Acoustic source localization in ports with different beamforming

algorithms

Luca Fredianelli 1 Institute for Chemical-Physical Processes of the Italian Research Council (IPCF-CNR) Via Giuseppe Moruzzi, 1, 56124 Pisa (Italy) Marco Bernardini 2 Ipool S.r.l. Via Cocchi, 7, 56121 Pisa (Italy) Francesca Tonetti 3 Graz University of Technology, 4420 Signal Processing and Speech Communication Laboratory Inffeldgasse 16c/EG, A-8010 Graz (Austria) Francesco Artuso, Francesco Fidecaro 4 University of Pisa – Physics dpt. Largo Bruno Pontecorvo, 3, 56127 Pisa (Italy) Gaetano Licitra 5 Environmental Protection Agency of Tuscany (ARPAT) Via Vittorio Veneto, 27, 56127, Pisa (Italy)

ABSTRACT Acoustic cameras have been used to investigate the origin of a noise and localize it on video for a couple of decades. This was made possible by applying beamforming techniques to the acoustic signals simultaneously acquired by a microphone array. The number of scientists working on improving the efficiency and accuracy of this technique increased over the years, leading to the design and production of different shapes for the antenna and microphone array. Moreover, in the last years many different algorithms for beamforming techniques have been published to improve the original “Delay and Sum” method. This field is evolving rapidly and, unfortunately, there is no clear view on the advantages of one method over another, both from a theoretical and a practical point of view. This work shows the different results obtained by different algorithms when applied to the same input acoustic signals, i.e., they can localize the noise source in different points of the screen. The input signals were acquired with acoustic camera measurements to investigate port noise, a topic that has been neglected for too long and on which only few studies have been carried out. The various sound sources acting on ships’ pass-by and the predominant emitters in a multi- source environment have therefore been localized using the different algorithms.

1 luca.fredianelli@ipcf.cnr.it 2 acustica@i-pool.it 3 f.tonetti@student.tugraz.at 4 francesco.fidecaro@unipi.it 5 g.licitra@arpat.toscana.it

1. INTRODUCTION

Microphone arrays have been firstly used by Billingsley and Kinns in 1976 to individuate the origin of a sound [1], mainly to study the sound field generated by a turbojet engine to subsequently make it less noisy. Acoustic Beamforming is a method of spatial filtering or localization of a specific sound source based on the direction of arrival acquired with a microphone array. The source localization is made possible by the time difference in which the sound reaches the different microphones.

Basically, a delay and an amplitude weight are applied to the output of each sensor, and the resulting signals are summed. Signals at particular angles experience constructive interference, while others experience destructive interference. The basic idea in beamforming, with its principal algorithm “Delay and Sum” [2], is to use the set of delays to steer the array to different directions or points in a scanning plane. When the steering direction coincides with a source, the maximum output power is observed. By interpolating the measured output power from all the scanning points, it is possible to color the power across the scanning plane and make an acoustic image. With synchronized contour result maps and spectral charts, Beamforming allows the visualization of sound and makes it easy to explore the source behavior in terms of frequency and position of noise sources. The directivity of an array depends on the wavelength λ, number of microphones, distance between the elements, geometry and diameter of the array and the used beamforming algorithm.

Years after its first use, the number of scientists working on improving the efficiency and accuracy of the technique increased over the years, leading to many new array shapes and beamforming algorithms that have been studied and created. In this way, beamforming has started to be applied in many different sectors such as high-speed trains [3], cars [4, 5], aircrafts [6], helicopters [7] and wind turbines [8], boosted by the fast development of computational efficiency that allowed to process bigger quantities of data in a shorter time. This, also led to the possibility of real-time beamforming. Nowadays the sampling frequency is improved and the number of microphones in the arrays is increased with respect to the original [9]. These developments comport more accurate results making beamforming a useful tool for sound sources localization [10].

However, different research teams or companies evolved the Delay and Sum into improved, and more complex algorithms, such as Music, Evob, Clean, Sonah, Capon, Damas. Their genesis apparently followed separated path, their theoretical explanation is difficult to find and applications are lacking. The present is a situation with no clear view on which algorithm to use in different conditions or which one is the most suitable. Only in 2022 [11] the authors conducted an experiment with an acoustic camera sited in a field distant from interfering sources and a sound source was moved to different positions in order to show the best algorithm for different measurement conditions.

The present work is aimed at showing the results of acoustic cameras measurements in different areas of an industrial port and showing how the application of different algorithms produces different results in the localization of noise sources.

2. NOISE IN PORT AREAS

During the last years, many citizens living around several European ports complained about noise [12-14], leading to several Interreg maritime projects and scientific studies dedicated to this pollutant [15-17]. While standardized methodologies and procedures for rail and road noise maps are present in literature, port noise, and specifically ships’ noise, faces difficulties due to the complexity and the wide number of sources involved. In order to fill the knowledge gap regarding port noise, the INTERREG Maritime Programme Italy-France 2014-2020 included different projects all aiming at describing the present situation and designing best practices looking for a long-term sustainability in the north part of the Tyrrhenian Sea. DECIBEL Project aimed at the

realization of infrastructures for noise reduction in small ports, L.I.S.T. PORT focused on vehicle traffic around ports, TRIPLO studied noise in the area included between the ports and the logistic platforms connected to them, RUMBLE had the target of mitigating noise emission in ports and annexed areas, REPORT looked at port noise from a more theoretical point of view and MON ACUMEN studied port noise monitoring and sustainable management. Among the results, a classification of noise sources acting in ports has been proposed by Fredianelli, et al. [18]. The authors divided the sources into five macro categories, each of different nature or use (road, railways, ship, port, and industrial sources), and subcategories according to their working operation modes. A guideline for the characterization of noise sources needed as inputs for port noise maps has been published for each category [19]. Few studies in literature were dedicated to the acoustic characterization of ships, marked as noise emitters from multiple spots and during different operations [20]. In fact, different ships have proper noise emissions while moving, maneuvering, mooring and performing ground operations [21-26]. Moving ships were studied and a proper characterization for small vessels [27], ferries [28] and for roll-on/roll-off (RORO), container ships, oil tankers and chemical tankers [29] has been published. Even if studies reported that the main source acting as a disturbance for citizens living around port areas is the annexed road traffic [30- 32], each ship and port source (cranes, forklifts, trucks) produces different noise levels and emits in various directions according to their type and their activity. Identifying the correct sources would be pivotal for correctly locate each emitting point in the acoustic models that are used for evaluating people exposure to port noise. In this regard, the most adequate tool is represented by an acoustic camera based on Beamforming technique.

Thus, in a wider context, the authors performed measurements with a commercial acoustic camera in different ports of the northern Tyrrhenian Sea. The camera height was always 2 m above the ground and the antenna was at least 1 m far from reflecting obstacles. The microphone array has a diameter of 1.7 m diameter, allowing beamforming from 150 Hz.

During the work, following measurements were made:

1. Ships at quay, in order to identify the sources such as generators, engines, fans. 2. Ships pass-by, in order to identify the positioning of the sound sources that act during

movement. 3. Loading/unloading activities of ships and container handling with the presence of a

multiplicity of sources. 4. Departure/arrival of ships. 5. Industrial sites characterized by multiple contributions determined by several sources. 3. BEAMFORMING ALGORITHMS

Over the past few decades many algorithms have been developed and tested for beamforming purposes. The beamforming idea relies on the possibility of writing the sound pressure measured by a microphone as the real source level multiplied by the so-called “steering vector”. This vector accounts for the sound attenuation due to the distance between the source and the microphone, and for the different delay with which the plane wave impacts every microphone. Sophisticated algorithms are needed because a real source generates phantom sources, and subsequently the main lobe always carries secondary lobes on the acoustic map. Therefore, often conventional beamforming algorithms relying only on the Cross Spectral Matrix would produce “dirty” acoustic maps. Therefore, the difficulty lies in simultaneously recognize the main lobe while ignoring the phantom sources.

The Conventional Beamforming algorithm, or Delay and Sum (DAS), [33] searches for the correct delays to be applied to the signals measured by the different microphone in order to make them in phase again and then maximize the output from the array. Then, source positions can be found exploiting the correlation between delays and different paths travelled by the plane wave.

An improvement of Conventional Beamforming is represented by the Functional Beamforming [34, 35], which is another algorithm that estimates the source autopower related to a particular

direction using the Cross Spectral Matrix (CSM) of the signal recorded by the microphones. Differently from the Conventional method, the CSM is raised to the power of an exponent and then of its inverse. This procedure may seem useless, but instead the result is that sidelobes are damped while main lobes are left unaltered. A higher exponent suppresses the side lobes more. EVOB (EigenValue Optimized Beamforming) is an optimization of functional beamforming developed by the owner of the commercial acoustic camera used in the present work, and details are not available.

A slightly different approach is brought by the MUSIC algorithm [36], as it is based on the eigenvectors of the covariance matrix of the signals recorded by the different microphones. In particular, MUSIC exploits the orthogonality relation between the eigenvectors related to the background noise and the steering vectors that exists when the last vectors are function of the correct arrival angle of the sound. Therefore, scanning the plane in front of the array can minimize the scalar product between the above-mentioned vectors in order to find the correct arrival angle.

A more sophisticated kind of algorithms uses deconvolution techniques, which is iterative extrapolation of useful information from a confusing and noisy scenario by eliminating misleading data. The main algorithm of this kind is represented by Clean [37]. Starting from an acoustic map obtained with conventional beamforming, Clean finds the maximum and subtract the Point Spread Function (PSF) of this maximum from the map, then substituting this PSF with a pure beam localized in the position of the maximum. It is worth noting that the PSF accounts not only for the main lobe generated by the real source, but also for the secondary lobes. In this way, subtracting and substituting it in the map, the fake sources are eliminated while the real one is preserved. The procedure is iterated and the final result is a map that should be cleaned from phantom sources. A further improvement of this algorithm is Clean-SC which is able to better identify sidelobes exploiting their coherence with the source which has generated them.

Another deconvolution method is DAMAS [38], whose aim is to determine the acoustic power coming from every point of the scanning plane by solving a set of linear equations. The known variables of this set of equations are the sound pressures recorded by microphones. Starting from an initial configuration of the grid, made of random acoustic power, the iterative method is implemented until a representative acoustic map is obtained.

The Capon method identifies the arrival angles of sound as the ones which minimize a so-called “cost function”. This function consists of a product between steering vectors and the inverse of the cross-spectral matrix of microphone outputs [39].

The theoretical and applicative aspects of each algorithm have been investigated by the authors in [40].

4. RESULTS

The present chapter shows, by way of example, a frame for each of the types of sources investigated in the port areas. For each of the frames, 6 different algorithms have been processed: DAS, Clean, MUSIC, EVOB, DAMAS and CAPON. The minimum dynamic range of 1 dB(A) has been chosen for each image, in order to facilitate viewing. The frequencies at which the frames are analyzed vary according to the case and the generated sounds, and are:

1. Ships at quay: Figure 1 with frequency range 450-600 Hz. 2. Ships pass-by: Figure 2 with frequency range 680-1600 Hz. 3. Loading/unloading activities of ships: Figure 3 with frequency range 267-389 Hz. 4. Departure/arrival of ships: Figure 4 with frequency range 470-530 Hz. 5. Industrial sites with several sources: Figure 5 with frequency range 800-1000 Hz. In the pass-by, the frame analyzed with MUSIC is voluntarily chosen different from the others in order to confirm the suspicion that the algorithm always places the source in the center of the image even when it should not.

Figure 1: Ship at quay. The frame is analyzed in the range 450-600 Hz with different algorithms.

Figure 2: Ship pass-by. The frame is analyzed in the range 680-1600 Hz with different algorithms.

Figure 3: Loading/unloading of ships. The frame is analyzed in the range 267-389 Hz with different algorithms.

Figure 4: Arrival of ships. The frame is analyzed in the range 470-530 Hz with different algorithms.

Figure 5: Industrial site. The frame is analyzed in the range 800-1000 Hz with different algorithms.

The active sources in the stationing of a ship strongly depend on the model itself. For the ferry shown in Fig. 1, the main source is represented by an air vent in the center of the side of the ship. Although the ship is different, the noise emitted by the RORO pass-by shown in Fig.2 is also produced by an air vent located on the side. The noise emitted by the ship-to-shore crane is very complex and of a different nature, but in the frame in Fig. 3 and at the low frequencies of 267-389 Hz it comes from the tail in the upper right corner. Even the arrival of a ferry has multiple sources, which add to those of the stationing, but the main source in Fig. 4 is at low frequencies and produced by the vibration of the rear bulkhead. In the frame of Fig. 5, the sources are a pump on a roof in the upper left and an unidentified source at the bottom right.

5. DISCUSSION AND CONCLUSIONS

Nowadays, there are multiple beamforming algorithms and no guides or explanations on which to use among those implemented by commercial software of the acoustic cameras are provided. Even the scientific literature is vague and not explanatory, leading to uncertainty in the choices of the algorithm with consequent difficulties for users of commercial antennas. In fact, the present work has shown how the application of different algorithms to the same acquisition frame can provide totally different results.

The measurements were carried out in port areas inside a larger research context aimed at characterizing the sound sources, which in ports are many and still not properly studied. Therefore, beamforming can be particularly useful in port noise for correctly locating the sound emission of very large, complex and varied sources and then provide accurate inputs for noise mapping models.

The present work has shown preliminary results that will serve as a starting point for further developments, despite the many difficulties in carrying out measurements in port areas due to the access and security issues, the presence and logistics of the vessels, the high background noise and the presence of multiple sources at the same time.

The results confirm that the basic algorithm, DAS, has a poor spatial resolution, which is improved by other more advanced algorithms. In the case of industrial noise, DAS was able to distinguish the simultaneous presence of two sound sources. However, the stationing measurement showed that DAS is very susceptible to sound wave reflections that affect the correct positioning of the source in certain geometries. Furthermore, DAS showed a considerable reaction time in the case of moving sources or in video elaboration, not single frame as reported in the paper. Clean produces a much more precise and focused localization, so precise that it is sometimes hard to see. This makes it the most suitable of the analyzed algorithms for the purpose of source localization, but it loses its effectiveness in presenting the results to the public. Clean does not seem to distinguish two simultaneously active sources, in accordance with its theory based on the selection of the maximum sound level. From what has been reported, no significant differences seem to emerge between the results obtained with Clean and DAMAS. These last two aspects require particular additional studies. Music has always localized the source in the center of the image, showing a bug in the software implementation. EVOB showed performances that improve DAS algorithm, as the source is identified in the same place, but with a better spatial resolution. CAPON does not seem to work at all or was not designed for applications of this nature. REFERENCES

1. Billingsley, J., & Kinns, R. (1976). The acoustic telescope. Journal of Sound and Vibration,

48(4), 485-510. 2. Rakotoarisoa, I., Fischer, J., Valeau, V., Marx, D., Prax, C., & Brizzi, L. E. (2014). Time-

domain delay-and-sum beamforming for time-reversal detection of intermittent acoustic sources in flows. The Journal of the Acoustical Society of America, 136(5), 2675-2686. 3. Noh, H. M., & Choi, J. W. (2015). Identification of low-frequency noise sources in high-

speed train via resolution improvement. Journal of Mechanical Science and Technology, 29(9), 3609-3615. 4. Ballesteros, J. A., Sarradj, E., Fernandez, M. D., Geyer, T., & Ballesteros, M. J. (2015).

Noise source identification with beamforming in the pass-by of a car. Applied Acoustics, 93, 106-119. 5. Bourgeois, J., & Minker, W. (Eds.). (2009). Time-domain beamforming and blind source

separation: speech input in the car environment. Boston, MA: Springer US. 6. Bu, H., Huang, X., & Zhang, X. (2021). An overview of testing methods for aeroengine fan

noise. Progress in Aerospace Sciences, 124, 100722. 7. Martin, G., Simon, F., & Biron, D. (2008). Detection of acoustic radiating areas of a generic

helicopter cabin by beamforming. Journal of the Acoustical Society of America, 123(5), 3310-3310. 8. Sun, S., Wang, T., Yang, H., & Chu, F. (2022). Damage identification of wind turbine

blades using an adaptive method for compressive beamforming based on the generalized minimax-concave penalty function. Renewable Energy, 181, 59-70. 9. Michel, U. (2006). History of acoustic beamforming. In 1st. Berlin Beamforming

Conference. 10. Chiariotti, P., Martarelli, M., & Castellini, P. (2019). Acoustic beamforming for noise

source localization–Reviews, methodology and applications. Mechanical Systems and Signal Processing, 120, 422-448. 11. Fredianelli, L., Tonetti, F., Bernardini, M., Artuso, F., Fidecaro, F., Licitra, G. (2022).

Accuracy and precision of beamforming algorithms in source localization with acoustic camera. Measurement. 12. Licitra, G.; Bolognese, M.; Palazzuoli, D.; Fredianelli, L.; Fidecaro, F. Port noise impact

and citizens’ complaints evaluation in RUMBLE and MON ACUMEN INTERREG projects. In Proceedings of the 26th International Congress on Sound and Vibration,

Montreal, QC, Canada, 7–11 July 2019. 13. Paschalidou, A.K.; Kassomenos, P.; Chonianaki, F. Strategic Noise Maps and Action Plans

for the reduction of population exposure in a Mediterranean port city. Sci. Total Environ. 2019, 654, 144–153. 14. Murphy, E., & King, E. A. (2012). Residential exposure to port noise: a case study of

Dublin, Ireland. In The 41st International Congress on Noise Control Engineering, New York: Institute of Noise Control Engineering (INCE-USA), August 19-22, 2012 15. Schenone, Corrado, et al. "The Port Noise Analysis and Control in Interreg Italy-France

Maritime Programme." INTER-NOISE and NOISE-CON Congress and Conference Proceedings. Vol. 259. No. 4. Institute of Noise Control Engineering, 2019. 16. Borelli, D., Gaggero, T., Pallavidino, E., Schenone, C., Kamdem, E. L. W., & Njiotang, C.

A. Y. (2020, December). Development of a Harbour Noise Monitoring Solution within the Interreg Maritime RUMBLE Project. In Forum Acusticum (pp. 1261-1262). 17. Fredianelli, L., Carpita, S., Bernardini, M., Del Pizzo, L. G., Brocchi, F., Bianco, F., &

Licitra, G. (2022). Traffic flow detection using camera images and machine learning methods in ITS for noise map and action plan optimization. Sensors, 22(5), 1929. 18. Fredianelli, L., Bolognese, M., Fidecaro, F., & Licitra, G. (2021). Classification of noise

sources for port area noise mapping. Environments, 8(2), 12. 19. Fredianelli, L., Gaggero, T., Bolognese, M., Borelli, D., Fidecaro, F., Schenone, C., &

Licitra, G. (2022). Source characterization guidelines for noise mapping of port areas. Heliyon, e09021. 20. Badino, A., Borelli, D., Gaggero, T., Rizzuto, E., & Schenone, C. (2012). Noise emitted

from ships: impact inside and outside the vessels. Procedia-Social and Behavioral Sciences, 48, 868-879. 21. Witte, J. Noise from moored ships. In INTER-NOISE and NOISE-CON Congress and

Conference Proceedings; Institute of Noise Control Engineering: Reston, VA, USA, 2010; pp. 3202–3211. 22. Di Bella, A., & Remigi, F. (2013, June). Prediction of noise of moored ships. In Proceedings

of Meetings on Acoustics ICA2013 (Vol. 19, No. 1, p. 010053). Acoustical Society of America. 23. Santander, A.; Aspuru, I.; Fernandez, P. OPS Master Plan for Spanish Ports Project. Study

of potential acoustic benefits of on-Shore power supply at berth. In Proceedings of the Euronoise 2018, Heraklion-Crete, Greece, 27–31 May 2018. 24. Badino, A.; Borelli, D.; Gaggero, T.; Rizzuto, E.; Schenone, C. Acoustical impact of the

ship source. In Proceedings of the 21st International Congress on Sound and Vibration, Beijing, China, 13–17 July 2014; pp. 13–17. 25. Di Bella, A.; Remigi, F.; Fausti, P.; Tombolato, A. Measurement methods for the

assessment of noise impact of large vessels. In Proceedings of the 23rd International Congress on Sound & Vibration, Athens, Greece, 10–14 July 2016. 26. Fausti, P.; Santoni, A.; Martello, N.Z.; Guerra, M.C.; Di Bella, A. Evaluation of airborne

noise due to navigation and manoeuvring of large vessels. In Proceedings of the 24th International Congress on Sound and Vibration, London, UK, 23–27 July 2017. 27. Bernardini, Marco, et al. "Noise assessment of small vessels for action planning in canal

cities." Environments 6.3 (2019): 31. 28. Nastasi, M., Fredianelli, L., Bernardini, M., Teti, L., Fidecaro, F., & Licitra, G. (2020).

Parameters Affecting Noise Emitted by Ships Moving in Port Areas. Sustainability, 12(20), 8742. 29. Fredianelli, L., Nastasi, M., Bernardini, M., Fidecaro, F., & Licitra, G. (2020). Pass-by

characterization of noise emitted by different categories of seagoing ships in ports. Sustainability, 12(5), 1740. 30. Schenone, C.; Pittaluga, I.; Borelli, D.; Kamali, W.; El Moghrabi, Y. The impact of

environmental noise generated from ports: Outcome of MESP project. Noise Mapp. 2016, 3.

31. Hanaoka, S., & Regmi, M. B. (2011). Promoting intermodal freight transport through the

development of dry ports in Asia: An environmental perspective. Iatss Research, 35(1), 16- 23. 32. Alsina-Pagès, R.M.; Socoró, J.C.; Barqué, S. Survey of Environmental Noise in the Port of

Barcelona. In Proceedings of the Euronoise—European Conference on Noise Control, Crete, Greece, 27–31 May 2018. 33. Grythe, J., & Norsonic, A. S. (2015). Beamforming algorithms-beamformers. Technical

Note, Norsonic AS, Norway. 34. Merino-Martinez, R., Snellen, M., & Simons, D. G. (2016, February). Functional

beamforming applied to full scale landing aircraft. In 6th Berlin Beamforming Conference, February. 35. Dougherty, R. P. (2014). Functional beamforming for aeroacoustic source distributions. In

20th AIAA/CEAS aeroacoustics conference (p. 3066). 36. Gupta, P., & Kar, S. P. (2015, April). MUSIC and improved MUSIC algorithm to estimate

direction of arrival. In 2015 International Conference on Communications and Signal Processing (ICCSP) (pp. 0757-0761). IEEE. 37. Sijtsma, P. (2007). CLEAN based on spatial source coherence. International journal of

aeroacoustics, 6(4), 357-374. 38. Brooks, T. F., & Humphreys, W. M. (2006). A deconvolution approach for the mapping of

acoustic sources (DAMAS) determined from phased microphone arrays. Journal of sound and vibration, 294(4-5), 856-879. 39. Handel, P., Stoica, P., & Soderstrom, T. (1993, January). Capon method for doa estimation:

accuracy and robustness aspects. In IEEE Winter Workshop on Nonlinear Digital Signal Processing (pp. P_7-1). IEEE. 40. Artuso, F., Fredianelli, L., Bernardini, M., Fidecaro, F., Licitra, G. (2022). An overview of

acoustic beamforming algorithms and their applications. Noise mapping.