Repository logo

Unattended acoustic sensor systems for noise monitoring in national parks




Yaremenko, Vladimir, author
Azimi-Sadjadi, Mahmood R., advisor
Pezeshki, Ali, committee member
Anderson, Charles, committee member

Journal Title

Journal ISSN

Volume Title


Detection and classification of transient acoustic signals is a difficult problem. The problem is often complicated by factors such as the variety of sources that may be encountered, the presence of strong interference and substantial variations in the acoustic environment. Furthermore, for most applications of transient detection and classification, such as speech recognition and environmental monitoring, online detection and classification of these transient events is required. This is even more crucial for applications such as environmental monitoring as it is often done at remote locations where it is unfeasible to set up a large, general-purpose processing system. Instead, some type of custom-designed system is needed which is power efficient yet able to run the necessary signal processing algorithms in near real-time. In this thesis, we describe a custom-designed environmental monitoring system (EMS) which was specifically designed for monitoring air traffic and other sources of interest in national parks. More specifically, this thesis focuses on the capabilities of the EMS and how transient detection, classification and tracking are implemented on it. The Sparse Coefficient State Tracking (SCST) transient detection and classification algorithm was implemented on the EMS board in order to detect and classify transient events. This algorithm was chosen because it was designed for this particular application and was shown to have superior performance compared to other algorithms commonly used for transient detection and classification. The SCST algorithm was implemented on an Artix 7 FPGA with parts of the algorithm running as dedicated custom logic and other parts running sequentially on a soft-core processor. In this thesis, the partitioning and pipelining of this algorithm is explained. Each of the partitions was tested independently to very their functionality with respect to the overall system. Furthermore, the entire SCST algorithm was tested in the field on actual acoustic data and the performance of this implementation was evaluated using receiver operator characteristic (ROC) curves and confusion matrices. In this test the FPGA implementation of SCST was able to achieve acceptable source detection and classification results despite a difficult data set and limited training data. The tracking of acoustic sources is done through successive direction of arrival (DOA) angle estimation using a wideband extension of the Capon beamforming algorithm. This algorithm was also implemented on the EMS in order to provide real-time DOA estimates for the detected sources. This algorithm was partitioned into several stages with some stages implemented in custom logic while others were implemented as software running on the soft-core processor. Just as with SCST, each partition of this beamforming algorithm was verified independently and then a full system test was conducted to evaluate whether it would be able to track an airborne source. For the full system test, a model airplane was flown at various trajectories relative to the EMS and the trajectories estimated by the system were compared to the ground truth. Although in this test the accuracy of the DOA estimates could not be evaluated, it was show that the algorithm was able to approximately form the general trajectory of a moving source which is sufficient for our application as only a general heading of the acoustic sources is desired.


Rights Access


noise monitoring
statistical signal processing
embedded systems
transient signals
remote sensing


Associated Publications