Audio Source Positioning Based on Angle of Arrival Measurements
Lager, Mikko (2020)
Lager, Mikko
2020
Teknis-luonnontieteellinen DI-tutkinto-ohjelma - Degree Programme in Science and Engineering, MSc (Tech)
Tekniikan ja luonnontieteiden tiedekunta - Faculty of Engineering and Natural Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2020-03-10
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202002272403
https://urn.fi/URN:NBN:fi:tuni-202002272403
Tiivistelmä
Estimating position is done in various contexts from locating phones with GPS to locating boats using hydrophones. In this thesis we study estimating audio source position based on angle of arrival measurements. Multiple different filters can be used on measured angles of arrival to deduce the position of the source. The filter to be used in this work was chosen to be the particle filter. Even though particle filter is computationally more heavy than many other filters, modern computers can simulate hundreds of particles in a short time without too much of an effort. We introduce the reader to the use of particle filter in positioning, along with theoretical background of it and positioning in a more general sense.
The data in this work is recorded in either an anechoic chamber or a room that has no special equipment installed to enhance audio quality in it. The measurements are done with a mobile device with four microphones. Audio source in the anechoic chamber is a loudspeaker playing speech or a person speaking and walking randomly in the room. If the data contains noise, it is played from loudspeakers in the same space as the source is located in. Another type of data handled in this work is measured outside in a racing event where multiple cars passed the measurement device as well as generated data with multiple sources.
The data is handled as a mixture between von Mises and uniform distribution. An important parameter of von Mises distribution is a variable called κ, which tells the concentration of the distribution. In this work we show and prove a way to estimate said variable with maximum likelihood method. Additionally, we introduce the reader to mathematical background of particle filter and positioning in more general sense. Results given by the particle filter depend on the chosen value of κ along with chosen q-value, which tells the smoothness of the result, and measurement model. Finally, we present and compare the results obtained by constant velocity and random walk models with several different q-values.
The data in this work is recorded in either an anechoic chamber or a room that has no special equipment installed to enhance audio quality in it. The measurements are done with a mobile device with four microphones. Audio source in the anechoic chamber is a loudspeaker playing speech or a person speaking and walking randomly in the room. If the data contains noise, it is played from loudspeakers in the same space as the source is located in. Another type of data handled in this work is measured outside in a racing event where multiple cars passed the measurement device as well as generated data with multiple sources.
The data is handled as a mixture between von Mises and uniform distribution. An important parameter of von Mises distribution is a variable called κ, which tells the concentration of the distribution. In this work we show and prove a way to estimate said variable with maximum likelihood method. Additionally, we introduce the reader to mathematical background of particle filter and positioning in more general sense. Results given by the particle filter depend on the chosen value of κ along with chosen q-value, which tells the smoothness of the result, and measurement model. Finally, we present and compare the results obtained by constant velocity and random walk models with several different q-values.