A Platform for Adaptive Audio Sensing in IoT
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2616171Utgivelsesdato
2017Metadata
Vis full innførselSamlinger
Sammendrag
The Internet of Things (IoT) is the technology which connects Things that surround us to the Internet. The interest in this technology has beenincreasing along with its fields of interest in recent years. This projectfocuses on checking the feasibility of a sensing audio system in the IoTenvironment. We think that an audio sensing system composed of lots ofsmall and low-cost devices has potential in several scenarios such as dBAnoise measurement and room occupation. We implement a prototype of asensor node able to detect the human audible spectrum, which we use astestbed. We selected hardware and software technologies in the way thatthe sensor node can work in an IoT environment: low-energy consumption,small footprint and wireless connection are the characteristics that thesensor node must have.
The key feature of our sensing system is the use of the audio spectrumanalysis. In fact it allows us to work with audio chunks of tens up tohundreds of milliseconds to be transmitted instead of streaming the audiocontinuously, reducing the information to send and the active time of thenode. The resulting sensing pipeline consists of Detection, Sampling, FastFourier Transform (FFT) and Transmission stages, continuously iteratingafter an idle state period. The firmware for the Detection, Samplingand FFT stages was developed. Random Access Memory (RAM) andRead Only Memory (ROM) occupation, time occupation and powerconsumption measurement were performed on different test cases andsample lengths. The FFT stage can also be performed in a more powerfulnode and the pros and cons were evaluated.
The measurement tests provided evidence on efficient RAM usage onthe testbed while simultaneously working with very small audio chunks.However, this deteriorated when we performed the FFT on board becauseit requested extra space on the RAM. From the gathered data from thetime occupation and power measurement tests, we developed an energymodel of our sensor node that helps to configure the energy profile of thesensing system. Performing the FFT is a hallmark of our sensing system,its performance on board is justified only if the data are needed locally orfor privacy-related reasons. We also suggest some proposals to improvethe FFT stage.