- Email: [email protected]
Directional monitoring terminal for aircraft noise
Journal of Sound and Vibration 374 (2016) 77–91
Contents lists available at ScienceDirect
Journal of Sound and Vibration journal homepage: www.elsevier.com/locate/jsvi
Directional monitoring terminal for aircraft noise M. Genescà a,b,n a Acoustics Research Center, Department of Electronics and Telecommunications, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway b Laboratory of Acoustics/Noise Control, Empa - Swiss Federal Laboratories for Materials Science and Technology, 8600 Dübendorf, Switzerland
a r t i c l e i n f o
abstract
Article history: Received 7 October 2015 Received in revised form 14 March 2016 Accepted 4 April 2016 Handling Editor: P. Joseph Available online 19 April 2016
This paper presents a concept of an aircraft noise monitoring terminal (NMT) that reduces background noise and the influence of ground reflection, in comparison with a single microphone. Also, it automatically identifies aircraft sound events based on the direction of arrival of the sound rather than on the sound pressure level (or radar data). And moreover, it provides an indicator of the quality of the sound pressure level measurement, i.e. if it is possibly disturbed by extraneous sources. The performance of this NMT is experimentally tested under real conditions in a measurement site close to Zurich airport. The results show that the NMT unambiguously identifies the noise events generated by the target aircraft, correctly detects those aircraft noise events that may be disturbed by the presence of other sources, and offers a substantial reduction in background and ground reflected sound. & 2016 Elsevier Ltd. All rights reserved.
Keywords: Aircraft noise monitoring Microphone array
1. Introduction Airport noise monitoring systems and prediction models are the two main tools for airport noise assessment and management. Prediction models are necessary to forecast the airport's acoustic footprint in future or hypothetical situations involving, for example, increases in flight volume or the modification of flight paths. Noise monitoring systems are mainly used to inform the public about the current airport noise situation and also to validate the output of the prediction models and reinforce its credibility in the eyes of the public. The noise monitoring systems include a network of noise measurement terminals (NMTs) distributed within the acoustic area of influence of the airport. These NMTs are typically equipped with an omnidirectional microphone that continuously monitors the noise levels at the site. Since this data is used for public information, the placement of these terminals is chosen so that the measured noise levels are representative of the acoustic influence of the airport on the population. Therefore they are located in urban, or at least inhabited, areas where other sound sources such as traffic or industrial noise may be present. This NMTs placement has two main disadvantages: First, the continuous sound pressure level recording contains noise events that may not be generated by aircraft. Therefore strategies to identify the aircraft noise events need to be put in place. The ISO standard 20906:2009(E) [1] describes the event identification process. It starts with a detection stage in which the parts of the recording that exceed a given threshold by a specified amount, for a duration which lies within specified limits, are detected. Next, it follows a n Correspondence address: Acoustics Research Center, Department of Electronics and Telecommunications, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway. E-mail address: [email protected]
http://dx.doi.org/10.1016/j.jsv.2016.04.004 0022-460X/& 2016 Elsevier Ltd. All rights reserved.
78
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
classification stage in which acoustical properties, that may help to separate non-aircraft sound events from aircraft events, of the detected events are analyzed. It finishes with an identification stage in which non-acoustical information such as radar or flight plan data is used to finally confirm whether an event is generated by an aircraft. The fact that the detection is based on a sound pressure level threshold causes a number of aircraft noise events to be missed in locations where the level of background noise is comparable to the aircraft noise level [2]. Moreover, the event identification error rate grows considerably in airports that do not have a radar data connected to the noise monitoring system because the final identification stage cannot be implemented. To overcome this first disadvantage, several algorithms to automatically classify noise events caused by aircraft have been developed [3–5]. The second disadvantage of the NMTs placement is that noise events caused by other sources may occur simultaneously to the aircraft noise events. In such a case the sound pressure level of the aircraft may be overestimated. As a consequence, NMTs cannot be used solely as a tool to penalize an airline when a particular flight exceeds the legal noise limits. Moreover, interference effects due to ground reflections are undesirable in aircraft noise monitoring applications because they depend on the height of the receiver and introduce uncertainty in the measurement [6]. To reduce these effects the ISO standard 20906:2009(E) [1] describes an ideal measurement site; but such sites are not always available. To overcome these disadvantages, the existing literature [7–10] provides examples of advanced NMTs that are integrated by a linear microphone array and that use beamforming to reduce the contribution of ground-borne sources and aircraft ground reflections on the continuous sound pressure level recording. Such terminals typically consist of a linear array of 12– 15 microphones with a length of around 3 m and provide a reduction of the ground-borne noise level between 5 dB and 10 dB. The NMT concept presented in this paper is also aimed at overcoming these disadvantages and consists of an array of nine microphones distributed in a 3D geometry with a diameter of approximately 1 m. This NMT provides an estimate of the direction of arrival (DOA) of the aircraft sound and a measurement of the aircraft sound pressure level (frequency spectrum and overall level) with a reduced contribution of the background noise and ground reflections compared to an omnidirectional microphone measurement. As in the case of the noise monitoring terminal NA-37 developed by RION [11], the sound DOA estimates are used for automatic aircraft event identification purposes so that non-acoustical information such as radar or flight plan information is not required. Thus it can be used in small airports without radar and by external (to the airport) entities or administrations. In addition, the DOA estimates provided by the NMT presented here are robust against ground reflections and, to an extent, to the presence of ground-borne noise sources. This feature is used to determine the quality of the measurement, i.e. if sources other than the target aircraft may have influenced the measurement. Furthermore, such a NMT can be used to obtain reliable aircraft spectral sound power data of individual flights. This data is required by the new generation of prediction models to be able to calculate the acoustic impact of individual flights [12,13]. This would allow the design of noise abatement operational procedures affecting not only flight paths but also the aircraft flight configuration (e.g. slats, flaps, landing gear and power settings). Initiatives to use NMTs integrated by single omnidirectional microphones as a basis to calculate aircraft spectral sound power data are reported in the literature [14]. However, they require the placement of NMTs in low background noise locations using them purely as a source of data for prediction models and removing their utility as a source of public information. The NMT presented here can be used in measurement campaigns dedicated to the collection of sound power spectral data such as those described in the literature [13,15] or as a permanent installation as part of the noise monitoring system of an airport. If at least two of these NMTs are used, the position of the aircraft could be estimated by triangulation. Once the aircraft position and the associated sound pressure levels at a set of NMTs are available, the aircraft sound power spectral data could be calculated by back propagating to the source. The remainder of this manuscript is structured as follows: In Section 2 the different modules of the NMT are described; Section 3 presents a case study used here to show the features of such a NMT; In Section 4 the results provided by the NMT
Fig. 1. Convention used to define the vertical θ and horizontal ϕ angle of arrival of the aircraft noise. The origin of coordinates represents the position of the array.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
79
in the context of the case study are compared to those obtained with a single omnidirectional microphone; Section 5 summarizes the main findings of the study.
2. Noise monitoring terminal The NMT presented here includes a compact microphone array of nine omnidirectional microphones divided, for signal processing purposes, in two sub-arrays. The placement of the compact array must guarantee that all ground-borne noise and ground reflections of the aircraft noise impinge on it from below, e.g. on a roof top terrace. The NMT consists of three main modules: the DOA estimation module; the directional monitoring module; and the event identification module. There are several outputs from the NMT. The first one, from the DOA estimation module, is a time series of aircraft noise DOA ^ ðτÞ, and defined as in Fig. 1. The second output, from the directional monitoring estimates, represented here by θ^ ðτÞ and ϕ module, is a time series of T seconds equivalent continuous sound pressure level measurements Lp;eq;T ðτÞ and the corresponding frequency spectra Pð f ; τÞ. The third output, from the event identification module, are the start time τistart and stop time τistop for each aircraft noise event i along with a flag indicating the quality of the measurement. Fig. 2 shows the data flow between these three modules.
Fig. 2. Data flow between the three modules that comprise the NMT.
Fig. 3. Microphone distribution in the NMT. The meaning of the four symbols used is the following: The solid black circles represent the omnidirectional microphones that belong exclusively to the first sub-array; The solid grey circle represents the omnidirectional microphone that belongs exclusively to the second sub-array; The grey circles with a black outline represent the microphones that belong to both sub-arrays; The cardioid shapes show the position and orientation of the virtual first-order microphones generated from the signals of the vertically aligned omnidirectional microphone pairs. All dimensions are in meters.
80
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
2.1. DOA estimation module As seen in Fig. 2, this module uses as its only input the signals of the microphones of the first sub-array. The first subarray consists of eight omnidirectional microphones depicted as black circles and as grey circles with a black outline in Fig. 3. Each of these eight microphones is located 0.15 m above or below one of the vertices of a regular tetrahedron of side length 0.85 m. The DOA estimation algorithm has been described in detail [16] and tested previously [17] and is not the focus of the present paper, thus it will not be discussed here. However, the main idea is to use each of the vertically aligned microphone pairs in the first sub-array to virtually generate a first-order directional microphone at each vertex of the regular tetrahedron, as shown in Fig. 3. A first-order microphone has the directivity gain DðθÞ that follows DðθÞ ¼ ð1 bÞ þ b cos θ;
(1)
where b A ½01 is a factor that determines the specific directivity pattern, e.g. b¼ 0.5 corresponds to a cardioid, b¼0.75 to a hypercardioid. As a first step in the DOA estimation algorithm, the specific directivity pattern of the four virtual first-order directional microphones is selected, according to the geometry of the measurement site, with the aim to reduce the influence of ground reflections which typically introduce error in the estimated vertical angle of arrival. Then, the time difference of arrival between all possible pairs of virtual microphones is estimated. Finally the DOA is estimated by exploiting the relationship between the time differences of arrival and the DOA. The accuracy of the vertical angle of arrival estimate depends mainly on the optimal choice of the directivity pattern of the virtual microphones according to the source–receiver geometry, and to a lesser extent on the ground properties [16,17]. In this module only the sound frequency content between 20 Hz and 200 Hz is considered. Due to the atmospheric attenuation most of the aircraft sound energy on the ground is concentrated below 200 Hz, and below 20 Hz the signal can be heavily influenced by wind. It is worth clarifying that when the aircraft sound is predominant in this frequency range, the vertical angle estimate provided by this module corresponds to the vertical angle of arrival of the aircraft sound. However, when ground-borne sources are predominant the vertical angle will be higher than 901 but may differ considerably from the real vertical angle of arrival of the ground-borne noise. This is due to the constrained search domain of the time difference of arrival estimates introduced to increase precision when aircraft sound is present [16]. 2.2. Directional monitoring module The goal of this module is to measure the aircraft sound pressure levels reducing the influence of the ground-borne sound and aircraft ground reflections with respect to an omnidirectional measurement. The inputs to this module are the discrete-valued signals pn(t) of the microphones n in the second sub-array, and the vertical angle of arrival estimates θ^ ðτÞ provided by the DOA estimation module. The second sub-array consists of three microphones vertically aligned which are represented by the solid grey circle and the two grey circles with a black outline in Fig. 3. The distance between the upper microphone pair is 0.30 m, and the distance between the lower pair is 0.03 m. Here the signals of these three microphones are processed to virtually obtain the output of a first-order directional microphone. The specific directivity pattern of such a microphone can be adapted to the geometry of the measurement site to reduce the influence of ground reflections and/or ground-borne sources. In the following paragraphs the signal processing algorithm is detailed. Let P n ð f ; τÞ and P m ð f ; τÞ be the discrete Fourier transform of the higher and lower microphone signals respectively of the microphone pair separated by d ¼0.30 m or alternatively of the microphone pair separated by d ¼0.03 m.T The discrete Pτ þ Fourier transform of a signal frame of length T received at microphone n is defined here as P n ð f ; τÞ ¼ t ¼ 2τ T pn ðt Þe j2π ft 2 where j is the imaginary unit. The output of a virtual first-order directional microphone Pð f ; τÞ is obtained as follows [18] P n ð f ; τ Þ þ P m ð f ; τÞ P n ð f ; τÞ P m ð f ; τÞ þb ; P ð f ; τÞ ¼ ð1 bÞ 2 jkd
(2)
where k is the wavenumber. P jkd cos θs If θs is the true value of the vertical angle of arrival of the source s, then P n ð f Þ ¼ s P s ð f Þe 2 þ Nn ð f Þ and P s jkd cos θs s 2 þ Nm ð f Þ where P ð f Þ is the Fourier transform of the signal from the source s at the mid point P m ð f Þ ¼ s P ð f Þe between n and m, and Nn(f) is the additive noise of microphone n (e.g. microphone self noise or wind-induced noise). Taking into account these expressions, Eq. (2) equals X s s P ð f ; τÞDð f ; θ ðτÞÞþ Nð f ; τÞ; (3) Pð f ; τÞ ¼ s
where Dð f ; θÞ is the directivity gain of the virtual directional microphone 2b sin kd cos θ=2 ; D f ; θ ¼ ð1 bÞ cos kd cos θ=2 þ kd
(4)
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
81
and Nð f ; τÞ is the additive noise of the virtual directional microphone Nð f ; τÞ ¼ ð1 bÞ
Nn ð f ; τÞ þ Nm ð f ; τÞ N n ð f ; τ Þ N m ð f ; τÞ þb : 2 jkd
(5)
If d⪡λ, where λ ¼ 2kπ is the wavelength, then Eq. (4) simplifies to Eq. (1), meaning that the sound pressure is measured with a first-order directivity filter. At the same time, if the separation between the microphone pair is much smaller than the wavelength, the second term on the right of Eq. (5) increases, also increasing the additive noise of the virtual first-order microphone. The frequency range of interest is from 88 Hz to 2818 Hz (octave bands from 125 Hz to 2 kHz) because the Aweighted aircraft sound pressure levels measured on the ground are quite low outside of this range [7,9]. The microphone pair with d ¼0.30 m is used to generate a virtual first-order microphone for frequencies from 88 Hz ðλ 3:8 mÞ to 355 Hz ðλ 1 mÞ, while the microphone pair with d ¼0.03 m is used for the remaining frequencies up to 2818 Hz ðλ 0:1 mÞ. In that way, the distance d is much smaller than the wavelength to get a first-order directivity pattern, but not so small so as to result in high additive noise amplification. The T seconds equivalent continuous sound pressure level Lp;eq;T ðτÞ for a particular frequency range is calculated as 0 ! 1 f2 ^ ðτÞÞj 2 X 2 jPð f ; τ Þ=Dð f ; θ A; (6) Lp;eq;T ðτÞ ¼ 10 log@ 2 Po Q f ¼ f1 where P o ¼ 2 10 5 Pa is the reference pressure value, and Q is the number of samples in T seconds of the signal pn(t). Note that the sound pressure spectrum of the virtual first-order microphone Pð f ; τÞ is divided by the directivity gain associated with the vertical angle estimate θ^ ðτÞ provided by the DOA estimation module. This is done to compensate for the attenuation imposed on the aircraft sound by the directivity pattern of the virtual first-order microphone. Assuming that the sources s in Eq. (3) are the target aircraft a and ground-borne sources gb (urban noise and ground reflections), this division can be written as P a gb P a ð f ; τÞDð f ; θ ðτÞÞ þ gb P gb ð f ; τÞDð f ; θ ðτÞÞ þ Nð f ; τÞ Pð f ; τÞ ¼ : (7) Dð f ; θ^ ðτÞÞ Dð f ; θ^ ðτÞÞ As described in Section 2.1, if the aircraft sound is the predominant source in the range within 20 Hz and 200 Hz then
θ^ ðτÞ C θa ðτÞ and therefore
XDð f ; θ ðτÞÞ gb Pð f ; τÞ Nð f ; τÞ CP a ð f ; τÞ þ P ð f ; τÞ þ : a a Dð f ; θ ðτÞÞ Dð f ; θ ðτÞÞ Dð f ; θ^ ðτÞÞ gb
(8)
gb
Under these circumstances it is also true that Dð f ; θ ðτÞÞ o Dð f ; θ ðτÞÞ, since the virtual microphones are vertically oriented as shown in Fig. 3. Therefore the noise from ground-borne sources is attenuated in comparison to the case of using an omnidirectional microphone. As explained in Section 2.1, when aircraft noise is not present it is not necessarily true that θ^ ðτÞ ¼ θgb ðτÞ and therefore the term in Eq. (7) corresponding to the ground-borne sources can even be amplified. However, this is not relevant since, for aircraft noise monitoring purposes, the interest lies only in the time intervals when an aircraft is detected. gb
a
Fig. 4. Upper and lower limit of the aircraft sound pressure level uncertainty caused by a vertical angle estimation uncertainty of ½ 4:41 4:41. These limits are plotted as a function of θa for different directivity patterns.
82
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
As also seen in Eq. (8), the noise term is amplified by a factor 1=Dð f ; θ ðτÞÞ. Regardless of the specific first-order a directivity pattern, Dð f ; θ ðτÞÞ is minimum when θa is close to 901, when a stronger influence of the microphone additive noise is expected. a As Eq. (8) relies on θ^ ðτÞ C θ ðτÞ it follows that the uncertainty in the vertical angle estimation translates into sound pressure level uncertainty in Eq. (6). In previous experimental tests [17] of the DOA estimation module it has been found a that θ^ ðτÞ θ ðτÞ A ½ 4:41 4:41. As an example, Fig. 4 shows the corresponding upper and lower limit of the aircraft sound a pressure level uncertainty as a function of θa for different first-order directivity patterns assuming θ^ ðτÞ θ ðτÞ A ½ 4:41 4:41. a
2.3. Event identification module The goal of this module is to identify the events i caused by target aircraft, and to provide a time stamp for the beginning
τistart and the end τistop of each event along with a flag indicating the quality of the measurement; this is done in three stages.
The first stage is the event detection in which the aircraft events are detected by analyzing the evolution of the aircraft ^ ðτÞ provided by the DOA estimation module. The second stage is the signal segmentation sound DOA estimates θ^ ðτÞ and/or ϕ in which the starting time of the event τistart is determined also based on the analysis of the sequence θ^ ðτÞ. The final stage is ^ ðτÞ and the the assessment of the measurement quality in which τistop is also determined. This is done by analyzing θ^ ðτÞ, ϕ evolution of Lp;eq;T ðτÞ provided by the directional monitoring module. As the specifics of the detection algorithm are largely dependent on the particular geometry of the measurement site they will be discussed in Section 3.3 in the context of the case study presented.
3. Measurement set-up In the present case study, the NMT is located 500 m away from the end of the runway RNW16 of the Zurich airport. The target aircraft are those taking off from this runway. Fig. 5 shows the NMT position (as a black dot) with respect to the runway (grey band), as well as the radar trajectories of 50 aircraft that took off during the measurement campaign carried out from 12:00 to 14:00 on the 15th of May and from 9:00 to 16:00 on the 16th of May 2014. Some of the aircraft make a turn of approximately 901 around the NMT, while some others make a turn of approximately 2701 and so are heard for a longer period of time. The vertical angle of sight of the aircraft with respect to the NMT ranges between 451 and 901. The NMT was placed on the roof top terrace of a 25 m tall building. The base of the tetrahedron depicted in Fig. 3 was at a height h ¼2.4 m over the flat terrace floor which was covered in wet gravel. Fig. 6 shows the array in the measurement site. The measurement site had a high background noise due to the motorway, tram and train line that can be seen in Fig. 6. The measured average background noise level considering only the traffic noise was 66.4 dB(A). The traffic and rail noise impinges on the microphone array with a vertical angle of arrival of approximately 1001. Six of the microphones of the first sub-array were B&K type 4190 and two were B&K type 4165, the microphone that belonged exclusively to the second sub-array was a Norsonic type 1225. All the preamplifiers were Norsonic type 1202 and were connected to Norsonic type 336 amplifiers that provide 200 V DC to the microphones. The microphone signals were acquired with three National Instruments acquisition cards type NI9234 mounted on a National Instruments cDAQ-9174
Fig. 5. Plot of the radar trajectories of 50 aircraft that took off from Runway RNWY16 of Zurich airport during the measurement campaign. The black dot represents the NMT terminal located 500 m away from the end of the runway (grey band).
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
83
Fig. 6. Microphone array in the measurement site.
chassis. The sampling frequency was set to 50 kHz. Prior to the measurement campaign, a B&K type 3541 calibrator for intensity probes was used to measure the transfer function between one reference channel and the others. These transfer functions were stored and used later to compensate for the phase and amplitude differences between channels. The wind speed measured at 6 m above the ground, by the airport weather station, was between 0 m/s and 4 m/s during the 15th of May, and between 3 m/s and 6 m/s, during the 16th of May. The average temperature measured at 2 m above the ground was 9 °C and the humidity was 70 percent the 15th of May and 12 °C and 55 percent the 16th of May.
3.1. DOA estimation module settings The DOA estimation algorithm has been implemented as described in Ref. [17] to provide a DOA estimate for each successive non-overlapping signal frame of length T ¼0.5 s. The factor b that describes the directivity of the virtual firstorder microphones generated by each vertically aligned microphone pair in the first sub-array has been set to b¼0.65. This is an appropriate choice according to the simulations presented in Ref. [16] for the expected range of vertical angles of sight of the aircraft with respect to the array.
3.2. Directional monitoring module settings The relevant setting regarding the directional monitoring module is the value of b. Assuming that the directivity pattern of the virtual first-order microphones created by the second sub-array follows the ideal equation Eq. (1), the goal is to gb a minimize the ratio Dðθ Þ=Dðθ Þ in Eq. (8). A straightforward solution is to choose the first-order directivity pattern whose gb zero directivity gain falls on θ . In this case study there are two types of ground-borne sources: the traffic/rail noise; and the ground reflections of the aircraft sound. For this particular measurement site, in spite of the high sound pressure levels of the traffic noise, the aircraft sound tends to prevail over the traffic due to the proximity of the measurement point to the aircraft flight path. Therefore, the factor b has been adaptively calculated so that the zero in the directivity gain falls on the DOA of the ground reflected noise. The described choice of b has only been applied while θ^ ðτÞ o 801. For 801 r θ^ ðτÞ r901, the a resulting values of Dðθ^ Þ Dðθ Þ are too low and, as predicted by Eq. (8), the additive noise would be largely amplified. Furthermore, in this range, values of b close to 1 need to be avoided to not raise the sound pressure level uncertainty (over 2 dB in the example in Fig. 4). Therefore, in this range, the factor b has been chosen adaptively so that Dðθ^ ðτÞÞ ¼ 0:5. For
84
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
Fig. 7. Flow diagram showing the stages (dashes boxes) and the steps (numbered blocs) involved in the event identification algorithm.
901o θ^ ðτÞ b is set to 0.5. In summary, b is defined here as 8 1 > > ; > > > > ð1 cos ð1801 θ^ ðτÞÞÞ > < 0:5 b¼ ; > > > ð1 cos θ^ ðτÞÞ > > > > : 0:5;
if θ^ ðτÞ o 801 if 801 r θ^ ðτÞ r901 if θ^ ðτÞ 4 901:
(9)
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
85
The directional monitoring module has also been configured to provide Lp;eq;T ðτÞ and the associated frequency spectrum Pð f ; τÞ for each successive non-overlapping signal frame of length T¼0.5 s.
3.3. Event identification module settings Fig. 7 shows, as a flow diagram, the stages and steps involved in the event identification algorithm. The first stage is the event detection and aims to detect aircraft noise events. As seen in Fig. 5, the target aircraft approach the NMT from the northwest and depart to the northeast (some of them turn to the west afterwards). This particular pattern is used here to unambiguously detect the presence of target aircraft as none of the other ground- or air-borne sources (aircraft departing from other runways) present a similar movement. Therefore, as shown in Fig. 7, the first step of the event ^ ðτÞ provided by the DOA identification algorithm consists of reading the value of the horizontal angle of arrival estimate ϕ estimation module. This angle is defined as shown in Fig. 1 within the interval ½ 1801 1801, meaning that when the aircraft first crosses the west axis in Fig. 1, the horizontal angle of arrival jumps from 1801 to 1801. Thus, the second step is to ^ ðτÞ with the previous one ϕ ^ ðτ TÞ to see if there is a relevant jump. If the jump is compare the most recent sample ϕ confirmed, the presence of an aircraft is unambiguously confirmed at the detection time τidetect. The second stage is the segmentation which is aimed at finding the initial time τistart of the aircraft sound event i that has just been detected. It includes the third and fourth steps in Fig. 7 which search for the closest time instant previous to τidetect when the vertical angle estimate is still lower than or equal to 901. This is the time τistart at which the DOA estimation module starts detecting an airborne source. The final stage is the quality assessment and aims to determine the final time τistop of the aircraft sound event i, and to collect data about the quality of the sound measurement of this event. Steps five to ten are repeated recursively over the subsequent time frames τo ¼ τistart þ zT (z¼ 0,1,2,3…) until τistop is found. Steps five to seven use the sound pressure level measurement provided by the directional monitoring module to calculate the equivalent sound pressure level over the five seconds surrounding the analyzed time τo. If this equivalent sound pressure level is 1 dB below the average background noise level τistop is reached. If this is the case, the algorithm returns “A” as a flag indicating that the measurement is undisturbed. Otherwise, step eight checks if at the analyzed time the vertical angle of arrival estimate is higher than 901. If this is the case τistop is reached and the returned flag “D” means that the measurement is potentially disturbed by the presence of a ground-borne source. If this is not the case, step nine checks if the next detection is within the following 30 s. If it is, τistop is reached and the returned flag “B” means that the measurement is potentially disturbed by the take-off of the next plane. Otherwise, the tenth step checks if at the analyzed time τo there is an irregular decrease in the slope of the sequence of horizontal angle of arrival estimates. This indicates that another airborne source is present. In this case, τistop is reached and the flag “C” indicates that the measurement is potentially disturbed by other air-borne sources. Finally, step eleven checks the duration of the event from τistart to τistop and if it is too short it overwrites the existing flag with “F”, indicating that the sound pressure level measurement is not reliable for an unknown reason.
Fig. 8. Example of the output of the NMT. The first and second plots correspond to the estimate of the vertical and horizontal angle of arrival, the third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels of the NMT (in black) and of a single microphone (in grey). The dashed lines mark the initial τistart and final τistop time of the detected event which has been flagged as “A” (undisturbed). There is a second event generated by the passage of a train.
86
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
4. Results The aircraft noise monitoring features of the NMT are the main focus of the present study, therefore the results from the DOA estimation module are not discussed here. A detailed analysis of the performance of this module can already be found in the literature [16,17]. 4.1. Event identification Figs. 8–10 show examples of the output of the NMT developed here compared to the information obtained from an omnidirectional microphone. The first and second plots in all these figures correspond to the estimate of the horizontal and vertical angle of arrival respectively. The third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels Lp;A;eq;0:5 s provided by the directional monitoring module compared to the that of a single microphone. The dashed lines mark the initial τistart and final τistop time of the detected events provided by the event identification module. The flag assigned to the event by this module is also shown.
Fig. 9. Example of the output of the NMT. The first and second plots correspond to the estimate of the vertical and horizontal angle of arrival, the third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels of the NMT (in black) and of a single microphone (in grey). The dashed lines mark the initial τistart and final τistop time of the detected events which has been flagged as “A” (undisturbed) and “C” (potentially disturbed by another airborne source) respectively.
Fig. 10. Example of the output of the NMT. The first and second plots correspond to the estimate of the vertical and horizontal angle of arrival, the third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels of the NMT (in black) and of a single microphone (in grey). The dashed lines mark the initial τistart and final τistop time of the detected events which has been flagged as “A” (undisturbed) and “F” (sound pressure measurement not reliable for unknown reasons) respectively.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
87
^ ðτÞ that allows for In the third plot of Fig. 8 two acoustic events can be seen. For the first event there is a clear jump in ϕ ^ ðτÞ indicating that it is not an the event to be identified as aircraft-generated. For the second event there is no jump in ϕ aircraft event. Indeed it corresponds to the passage of a train while the aircraft was still distinguishable. If only the acoustic information provided by an omnidirectional microphone is available, an automatic identification of these events is not possible and radar data is needed for clarification, whereas with the presented NMT these events are automatically identified. Note also, that θ^ ðτÞ tends to 901 at the time of the second event which is a good indicator that this is likely a groundborne source. Fig. 9 shows two consecutive detected aircraft events flagged “A” (undisturbed) and “C” (potentially disturbed by another ^ ðτÞ which indicates that there airborne source). The “C” flag of the second event is due to the sudden change of the slope of ϕ is a non-target source interfering with the estimation of the horizontal angle of arrival of the target source. Moreover, at the time of this disturbance, θ^ ðτÞ is still well below 901 indicating that the interfering source is an airborne one. Actually, the interfering source is the aircraft that generated the first acoustic event whose trajectory contains a 2701 turn (see Fig. 5) and is heard again when flying west. The third plot in Fig. 10 shows three acoustic events (around times 60 s, 120 s, and 145 s); only the first and third are identified as aircraft events. The first is flagged “A”, but the third is flagged “F” indicating that the sound pressure measurement is not reliable for unknown reasons. At the time of the second event, θ^ ðτÞ is higher than 901 indicating that this event is caused by a ground-borne source. Indeed, this event corresponds to the passage of a train overlapped to the beginning of the third event. The third event is identified as an aircraft but shortly after the detection time, θ^ ðτÞ becomes greater than 901 indicating that the end of the event is reached because a ground-borne source is detected. This happens because of the presence of the train. As explained in Section 3.3 such a situation should be flagged as “D” (potentially disturbed by a ground-borne source) but since there are less than 15 s between the start and the end of the third event the flag “F” is used instead. Note that an important advantage of this NMT is the fact that, even when the sound pressure measurement is flagged as potentially disturbed by other sources, the targeted aircraft events are still detected and identified. This means that, if desired, these events can be inspected off-line by an expert and a final decision could be made regarding whether to use or not the sound pressure data. In this particular case study, all the aircraft events have been satisfactorily identified. Out of the 50 recorded flights, 45 have been flagged as “A” (undisturbed), three have been flagged as “C” (potentially disturbed by another airborne source) and two as “F” (sound pressure measurement not reliable for unknown reasons). 4.2. Directional monitoring In this section the results related to the directional monitoring module are presented and discussed. 4.2.1. Analysis of the background noise suppression Fig. 11 shows the difference between the background noise levels measured by the NMT and those measured by an omnidirectional microphone. The difference is shown as a function of the frequency and as a function of the vertical angle of arrival of the aircraft sound. These values have been calculated by averaging the sound pressure level differences over 314
Fig. 11. Difference between the average background noise levels measured by the NMT and by an omnidirectional microphone as a function of the frequency and the vertical angle of arrival of the aircraft sound. Negative values indicate that the levels measured by the NMT are lower than those measured by an omnidirectional microphone, positive values indicate the contrary.
88
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
signal frames of 0.5 s length of continuously recorded background noise (traffic noise and wind-induced noise). To obtain the sound pressure levels corresponding to the NMT, the settings described in Section 3.2 have been applied. Since the recording contains only background noise, the estimates θ^ provided by the DOA estimation module have been disregarded and artificially reset to represent the vertical angle of arrival of the sound of a hypothetical aircraft. For angles higher than 751 it can be seen that at low frequencies (up to 140 Hz) and also between 355 Hz and 400 Hz, the background noise level measured by the NMT is slightly higher than the one measured by an omnidirectional microphone. This is due to the high wind speed during the measurements that induces a high level of noise in the microphone signals. At those frequencies, the distance d between the microphone pairs (d¼0.30 m for 100 Hz and 125 Hz, and d ¼0.03 m for 400 Hz) is quite small compared to the wavelength. Under these circumstances, as described by Eq. (5), any kind of additive noise such as wind-induced noise gets amplified. Moreover, this phenomenon is more acute at vertical angles close to 901 a because in this range Dðθ Þ is smaller and therefore the wind-induced noise is further amplified according to Eq. (8). All these suggest that the NMT presented here is more sensitive to wind-induced noise than an omnidirectional microphone. However, this problem could be solved by using one more microphone in the directional noise monitoring module so that three microphone pairs instead of two are used to cover the whole frequency. The idea is to ensure that the distance d is never too small compared with the wavelength over the whole frequency range. Above 801, excluding the frequencies mentioned above, the background noise measured by the NMT is lower than that measured by an omnidirectional microphone. However, the difference is limited because in this range the restriction Dðθ^ ðτÞÞ ¼ 0:5 has been imposed to avoid further amplification of the additive noise and as a consequence the ratio gb a Dð f ; θ ðτÞÞ=Dð f ; θ Þ in Eq. (8) is not low enough to provide a greater reduction of the background noise. For the angles immediately below 801 the background noise measured by the NMT is more than 15 dB lower than that measured by an omnidirectional microphone. This is because at those angles, the DOA of the ground reflections coincides with that of the traffic/rail noise, and since the directivity pattern of the directional monitoring module is chosen so that the zero in directivity gain falls in the DOA of the ground reflections it also falls in the DOA of the traffic/rail noise. Considering the overall sound pressure levels Lp;A;eq;0:5 s , the difference between the background noise measured by the NMT and by an omnidirectional microphone is within 6:7 dB and 7:6 dB for aircraft noise vertical angles of arrival below 801, and within 3 dB and 0:8 dB for aircraft noise vertical angles of arrival above 801. 4.2.2. Analysis of the spectrogram Fig. 12 shows the spectrogram of a continuous recording including two aircraft events flagged as “A”. As additional information, the top plot shows the Lp;A;eq;0:5 s provided by the omnidirectional microphone (grey line) and by the NMT (black line), and the vertical angle estimate provided by the NMT (black thin line). The middle plot shows the A-weighted spectrogram provided by an omnidirectional microphone, and the bottom plot shows the same provided by the NMT. The effect of the ground reflection can be seen in the omnidirectional microphone spectrogram (middle plot) in the form of a comb filter below 1 kHz, it shows the constructive and destructive interference frequencies. This effect is not present in
Fig. 12. Spectrogram of a continuous recording including two aircraft events flagged as “A”. The top plot shows the Lp;A;eq;0:5 s provided by the omnidirectional microphone (grey line) and by the NMT (black line). It also shows the vertical angle estimate provided by the NMT (black thin line). The middle plot shows the A-weighted spectrogram provided by an omnidirectional microphone and the bottom plot shows the same provided by the NMT.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
89
the NMT spectrogram (bottom plot) due to the attenuation imposed on the reflected wave by the directivity pattern of the virtual first-order microphone. Furthermore, the effect of the ground reflection in the omnidirectional spectrogram cannot be noticed before the aircraft sound impinges on the NMT with a vertical angle of arrival of approximately 751. This is because before this angle is reached, the reflection point that satisfies the law of specular reflection is located outside of the roof terrace where the NMT is placed. As a consequence, it can be seen that the curves Lp;A;eq;0:5 s of both the NMT and the omnidirectional microphone in the top plot are very similar until a vertical angle of arrival of 751 is reached. Afterwards the curve corresponding to the omnidirectional microphones exhibits higher values due mainly to the effect of the ground reflections. As the reflection coefficient of the ground increases as the vertical angle of arrival (as defined in Fig. 1) decreases, the constructive and destructive interference effects are stronger at the point of the flight path at which the vertical angle of arrival is the lowest. In the case of the first aircraft noise event, the maximum in Lp;A;eq;0:5 s happens precisely at that point. This can be seen in the grey curve of the top plot and it is also marked in the omnidirectional spectrogram as t Lmax . In the NMT spectrogram of this same event, in which the ground effect is suppressed, t Lmax happens earlier in time. Therefore, it is shown that the ground reflection may affect both the level and the time of occurrence of the maximum sound pressure level. In the second aircraft noise event the ground reflections have not such an acute effect on the maximum level, probably, due to the fact that minimum vertical angle of arrival is only of 63:31 and thus the ground reflection coefficient stays lower than in the first event in which a minimum vertical angle of arrival of 46:51 is reached. The NMT ground-borne noise suppression shows up as a general lightening of the NMT spectrogram with respect to the omnidirectional microphone one. It is more noticeable towards the end of the event because the aircraft noise contribution gb a decreases, but mainly because at the beginning of the event (vertical angles of arrival over 801) the ratio Dð f ; θ ðτÞÞ=Dð f ; θ Þ is not low enough to provide a greater reduction of the background noise as explained in Section 4.2.1. 4.2.3. Analysis of the maximum sound pressure level and the sound exposure level Fig. 13 shows the difference of the single event sound exposure level LE;A and the maximum sound pressure Lp;eq;1 s;max as a function of the minimum vertical angle of arrival of each flight when measuring with the NMT with respect to an omnidirectional microphone. The metrics LE;A and Lp;eq;1 s;max have been calculated as described in Ref. [1] and only the flights flagged as “A” have been considered. In this case study, the difference in Lp;eq;1 s;max measured by the NMT and by an omnidirectional microphone is due to the effects of the ground reflections rather than to the ground-borne sources. This difference increases as the minimum vertical angle of arrival decreases. This is explained by the fact that for flights with a low minimum vertical angle of arrival, the effect of ground reflections is important enough that sometimes it even results in a displacement of the time in which Lp;eq;1 s;max occurs as seen in Fig. 12. As an example, the minimum angle of arrival of the first event in Fig. 12 is 46:51 and the difference of the Lp;eq;1 s;max measured by the NMT with respect to an omnidirectional microphone is of 2.5 dB. In the case of the second event, the minimum angle of arrival is 63:31 and the difference in Lp;eq;1 s;max is 1.5 dB. Regarding LE;A the difference also increases as the minimum vertical angle of arrival decreases. Also in this case, for the present case study, the effect of the ground reflections is mainly responsible for this difference. As can be seen in Fig. 12, after the vertical angle of arrival of 751 is reached, the curve Lp;A;eq;0:5 s corresponding to the omnidirectional microphone has substantially higher values than that of the NMT due to the effect of the ground reflections. The difference between both curves increases as the angle of arrival of the aircraft noise decreases. Only towards the end of the event does the difference between both curves increase due to the ground-borne noise suppression of the NMT. As an example, for the first event in
Fig. 13. Increase of the single event sound exposure level LE;A and the maximum sound pressure Lp;eq;1 s;max as a function of the minimum vertical angle of arrival of the flight when measuring with the NMT with respect to a single microphone. Only the flights flagged as “A” have been considered.
90
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
Fig. 12, the difference of the LE;A measured by the NMT with respect to an omnidirectional microphone is 2.6 dB, and in the case of the second event it is 1.8 dB.
5. Conclusions This study presents a concept of a NMT that provides: an estimate of the aircraft sound DOA that is used here for automatic aircraft noise events identification; a background and ground reflected noise reduction with respect to a single microphone; and an indicator of the quality of the sound pressure level measurement. This NMT consists of a compact microphone array of nine omnidirectional microphones divided, for signal processing purposes, into two sub-arrays. The microphones are arranged in a tridimensional geometry with a radius of approximately 1 m. The placement of the compact array must guarantee that all ground-borne noise and aircraft noise reflections impinge on it from below. The described NMT is tested in this study at the roof top terrace of a 25 m tall building 500 m away from the end of the runway RNW16 of the Zurich airport; the goal is to monitor aircraft taking off from this runway. The experimental results show several findings. First, all the noise events caused by the target aircraft are correctly identified based on the evolution of the horizontal angle estimates. Other noise events such as those generated by a train passage or the fly-over of a non target aircraft are not misidentified. This shows that this NMT can operate without requiring radar or flight plan data. Second, those aircraft noise events that may have been disturbed by non target sources are correctly flagged. Those events identified as aircraft related but flagged as potentially disturbed by other sources can be inspected off-line by an expert to decide whether to use the sound pressure level data or not. Third, the background noise suppression is between 0.8 dB and 7.6 dB depending on the vertical angle of arrival of the aircraft sound. Fourth, the effect of the ground reflections is largely suppressed. The ground reflections, if not suppressed, are in this case study responsible for overestimations of Lp;eq;1 s;max up to 3.2 dB, and of LE;A up to 2.8 dB. The experimental results also indicate that the system has limitations. The first one is a limited reduction of the ground reflected sound and the ground-borne noise impinging with a vertical angle of arrival close to 901, when the aircraft vertical angle of sight is close to 901. To obtain good reduction in these situations, directivity patterns between hypercardioid and figure-of-eight are optimal. These directivity patterns attenuate considerably the direct aircraft sound and, although this attenuation is compensated later, diminish dramatically the signal-to-noise ratio between the direct aircraft sound and the self noise and wind-induced noise of the microphones. To avoid this, non-optimal directivity patterns need to be used for angles of arrival of the aircraft noise over 801 limiting the reduction of background and ground reflected sound in this angular range. The second limitation is a higher sensitivity to wind-induced noise than a single microphone. The amplification of the wind-induced noise occurs when the distance of a microphone pair of the second sub-array is much smaller than the wavelength of the sound. However, this problem could be mitigated by using one more microphone so that three microphone pairs instead of two are used to cover the whole frequency range.
Acknowledgments The work presented here is part of the project SOUNDTRACK funded by the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 299620. Thanks to U.P. Svensson at NTNU and R. Pieren, J.M. Wunderli and K. Heutschi at Empa for their critical comments on this manuscript. Thanks to D.M. Arthur for proof reading this manuscript (and many others).
References [1] ISO 20906:2009(E), Acoustics—unattended monitoring of aircraft sound in the vicinity of airports, 2009. [2] B. Schäffer, R. Bütikofer, S. Pluss, G. Thomann, Aircraft noise: accounting for changes in air traffic with time of day, Journal of the Acoustical Society of America 129 (1) (2011) 185–199. [3] C. Asensio, M. Ruiz, M. Recuero, Real-time aircraft noise likeness detector, Applied Acoustics 71 (2010) 539–545. [4] M. Tarabini, G. Moschioni, C. Asensio, D. Bianchi, B. Saggin, Unattended acoustic events classification at the vicinity of airports, Applied Acoustics 89 (2014) 91–98. [5] G. Moschioni, B. Saggin, M. Tarabini, The use of sound intensity for the determination of the acoustic trajectory of aircrafts, International Congress on Acoustics, Vol. 2007, May 2007, pp. 2591–2596. [6] R. Bütikofer, G. Thomann, Uncertainty and level adjustments of aircraft noise measurements, INTER-NOISE and NOISE-CON Congress and Conference Proceedings, Vol. 2009, July 2009, pp. 483–490. [7] M.M. Boone, T.C. van den Dool, P. Sijtsma, U. Stelwagen, Microphone array technology for aircraft noise measurements, Internoise 2000, Nice, France, 2000, pp. 28–30. [8] A. Mast, M.M. Boone, T.C. van den Dool, An experimental microphone antenna array for measuring aircraft noise, Internoise 2001, The Hague, Netherlands, 2001, pp. 2115–2120. [9] M. Genescà, J. Romeu, T. Pàmies, A. Sánchez, Real time aircraft fly-over noise discrimination, Journal of Sound and Vibration 323 (2009) 112–129. [10] M. Genescà, J. Romeu, R. Arcos, S. Martín, Measurement of aircraft noise in a high background noise environment using a microphone array, Transportation Research Part D 18 (2013) 70–77. [11] Rion (Japan), Environmental sound monitor (for aircraft noise measurement). 〈http://svmeas.rion.co.jp/download/catalog/NA-37#NA-37〉, 2014.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
91
[12] I.L.N. Granien, R.T. Randeberg, H. Olsen, Next generation aircraft noise models - needs and requirements, Internoise 2010, Lisboa, Portugal, 2010, pp. 1–8. [13] B. Schaffer, C. Zellmann, W. Krebs, S. Pluss, K. Eggenschwiler, R. Butikofer, J. Wunderli, Sound source data for aircraft noise calculations state of the art and future challenges, Euronoise 2012, Prague, Czech Republic, 2012, pp. 589–594. [14] F. van den Berg, A.R. Eisses, P.J.G. van Beek, A model based monitoring system for aircraft noise, Acoustics 2008, Paris, France, 2008, pp. 881–884. [15] C. Zellmann, J.M. Wunderli, B. Schaffer, sonAIR - data acquisition for a next generation aircraft noise simulation model, Internoise 2013, Innsbruck, Austria, 2013, pp. 1–9. [16] M. Genescà, U.P. Svensson, G. Taraldsen, Estimation of aircraft sound direction of arrival using directional-microphone arrays, Acta Acustica united with Acustica 101 (1) (2015) 113–123. [17] M. Genescà, U.P Svensson, G. Taraldsen, Estimation of aircraft location using a directional-microphone array - an experimental study, Journal of the Acoustical Society of America 137 (4) (2015) 1914–1922. [18] G.W. Elko, A.N. Pong, A steerable and variable first-order differential microphone array, IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 1, 1997, pp. 223–226.
Contents lists available at ScienceDirect
Journal of Sound and Vibration journal homepage: www.elsevier.com/locate/jsvi
Directional monitoring terminal for aircraft noise M. Genescà a,b,n a Acoustics Research Center, Department of Electronics and Telecommunications, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway b Laboratory of Acoustics/Noise Control, Empa - Swiss Federal Laboratories for Materials Science and Technology, 8600 Dübendorf, Switzerland
a r t i c l e i n f o
abstract
Article history: Received 7 October 2015 Received in revised form 14 March 2016 Accepted 4 April 2016 Handling Editor: P. Joseph Available online 19 April 2016
This paper presents a concept of an aircraft noise monitoring terminal (NMT) that reduces background noise and the influence of ground reflection, in comparison with a single microphone. Also, it automatically identifies aircraft sound events based on the direction of arrival of the sound rather than on the sound pressure level (or radar data). And moreover, it provides an indicator of the quality of the sound pressure level measurement, i.e. if it is possibly disturbed by extraneous sources. The performance of this NMT is experimentally tested under real conditions in a measurement site close to Zurich airport. The results show that the NMT unambiguously identifies the noise events generated by the target aircraft, correctly detects those aircraft noise events that may be disturbed by the presence of other sources, and offers a substantial reduction in background and ground reflected sound. & 2016 Elsevier Ltd. All rights reserved.
Keywords: Aircraft noise monitoring Microphone array
1. Introduction Airport noise monitoring systems and prediction models are the two main tools for airport noise assessment and management. Prediction models are necessary to forecast the airport's acoustic footprint in future or hypothetical situations involving, for example, increases in flight volume or the modification of flight paths. Noise monitoring systems are mainly used to inform the public about the current airport noise situation and also to validate the output of the prediction models and reinforce its credibility in the eyes of the public. The noise monitoring systems include a network of noise measurement terminals (NMTs) distributed within the acoustic area of influence of the airport. These NMTs are typically equipped with an omnidirectional microphone that continuously monitors the noise levels at the site. Since this data is used for public information, the placement of these terminals is chosen so that the measured noise levels are representative of the acoustic influence of the airport on the population. Therefore they are located in urban, or at least inhabited, areas where other sound sources such as traffic or industrial noise may be present. This NMTs placement has two main disadvantages: First, the continuous sound pressure level recording contains noise events that may not be generated by aircraft. Therefore strategies to identify the aircraft noise events need to be put in place. The ISO standard 20906:2009(E) [1] describes the event identification process. It starts with a detection stage in which the parts of the recording that exceed a given threshold by a specified amount, for a duration which lies within specified limits, are detected. Next, it follows a n Correspondence address: Acoustics Research Center, Department of Electronics and Telecommunications, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway. E-mail address: [email protected]
http://dx.doi.org/10.1016/j.jsv.2016.04.004 0022-460X/& 2016 Elsevier Ltd. All rights reserved.
78
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
classification stage in which acoustical properties, that may help to separate non-aircraft sound events from aircraft events, of the detected events are analyzed. It finishes with an identification stage in which non-acoustical information such as radar or flight plan data is used to finally confirm whether an event is generated by an aircraft. The fact that the detection is based on a sound pressure level threshold causes a number of aircraft noise events to be missed in locations where the level of background noise is comparable to the aircraft noise level [2]. Moreover, the event identification error rate grows considerably in airports that do not have a radar data connected to the noise monitoring system because the final identification stage cannot be implemented. To overcome this first disadvantage, several algorithms to automatically classify noise events caused by aircraft have been developed [3–5]. The second disadvantage of the NMTs placement is that noise events caused by other sources may occur simultaneously to the aircraft noise events. In such a case the sound pressure level of the aircraft may be overestimated. As a consequence, NMTs cannot be used solely as a tool to penalize an airline when a particular flight exceeds the legal noise limits. Moreover, interference effects due to ground reflections are undesirable in aircraft noise monitoring applications because they depend on the height of the receiver and introduce uncertainty in the measurement [6]. To reduce these effects the ISO standard 20906:2009(E) [1] describes an ideal measurement site; but such sites are not always available. To overcome these disadvantages, the existing literature [7–10] provides examples of advanced NMTs that are integrated by a linear microphone array and that use beamforming to reduce the contribution of ground-borne sources and aircraft ground reflections on the continuous sound pressure level recording. Such terminals typically consist of a linear array of 12– 15 microphones with a length of around 3 m and provide a reduction of the ground-borne noise level between 5 dB and 10 dB. The NMT concept presented in this paper is also aimed at overcoming these disadvantages and consists of an array of nine microphones distributed in a 3D geometry with a diameter of approximately 1 m. This NMT provides an estimate of the direction of arrival (DOA) of the aircraft sound and a measurement of the aircraft sound pressure level (frequency spectrum and overall level) with a reduced contribution of the background noise and ground reflections compared to an omnidirectional microphone measurement. As in the case of the noise monitoring terminal NA-37 developed by RION [11], the sound DOA estimates are used for automatic aircraft event identification purposes so that non-acoustical information such as radar or flight plan information is not required. Thus it can be used in small airports without radar and by external (to the airport) entities or administrations. In addition, the DOA estimates provided by the NMT presented here are robust against ground reflections and, to an extent, to the presence of ground-borne noise sources. This feature is used to determine the quality of the measurement, i.e. if sources other than the target aircraft may have influenced the measurement. Furthermore, such a NMT can be used to obtain reliable aircraft spectral sound power data of individual flights. This data is required by the new generation of prediction models to be able to calculate the acoustic impact of individual flights [12,13]. This would allow the design of noise abatement operational procedures affecting not only flight paths but also the aircraft flight configuration (e.g. slats, flaps, landing gear and power settings). Initiatives to use NMTs integrated by single omnidirectional microphones as a basis to calculate aircraft spectral sound power data are reported in the literature [14]. However, they require the placement of NMTs in low background noise locations using them purely as a source of data for prediction models and removing their utility as a source of public information. The NMT presented here can be used in measurement campaigns dedicated to the collection of sound power spectral data such as those described in the literature [13,15] or as a permanent installation as part of the noise monitoring system of an airport. If at least two of these NMTs are used, the position of the aircraft could be estimated by triangulation. Once the aircraft position and the associated sound pressure levels at a set of NMTs are available, the aircraft sound power spectral data could be calculated by back propagating to the source. The remainder of this manuscript is structured as follows: In Section 2 the different modules of the NMT are described; Section 3 presents a case study used here to show the features of such a NMT; In Section 4 the results provided by the NMT
Fig. 1. Convention used to define the vertical θ and horizontal ϕ angle of arrival of the aircraft noise. The origin of coordinates represents the position of the array.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
79
in the context of the case study are compared to those obtained with a single omnidirectional microphone; Section 5 summarizes the main findings of the study.
2. Noise monitoring terminal The NMT presented here includes a compact microphone array of nine omnidirectional microphones divided, for signal processing purposes, in two sub-arrays. The placement of the compact array must guarantee that all ground-borne noise and ground reflections of the aircraft noise impinge on it from below, e.g. on a roof top terrace. The NMT consists of three main modules: the DOA estimation module; the directional monitoring module; and the event identification module. There are several outputs from the NMT. The first one, from the DOA estimation module, is a time series of aircraft noise DOA ^ ðτÞ, and defined as in Fig. 1. The second output, from the directional monitoring estimates, represented here by θ^ ðτÞ and ϕ module, is a time series of T seconds equivalent continuous sound pressure level measurements Lp;eq;T ðτÞ and the corresponding frequency spectra Pð f ; τÞ. The third output, from the event identification module, are the start time τistart and stop time τistop for each aircraft noise event i along with a flag indicating the quality of the measurement. Fig. 2 shows the data flow between these three modules.
Fig. 2. Data flow between the three modules that comprise the NMT.
Fig. 3. Microphone distribution in the NMT. The meaning of the four symbols used is the following: The solid black circles represent the omnidirectional microphones that belong exclusively to the first sub-array; The solid grey circle represents the omnidirectional microphone that belongs exclusively to the second sub-array; The grey circles with a black outline represent the microphones that belong to both sub-arrays; The cardioid shapes show the position and orientation of the virtual first-order microphones generated from the signals of the vertically aligned omnidirectional microphone pairs. All dimensions are in meters.
80
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
2.1. DOA estimation module As seen in Fig. 2, this module uses as its only input the signals of the microphones of the first sub-array. The first subarray consists of eight omnidirectional microphones depicted as black circles and as grey circles with a black outline in Fig. 3. Each of these eight microphones is located 0.15 m above or below one of the vertices of a regular tetrahedron of side length 0.85 m. The DOA estimation algorithm has been described in detail [16] and tested previously [17] and is not the focus of the present paper, thus it will not be discussed here. However, the main idea is to use each of the vertically aligned microphone pairs in the first sub-array to virtually generate a first-order directional microphone at each vertex of the regular tetrahedron, as shown in Fig. 3. A first-order microphone has the directivity gain DðθÞ that follows DðθÞ ¼ ð1 bÞ þ b cos θ;
(1)
where b A ½01 is a factor that determines the specific directivity pattern, e.g. b¼ 0.5 corresponds to a cardioid, b¼0.75 to a hypercardioid. As a first step in the DOA estimation algorithm, the specific directivity pattern of the four virtual first-order directional microphones is selected, according to the geometry of the measurement site, with the aim to reduce the influence of ground reflections which typically introduce error in the estimated vertical angle of arrival. Then, the time difference of arrival between all possible pairs of virtual microphones is estimated. Finally the DOA is estimated by exploiting the relationship between the time differences of arrival and the DOA. The accuracy of the vertical angle of arrival estimate depends mainly on the optimal choice of the directivity pattern of the virtual microphones according to the source–receiver geometry, and to a lesser extent on the ground properties [16,17]. In this module only the sound frequency content between 20 Hz and 200 Hz is considered. Due to the atmospheric attenuation most of the aircraft sound energy on the ground is concentrated below 200 Hz, and below 20 Hz the signal can be heavily influenced by wind. It is worth clarifying that when the aircraft sound is predominant in this frequency range, the vertical angle estimate provided by this module corresponds to the vertical angle of arrival of the aircraft sound. However, when ground-borne sources are predominant the vertical angle will be higher than 901 but may differ considerably from the real vertical angle of arrival of the ground-borne noise. This is due to the constrained search domain of the time difference of arrival estimates introduced to increase precision when aircraft sound is present [16]. 2.2. Directional monitoring module The goal of this module is to measure the aircraft sound pressure levels reducing the influence of the ground-borne sound and aircraft ground reflections with respect to an omnidirectional measurement. The inputs to this module are the discrete-valued signals pn(t) of the microphones n in the second sub-array, and the vertical angle of arrival estimates θ^ ðτÞ provided by the DOA estimation module. The second sub-array consists of three microphones vertically aligned which are represented by the solid grey circle and the two grey circles with a black outline in Fig. 3. The distance between the upper microphone pair is 0.30 m, and the distance between the lower pair is 0.03 m. Here the signals of these three microphones are processed to virtually obtain the output of a first-order directional microphone. The specific directivity pattern of such a microphone can be adapted to the geometry of the measurement site to reduce the influence of ground reflections and/or ground-borne sources. In the following paragraphs the signal processing algorithm is detailed. Let P n ð f ; τÞ and P m ð f ; τÞ be the discrete Fourier transform of the higher and lower microphone signals respectively of the microphone pair separated by d ¼0.30 m or alternatively of the microphone pair separated by d ¼0.03 m.T The discrete Pτ þ Fourier transform of a signal frame of length T received at microphone n is defined here as P n ð f ; τÞ ¼ t ¼ 2τ T pn ðt Þe j2π ft 2 where j is the imaginary unit. The output of a virtual first-order directional microphone Pð f ; τÞ is obtained as follows [18] P n ð f ; τ Þ þ P m ð f ; τÞ P n ð f ; τÞ P m ð f ; τÞ þb ; P ð f ; τÞ ¼ ð1 bÞ 2 jkd
(2)
where k is the wavenumber. P jkd cos θs If θs is the true value of the vertical angle of arrival of the source s, then P n ð f Þ ¼ s P s ð f Þe 2 þ Nn ð f Þ and P s jkd cos θs s 2 þ Nm ð f Þ where P ð f Þ is the Fourier transform of the signal from the source s at the mid point P m ð f Þ ¼ s P ð f Þe between n and m, and Nn(f) is the additive noise of microphone n (e.g. microphone self noise or wind-induced noise). Taking into account these expressions, Eq. (2) equals X s s P ð f ; τÞDð f ; θ ðτÞÞþ Nð f ; τÞ; (3) Pð f ; τÞ ¼ s
where Dð f ; θÞ is the directivity gain of the virtual directional microphone 2b sin kd cos θ=2 ; D f ; θ ¼ ð1 bÞ cos kd cos θ=2 þ kd
(4)
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
81
and Nð f ; τÞ is the additive noise of the virtual directional microphone Nð f ; τÞ ¼ ð1 bÞ
Nn ð f ; τÞ þ Nm ð f ; τÞ N n ð f ; τ Þ N m ð f ; τÞ þb : 2 jkd
(5)
If d⪡λ, where λ ¼ 2kπ is the wavelength, then Eq. (4) simplifies to Eq. (1), meaning that the sound pressure is measured with a first-order directivity filter. At the same time, if the separation between the microphone pair is much smaller than the wavelength, the second term on the right of Eq. (5) increases, also increasing the additive noise of the virtual first-order microphone. The frequency range of interest is from 88 Hz to 2818 Hz (octave bands from 125 Hz to 2 kHz) because the Aweighted aircraft sound pressure levels measured on the ground are quite low outside of this range [7,9]. The microphone pair with d ¼0.30 m is used to generate a virtual first-order microphone for frequencies from 88 Hz ðλ 3:8 mÞ to 355 Hz ðλ 1 mÞ, while the microphone pair with d ¼0.03 m is used for the remaining frequencies up to 2818 Hz ðλ 0:1 mÞ. In that way, the distance d is much smaller than the wavelength to get a first-order directivity pattern, but not so small so as to result in high additive noise amplification. The T seconds equivalent continuous sound pressure level Lp;eq;T ðτÞ for a particular frequency range is calculated as 0 ! 1 f2 ^ ðτÞÞj 2 X 2 jPð f ; τ Þ=Dð f ; θ A; (6) Lp;eq;T ðτÞ ¼ 10 log@ 2 Po Q f ¼ f1 where P o ¼ 2 10 5 Pa is the reference pressure value, and Q is the number of samples in T seconds of the signal pn(t). Note that the sound pressure spectrum of the virtual first-order microphone Pð f ; τÞ is divided by the directivity gain associated with the vertical angle estimate θ^ ðτÞ provided by the DOA estimation module. This is done to compensate for the attenuation imposed on the aircraft sound by the directivity pattern of the virtual first-order microphone. Assuming that the sources s in Eq. (3) are the target aircraft a and ground-borne sources gb (urban noise and ground reflections), this division can be written as P a gb P a ð f ; τÞDð f ; θ ðτÞÞ þ gb P gb ð f ; τÞDð f ; θ ðτÞÞ þ Nð f ; τÞ Pð f ; τÞ ¼ : (7) Dð f ; θ^ ðτÞÞ Dð f ; θ^ ðτÞÞ As described in Section 2.1, if the aircraft sound is the predominant source in the range within 20 Hz and 200 Hz then
θ^ ðτÞ C θa ðτÞ and therefore
XDð f ; θ ðτÞÞ gb Pð f ; τÞ Nð f ; τÞ CP a ð f ; τÞ þ P ð f ; τÞ þ : a a Dð f ; θ ðτÞÞ Dð f ; θ ðτÞÞ Dð f ; θ^ ðτÞÞ gb
(8)
gb
Under these circumstances it is also true that Dð f ; θ ðτÞÞ o Dð f ; θ ðτÞÞ, since the virtual microphones are vertically oriented as shown in Fig. 3. Therefore the noise from ground-borne sources is attenuated in comparison to the case of using an omnidirectional microphone. As explained in Section 2.1, when aircraft noise is not present it is not necessarily true that θ^ ðτÞ ¼ θgb ðτÞ and therefore the term in Eq. (7) corresponding to the ground-borne sources can even be amplified. However, this is not relevant since, for aircraft noise monitoring purposes, the interest lies only in the time intervals when an aircraft is detected. gb
a
Fig. 4. Upper and lower limit of the aircraft sound pressure level uncertainty caused by a vertical angle estimation uncertainty of ½ 4:41 4:41. These limits are plotted as a function of θa for different directivity patterns.
82
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
As also seen in Eq. (8), the noise term is amplified by a factor 1=Dð f ; θ ðτÞÞ. Regardless of the specific first-order a directivity pattern, Dð f ; θ ðτÞÞ is minimum when θa is close to 901, when a stronger influence of the microphone additive noise is expected. a As Eq. (8) relies on θ^ ðτÞ C θ ðτÞ it follows that the uncertainty in the vertical angle estimation translates into sound pressure level uncertainty in Eq. (6). In previous experimental tests [17] of the DOA estimation module it has been found a that θ^ ðτÞ θ ðτÞ A ½ 4:41 4:41. As an example, Fig. 4 shows the corresponding upper and lower limit of the aircraft sound a pressure level uncertainty as a function of θa for different first-order directivity patterns assuming θ^ ðτÞ θ ðτÞ A ½ 4:41 4:41. a
2.3. Event identification module The goal of this module is to identify the events i caused by target aircraft, and to provide a time stamp for the beginning
τistart and the end τistop of each event along with a flag indicating the quality of the measurement; this is done in three stages.
The first stage is the event detection in which the aircraft events are detected by analyzing the evolution of the aircraft ^ ðτÞ provided by the DOA estimation module. The second stage is the signal segmentation sound DOA estimates θ^ ðτÞ and/or ϕ in which the starting time of the event τistart is determined also based on the analysis of the sequence θ^ ðτÞ. The final stage is ^ ðτÞ and the the assessment of the measurement quality in which τistop is also determined. This is done by analyzing θ^ ðτÞ, ϕ evolution of Lp;eq;T ðτÞ provided by the directional monitoring module. As the specifics of the detection algorithm are largely dependent on the particular geometry of the measurement site they will be discussed in Section 3.3 in the context of the case study presented.
3. Measurement set-up In the present case study, the NMT is located 500 m away from the end of the runway RNW16 of the Zurich airport. The target aircraft are those taking off from this runway. Fig. 5 shows the NMT position (as a black dot) with respect to the runway (grey band), as well as the radar trajectories of 50 aircraft that took off during the measurement campaign carried out from 12:00 to 14:00 on the 15th of May and from 9:00 to 16:00 on the 16th of May 2014. Some of the aircraft make a turn of approximately 901 around the NMT, while some others make a turn of approximately 2701 and so are heard for a longer period of time. The vertical angle of sight of the aircraft with respect to the NMT ranges between 451 and 901. The NMT was placed on the roof top terrace of a 25 m tall building. The base of the tetrahedron depicted in Fig. 3 was at a height h ¼2.4 m over the flat terrace floor which was covered in wet gravel. Fig. 6 shows the array in the measurement site. The measurement site had a high background noise due to the motorway, tram and train line that can be seen in Fig. 6. The measured average background noise level considering only the traffic noise was 66.4 dB(A). The traffic and rail noise impinges on the microphone array with a vertical angle of arrival of approximately 1001. Six of the microphones of the first sub-array were B&K type 4190 and two were B&K type 4165, the microphone that belonged exclusively to the second sub-array was a Norsonic type 1225. All the preamplifiers were Norsonic type 1202 and were connected to Norsonic type 336 amplifiers that provide 200 V DC to the microphones. The microphone signals were acquired with three National Instruments acquisition cards type NI9234 mounted on a National Instruments cDAQ-9174
Fig. 5. Plot of the radar trajectories of 50 aircraft that took off from Runway RNWY16 of Zurich airport during the measurement campaign. The black dot represents the NMT terminal located 500 m away from the end of the runway (grey band).
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
83
Fig. 6. Microphone array in the measurement site.
chassis. The sampling frequency was set to 50 kHz. Prior to the measurement campaign, a B&K type 3541 calibrator for intensity probes was used to measure the transfer function between one reference channel and the others. These transfer functions were stored and used later to compensate for the phase and amplitude differences between channels. The wind speed measured at 6 m above the ground, by the airport weather station, was between 0 m/s and 4 m/s during the 15th of May, and between 3 m/s and 6 m/s, during the 16th of May. The average temperature measured at 2 m above the ground was 9 °C and the humidity was 70 percent the 15th of May and 12 °C and 55 percent the 16th of May.
3.1. DOA estimation module settings The DOA estimation algorithm has been implemented as described in Ref. [17] to provide a DOA estimate for each successive non-overlapping signal frame of length T ¼0.5 s. The factor b that describes the directivity of the virtual firstorder microphones generated by each vertically aligned microphone pair in the first sub-array has been set to b¼0.65. This is an appropriate choice according to the simulations presented in Ref. [16] for the expected range of vertical angles of sight of the aircraft with respect to the array.
3.2. Directional monitoring module settings The relevant setting regarding the directional monitoring module is the value of b. Assuming that the directivity pattern of the virtual first-order microphones created by the second sub-array follows the ideal equation Eq. (1), the goal is to gb a minimize the ratio Dðθ Þ=Dðθ Þ in Eq. (8). A straightforward solution is to choose the first-order directivity pattern whose gb zero directivity gain falls on θ . In this case study there are two types of ground-borne sources: the traffic/rail noise; and the ground reflections of the aircraft sound. For this particular measurement site, in spite of the high sound pressure levels of the traffic noise, the aircraft sound tends to prevail over the traffic due to the proximity of the measurement point to the aircraft flight path. Therefore, the factor b has been adaptively calculated so that the zero in the directivity gain falls on the DOA of the ground reflected noise. The described choice of b has only been applied while θ^ ðτÞ o 801. For 801 r θ^ ðτÞ r901, the a resulting values of Dðθ^ Þ Dðθ Þ are too low and, as predicted by Eq. (8), the additive noise would be largely amplified. Furthermore, in this range, values of b close to 1 need to be avoided to not raise the sound pressure level uncertainty (over 2 dB in the example in Fig. 4). Therefore, in this range, the factor b has been chosen adaptively so that Dðθ^ ðτÞÞ ¼ 0:5. For
84
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
Fig. 7. Flow diagram showing the stages (dashes boxes) and the steps (numbered blocs) involved in the event identification algorithm.
901o θ^ ðτÞ b is set to 0.5. In summary, b is defined here as 8 1 > > ; > > > > ð1 cos ð1801 θ^ ðτÞÞÞ > < 0:5 b¼ ; > > > ð1 cos θ^ ðτÞÞ > > > > : 0:5;
if θ^ ðτÞ o 801 if 801 r θ^ ðτÞ r901 if θ^ ðτÞ 4 901:
(9)
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
85
The directional monitoring module has also been configured to provide Lp;eq;T ðτÞ and the associated frequency spectrum Pð f ; τÞ for each successive non-overlapping signal frame of length T¼0.5 s.
3.3. Event identification module settings Fig. 7 shows, as a flow diagram, the stages and steps involved in the event identification algorithm. The first stage is the event detection and aims to detect aircraft noise events. As seen in Fig. 5, the target aircraft approach the NMT from the northwest and depart to the northeast (some of them turn to the west afterwards). This particular pattern is used here to unambiguously detect the presence of target aircraft as none of the other ground- or air-borne sources (aircraft departing from other runways) present a similar movement. Therefore, as shown in Fig. 7, the first step of the event ^ ðτÞ provided by the DOA identification algorithm consists of reading the value of the horizontal angle of arrival estimate ϕ estimation module. This angle is defined as shown in Fig. 1 within the interval ½ 1801 1801, meaning that when the aircraft first crosses the west axis in Fig. 1, the horizontal angle of arrival jumps from 1801 to 1801. Thus, the second step is to ^ ðτÞ with the previous one ϕ ^ ðτ TÞ to see if there is a relevant jump. If the jump is compare the most recent sample ϕ confirmed, the presence of an aircraft is unambiguously confirmed at the detection time τidetect. The second stage is the segmentation which is aimed at finding the initial time τistart of the aircraft sound event i that has just been detected. It includes the third and fourth steps in Fig. 7 which search for the closest time instant previous to τidetect when the vertical angle estimate is still lower than or equal to 901. This is the time τistart at which the DOA estimation module starts detecting an airborne source. The final stage is the quality assessment and aims to determine the final time τistop of the aircraft sound event i, and to collect data about the quality of the sound measurement of this event. Steps five to ten are repeated recursively over the subsequent time frames τo ¼ τistart þ zT (z¼ 0,1,2,3…) until τistop is found. Steps five to seven use the sound pressure level measurement provided by the directional monitoring module to calculate the equivalent sound pressure level over the five seconds surrounding the analyzed time τo. If this equivalent sound pressure level is 1 dB below the average background noise level τistop is reached. If this is the case, the algorithm returns “A” as a flag indicating that the measurement is undisturbed. Otherwise, step eight checks if at the analyzed time the vertical angle of arrival estimate is higher than 901. If this is the case τistop is reached and the returned flag “D” means that the measurement is potentially disturbed by the presence of a ground-borne source. If this is not the case, step nine checks if the next detection is within the following 30 s. If it is, τistop is reached and the returned flag “B” means that the measurement is potentially disturbed by the take-off of the next plane. Otherwise, the tenth step checks if at the analyzed time τo there is an irregular decrease in the slope of the sequence of horizontal angle of arrival estimates. This indicates that another airborne source is present. In this case, τistop is reached and the flag “C” indicates that the measurement is potentially disturbed by other air-borne sources. Finally, step eleven checks the duration of the event from τistart to τistop and if it is too short it overwrites the existing flag with “F”, indicating that the sound pressure level measurement is not reliable for an unknown reason.
Fig. 8. Example of the output of the NMT. The first and second plots correspond to the estimate of the vertical and horizontal angle of arrival, the third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels of the NMT (in black) and of a single microphone (in grey). The dashed lines mark the initial τistart and final τistop time of the detected event which has been flagged as “A” (undisturbed). There is a second event generated by the passage of a train.
86
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
4. Results The aircraft noise monitoring features of the NMT are the main focus of the present study, therefore the results from the DOA estimation module are not discussed here. A detailed analysis of the performance of this module can already be found in the literature [16,17]. 4.1. Event identification Figs. 8–10 show examples of the output of the NMT developed here compared to the information obtained from an omnidirectional microphone. The first and second plots in all these figures correspond to the estimate of the horizontal and vertical angle of arrival respectively. The third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels Lp;A;eq;0:5 s provided by the directional monitoring module compared to the that of a single microphone. The dashed lines mark the initial τistart and final τistop time of the detected events provided by the event identification module. The flag assigned to the event by this module is also shown.
Fig. 9. Example of the output of the NMT. The first and second plots correspond to the estimate of the vertical and horizontal angle of arrival, the third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels of the NMT (in black) and of a single microphone (in grey). The dashed lines mark the initial τistart and final τistop time of the detected events which has been flagged as “A” (undisturbed) and “C” (potentially disturbed by another airborne source) respectively.
Fig. 10. Example of the output of the NMT. The first and second plots correspond to the estimate of the vertical and horizontal angle of arrival, the third plot shows the A-weighted 0.5 s equivalent continuous sound pressure levels of the NMT (in black) and of a single microphone (in grey). The dashed lines mark the initial τistart and final τistop time of the detected events which has been flagged as “A” (undisturbed) and “F” (sound pressure measurement not reliable for unknown reasons) respectively.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
87
^ ðτÞ that allows for In the third plot of Fig. 8 two acoustic events can be seen. For the first event there is a clear jump in ϕ ^ ðτÞ indicating that it is not an the event to be identified as aircraft-generated. For the second event there is no jump in ϕ aircraft event. Indeed it corresponds to the passage of a train while the aircraft was still distinguishable. If only the acoustic information provided by an omnidirectional microphone is available, an automatic identification of these events is not possible and radar data is needed for clarification, whereas with the presented NMT these events are automatically identified. Note also, that θ^ ðτÞ tends to 901 at the time of the second event which is a good indicator that this is likely a groundborne source. Fig. 9 shows two consecutive detected aircraft events flagged “A” (undisturbed) and “C” (potentially disturbed by another ^ ðτÞ which indicates that there airborne source). The “C” flag of the second event is due to the sudden change of the slope of ϕ is a non-target source interfering with the estimation of the horizontal angle of arrival of the target source. Moreover, at the time of this disturbance, θ^ ðτÞ is still well below 901 indicating that the interfering source is an airborne one. Actually, the interfering source is the aircraft that generated the first acoustic event whose trajectory contains a 2701 turn (see Fig. 5) and is heard again when flying west. The third plot in Fig. 10 shows three acoustic events (around times 60 s, 120 s, and 145 s); only the first and third are identified as aircraft events. The first is flagged “A”, but the third is flagged “F” indicating that the sound pressure measurement is not reliable for unknown reasons. At the time of the second event, θ^ ðτÞ is higher than 901 indicating that this event is caused by a ground-borne source. Indeed, this event corresponds to the passage of a train overlapped to the beginning of the third event. The third event is identified as an aircraft but shortly after the detection time, θ^ ðτÞ becomes greater than 901 indicating that the end of the event is reached because a ground-borne source is detected. This happens because of the presence of the train. As explained in Section 3.3 such a situation should be flagged as “D” (potentially disturbed by a ground-borne source) but since there are less than 15 s between the start and the end of the third event the flag “F” is used instead. Note that an important advantage of this NMT is the fact that, even when the sound pressure measurement is flagged as potentially disturbed by other sources, the targeted aircraft events are still detected and identified. This means that, if desired, these events can be inspected off-line by an expert and a final decision could be made regarding whether to use or not the sound pressure data. In this particular case study, all the aircraft events have been satisfactorily identified. Out of the 50 recorded flights, 45 have been flagged as “A” (undisturbed), three have been flagged as “C” (potentially disturbed by another airborne source) and two as “F” (sound pressure measurement not reliable for unknown reasons). 4.2. Directional monitoring In this section the results related to the directional monitoring module are presented and discussed. 4.2.1. Analysis of the background noise suppression Fig. 11 shows the difference between the background noise levels measured by the NMT and those measured by an omnidirectional microphone. The difference is shown as a function of the frequency and as a function of the vertical angle of arrival of the aircraft sound. These values have been calculated by averaging the sound pressure level differences over 314
Fig. 11. Difference between the average background noise levels measured by the NMT and by an omnidirectional microphone as a function of the frequency and the vertical angle of arrival of the aircraft sound. Negative values indicate that the levels measured by the NMT are lower than those measured by an omnidirectional microphone, positive values indicate the contrary.
88
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
signal frames of 0.5 s length of continuously recorded background noise (traffic noise and wind-induced noise). To obtain the sound pressure levels corresponding to the NMT, the settings described in Section 3.2 have been applied. Since the recording contains only background noise, the estimates θ^ provided by the DOA estimation module have been disregarded and artificially reset to represent the vertical angle of arrival of the sound of a hypothetical aircraft. For angles higher than 751 it can be seen that at low frequencies (up to 140 Hz) and also between 355 Hz and 400 Hz, the background noise level measured by the NMT is slightly higher than the one measured by an omnidirectional microphone. This is due to the high wind speed during the measurements that induces a high level of noise in the microphone signals. At those frequencies, the distance d between the microphone pairs (d¼0.30 m for 100 Hz and 125 Hz, and d ¼0.03 m for 400 Hz) is quite small compared to the wavelength. Under these circumstances, as described by Eq. (5), any kind of additive noise such as wind-induced noise gets amplified. Moreover, this phenomenon is more acute at vertical angles close to 901 a because in this range Dðθ Þ is smaller and therefore the wind-induced noise is further amplified according to Eq. (8). All these suggest that the NMT presented here is more sensitive to wind-induced noise than an omnidirectional microphone. However, this problem could be solved by using one more microphone in the directional noise monitoring module so that three microphone pairs instead of two are used to cover the whole frequency. The idea is to ensure that the distance d is never too small compared with the wavelength over the whole frequency range. Above 801, excluding the frequencies mentioned above, the background noise measured by the NMT is lower than that measured by an omnidirectional microphone. However, the difference is limited because in this range the restriction Dðθ^ ðτÞÞ ¼ 0:5 has been imposed to avoid further amplification of the additive noise and as a consequence the ratio gb a Dð f ; θ ðτÞÞ=Dð f ; θ Þ in Eq. (8) is not low enough to provide a greater reduction of the background noise. For the angles immediately below 801 the background noise measured by the NMT is more than 15 dB lower than that measured by an omnidirectional microphone. This is because at those angles, the DOA of the ground reflections coincides with that of the traffic/rail noise, and since the directivity pattern of the directional monitoring module is chosen so that the zero in directivity gain falls in the DOA of the ground reflections it also falls in the DOA of the traffic/rail noise. Considering the overall sound pressure levels Lp;A;eq;0:5 s , the difference between the background noise measured by the NMT and by an omnidirectional microphone is within 6:7 dB and 7:6 dB for aircraft noise vertical angles of arrival below 801, and within 3 dB and 0:8 dB for aircraft noise vertical angles of arrival above 801. 4.2.2. Analysis of the spectrogram Fig. 12 shows the spectrogram of a continuous recording including two aircraft events flagged as “A”. As additional information, the top plot shows the Lp;A;eq;0:5 s provided by the omnidirectional microphone (grey line) and by the NMT (black line), and the vertical angle estimate provided by the NMT (black thin line). The middle plot shows the A-weighted spectrogram provided by an omnidirectional microphone, and the bottom plot shows the same provided by the NMT. The effect of the ground reflection can be seen in the omnidirectional microphone spectrogram (middle plot) in the form of a comb filter below 1 kHz, it shows the constructive and destructive interference frequencies. This effect is not present in
Fig. 12. Spectrogram of a continuous recording including two aircraft events flagged as “A”. The top plot shows the Lp;A;eq;0:5 s provided by the omnidirectional microphone (grey line) and by the NMT (black line). It also shows the vertical angle estimate provided by the NMT (black thin line). The middle plot shows the A-weighted spectrogram provided by an omnidirectional microphone and the bottom plot shows the same provided by the NMT.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
89
the NMT spectrogram (bottom plot) due to the attenuation imposed on the reflected wave by the directivity pattern of the virtual first-order microphone. Furthermore, the effect of the ground reflection in the omnidirectional spectrogram cannot be noticed before the aircraft sound impinges on the NMT with a vertical angle of arrival of approximately 751. This is because before this angle is reached, the reflection point that satisfies the law of specular reflection is located outside of the roof terrace where the NMT is placed. As a consequence, it can be seen that the curves Lp;A;eq;0:5 s of both the NMT and the omnidirectional microphone in the top plot are very similar until a vertical angle of arrival of 751 is reached. Afterwards the curve corresponding to the omnidirectional microphones exhibits higher values due mainly to the effect of the ground reflections. As the reflection coefficient of the ground increases as the vertical angle of arrival (as defined in Fig. 1) decreases, the constructive and destructive interference effects are stronger at the point of the flight path at which the vertical angle of arrival is the lowest. In the case of the first aircraft noise event, the maximum in Lp;A;eq;0:5 s happens precisely at that point. This can be seen in the grey curve of the top plot and it is also marked in the omnidirectional spectrogram as t Lmax . In the NMT spectrogram of this same event, in which the ground effect is suppressed, t Lmax happens earlier in time. Therefore, it is shown that the ground reflection may affect both the level and the time of occurrence of the maximum sound pressure level. In the second aircraft noise event the ground reflections have not such an acute effect on the maximum level, probably, due to the fact that minimum vertical angle of arrival is only of 63:31 and thus the ground reflection coefficient stays lower than in the first event in which a minimum vertical angle of arrival of 46:51 is reached. The NMT ground-borne noise suppression shows up as a general lightening of the NMT spectrogram with respect to the omnidirectional microphone one. It is more noticeable towards the end of the event because the aircraft noise contribution gb a decreases, but mainly because at the beginning of the event (vertical angles of arrival over 801) the ratio Dð f ; θ ðτÞÞ=Dð f ; θ Þ is not low enough to provide a greater reduction of the background noise as explained in Section 4.2.1. 4.2.3. Analysis of the maximum sound pressure level and the sound exposure level Fig. 13 shows the difference of the single event sound exposure level LE;A and the maximum sound pressure Lp;eq;1 s;max as a function of the minimum vertical angle of arrival of each flight when measuring with the NMT with respect to an omnidirectional microphone. The metrics LE;A and Lp;eq;1 s;max have been calculated as described in Ref. [1] and only the flights flagged as “A” have been considered. In this case study, the difference in Lp;eq;1 s;max measured by the NMT and by an omnidirectional microphone is due to the effects of the ground reflections rather than to the ground-borne sources. This difference increases as the minimum vertical angle of arrival decreases. This is explained by the fact that for flights with a low minimum vertical angle of arrival, the effect of ground reflections is important enough that sometimes it even results in a displacement of the time in which Lp;eq;1 s;max occurs as seen in Fig. 12. As an example, the minimum angle of arrival of the first event in Fig. 12 is 46:51 and the difference of the Lp;eq;1 s;max measured by the NMT with respect to an omnidirectional microphone is of 2.5 dB. In the case of the second event, the minimum angle of arrival is 63:31 and the difference in Lp;eq;1 s;max is 1.5 dB. Regarding LE;A the difference also increases as the minimum vertical angle of arrival decreases. Also in this case, for the present case study, the effect of the ground reflections is mainly responsible for this difference. As can be seen in Fig. 12, after the vertical angle of arrival of 751 is reached, the curve Lp;A;eq;0:5 s corresponding to the omnidirectional microphone has substantially higher values than that of the NMT due to the effect of the ground reflections. The difference between both curves increases as the angle of arrival of the aircraft noise decreases. Only towards the end of the event does the difference between both curves increase due to the ground-borne noise suppression of the NMT. As an example, for the first event in
Fig. 13. Increase of the single event sound exposure level LE;A and the maximum sound pressure Lp;eq;1 s;max as a function of the minimum vertical angle of arrival of the flight when measuring with the NMT with respect to a single microphone. Only the flights flagged as “A” have been considered.
90
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
Fig. 12, the difference of the LE;A measured by the NMT with respect to an omnidirectional microphone is 2.6 dB, and in the case of the second event it is 1.8 dB.
5. Conclusions This study presents a concept of a NMT that provides: an estimate of the aircraft sound DOA that is used here for automatic aircraft noise events identification; a background and ground reflected noise reduction with respect to a single microphone; and an indicator of the quality of the sound pressure level measurement. This NMT consists of a compact microphone array of nine omnidirectional microphones divided, for signal processing purposes, into two sub-arrays. The microphones are arranged in a tridimensional geometry with a radius of approximately 1 m. The placement of the compact array must guarantee that all ground-borne noise and aircraft noise reflections impinge on it from below. The described NMT is tested in this study at the roof top terrace of a 25 m tall building 500 m away from the end of the runway RNW16 of the Zurich airport; the goal is to monitor aircraft taking off from this runway. The experimental results show several findings. First, all the noise events caused by the target aircraft are correctly identified based on the evolution of the horizontal angle estimates. Other noise events such as those generated by a train passage or the fly-over of a non target aircraft are not misidentified. This shows that this NMT can operate without requiring radar or flight plan data. Second, those aircraft noise events that may have been disturbed by non target sources are correctly flagged. Those events identified as aircraft related but flagged as potentially disturbed by other sources can be inspected off-line by an expert to decide whether to use the sound pressure level data or not. Third, the background noise suppression is between 0.8 dB and 7.6 dB depending on the vertical angle of arrival of the aircraft sound. Fourth, the effect of the ground reflections is largely suppressed. The ground reflections, if not suppressed, are in this case study responsible for overestimations of Lp;eq;1 s;max up to 3.2 dB, and of LE;A up to 2.8 dB. The experimental results also indicate that the system has limitations. The first one is a limited reduction of the ground reflected sound and the ground-borne noise impinging with a vertical angle of arrival close to 901, when the aircraft vertical angle of sight is close to 901. To obtain good reduction in these situations, directivity patterns between hypercardioid and figure-of-eight are optimal. These directivity patterns attenuate considerably the direct aircraft sound and, although this attenuation is compensated later, diminish dramatically the signal-to-noise ratio between the direct aircraft sound and the self noise and wind-induced noise of the microphones. To avoid this, non-optimal directivity patterns need to be used for angles of arrival of the aircraft noise over 801 limiting the reduction of background and ground reflected sound in this angular range. The second limitation is a higher sensitivity to wind-induced noise than a single microphone. The amplification of the wind-induced noise occurs when the distance of a microphone pair of the second sub-array is much smaller than the wavelength of the sound. However, this problem could be mitigated by using one more microphone so that three microphone pairs instead of two are used to cover the whole frequency range.
Acknowledgments The work presented here is part of the project SOUNDTRACK funded by the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 299620. Thanks to U.P. Svensson at NTNU and R. Pieren, J.M. Wunderli and K. Heutschi at Empa for their critical comments on this manuscript. Thanks to D.M. Arthur for proof reading this manuscript (and many others).
References [1] ISO 20906:2009(E), Acoustics—unattended monitoring of aircraft sound in the vicinity of airports, 2009. [2] B. Schäffer, R. Bütikofer, S. Pluss, G. Thomann, Aircraft noise: accounting for changes in air traffic with time of day, Journal of the Acoustical Society of America 129 (1) (2011) 185–199. [3] C. Asensio, M. Ruiz, M. Recuero, Real-time aircraft noise likeness detector, Applied Acoustics 71 (2010) 539–545. [4] M. Tarabini, G. Moschioni, C. Asensio, D. Bianchi, B. Saggin, Unattended acoustic events classification at the vicinity of airports, Applied Acoustics 89 (2014) 91–98. [5] G. Moschioni, B. Saggin, M. Tarabini, The use of sound intensity for the determination of the acoustic trajectory of aircrafts, International Congress on Acoustics, Vol. 2007, May 2007, pp. 2591–2596. [6] R. Bütikofer, G. Thomann, Uncertainty and level adjustments of aircraft noise measurements, INTER-NOISE and NOISE-CON Congress and Conference Proceedings, Vol. 2009, July 2009, pp. 483–490. [7] M.M. Boone, T.C. van den Dool, P. Sijtsma, U. Stelwagen, Microphone array technology for aircraft noise measurements, Internoise 2000, Nice, France, 2000, pp. 28–30. [8] A. Mast, M.M. Boone, T.C. van den Dool, An experimental microphone antenna array for measuring aircraft noise, Internoise 2001, The Hague, Netherlands, 2001, pp. 2115–2120. [9] M. Genescà, J. Romeu, T. Pàmies, A. Sánchez, Real time aircraft fly-over noise discrimination, Journal of Sound and Vibration 323 (2009) 112–129. [10] M. Genescà, J. Romeu, R. Arcos, S. Martín, Measurement of aircraft noise in a high background noise environment using a microphone array, Transportation Research Part D 18 (2013) 70–77. [11] Rion (Japan), Environmental sound monitor (for aircraft noise measurement). 〈http://svmeas.rion.co.jp/download/catalog/NA-37#NA-37〉, 2014.
M. Genescà / Journal of Sound and Vibration 374 (2016) 77–91
91
[12] I.L.N. Granien, R.T. Randeberg, H. Olsen, Next generation aircraft noise models - needs and requirements, Internoise 2010, Lisboa, Portugal, 2010, pp. 1–8. [13] B. Schaffer, C. Zellmann, W. Krebs, S. Pluss, K. Eggenschwiler, R. Butikofer, J. Wunderli, Sound source data for aircraft noise calculations state of the art and future challenges, Euronoise 2012, Prague, Czech Republic, 2012, pp. 589–594. [14] F. van den Berg, A.R. Eisses, P.J.G. van Beek, A model based monitoring system for aircraft noise, Acoustics 2008, Paris, France, 2008, pp. 881–884. [15] C. Zellmann, J.M. Wunderli, B. Schaffer, sonAIR - data acquisition for a next generation aircraft noise simulation model, Internoise 2013, Innsbruck, Austria, 2013, pp. 1–9. [16] M. Genescà, U.P. Svensson, G. Taraldsen, Estimation of aircraft sound direction of arrival using directional-microphone arrays, Acta Acustica united with Acustica 101 (1) (2015) 113–123. [17] M. Genescà, U.P Svensson, G. Taraldsen, Estimation of aircraft location using a directional-microphone array - an experimental study, Journal of the Acoustical Society of America 137 (4) (2015) 1914–1922. [18] G.W. Elko, A.N. Pong, A steerable and variable first-order differential microphone array, IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 1, 1997, pp. 223–226.