Improving Loudspeaker Characteristics in a Low Power Environment

Abstract

This master's thesis explores the improvement of loudspeaker characteristics by using digital signal processing. The main aim of the thesis is to improve the overall audio quality with regards to limited processing power and power supply. Improving the audio quality is important since the vast expansion of the usage of loudspeakers has led to demands on cost, size and design choices all leading to different loudspeaker characteristics. A software algorithm capable of running in real-time on the loudspeaker is used to achieve the aim. The thesis includes a literature study as well as a practical implementation with tests to find out which algorithm performs the best. The result of the literature study concluded that a virtual bass enhancement algorithm, as well as a dynamic equalizer algorithm, are the best alternatives to enhance the audio quality of the network loudspeakers. From the many different algorithms found in the literature study, the virtual bass enhancement algorithm, arc-tangent square root, was determined to give the best result by both scoring the highest of the algorithms tested in a listening test and showing good performance in performance tests. The simplicity of the algorithm together with the improvement it offers, leads to it being a good option for any network speaker. Future improvements on the algorithms should focus on finding a better way of reducing noise on the input signal before processing it with the algorithms.

Popular Abstract

In today's society, speakers play an integral part of everyday life. Due to speakers existing everywhere including your phone, at the train station, and in the elevator, there is value in researching how to improve the audio quality for all loudspeakers. The problems that occur when design choices, such as form factor or economical considerations, impair the ability of the loudspeaker to produce bass has historically been solved using a sub-woofer, a specialized bass speaker. However, with the demand for smaller speakers ever increasing, using a sub-woofer to enhance the bass is not feasible because they need to be bigger than a regular speaker. This project focuses on how the audio quality can be improved on a network speaker with the help of software.

By using digital signal processing the quality can be improved with a few methods.
In this project two of these methods were used. One method focuses on making the bass more pronounced. The other method focuses on tricking the brain that there is more bass than there actually is.

A simple way to make the bass more pronounced is with the use of digital filters, in this case a low-shelf filter, which will boost a certain range of frequencies, while leaving the rest unmodified. However, the signal can only reach a certain level before information is lost. The space left between the peaks and the maximum level is commonly called headroom. Processing the signal with a low-shelf filter will decrease the amount of available headroom. By dynamically changing the amount boosted with regards to the amount of available headroom a more pronounced bass can be achieved, while not disturbing the rest of the audio content. This approach is a so called dynamic equalizer (EQ).

Tricking the brain into thinking there is more bass can be done by using a psycho-acoustic phenomenon called "Missing Fundamental". This phenomena occurs because pure tones are not natural. Natural tones consists of a fundamental tone and the fundamental's harmonic series, also known as overtone series. In the overtone series the fundamental is the strongest and the overtones are slowly decreasing in strength. The "Missing Fundamental"-phenomenon makes it so, if a tone's harmonic series can be heard then the fundamental can be heard as well, even if the fundamental does not exist in the signal. Because of that, if the harmonic series could be generated the brain would think the bass was more pronounced than it actually is. With the help of a non-linear device (NLD), the harmonic series of a signal can be generated.

A basic way to create a better perceived bass with the help of an NLD is by simply allowing the bass content of a signal to be processed by an NLD. This was implemented with the help of the NLD arc-tangent square root (ATSR). The ATSR takes an input signal and applies a mathematical function to calculate the resulting signal.

Another more advanced way to improve the perceived bass is to generate the harmonics with regard to different instruments instead of the signal as a whole. This method consists of trying to "demix" the signal into multiple signals, each representing different instruments and then applying the NLD to the different signals. In the implementation that was done the signal is not "demixed" and instead it is treated as one single instrument.

The two implementations mentioned earlier together with a dynamic EQ were evaluated by participants of a listening test. In the listening tests, the participants were instructed to rate each algorithm for six different music clips and six different announcements. The results showed that the basic ATSR implementation improves the perceived audio quality for music compared to the unaltered reference signal. While the dynamic EQ algorithm shows improvement on some of the music clips compared to the reference, it is not rated as high as the basic ATSR implementation. The music demixing implementation did not show any significant improvement to the audio quality. Neither of the implementations showed any positive change in the audio quality on the announcements.
The three implementations were also tested with a performance test measuring the processing power required to run the implementation. The performance test showed that the dynamic EQ had the best performance, closely followed by the basic ATSR implementation. The music demixing implementation had the worst performance on the performance test.

Finally the conclusion of this project is that the audio quality of any network loudspeaker could be improved by implementing the basic ATSR implementation presented. For scheduled announcements, the algorithm could be turned off momentarily. Therefore, the ATSR should be implemented for any network loudspeaker.

Further research should focus on implementing a noise reduction filter. By reducing the noise in a better and more effective way all of the algorithms tested in this project would be improved and the audio quality would be improved even more. Additionally, testing more types of speakers than the ones in this thesis would be beneficial to achieve a wider understanding of the algorithms.

• author: Andreasson, Jonas and Olsson, Love
• supervisor: Kaan Bür
• organization: Department of Electrical and Information Technology
• course: EITM01 20241
• year: 2024
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• report number: LU/LTH-EIT 2024-999
• language: English
• id: 9165526
• date added to LUP: 2024-06-26 10:56:29
• date last changed: 2024-06-26 10:56:29