STEM 1

Multi-source noise cancellation allows for increased cancellation performance in headphones because of the ability to recognize noise in the user's surroundings before it reaches the headphones. Current ANC (Active Noise Cancellation) solutions only work at lower frequencies and multi-source solutions are immobile. This project aims to develop a mobile app to make multi-source noise cancellation portable and improve ANC performance in the industry. ANC technology works by using a digital signal processing unit that can take audio and background noise coming from a microphone as input and produce an anti-noise signal played simultaneously as background noise comes in, effectively canceling noise. The solution works well at lower frequencies but loses its efficiency in canceling higher-frequency noises. A mobile app was developed that enhanced the capability of phones to be used as reference microphones for a multi-source noise cancellation algorithm. Connected phones would communicate microphone information to feed the algorithm. It was found that while the app was in use, the mean connection relay time of the app was greater than the mean time for noise to travel to the user within the tested distances. These findings provide evidence that noise cancellation produced from this input may not show improvement over the industry standard, leaving the user with a standard or slightly improved noise cancellation experience. In the future, it may be possible to implement other connection protocols to improve connection relay time, leading to improved cancellation.

Noise-cancelling headphones are headphones that are meant to reduce unwanted background noise while in use. The demand for such technology has increased in recent years leading manufacturers to release many products featuring ANC (Active Noise Cancellation). ANC technology works by producing sound waves of equal amplitude, but an opposite signal of the noise perceived by the headphones; they produce these waves by using microphones on the outside shell of the headphones to observe background noise and process the opposite, or anti-noise, through a digital signal processing (DSP) unit. When these two opposite sound waves are heard in conjunction by the user's ears, any background noise is effectively canceled (Kuo & Morgan, 1999). Although this technology works well in theory, when put in practice ANC tends to struggle in attenuating more variable higher-frequency noise. This limiting factor leads to subpar cancellation performance when in an environment with high-frequency noise (Kajikawa et al., 2012). Previous research has been done by researchers at UIUC, within a project called MUTE, to identify why this issue occurs and to propose a possible solution. The inefficiency at higher frequencies is due to the small window of time between the receiving of the input signal and the production of the anti-noise signal. Within their project, an external microphone was leveraged to connect to headphones through an IoT (Internet of Things) relay. The concept of lookahead was introduced because wireless signals travel faster than sound. Therefore, the headphones receive the microphone signal before the noise reaches the headphones. As a result, the DSP could begin producing an anti-noise signal before hearing the noise allowing for sufficient time to produce an accurate signal (Shen et al., 2018).

To develop the proposed mobile app, multiple software tools were used. The app was built in the Dart programming language on the Flutter framework. Flutter was chosen for its cross-platform interoperability, allowing the mobile app to be run on both iOS and Android operating systems. The Socket.io server API was used within the application to establish a WebSocket connection to a server hosted using the Heroku platform. Two Android phones were used for testing of the application, specifically its data transfer time. As input for a given noise cancellation algorithm, an audio signal of noise must be taken. To capture or record this audio from mobile phones with the app installed, a recorder package was used within Flutter. The audio would be recorded and saved as an m4a file locally, constantly updating as the recording continues. The recording continues to update until the user manually stops it through the User Interface (UI). While the recording is updating the audio file is being read as bit data and is being chunked into 10 kb pieces. The purpose for audio chunking is to lighten the load on the server for uploading the audio allowing the phone connection and transfer time to be faster. The audio chunking also guarantees that there is a high flow of data, ensuring that new data is being collected and uploaded in a timely manner. Generally, the lower the chunk size is the faster the connection will be, but this does make it harder to process and convert back to audio. However, this tradeoff is worthwhile given the speed of today's processors in phones. To transfer or relay the audio input signal from one phone to another, a WebSocket connection was used. WebSocket was chosen as it is a faster and a more standard transfer protocol when compared to other methods over WiFi (Pimentel & Nickerson, 2012). WiFi was considered to provide the fastest connection speeds when a user is in a public environment. Overall, connection and transfer speed were the highest priority and WiFi was deemed to be the most viable option for this. For a WebSocket connection between phones, a WebSocket server is necessary to store and send audio data. Heroku was used to setup a WebSocket server, and within the Flutter app the Socket.io library was used to establish a connection to the server. Whenever a new audio chunk would be saved it would be pushed to the Heroku server. The other phone retrieving the audio data, would also establish a connection to the server, retrieve the chunked audio data, and convert it back to an audio file locally.