An Active Indoor Noise Control System Based on CS Algorithm

Abstract

Environmental noise causes enormous harm to human health. This research suggests an active noise control (ANC) system based on the cuckoo search (CS) algorithm to reduce indoor noise pollution. The indoor environment’s ANC is more complicated than the conventional linear ANC system used in cars, aircraft, and other confined spaces. The suggested system architecture considers noise from many directions in order to reduce the impact of external noise on persons within a given environment. The effective search strategy of the CS algorithm enhances the approach of updating filter coefficients based on the LMS/NLMS algorithm in the conventional ANC systems. The results from the simulations demonstrate that the suggested strategy successfully validates the hypothesis and provides significant noise reduction. Additionally, we developed the system’s hardware, which is based on a digital signal processor (DSP). The experimental results show that the proposed technology could perform well with respect to ANC.

1. Introduction

Human health is substantially impacted by noise. Working in a noisy setting for an extended period of time will harm human hearing [1,2], and the noise in the work environment can be called occupational noise. Non-occupational noise includes noise from house repairs, public transportation, and electrical equipment. Noise levels below 85 dB are typically regarded as safe for the general population [3]; however, a prolonged exposure to 85 decibels can result in hearing damage. The WHO recommends a daytime noise level of no more than 40 dB [4], although the most recent research indicates that indoor ambient noise levels of no more than 49.6 dB [5] are generally considered to be acceptable for human health. As a result, noise management is critical in daily life.

Acoustic noise may be controlled in two ways: passively [6] or actively [7]. The classic method of acoustic noise reduction employs passive measures such as enclosures, barriers, and silencers to reduce unwanted noise. Passive silencers employ either the notion of impedance change generated by a combination of baffles and tubes (reactive silencers), or the concept of energy loss caused by sound propagation in a duct lined with sound-absorbing material (resistive silencers) to achieve quietness. In order to address these issues, there has been a great deal of interest in active noise control (ANC) [8]. An electroacoustic device in the ANC system suppresses the undesirable sounds by creating an anti-sound (anti-noise) of equal volume and with an opposite phase. Acoustically, the original, undesirable sound and the anti-noise merge, resulting in the cancellation of both sounds. The efficacy of primary noise cancellation is determined by the correctness of the amplitude and phase of the produced anti-noise. The efficacy of active control in comparison to passive attenuation approaches is used to assess its success. Active attenuation is an appealing way to provide significant noise reduction in a tiny device, especially at low frequencies (below 600 Hz). Active control has significant advantages at low frequencies when lower sampling rates are appropriate and only plane wave propagation is permitted.

Due to the maturation of adaptive-filtering theory and digital signal processors in the 1980s, the development of ANC was rapid. Elliott and Nelson [9], George and Panda [10], and the references therein all contributed to the development of ANC. The filtered-x least-mean-square (FxLMS) method [11] is the most widely used adaptive algorithm in ANC. Since the FxLMS method has a basic structure, it has been widely investigated and expanded. Moreover, the FxLMS method is widely used because of its cheap overhead processing and superior performance than the Volterra filtered-x LMS (VFXLMS) technique [12]. Improved nonlinear ANC algorithms based on the VFXLMS and FSLMS algorithms have recently been developed [13]. The researchers of all these publications assess the performance of their algorithms in the presence of nonlinear main routes and chaotic noise. Many naturally energized algorithms exhibit certain swarm intelligence traits, and in applied research, nature/bit-inspired approaches have been widely exploited for the optimization of many difficult problems. Optimization algorithms must be used to determine the most accurate and optimal solution to a problem when numerical approaches fail. As a result, when the noise level is raised, the FxLMS/FxNLMS algorithm fails to work. To mitigate the impact of nonlinearity, a nonlinear-based ANC method is presented, in which the Cuckoo search (CS) algorithm [14] is used to control the parameters of a filter bank structure. For multimodal objective functions, CS was shown to outperform particle swarm optimization (PSO) [15] and genetic algorithms (GA) [16]. This was due in part to the fact that CS has fewer parameters to fine-tune than PSO and GA. As a result, when compared to other metaheuristic algorithms, CS is more flexible and resilient for many optimization tasks. To summarize, this study applies the CS approach to an ANC system. Unlike other typical gradient-based methods, the suggested strategy does not require any secondary path estimate-filtering calculation and so delivers a computational advantage.

Many people use headphones with ANC technology at present [17]. ANC may minimize engine noise in high-speed automobiles [18], cars [19], planes [20], and other vehicles. However, there have been few studies on noise reduction in an interior scenario. The causes of indoor ambient noise are discussed in this study, and an ANC system for an indoor application is presented. According to the simulation, it performs better than the suggested system in terms of saturation in tone signals, multi-tone signals, and real noise signals. l. Finally, we designed the hardware of the system, and carried out experiments to verify the results. The following are some highlights of the study’s contributions to innovation:

1. An ANC system is supposed for use indoors, taking outdoor interference noise into consideration;

2. A CS-based animal dynamic method is developed for nonlinear adaptive control systems with accuracy, robustness, and stability;

3. The adaptive control method proposed in this study, based on the CS algorithm, successfully simulates sinusoidal, random, and complex random noise interference in various situations of linear/nonlinear and primary/secondary routes, demonstrating the scheme’s worth.

2. The Indoor ANC System Based on CS Algorithm

2.1. System Framework

Feedback ANC is now a well-known technique [21], and it was a reasonably simple feedforward control system for a long, narrow duct. Before passing via a loudspeaker, an input microphone near the noise source detects a reference signal x(n). The noise canceler generates a signal y(n) with the same amplitude but that is 180° out of phase using the reference input signal. This anti-noise signal is utilized to drive the loudspeaker, which produces a canceling sound that reduces the duct’s primary acoustic noise. The noise in [21] is not complicated, and the distance between the source and canceling speaker is short; thus, its implementation is not complex. However, when ANC technology is applied to indoor environments, it is a bit more complicated.

The indoor ANC system design proposed in this paper can be seen in Figure 1. Outside noise can be produced by a variety of sources, including air conditioner noise, construction noise, traffic noise, and the voices of passersby. These noises can be unpredictable in addition to being periodic. In order to identify outside noise sources, the system discussed in this study incorporates input microphones placed at each window. The noise canceler makes use of the incoming signal. The noise canceler generates a signal y(n) with the same amplitude but that is 180° out of phase with the input signal x(n). This noise is sent to a loudspeaker to cancel out the undesirable noise. The error microphone detects the error (or residual) signal e(n), which is then utilized to optimize the filter coefficients. Since the error signal is not compared to the reference input, using a downstream error signal to alter the adaptive filter coefficients does not constitute feedback. Actual implementations need additional considerations with respect to the ability to engage with acoustic issues in the duct. In the suggested system, we utilize a house with four windows as an example, installing four input microphones and error microphones in each office position near the human ear to improve the noise reduction performance as much as possible. When the primary noise is periodic (or nearly so) and produced by rotating or reciprocating machinery, the input microphone can be substituted with a non-acoustic sensor such as a tachometer, accelerometer, or optical sensor. This alternative solves the acoustic feedback problem. However, in the proposed system, such sensors are not as versatile; thus, they are not employed.

2.2. The Preknowledge of ANC Algorithm

Broadband ANC may be described using mathematical model notion. For linear ANC, the filtered-X LMS (FXLMS) method [13] has a minimal computing cost, and it serves as the foundation for many additional algorithms. It is compatible with feedforward, feedback, and hybrid ANC systems. Figure 2 depicts the block diagram of the FxLMS algorithm, where the output y(n) is computed as follows:

$$y(n)=w^T(n)\cdot x(n)={\displaystyle \sum _{i=0}^{N-1}{w}_i(n)\cdot x(n-i)},$$

(1)

where $w^T(n)={\left[w_0(n),w_1(n),\cdots ,w_{N-1}(n)\right]}^{\mathrm{T}}$ is the coefficient vector of $\mathrm{W}(z)=-\mathrm{P}(z)/\mathrm{H}(z)$ at time n; $\mathrm{P}(z)$ and $\mathrm{H}(z)$ denote the function of the primary-path and the secondary-path, respectively. $x(n)={\left[x(n),x(n-1),\cdots ,x(n-N+1)\right]}^{\mathrm{T}}$ is the reference signal vector at time n. (The meaning of unmentioned symbols is explained in the Appendix A). The residual noise (error signal) can be expressed as

$$e(n)=d(n)-y^{\prime }(n),$$

(2)

where the acoustic signal transmitted to the error microphone through the primary channel is represented by

$$d(n)=x^{\mathsf{{\rm T}}}(n)\mathrm{P}.$$

(3)

The basic FxLMS algorithm’s update equation can be written as

$$w_i(n+1)=w_i(n)-\mu \cdot x\prime (n-i)\cdot e(n),i=0,1,...,N-1,$$

(4)

where μ is the step size that should satisfy the following equation

$$0<\mu <\frac{1}{N{P}_y},$$

(5)

where $P_y$ is the power of y(n). $x\prime (n)={\left[x\prime (n),x\prime (n-1),\cdots ,x\prime (n-N+1)\right]}^{\mathrm{T}}$ represents the filtered version of the reference input, and it can be denoted as

$$x\prime (n)=c^{\mathrm{T}}x(n)={\displaystyle \sum _{i=0}^{M-1}{c}_{i}x(n-i)},$$

(6)

where $c$ is the coefficient vector of the secondary-path estimate $C(z)$. Normalized LMS (NLMS) [22] improves the step size of the LMS to stabilize the filter convergence. The NLMS algorithm’s weight vector update formula is as follows:

$$W(n+1)=W(n)+\mu (n)\cdot X(n)\cdot e(n),$$

(7)

where

$$\mu (n)=\frac{\alpha }{X^{\mathrm{T}}(n)X(n)+\gamma }.$$

(8)

where $\alpha \in \left(0,2\right)$ is a constant value that controls the convergence, and $\gamma \in \left[0,1\right]$ is used to control the step size. The FxLMS/FxNLMS methodology has been widely applied to ANC systems in current typical scenarios; however, both approaches must be updated to allow for a larger range of processing in the indoor environment proposed in this work.

2.3. The CS Algorithm for ANC System

A strategy for overcoming the nonlinearity associated with the microphone and speaker of an active noise control (ANC) system is proposed for the system presented in this paper. When the noise from the source exceeds the dynamic limit of the microphone, the reference microphone enters saturation mode. Nonlinearity introduced by ANC might be due to nonlinearity in the primary and secondary paths as well as chaotic noise. Nonlinearity is introduced into the system when the controller attempts to drive the loudspeaker system beyond its dynamic limitations. The secondary route, which is often calculated by a linear transfer function with low-level auxiliary noise, does not model such nonlinearity.

Figure 3 depicts the suggested ANC system paradigm. The input sensor detects the source noise signal in this suggested technique. It sends the input signal to the ANC system, which employs a series of methods to analyze the input noise signal in order to generate an artificial noise signal with the opposite or inverse phase but the same amplitude and frequency as the input noise signal. The speaker is used to generate an artificial noise signal known as an anti-noise signal because it has the same amplitude and frequency as a noise signal but is in the opposite phase. When manufactured noise signals and source noise signals are mixed, their effects cancel each other out, resulting in the creation of a quiet zone. The error microphone measures the outcome of these signals. If we achieve the intended result, i.e., establishing a silent zone, the process will terminate. If we need to process further, the CS algorithm is used to update the filter coefficients until we receive the desired output. Noise can be decreased by combining artificial and source noise to create a comfort zone around the ear. The meaning of each function in Figure 3 is the same as in Figure 2, except that the CS algorithm is used to update the filter coefficients, as detailed below.

The CS algorithm is employed in this approach to update the filter coefficient, and the CS algorithm’s updating process is presented in Algorithm 1. The filter coefficient is first initialized. The filter coefficient is tuned as the host nest in the algorithm. The fitness value is then calculated by substituting the coefficients of filter into the fitness function. Lévy flying is continued along with evaluating the fitness function when the number of runs is less than the predetermined number of K. At the same time, a nest is chosen at random and replaced if its fitness value is lower than the preceding one. Then, there is the possibility of eliminating the poor ones and constructing new ones. Finally, the best solution is saved and ranked. Figure 4 is the flow chart of the CS algorithm. After initialization, the CS algorithm will calculate the fitness value and then update the position. Based on the pattern of the CS algorithm, it will judge whether the nest is exposed or not. If the exposure means that the position is not good, it will update the position again; if not, it means that the position is not bad, that is, the fitness function will be calculated again. Finally, check whether the end condition is met. If so, the end has been achieved. Otherwise, continue with the previous step. It is worth noting that different functions can be used to update the position [14], which we distinguish between A and B in Figure 4.

Algorithm 1. The CS algorithm for ANC system

Begin:    Initialization: Random real numbers are used to generate initial population of n host nests xi (i = 1, 2,..., N), and xi represent the coefficients of the FIR filter    Fitness calculation: Substituting each host nests into the fitness function $f_i$ to calculate the corresponding fitness value    While (t < K)     Obtain a cuckoo randomly by Lévy flight     evaluating its fitness, $f_i$     Choose a nest randomly   if ($f_i>f_j$)     replace j by the new solution;   end    A fraction ($P_a\in \left[0,1\right]$) of the worse nests is abandoned and new ones are built;   Keep the best solutions (or nests with quality solutions);   Rank the solutions and find the current best   end while    Postprocess results and visualization end

The fitness function can be expressed as

$$f_i=\min \left\{E\left[e_i^2\left(n\right)\right]\right\},$$

(9)

where $E[\cdot ]$ is the mean squared error. When generating new solutions x_t+1, a cuckoo i, a Lévy flight is performed:

$$x_{i+1}=x_i+\alpha \oplus Lévy\left(\lambda \right),$$

(10)

where $\alpha >0$ represents the step size, which is normally set to 1. $\oplus$ denotes entry wise multiplications. This entry wise product is similar to those used in PSO, but the random walk through the Lévy flight is more efficient in traversing the search space in the long run since its step length is significantly greater. The preceding equation is the stochastic equation for a random process. A random walk is a Markov chain in which the future status/position is determined only by the present location (the first component in the above equation) and the transition probability (the second term). The Lévy flight basically yields a random walk, with the random step length selected from a Lévy distribution, which can be written as follows:

$$Lévy\left(\lambda \right)\sim u=t^{-\lambda },\left(1<\lambda \le 3\right).$$

(11)

It has an infinite variance and an infinite mean; the steps here effectively constitute a random walk process with a power-law step-length distribution and a hefty tail. Lévy should explore the best solution he has so far to conceive of some fresh solutions; this would expedite the local search. However, a significant percentage of the new solutions should be created by far field randomization and their locations should be sufficiently far from the present best solution to ensure that the system does not become caught in a local optimum.

3. Simulation

Numerous simulation experiments are performed in this section to validate the proposed method, and the classic FXLMS and FXNLMS algorithms are compared. The input microphone is 180 cm away from the error microphone in the simulation, and the canceling speaker is 50 cm away from the error microphone. On the same plane lies the input microphone, the canceling speaker, and the error microphone. The main parameters of the simulation are given in

Table 1. The key parameters of simulation.

Parameters	Value
Frequency of signal	200–20 KHz
The number of iterations	35,000–40,000
Sampling frequency	16 KHz
μ (for FXLMS and FXNLMS)	0.05
K (for CS)	400
N (Length of Filter coefficient)	32

.

The 200 Hz single tone signal is first investigated, with the simulation results given in Figure 5a, which represents the power spectrum of a 200 Hz single tone signal, demonstrating that the signals are focused in the 200 Hz frequency region; Figure 5b shows the change in the mean square error (MSE) during the iteration process of FxLMS and FxNLMS, demonstrating that the change is extremely minimal after 16,000 generations. The original signal’s normalized frequency spectrum is shown in Figure 5c. The normalized frequency spectra of the FxLMS, FxNLMS, and CS techniques are shown in Figure 5d–f, respectively. The iterative phase of the CS algorithm is depicted in Figure 5g. It can be foundthat the CS algorithm has a faster convergence time than the LMS and NLMS algorithms. Finally, Figure 5f compares the three approaches. On 200 Hz signals, the three approaches show strong noise reduction results, with the FXLMS algorithm and the CS algorithm having the best impacts.

Subsequently, we tried various single-tone signals given simultaneously. Figure 6a represents the spectrum of the noise. We set four frequencies, 1 KHz, 5 KHz, 10 KHz, and 20 KHz, accordingly. Figure 6d illustrates the frequency spectrum filtered using the FXLMS algorithm. It can be observed that most of the noise frequency bands are substantially removed; however, there are still sounds after filtering using the FXNLMS algorithm. Finally, the frequency generated by the CS algorithm is quite pure. As can be observed from Figure 6h, the noise reduction impact of the CS algorithm is superior.

Finally, we tested some background noise. The signal spectrum of the noise is more chaotic, as seen in Figure 7a,c. In this case, the noise can be successfully eliminated after filtering, as shown in Figure 7d–f, and as shown in Figure 7h, the CS algorithm has a stronger noise suppression effect.

To summarize the above simulations, we compared the noise reduction effects of each technique, as shown in

Table 2. Linear noise suppression performance of various methods for various frequencies.

Frequency (Hz)	Reduction by FXLMS (dB)	Reduction by FXNLMS (dB)	Reduction by CS (dB)
200	66.54 dB	66.63	66.76 dB
1000	48.2	43.986	47.33
5000	57.4	57.22	57.8
10,000	58	57.77	58
20,000	66.4	64.78	66.4

. Linear noise may be effectively denoised using all three approaches, with the CS slightly outperforming LMS and NLMS. At the same time, the CS algorithm has fewer repetitions than FXLMS and FXNLMS; hence, its complexity is smaller.

Finally, we tested the nonlinear noise scenario, as shown in Figure 8. The NLMS method and CS algorithm can decrease the noise to 28.7 dB–39 dB, and the CS algorithm is 2.7 dB–4.5 dB better than the NLMS algorithm. The aforementioned simulation shows that the suggested technique performs better than the conventional method and can minimize both linear and nonlinear noise. At this moment, the LMS algorithm cannot function effectively owing to nonlinear noise. In this section, a simulation is used to confirm the algorithm’s performance. To further assess the system’s performance, experimental tests based on the hardware platform are also conducted in the next section.

4. Experiment and Discussion

4.1. Hardware Implementation

Referring to [23],

Table 3. Noise suppression performance of various methods for various frequencies.

Parts	Types	Note
DSP	TMS320VC5509A	Used to run the algorithm
ADC/DAC	TLV320AIC23B	Integrated ADC and DAC functions
Power amplifier	MAX4298	-
Preamplifier	MAX4252	-
EEPROM	AT25256	Used to store data

displays the parts that make up the system. Two input signals from the input microphone and the error microphone can each be accepted by the device. Using a power amplifier, the input signal is transmitted to an antialiasing filter, and the output signals are transformed to an analog form and a power amplifier used to drive a canceling speaker. The requirements for the input and error signals are marginally different. The error signal significantly lowers after the noise cancellation option is used. In some systems, acoustic feedback from the canceling speaker may cause a slight increase in the input signal. The maximum amplitude of a broadband signal that does not saturate a particular system is lower than the comparable number for a sinusoid; additionally, signal statistics will also impact the dynamic range. The DSP used here, the TMS320VC5509 from the Texas Instruments’ TMS family, controls the algorithm.

The resultant hardware, including the input/output board and DSP board, is shown in Figure 9. The input/output board is mostly used for output and output signal amplification and filtering, while the DSP board, which includes the TLV320AIC23B, can gather audio signals and process algorithms using the TMS320VC5509A.

4.2. Experimental Setup

The setup of experiments are shown in Figure 10 along with their configurations. We have a workstation with a window next to it in the lab, and we have set the noise generators outside the window to imitate different noises. A noise microphone and a canceling speaker are installed next to the office chair; the DSP board controls both devices, and the input/output board amplifies and filters the signals coming from them.

The following is an introduction to the experimental setting and the instruments utilized. The lab is shown in Figure 11a, the system is shown in Figure 11b, and the noise collection microphone and noise generator are shown in Figure 11c. The noise generator is outdoors, and the noise acquisition microphone is mounted close to the window. The input/output plate installed within the box is shown in Figure 11d.

4.3. Experimental Test and Discussion

We activated the artificial noise source outside the room during the test. For comparative purposes, the error microphone was utilized to assess the indoor ambient noise at this time without activating the ANC system, and after activating the ANC system, the error microphone’s influence on the audio noise reduction was observed. First, the impact of a single tone signal on noise reduction was evaluated. Figure 12 displays the single-tone signal’s noise reduction outcome (500 Hz). It has been discovered that the system’s noise reduction impact is around 13 dB more pronounced for single tone signals. The findings of the further testing of monophonic sounds at various frequencies are shown in

Table 4. Noise suppression performance for various frequencies in experiment.

Frequency (Hz)	Reduction (dB)
200	9
500	8.8
1000	10.8
2000	11.5
3000	11.2
5000	8.3
10,000	7.8
20,000	7.6

. The test findings demonstrate that a single tone signal has a greater than 7 dB noise reduction impact. The actual environment has many unpredictable characteristics; therefore, the outcomes are significantly different from the simulated findings. To reduce the discrepancy between the simulation and the actual scenario, we can further enhance the system model in the future.

Finally, noises that replicated the real environment were produced using a synthetic noise source. The difference in audio levels between the interior environment before and after the ANC system was activated revealed a 6–7 dB decrease in noise, as shown in Figure 13. It is important to note that the benchmark for determining the effectiveness of noise reduction should be revised properly when the interior volume increases. People may be positioned randomly across the room (this paper presumes that a person spends most of his/her time at his/her desk), and noises may originate from various locations. Given the aforementioned elements, more thought must be given to the actual application design. To enhance the effectiveness of noise suppression, the number of sensors may be raised while the canceling speaker’s power may be increased. This study examined a single case. Finally, there is a strong correlation between the microphone’s performance and the noise the system collects. Different microphones have varying effects on the sensitivity of the sound and the caliber of the sound captured during the real test procedure. As a result, the aim with respect to audio acquisition should be the microphone with the best performance. Meanwhile, the performance that the sound box produces also has a certain impact on how well the noise is reduced. For example, single-directional loudspeakers can only emit sound waves in one direction but surround stereo speakers can emit sound waves in all directions.

5. Conclusions

An ANC system for reducing indoor noise is proposed in this research. The real application scenario was established to improve the conventional ANC application space after first considering and analyzing the source of the indoor noise. Then, on the foundation of the traditional ANC algorithm, a CS-based ANC noise reduction algorithm was developed, and it was demonstrated to considerably improve the system’s performance. The simulation results tested single frequency noise and real noise to determine whether the method was successful. The system’s effectiveness with respect to reducing actual noise levels and providing people with a comfortable indoor atmosphere underwent one more test in the lab.