, it prefers to enhance some preselected noise components
to extract important sound information. For example, the driver may want to
know how the engine is working when driving. Due to its flexibility of
amplifying or attenuating noises with predetermined levels at certain
frequencies, active noise equalizer (ANE) [4] systems and other similar algorithms [5–7] have
potential applications.
High-quality audio reproduction in cabins can be
difficult due to the engine noise and low-frequency performance of the
loudspeakers. With the flexibility of ANE system, we propose a novel method to
solve this problem. Instead of trying to cancel the engine noise
entirely, the proposed integrated system equalizes
the engine-noise harmonics based on the bass information to enhance the low-
frequency
part of audio signal. The main challenges are to track the frequencies of engine harmonics
and to tune these harmonics to match the bass components
of audio signal.In
order to integrate active noise control with bass enhancement, the proposed
system uses frequency-sampling filter
(FSF) [8] and multifrequency ANE [4] to tune the engine harmonics, and
convert the annoying low-frequency noise into desired audio bass components.
The remainder of this paper is
structured as follows. Section 2
presents the narrowband ANE system, followed by a description of the proposed
system in Section 3.
Simulation results under different driving conditions are given in Section 4, and Section 5 concludes this paper.
2. Narrowband Active Noise Equalizer
The single-frequency narrowband
ANE [4] system is based on an adaptive notch filter using the filtered-X least
mean square (FXLMS) [1] algorithm. As shown in Figure 1, the secondary output
is split into two branches: the canceling branch and the balancing branch. A
pseudoerror is used to trick the adaptive filter to converge to a desirable state determined
by the user. The pseudoerror can be expressed as
After convergence, the pseudoerror approaches zero. However, the actual residual noise converges to
where is known as the gain factor determined by the
user.
Figure 1: Block diagram of single-frequency ANE system.
Depending
on the gain factor ,
ANE can be classified into four operation modes [4]:
(i)
cancellation
mode (): ANE
functions as the conventional narrowband ANC;
(ii)
attenuation mode (): the amount of attenuation is
determined by .
Therefore, it is possible to retain some portion of the noise at the selected
frequency;
(iii)
neutral
mode (): the noise passes through the ANE
system without attenuation;
(iv)
enhancement
mode (): the ANE functions as an amplifier
that enhances the noise component with amount determined by .
3. Proposed System in Automobile Cabins
A proposed system in car cabins that integrates bass enhancement and
active noise equalizer is shown in Figure 2. This system can be divided into
three subsystems: (i) the “bass extraction” block extracts bass components from
the car audio system based on the engine speed; (ii) the “postprocessing” block
processes; these bass components to match with frequencies of engine harmonics;
and (iii) the “multifrequency ANE” block implements a multifrequency ANE that
enhances desired low-frequency audio components using equalized engine
harmonics. A detailed overview of these subsystems is described as follows.
Figure 2: System block diagram inside the automobile cabin.
3.1. Bass Extraction
The audio signal components that will be
enhanced are those close to the engine-noise components, which are related to
the engine revolutions per minute (RPM). Because the engine RPM is time
varying, the engine-noise components change accordingly, thus the filters must
self-configure according to the engine RPM to extract the desired audio signal components.
In other words, the filter’s center frequency should be tuned by the engine
RPM.
As shown in Figure 3,
the audio signal is passed through a low pass filter with a cutoff frequency at
500 Hz, and the audio signal is
decimated to a lower sampling frequency of 1.5 kHz. Therefore, a lower computational load is achieved for
processing bass information of the audio signal.
Figure 3: Audio signal extraction
block diagram.
To utilize engine
noise for enhancing bass reproduction, extraction of the audio signal at
frequencies of engine harmonics is needed. This requires a bank of passband
filters align with predominant engine harmonics. Fast online reconfiguration
and computational efficiency are important considerations for designing the
filter bank. The FSF is chosen to meet these requirements. It is based on
sampling a desired amplitude spectrum to obtain the corresponding filter
coefficients. The number of FSF channels equals to the number of predominant
engine-noise harmonics, where each
channel corresponds to one engine harmonic. As shown in Figure 4, the unique characteristic of the FSF structure
allows recursive implementation of finite-impulse response filters, leading to
both computational efficiency and fast online reconfiguration. The transfer function of the FSF is
expressed as
where is the filter length, is frequency sample
value at channel , and is a radius of pole that is slightly less than
unity. Equation (3) shows that the FSF has parallel bandpass filters with center frequencies at , where . Therefore, the
parameter controls center
frequencies of all bandpass filters. The following sections further describe
how to design an FSF for a particular engine.
Figure 4: Frequency-sampling filter
block diagram.
3.1.1. Engine
RPM and the Fundamental Frequency of Engine Noise
This section investigates the fundamental and firing frequencies of a
4-stroke engine. A sampling frequency of 1.5 kHz is selected for the FSF
processing block. This sampling frequency restricts the range of engine noise
to 600 Hz. For a 4-stroke engine, the fundamental frequency is the product of
the firing frequency and number of the cylinders, where the firing frequency is
The fundamental frequency of engine
noise is the fourth harmonic of the firing frequency. Depending on the engine
noise profile, the harmonics selected can be different. When higher frequency
harmonics are selected, this range will be lowered accordingly. For most cars
and with the objective of bass enhancement, the sampling frequency of
1.5 kHz
is reasonable.
3.1.2. Parametric Factor
There are two methods in determining the main parameters to control the
filtering and center frequencies of FSF. One is to set the filter length as a constant value and change each of
the frequency sample values . However, this approach requires
changing multiple sample values during online filter reconfiguration. On the
other hand, if we first set the relative frequency samples at certain values,
it is possible to achieve the reconfiguration by changing only the FSF filter
length . For example, when we set
the filter at to
coincide with the fundamental frequency of noise, the filter length can be
derived as
When the RPM is 2500, the corresponding
filter length is 180. It is also important to point out that the FSF does not
incur a higher computational load when the filter length increases. This is
because most frequency samples are zero and only few frequency samples
defined in the passband require computation.
3.1.3. Transition Band Sample Value
Rabiner et al. proposed some typical values for the coefficients in the
transition band [9]. In the case of designing the FSF for handling typical RPM
from 1000 to 2500, the filter length ranges from 180 to 450. If three samples
are used to define the frequency samples in the passband, the optimum value for
transition band is found to be 0.4 [10] The illustration is shown in Figure 5.
Figure 5: Diagram of FSF filter setting for fundamental engine noise
frequency.
3.1.4. Selecting Suitable Filter Length/Frequency Resolution
As the
sampling frequency is 1500 Hz, the frequency resolution for FSF
is . According to the relationship:
where is the sample index that is selected to align
at the engine noise frequency. As a result, index controls the resolution of the filter.
Therefore, the optimal resolution is determined by the frequency range of the
engine noise. Offline calibration is required for different engines to select
the proper value of ,
which is set to the center frequency of fundamental engine noise, and
correspondingly determine the frequency resolution.
3.2. Postprocessing
The signal power estimation is performed before sending to postprocessing
block. The process can be expressed as
where is the signal power, is the current sample, and is known as the smoothing parameter or
forgetting factor, typically set between 0.9 to 0.999. There are many options
for the postprocessing block. Users can perform different kinds of equalization. This paper proposes two schemes. The bass enhancement scheme is
designed for higher amplification of equalized engine noise, and the noise
cancellation scheme is designed for more engine noise reduction.
3.2.1. Bass
Enhancement Scheme
The bass enhancement scheme emphasizes on
the enhancement of bass components in the audio signal. Using the power
estimation results obtained
from previous block, the gain factors , in the ANE systems can be the calculated as
where is the power of the FSF’s output that
corresponding to the engine harmonic frequency, and is a constant that controls the volume of
the sound in order to mix the tuned engine noise with the original audio
output. Users can tune to different levels of bass enhancement. The variable is the number of predominant engine noise
harmonics which is dependent on the
particular engine type. If the in cabin loudspeakers are incapable in reproducing the signal at
engine noise fundamental frequency, the perception of bass can still be
enhanced by other harmonics due to the famous “missing fundamental” phenomenon.
In order to set the value
of that determines ,
it is important to derive the relationship between the sound pressure level of
the audio signal and engine noise. In typical audio system, the sound pressure
level ranges from 50 dB to 80 dB. On the other hand, the engine noise level in a
cabin ranges from 45 dB to 75 dB [9]. For a 16-bit audio signal, which is
normalized to unit, the sound pressure level is stated as
This equation sets the maximum sound pressure level to 96 dB when the amplitude of equals to 1.
To calibrate
the value of factor ,
it is assumed that if the signal is 60 dB, the engine noise should be neither
amplified nor attenuated. According to (9) and setting to 60 dB, the amplitude of the signal is computed
as
The power of the signal is
approximately .
Setting to 1 results in .
3.2.2. Noise
Cancellation Scheme
It can be
seen from the previous scheme that by tuning the factor ,
higher enhancement at the low frequency can be achieved. However, at the same
time, the timbre of the original signal will also change. To fulfill the needs
of enhancing bass reproduction while maintaining a balanced timbre with
significant noise cancellation, we propose another scheme known as the noise
cancellation scheme.
In this
scheme, when engine noise is louder than the audio signal, a proper equalized
engine noise is used to enhance the audio signal. In order to maintain a better
timber, this scheme does not allow any amplification of the engine noise, or
the gain factors for engine noise harmonics should be always smaller than one. The
rationale behind this scheme is to make the amplitude of the engine harmonics equals
to the corresponding amplitude of the audio signal at that frequency. In this
way, when there is audio signal present at the engine noise harmonics, the ANE system
amplifies the amplitude of the engine noise to produce a 3 dB enhancement of
audio signal.
When the
engine noise is lower than the audio signal, we keep or cancel the engine noise
harmonics depending on whether the audio signal is present or not. As a result,
the gain factor for the ANE system is either one or zero. The maximum gain of
3 dB is achieved when the engine noise level equals the audio signal level.
Therefore, to achieve the desired gain adjustment in Section 2, a new gain
scheme is proposed as follows:
where is the sound pressure level of audio at the corresponding
engine noise harmonic frequency, is the sound pressure level of the engine
noise harmonic, is used as a threshold and is set to 45 dB,
and is a constant governing the equalization
between the gain factor and difference between the sound pressure level of
audio signal and engine noise.
To equalize
the engine noise when ,
the gain factor is chosen such that
where is the amplitude of the engine noise and is the amplitude of the audio signal.
Substituting (9) and (11) into (12), we have
Taking logarithm of both sides, we obtain
This results
in
and
According to this gain factor scheme
under a loud engine noise condition, it is expected to achieve both reduction
of engine noise and a 3 dB bass enhancement at certain frequencies.
3.3. Multifrequency ANE System
To perform the active
control of the engine noise, we designed a multifrequency ANE system consisting
of several independent single-frequency
ANE systems connected in parallel. Each single-frequency ANE is tuned to the corresponding harmonic frequency of the
engine noise. The overall block diagram of the
multichannel ANE is shown in Figure 6. The number of the
single-frequency ANE system is determined by the number of the selected
predominant engine noise harmonics. Each ANE block has its own gain factor
tuned to the power of the related audio component. When the audio signal is
changing with time, the equalization of the low-frequency signal responds
accordingly.
Figure 6: System block diagram of the
multifrequency ANE.
4. Simulation Results
Performance of the proposed system is evaluated by both a synthesized
engine noise and a recorded in cabin engine noise (Toyota Crown at passenger
seat with the engine running at around 2600 RPM). The reference signal is
generated using cosine wave with the center frequency at the corresponding
engine noise harmonic. Kim and Park showed in [11] that the self-generated
reference could achieve good performance in ANC applications. Figures 7 and 8 show the spectrogram and power distribution of the engine noise, respectively. For this recorded engine noise, we select two
predominant frequency components and an FSF is used to extract the bass audio
information.
Figure 7: Spectrogram of the
recorded
engine noise.
Figure 8: Power distribution of the
recorded
engine noise.
The audio signal used for the simulation is “Hotel California”
by The Eagles (live version). The
sound clip was taken from the start of the track, which consists of a
bass drum with some audience
noise. This track makes it easier to focus on the bass. The sound clip and
simulation results wave files are available at [12].
4.1. Bass Enhancement Scheme
The results shown in Figures 9
and 10 are the spectrograms that show bass components of audio signal
before and after the process, respectively. The predominant engine noise
harmonics are attenuated (marked as circles in diagrams) when the audio is
absent, and tuned according to the gain factor shown in Figure 11, when the audio is present.
Figure 9: Spectrogram of the sound in cabin when system off.
Figure 10: Spectrogram of the sound in cabin when system on.
Figure 11: Gain factor for fundamental frequency.
To display the tuned
engine noise more clearly, the spectrogram of the tuned engine noise is shown
in Figure 12. It is observed
that the tuned engine noise has a similar spectrogram distribution as the audio signal.
Figure 12: Spectrogram of the tuned
noise.
The proposed
system is also evaluated using synthesized engine noise to test the effectiveness
at defined harmonics. In the following simulation, the synthesized engine is
running at 3000 RPM, with its predominant harmonic frequencies at 100, 200, 300,
and 400 Hz. As seen from Figure 13, the engine noise components at 100, 300,
and 400 Hz are attenuated by 5, 8, and 15 dB. However, a 3 dB enhancement is
achieved at 200 Hz. The equalized engine noise is equalized to enhance the bass
component of the audio signal. The gain factor value for the 200 Hz harmonic
over the duration of simulation is shown in Figure 14.
Figure 13: Bass enhancement scheme with synthesized engine noise.
Figure 14: Gain factor (at 200 Hz) in bass enhancement scheme.
4.2. Noise Cancellation Scheme
In this
simulation, we investigate the performance of the proposed system under noise
cancellation scheme. The system is tested with the recorded engine noise
(running at 2600 RPM) and with SPL of 75 dB. The spectrogram of this engine
noise is similar with those under bass enhancement mode.
The tested
audio file is extracted from a short speech clip. We simulate the case when the
driver is listening to news or making a phone call. The system adapts to cancel
the engine noise to achieve a better SNR for speech perception in the car
cabin. Engine noise before and after processing is shown in Figure 15. It can
be clearly observed that the most prominent engine noise harmonics are reduced
by 6 dB. Gain factor for the fundamental frequency over the period of
simulation is shown in Figure 16.
Figure 15: Engine noise before and
after processing.
Figure 16: Gain factor using noise cancellation scheme.
Similar to
the bass enhancement scheme, we evaluate the system using audio signal and the
synthesized engine noise. As seen from Figure 17, the engine noise components
are significantly reduced, especially at 400 Hz since there is very little
audio component. The gain factor value for 200 Hz harmonic over the duration of
simulation is shown in Figure 18. Compared with the result obtained in bass
enhancement scheme, it clearly shows that the gain factor value is confined in
the range of 0 to 1, and engine noise is never been amplified.
Figure 17: Noise cancellation scheme with synthesized engine noise.
Figure 18: Gain factor (at 200 Hz) in noise cancellation scheme.
5. Conclusion
Instead of canceling the engine noise entirely, this paper presented a
system that utilizes the engine noise to enhance the bass reproduction of audio
signal in automobile cabins. The proposed system integrated bass extraction,
audio signal processing, and active noise equalization to enhance desired bass
signal and reduce noise. Several engine noises and audio signals are used to evaluate
the performance of integrated audio and active noise equalization system.
Simulation results showed that the proposed system can achieve audio bass
reproduction and noise reduction inside the car cabins.