Design and implementation of steered-beamformer and cross-correlation method for acoustic-based position estimation of an object and a person using reflected waves

Abstract

In conventional acoustic control, reflected sound waves are used to analyze the room characteristics, estimate the room geometry, and detect the location of a person. In this paper, active localization is proposed to estimate a person’s position using the reflected sound waves. Echolocation and biomimetic sonar system are the technology that detects a location through an echo emitted from a sound source and returns, and it has been primarily implemented using an ultrasonic sensor with the straightness of the signal. However, the audible frequency-based echolocation demonstrated in this paper uses the scattering phenomenon in a reverberant environment, which also has the advantage of detecting an object or a person positioned behind an obstacle. This work introduces the concept of the sound field variation, which is the difference between the event and reference sound field that are measured by a microphone array. Event sound field is defined as the sound field where there is a person in the control zone, and reference sound field defines the sound field formed when there is no person. From the viewpoint of the room impulse response function, when a person or an object exists in the control space, new echoes are generated. In addition, some echoes are reduced by the person or object blocking the previous wavefront. In other words, the sound field variation represents all changed echoes. In order to establish the methodology of acoustic-based position estimation using reflected sound waves, four main aspects were studied. First, position estimation using reflected waves considers sound field variation as a new sound source, and represents energy distribution of the change in signals which are reproduced by a loudspeaker as an acoustic image. The changed signals measured by a linear microphone array are depicted by a two-dimensional acoustic image with the moving averaged Steered-Response-Power proposed in this paper, and the distance and angle are estimated through the maximum value of the acoustic image. Second aspect concerns the operating condition of the proposed active localization in reverberant environment. The performance of position estimation varies according to wall absorption which also varies the acoustic image. The position error by the various absorption condition was analyzed using the acoustical finite-difference time-domain method, and the operating conditions were represented by early decay time (EDT) and reverberation time (RT20). The third aspect is regarding the improvement of acoustic images used for position estimation. Acoustic image enhancement can be defined as a de-reverberation problem in which the first-order reflections generated by a person or an object appear clearly in the acoustic image, and higher-order reflections are suppressed. However, there are a couple of difficulties with respect to the de-reverberation problem, which are linked to the ambiguity of the phase and the overlapped magnitude, respectively. Above mentioned difficulties arise due to following reasons. First, since the signal of sound field variation represents the difference between the event and reference sound field signals, generated reflections from a person or an object are in a positive phase, but the reflected signal that changes by blocking the previous sound wave form a negative phase. Therefore, the phase may have overlapped components. Also, it is unknown where this time sample will occur. Second, in the control space, there is a change in the amplitude of the reflected sound depending on where a person or an object is located and the relative position of the boundaries. Since the signal measured by the microphone is a signal in which all the reflected sounds are superimposed, it is unknown how many reflections are of the relative magnitude and overlap. Accordingly, the application of de-reverberation techniques to suppress more than second order reflection was studied in combination with the blind system identification technique. Finally, advantages in using audible frequencies over ultrasonic sensor are shown for a problem of position estimation of a hidden person and object. The audible frequency, which is a relatively low frequency than the ultrasonic sensor, hits the person or object in the reverberant space and the scattering phenomenon can occur. As a result, the proposed active localization can also find an object or a person behind an obstacle in an anechoic environment. Through a feasibility test, it was verified that the acoustic image for position estimation in reverberant environment has more distinct features compared to the anechoic environment. Therefore, in this paper, the new active localization technique using the audible frequency is proposed, and a sound field analysis technique and signal processing method in reverberant environment were verified. Consequently, it was possible to estimate the position of a PVC pipe and a person by using their reflection in a classroom.