Citation
Proposed method for recording and analyzing cell phone adaptive noise reduction filters

Material Information

Title:
Proposed method for recording and analyzing cell phone adaptive noise reduction filters
Creator:
Nelson, Anthony
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Master's ( Master of science)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
Department of Music and Entertainment Industry Studies, CU Denver
Degree Disciplines:
Recording arts
Committee Chair:
Grigoras, Caitlin
Committee Members:
Smith, Jeff
Whitecotton, Cole

Notes

Abstract:
During cell phone conversations, the noises from the world around us can interfere with call clarity; noise being an unwanted sound that ruins the quality of the signal. One solution to minimize the noise is to utilize adaptive noise cancellation filters. The majority of cell phone manufacturers have implemented this technology [1][2] into their products, aiming to increase caller intelligibility by reducing ambient background noise. Many aspects of a cell phone's noise filtering features are proprietary, [3] which is unfortunate from a forensic standpoint. The algorithms used in many cell phone adaptive noise cancellation systems are kept private from the public. There is a void in forensic audio analysis of cell phones [4]. Very little forensic research has been devoted to the experimentation of adaptive noise filters, especially in regards to their use in cell phones. Since a large number of cell phones utilize such filtering, it is important that the forensic community considers this key attribute offered in cell phones today. The adaptive filters’ unique qualities could prove to be an asset to forensic sciences. This thesis provides in-depth research into the performance of adaptive noise cancellation filters in cell phones and proposes methods to test and analyze them. A method to capture the adaptive noise filters via audio recording was tested and proven to be successful. In addition to this, signal to noise ratio (SNR) values were recorded and categorized by cell phone maker and model.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
Copyright Anthony Nelson. Permission granted to University of Colorado Denver to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item has the following downloads:


Full Text
PROPOSED METHOD FOR RECORDING AND ANALYZING CELL PHONE
ADAPTIVE NOISE REDUCTION FILTERS by
ANTHONY NELSON B.S., University of Colorado Denver, 2018
A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfilment of the requirements for the degree of Master of Science Recording Arts Program
2018


This thesis for the Master of Science degree by Anthony Nelson Has been approved for the Recording Arts Program by
Catalin Grigoras, Chair Jeff Smith Cole Whitecotton
Date: December 15, 2018


Nelson, Anthony (M.S., Recording Arts Program)
Proposed Method for Recording and Analyzing Cell Phone Adaptive Noise Reduction Filters Thesis directed by Associate Professor Catalin Grigoras
ABSTRACT
During cell phone conversations, the noises from the world around us can interfere with call clarity; noise being an unwanted sound that ruins the quality of the signal. One solution to minimize the noise is to utilize adaptive noise cancellation filters. The majority of cell phone manufacturers have implemented this technology [1][2] into their products, aiming to increase caller intelligibility by reducing ambient background noise. Many aspects of a cell phone's noise filtering features are proprietary, [3] which is unfortunate from a forensic standpoint. The algorithms used in many cell phone adaptive noise cancellation systems are kept private from the public.
There is a void in forensic audio analysis of cell phones [4], Very little forensic research has been devoted to the experimentation of adaptive noise filters, especially in regards to their use in cell phones. Since a large number of cell phones utilize such filtering, it is important that the forensic community considers this key attribute offered in cell phones today. The adaptive filters’ unique qualities could prove to be an asset to forensic sciences. This thesis provides in-depth research into the performance of adaptive noise cancellation filters in cell phones and proposes methods to test and analyze them. A method to capture the adaptive noise filters via audio recording was tested and proven to be successful. In addition to this, signal to noise ratio (SNR) values were recorded and categorized by cell phone maker and model.
The form and content of this abstract are approved. I recommend its publication.
Approved: Catalin Grigoras
ii


DEDICATION
I would like to dedicate this thesis to my friends and family.


ACKNOWLEDGEMENTS
I would like to thank Catalin Grigoras, Jeff Smith, Cole Whitecotton and the NCMF program for the amazing help throughout this program. I would also like to acknowledge my friends and family for volunteering their time and cell phones for testing purposes. Especially the one and only Paul Malatesta for helping with multiple tests. Lastly, I’d like to give thanks to my fiancee Nicole Pacheco, who supported me through this entire process. Without the involvement from these people my drive and ability to finish this thesis would have been altered.
IV


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION..........................................................1
Background............................................................2
Scope and Limitations.................................................3
II. ADAPTIVE NOISE REDUCTION FILTER.......................................4
Introduction Into Filters.............................................4
How Adaptive Filters Work.............................................4
Adaptive Noise Cancellation Filters In Cell Phones....................5
Adaptive Filter Algorithms............................................6
III. CELL PHONE LAYOUT.....................................................8
General Cell Phone Microphone Layout..................................8
Apple iPhone..........................................................9
LG...................................................................11
Samsung..............................................................12
IV. TESTING..............................................................14
Lab Testing..........................................................15
Live Scenario Testing................................................16
V. ANALYSIS AND RESULTS.................................................18
Spectrogram..........................................................20
v


SNR Results..................................................22
FFT Analysis .................................................28
VF CONCLUSION....................................................32
VII. FURTHER RESEARCH..............................................35
REFERENCES.........................................................36
vi


LIST OF TABLES
TABLE
5.1 Average SNR Values at 70 dB.............................................26
5.2. Average SNR Values at 80 dB ...........................................26
5.3 Average SNR Values Live.................................................26
5.4 Average SNR Values Noise of 70 dB ......................................27
5.5 Average SNR Values Noise of 80 dB ......................................27
5.6 Average SNR Values Live Noise...........................................27
vii


LIST OF FIGURES
FIGURE
2.1 Adaptive Filter Signal Flow....................................................5
2.2 ANC Signal Flow................................................................6
3.1 General Cell Phone Microphone Array............................................8
3.2 iPhone 6 Microphone Array......................................................9
3.3 iPhone 7 Microphone Array.....................................................10
3.4 iPhone 8 Microphone Array.....................................................10
3.5 LG Stylo3 Microphone Array....................................................11
3.6 LGK20 Microphone Array........................................................11
3.7 Samsung Galaxy S6 Microphone Array............................................12
3.8 Samsung Galaxy S9 Microphone Array............................................13
4.1 Master Signal.................................................................16
5.1 Sample Recording Before Speech Removal........................................19
5.2 Sample Recording After Speech Removal.........................................19
5.3 LG Stylo3 With Noise Suppression..............................................20
5.4 LG Stylo3 No Noise Suppression................................................21
5.5 LG K20 With Noise Suppression.................................................21
5.6 LG K20 No Noise Suppression...................................................22
5.7 SNR Averages at 70 dB.........................................................23
viii


5.8. SNR Averages at 80 dB.........................................................23
5.9 SNR Averages Live..............................................................24
5.10 SNR Averages of Noise at 70 dB................................................24
5.11 SNR Averages of Noise at 80 dB................................................25
5.12 SNR Averages of Live Noise....................................................25
5.13. FFT iPhone 6 80 dB ANC On vs Off.............................................28
5.14 FFT iPhone 7 80 dB ANC On vs Off..............................................29
5.15 FFT iPhone 8 80 dB ANC On vs Off..............................................29
5.16 FFT iPhone 7 Live ANC On vs Off...............................................30
5.17 FFT LG K20 80 dB ANC On vs Off................................................30
5.18 FFT LG K20 Live ANC On vs Off.................................................31
IX


LIST OF TERMS
TERM
SNR - Signal to Noise Ratio ANC - Adaptive Noise Cancellation FFT - Fast Fourier Transform Hz - Hertz
PCM - Pulse Code Modulation SPL - Sound Pressure Level
x


CHAPTER I
INTRODUCTION
The first active noise control system was patented in 1936 by developer Paul Lueg. Lueg’s system [5] was known as the “Process of Silencing Sound Oscillation.” Since then, active noise control systems have made their way into a variety of devices, most notably headphones, automobiles, and cell phones. In 2008, Audience introduced a development in cell phones that enabled them to utilize adaptive noise control in voice calls. The feature was aimed at improving caller intelligibility and reducing background noise [6], Soon after the market began to change and adopted the feature for their products.
Noise cancellation is a feature that can be found in most cell phones. Depending on the manufacturer, the active noise cancellation can be referred to as “noise cancellation” or “noise suppression.” Since this feature is so common, it is imperative to study the abilities and unique qualities of noise cancellation for forensic research purposes. This thesis considers seven different cell phones, as follows: iPhone 6 iPhone 7 iPhone8 LG Stylo3 LG K20 Samsung S6 Samsung S9
1


Background
The purpose of noise cancellation in a cell phone conversation is to increase caller clarity and reduce background noise. Cell phone conversations occur in a multitude of acoustic noise environments and as a result, cell phone conversation can be plagued with unwanted background noise that can deteriorate the call quality.
The solution is to use adaptive noise cancellation. For noise cancellation to be possible a cell phone needs to have multiple microphones to access the incoming signals. The first cell phones to offer adaptive noise cancellation did not have an additional microphone. Headphones were needed to activate the feature by providing a second microphone. Later designs in the early 2010’s provided a microphone array for filtering. Often times, a cell phone manufacturer will license through a third party to develop the active noise canceling application. Like how Apple used the company Cirrus Logic to develop audio chips for their product [7],
Since cell phone manufacturers, such as Apple and Samsung, use a proprietary algorithm [6] for their products noise cancellation feature, this could indicate that different cell phone manufacturers adaptive noise cancellation filters would hinder different results in caller quality. An analysis of the speech quality could help identify cell phone brand or manufacture. SNR can be used as a mathematical measurement of filter quality. This could offer a way to identify a cell phone by brand or model. Recording SNR values in how they pertain to cell phone brand and model is beneficial for forensic research, since the data could later be used in a database for further forensic research.
Some notable third-party companies that produce adaptive noise cancellation technologies for cell phone manufacturers and their chipsets include:
Qualcomm Snapdragon
2


Audience A1010 Voice Processor
Cirrus Logic
A large number of cell phone’s utilize Qualcomm Snapdragon chips, which offers an adaptive noise cancellation feature [8], However, it appears that the cell phones user interface might be responsible for the features and quality of the adaptive noise filter. For instance, Sony offers an adaptive noise cancellation filter that allows the user to change the setting based on the type of noisy environment they are in [9], This feature seems to be unique to the Sony Xperia user interface. The Samsung’s user interface no longer allows the user to turn on or off the adaptive noise cancellation filter. Cell phone manufacturers develop their own user interfaces to control and change how the Operating System performs certain tasks [10],
Scope and Limitations
The scope of the proposed methods has been limited to three individual cell phone manufacturers. LG, Apple, and Samsung. The limitation of this study is that it has not yet categorized cell phone brand by the chipsets. In addition, this thesis does not actively assume which algorithm for adaptive filter a manufacturer uses. It does attempt to interpret and record key attributes in a cell phone manufacturers’ use of adaptive noise cancellation filters.
3


CHAPTER II
ADAPTIVE NOISE REDUCTION FILTER Introduction Into Filters
The term “filter” is a derivative of the electrical engineering world, where a filter performs a transformation of an electrical signal into something different [11], The point of using a filter is to remove unwanted parts of a signal. The unwanted signal could be noise or a certain frequency range. For this paper, we can separate filters into two different categories known as fixed and adaptive filters. A fixed filter estimates the input signal and removes the unwanted noise and is dependent on prior knowledge of the input signals. Since prior knowledge of the noise characteristics is vital in the design of a fixed filter, it limits the filters used in a real-life scenario. For instance, a fixed filter would yield minimal results in filtering out the noise of a busy city, since the noisy environment of a city generally has constantly changing noise. However, adaptive filters require minimal or no prior knowledge of the input signals’ noisy characteristics. This makes adaptive filters perfect for noise cancellation applications. [12]
How Adaptive Filters Work
Figure 1 shows a general layout for an adaptive filter, with x(k) being the input signal, d(k) being the desired response, h(k) being the impulse response of the filter, and y(k) being the filtered output. If filter x(k) filtered we get y(k), we subtract y(k) from the desired input d(k).
This would result in e(k)[13]. From there, we pass the signal into a coefficient adjustment algorithm, labeled in figure 1 “Adaptive Algorithm.” After the signal passes through the adaptive algorithm, it is fed back into the adaptive filter to make an adjustment to the signal coefficients.
In essence, it changes how the filter is filtering the signal. The coefficients in an adaptive filter
4


are the various impulse responses from the signal, the most common algorithms being least mean
square, recursive least square and normalized least mean square.
X(K)
Figure 2.1 Adaptive Filter Signal Flow
Adaptive Noise Cancellation Filters In Cell Phones
Adaptive noise cancellation filters acclimate to the noise of the changing environment and filter out unwanted sounds. Adaptive noise cancellation works by utilizing two inputs, which are the primary and references inputs [13], An adaptive noise cancellation filter application on a cell phone can be demonstrated in figure 2 - ANC Signal Flow. The primary signal in a cell phone conversation should be comprised of the caller’s voice, and the background noise would be comprised of a noisy city street. The background noise is decreasing the intelligibility of the caller’s voice. The reference input is now comprised of the source of the noise. It is important to note that the reference input noise source has no relation to the caller’s voice; the noise source is associated with the noisy city street. The noise source is passed through an adaptive filter where
5


it produces a filtered output noise. The filtered output noise is then subtracted from the primary
signal and creates a desired output signal.
PJdMAlfy SIGNAL=VOICE+NOISE „ +- DESIRED SIGNAL OUTPUT
----------------------------—r—
FILTERED OUTPUT NOISE
REFERENCE=NOISE SOURCE
ADAPTIVE FILTER
Figure 2.2 ANC Signal Flow
Adaptive Filter Algorithms
There are many different types of algorithms used for adaptive filters. The most common types consist of LMS (least mean square), NLMS (normalized least mean square), and RLS (recursive least square) [14], The implementation of the algorithm is used to generate desired results. Different algorithms poses both wanted and unwanted outcomes. In the study Analysis of Adaptive Filter Approach for Speech Enhancement Using Simulink, International Journal of Advance Research, Ideas and Innovations in Technology [15], the authors found that the RLS algorithm provides a faster and smaller error rate when processing coefficients in an unknown setting. The RLS algorithm was recommended for adaptive filtering. The study Comparison between Adaptive filter Algorithms (LMS, NLMS and RLS) [16], also found the same results involving the RLS algorithm, and recommended this algorithm for speech enhancement adaptive
6


filters. It is important to understand the attributes of these algorithms due to the fact that they play a crucial role in the adaptive filter design.
7


CHAPTER III
CELL PHONE LAYOUT General Cell Phone Microphone Layout
Most cell phones contain an array of microphones in order to record the multiple signals involved in a phone call and active noise reduction. Figure 3 demonstrates a general layout for a cell phone's microphone array. The microphone on the bottom is used to record the voice signal while the microphone at the top of the phone near the camera is used to record the noise signal source. In figures 4-10 we can see the general microphone array layout for iPhones, LG, and Samsung cell phones. This appears to be the optimal layout for a cell phone microphone array.
LED indicator Front camera Earpiece-------
Home key Back key —
< Front >
Proximity/ Light sensor
Touchscreen
Recent apps key Speakers
< Top >
Speaker key
< Bottom >
Headset jack
Charger/USB port
< Sides »
Charging
contacts
Back camera Flash Wide camera
< Back >
Figure 3.1 General Cell Phone Microphone Array
8


Apple iPhone
Apple first introduced the noise cancellation feature on the iPhone 4. Since then, various brands have implemented the noise cancellation or reduction feature in their devices.
Apple licensed the company Audience A1010 Voice Processor [17], However, as of 2010, Apple has employed the company Cirrus Logic to develop their noise reduction chips. This thesis examines the adaptive noise cancellation filter for the iPhone 6, iPhone 7 and iPhone 8. The microphone layout in figures 4, 5 and 6 shows the basic layout used. One microphone is located at the bottom of the phone and the other two microphones are located on the top of the phone.
[Receive; .'â– >.â– >
ifnlOMptlr.:.-> â– â– .:
FaceTir? camra
rang/S«t.- ■ >’ ttiwiboh
vwuwia -buttons
Ap-p icon.-.
MmIM-TO:.. -.. •:
M-ea-dsel jack----
| Bottom r c:
Figure 3.2 iPhone 6 Microphone Array. By
https://www.att.com/devicehowto/index.html#!/devicediagram?make=Apple&model=iPhone6
9


iPhone 7
External Buttons and Connectors
iPhone 8
External Buttons and Connectors
Home/Touch ID sensor Volume up/down Ring/Silent switch
Side button Built-in stereo speaker Built-in microphone
Lightning connector
Built-in microphone Built-in stereo speaker
Figure 3.3 iPhone 7 Microphone Array, by https://www.apple.com/iphone-8/specs/
Tech Specs
Home/Touch ID sensor Volume up/down Ring/Silent switch
Side button Built-in stereo speaker Built-in microphone
Lightning connector
Built-in microphone Built-in stereo speaker
Figure 3.4 iPhone 8 Microphone Array, by https://www.apple.com/iphone-8/specs/
10


LG
LG cell phones utilize Qualcomm Snapdragon processor chips for active noise cancellation filtering. For this study, an LG Stylo 3 and K20 phone were examined. Their microphone arrays are shown in figures 7 and 8. In LG cell phones, the noise cancellation filter is referred to as “noise suppression.” Please note that the microphone array in figure 3.5 dictates that the LG Stylo 3 only has one microphone. This could indicate that only one microphone records both the voice signal and the background noise signal.
PrOKimity/Ambient Light Sensor
LED Light
Touchscreen
Back Key
Microphone
Earpiece
Front-Facing Camera Lens
Overview Key Home Key
Stylus Pen
Power/Lock
Button,
Fingerprint
Sensor
Speaker
Back-Facing Camera Lens
Flash
Vtolume Buttons
3.5mm Headset Jack Charger/Accessory Pori
Figure 3.5 LG Stylo 3 Microphone Array by LG Stylo 3 manual
Top and Bottom Views
0 Headset jack
(E) Microphone
0 USB/Charger port 0 Microphone
Figure 3.6 LG K20 Microphone Array by LGK20 manual
11


Samsung
Samsung deploys Qualcomm Snapdragon chips into their devices, is responsible for the adaptive noise cancellation. Previous Samsung Galaxy phones used to allow the user to specify if they would like to use noise suppression or not. This was provided by a toggle switch in the call setting menu. With the introduction of the Samsung Galaxy S6, this option is no longer available. All Samsung cellular products now have the noise suppression turned on by default. This information was provided by Samsung customer support. The noise suppression features were tested on Samsung Galaxy S9 and Samsung Galaxy S9. Figure 9 demonstrates the microphone layout for Samsung Galaxy S6, and figure 10 demonstrates the microphone layout for Samsung Galaxy S9. Both phones utilize a microphone on the bottom and the top.
Earpiece Proximity and Light Sensors
LED Indicator----
Volume key—
Touchscreen
Recents key
Front
Camera
—Power key
Back key
Home key / Fingerprint Sensor
Infrared Transmitter
Microphone
I
SIM Card
I
r

Rear-
Camera
•8
Speaker-

Microphone
•Flash
â– Heart Rate Monitor
Headset Jack
USB Charger/Accessory Port
Figure 3.7 Samsung Galaxy S6 Microphone Array by Samsung Galaxy S6 manual
12


Rear camera
SIM card / Memory card tray
Flash
Heart rate sensor
Earphone jack
Figure 3.8 Samsung Galaxy S9 Microphone Array by Samsung Galaxy S6 manual
13


CHAPTER IV
TESTING
It is imperative to implement and test a cell phone’s adaptive noise cancellation feature in different environments. Two tests were conducted for the purpose of this thesis. Both Test 1 and Test 2 were replicated three different times with the adaptive noise cancellation feature on and off. This was done for comparison reasons. Two cell phones were needed to conduct the test.
Cell Phone A represents the caller's cell phone. Cell phone B represents the receivers signal cell phone. To acquire the recordings the application an named Another Call Recorder by the company NLL APPS was installed on the receiver signal cell phone. To mitigate, bias both tests were deployed in different locations and studios. It is important to note that cell phone noise cancellation feature needs to signal to have both sources active at the same time. The first signal was the caller's voice and the second signal was the noise. In addition, it is important to note that Samsung cell phones models from Galaxy S4 to Galaxy S5 allow the consumer to turn the active noise cancellation on and off. However, turning adaptive noise cancellation off for Samsung Galaxy S to S9 was not an option due to the fact that the noise suppression feature is built into the phone. Without the option to turn it off. This was confirmed by Samsung Technical Support.
Both tests were modeled after the tests found in the studies Dual-microphone noise reduction for mobile phone application [18], and Background Noise Reduction Design for Dual Microphone Cellular Phones: Robust Approach [19], Both studies aimed to test the adaptive noise filter of a cell phone and its enhancement quality of a caller’s speech. In Dual-microphone noise reduction for mobile phone application, the approach to testing was to record and compare the adaptive noise filter of a cell phone in three different live scenarios. The first scenario was recorded in an office space, the second was recorded in a cafeteria, and the third scenario was
14


recorded in a bus. In Background Noise Reduction Design for Dual Microphone Cellular Phones: Robust Approach, the approach for testing was to examine different cell phones’ adaptive filters in a lab scenario. Noise generators were introduced in different locations while the source cell phone recorded the noise and filtered it out. These two tests provided a fantastic approach to produce and examine a cell phone's adaptive noise filter. The live and lab testing influenced the approach in the following test; Lab testing and Live Scenario Testing.
Lab Testing
The laboratory test consisted of playing a broadband noise labeled as “Master Waveform” over loudspeakers, while a human spoke into cell phone A.
Cell phone A transmitted the signals to Cell phone B, where the application Another Call Recorder recorded the signals. Each sample recording consisted of 20 seconds. Seven different cell phones were used in Test 1. Each cell phone was recorded 3 times with the noise cancellation filter and 3 times without. In addition, the source cell phone was recorded at two different volume levels. The first volume level for the Master Waveform was set at 70 dBa. 70 dBa fs would be the average volume level for a noisy environment [20], such as traffic and that of everyday life. The second volume level for Master Waveform was set at 80 dBa. 80 dBa fs would be the above average volume level for a noisy environment like a vacuum cleaner or garbage disposal. Please note that a volume threshold is needed to trigger the noise cancelation filter. All volume dB limits were measured with an SPL meter at A weighted scaled.
The Master Waveform was needed to simulate a controlled noisy signal. White noise was used as the noise due to the fact that it disperses noise evenly across the frequency spectrum [21.] In addition to this, impulses were added at the 1-second and 19-second mark to help create a reference for future synchronization.
15


Figure 4.1 Master Signal
ARC recorded all test recordings at 16bit, 44100 kHz, WAV PCM. Cell phone communication only transmits at the frequency range of 8bit 8 kHz, This reaction then caused all sample recordings to be downsampled to 8bit, Wav PCM, and all redundant data above 8 kHz to be removed. Comparison of the difference between the noise cancelation filter being turned on or off and by manufacture model can be seen in Chapter 5. To mitigate bias, Test 1 was performed at different labs in different parts of the country. The main location for lab testing was performed in Denver, CO Other locations included Colorado Springs, CO, Davis, CA and Berkeley, CA.
The replication of Test 1 in different environments hindered similar results.
Live Scenario Testing
The implementation of the live scenario testing was performed to replicate the methods of the lab test in a real-life scenario. The caller's cell phone was operated by an individual providing the speech signal. The noise signal is the natural acoustical environment of the marketplace. Implementation of a real-life scenario was performed at a marketplace during rush hour, which can result in a volume level of 70 dBa. The same method of recording between the caller's cell phone and the receiving cell phone from the lab test was replicated in the live scenario. Each cell phone recorded three 20-second test recordings with the noise cancellation filter on and 3 test recordings with the noise cancellation filter off. Since playing broadband noise is not possible in a public location, a twenty-second sample was recorded as an exemplary
16


reference noise sample. In addition to this, clicks by a finger-snap were implemented into the background at the 1-second and 19-second mark of test sample recordings. This was done to mimic the impulses in the Master Waveform from the lab test for synchronization in later analysis. Results for the real-life scenario can be seen in chapter 5.
17


CHAPTER V
ANALYSIS AND RESULTS
All sample recordings from the lab test were imported into Adobe Audition, where each sample recording was synced in the timeline by utilizing the impulses at the 1-second and 19-second mark. After syncing, all test recordings were trimmed to 20-second increments. All recordings from the Live Test were also imported into Adobe Audition. The Live Test recordings do not have impulses at the 1- and 19-seconds marks. They were synchronized by the snaps at the beginning of the live test recordings.
In addition to trimming, all test recordings underwent a speech signal removal process. Figure 12 demonstrates the test recording in its full form. Figure 13 represents the test recording after the speech signal was removed. Removing the speech signal from the sample recording allowed further analysis of the effects of adaptive filter on the noise signal in the original sample recordings.
18


Figure 5.1 Sample Recording Before Speech Removal
AFTER
;.IO +odB
Figure 5.2 Sample Recording After Speech Removal
19


Spectrogram
A spectrogram is a great way to capture and visualize the adaptive noise filter at work. The spectrograms in figures 5.3 and 5.4 are from two recordings made by an LG Stylo by utilizing test/method 1. The recording in figure 5.3 is with noise cancellation and the recording in figure 5.4 is without. In figure 14, we can see that the noise is evenly reduced as indicated by the breaks in the red color. This tells us that the noise is filtering out the noise signal and allowing the voice signal to pass through. We can also see these results for the test recordings in figures 5.5 and 5.6.
Figure 5.3 LG Stylo3 With Noise Suppression
20



Figure 5.4 LG Stylo3 No Noise Suppression
Figure 5.5 LG K20 With Noise Suppression
21


Figure 5.6 LG K20 No Noise Suppression
SNR Results
SNR or Signal to Noise Ratio was used to plot every test cell phone with and without the adaptive noise filter on and off. Similarly, in the paper Dual-microphone noise reduction for mobile phone application [18], the authors used SNR to analyze the SNR relationship between their test cell phone’s adaptive noise filtering. For the SNR analysis, it was important to compare an SNR relationship between a cell phones noise-canceling feature turned on and off. In addition, the SNR was compared between different cell phone brands. The SNR helps to apply a calculated representation of the ANC filter in use over time.
Figures 16 to 45 plots signal to noise over time against the originally recorded sample. SNR averages between cell phones and volume threshold, in relation to the adaptive noise filter on vs off can be seen in figures 18 through 23. In addition to this SNR average values of the
22


sample, recording can be seen in tables 5.1 through 5.6. The error bar in the following figures 5.1
through 5.6 represents the standard deviation.
SNR 70dB Averages ANC Filter On vs ANC Filter Off
iPhone6 iPhone7 iPhone8 LGStylo3 LGK20 SamsungS6 SamsungS9
Cell Phone
Figure 5.7 SNR Averages at 70 dB
SNR 80dB AVERAGES ANC Filter On vs ANC Filter Off
iPhone6 iPhone7 iPhone8 LGStylo3 LGK20 SamsungS6 SamsungS9
Cell Phone
Figure 5.8 SNR Averages at 80 dB
23


Live SNR Averages ANC Filter On vs ANC Filter Off
70 r
60 -
iPhone7 LG Stylo3 LG K20 iPhone7
Cell Phone
I ANC Filter On
I I ANC Filter Off
Figure 5.9 SNR Averages Live
SNR 70dB Noise Averages ANC Filter On vs ANC Filter Off
iPhone6 iPhone7 iPhone8 LGStylo3 LGK20 SamsungS6 SamsungS9
Cell Phone
Figure 5.10 SNR Averages of Noise at 70 dB


SNR 80dB Noise Averages ANC Filter On vs ANC Filter Off
Cell Phone
Figure 5.11 SNR Averages of Noise at 80 dB
Live Noise SNR Averages ANC Filter On vs ANC Filter Off
20 -
iPhone7 LG Stylo3 LG K20 iPhone7
Cell Phone
|ANC Filter On
^â– ANC Filter Off
Figure 5.12 SNR Averages of Live Noise


Tables
Table 5.1 Average SNR Values at 70 dB
ANC_on_SNR ANC_off_SNR
iPhone 6 33.1707 35.0946
iPhone 7 25.3598 32.0819
iPhone 8 32.9037 33.1431
LG Stylo3 18.9703 31.9403
LGK20 36.6622 34.3940
Samsung Galaxy S6 34.9594 N/A
Samsung Galaxy_S9 21.9771 N/A
Table 5.2 Average SNR Values at 80 dB
ANC_on_SNR ANC_off_SNR
iPhone 6 33.1707 35.0946
iPhone 7 25.3598 32.0819
iPhone 8 32.9037 33.1431
LG Stylo3 18.9703 31.9403
LGK20 36.6622 34.3940
Samsung Galaxy S6 34.9594 N/A
Samsung Galaxy S9 21.9771 N/A
Table 5.3 Average SNR Values Live
ANC_on_SNR ANC_off_SNR
iPhone 7 25.1318 26.1005
LG Stylo3 38.2109 41 .2942 51 .2349
LGK20 49.2333
26


Table 5.4 Average SNR Values Noise of 70 dB
ANC_on_SNR ANC_off_SNR
iPhone 6 10.9489 11.3862
iPhone 7 6.2233 7.7634
iPhone 8 7.3209 9.3229
LG Stylo 5.2909 4.1179
LGK20 5.0634 6.5641
Samsung Galaxy S6 6.0795 N/A
Samsung Galaxy S9 6.3012 N/A
Table 5.5 Average SNR Values Noise of 80 dB
ANC_on_SNR ANC_off_SNR
iPhone 6 9.1001 11.4825
iPhone 7 6.5654 9.2167
iPhone 8 6.7051 11.7282
LG Stylo3 1.5742 2.8837
LGK20 5.0634 19.6365
Samsung Galaxy S6 10.3680 N/A
Samsung Galaxy S9 8.1867 N/A
Table 5.6 Average SNR Values Live Noise
ANC_on_SNR ANC_off_SNR
iPhone 7 6.9273 7.0534
LG Stylo3 6.5650 9.2167 16.4150 16.0683
LGK2Q
27


FFT Analysis
The Fast Fourier transform (FFT) is a mathematical expression that can be used to convert an audio signal from the time domain to the frequency domain [22], This allows us to visualize an audio signal through frequency (Hz) over gain (dB). Where frequency utilizes the x-axis and gain(dB) utilizes the Y-axis, as displayed in figure 5.14.
All background noise samples discussed previously in this chapter were processed through FFT analysis. Each recording was classed into categories by phone, adaptive noise filter setting and volume threshold of the noise. For instance, iPhone test samples that were recordings at 70 dB were compared between the adaptive noise filter being on and off. The blue frequency in figures 5.14 through 5.19 represents the FFT frequency analysis with the adaptive noise filter on, where all red frequencies in figures 5.14 through 5.19 represent the FFT frequency analysis with the adaptive filter off.
I ADAPTIVE FILTER ON
â–¡ ADAPTIVE FILTER OFF
- -90
E-95
--100
-105
E-no
-115
28


Figure 5.13 FFT iPhone 6 80 dB ANC On vs Off
Figure 5.14 FFT iPhone 7 80 dB ANC On vs Off
29


Figure 5.16 FFT iPhone 7 Live ANC On vs Off
Figure 5.17 FFT LG K20 80 dB ANC On vs Off
30


31


CHAPTER VI
CONCLUSION
With the lab test, it is apparent that after analyzing the SNR average values and viewing a spectrogram that one could detect if a cell phone has the adaptive noise filter or noise cancellation on or off. For instance, the LG Stylo3, at 80 dB, has an increase in its SNR when noise reduction is off. This would indicate that the noise is bleeding into the desired signal of the sample recording. One can see similar results with the plotted SNR in figures and tables listed in Chapter 5. The lab test hindered similar SNR average differences between filter being on and off across all test cell phones. The increase in SNR average when the adaptive noise filters off makes sense, due to the fact that more of the noise signal is not being suppressed.
The FFT analysis of the noise signals also yielded some interesting results. All iPhone cell phones’ noise samples performed a consistent dip at 4,000 Hz and remained at a volume level of around 120 dB. We can see this consistency between all iPhones in figures 5.15, 5.16, 5.17 and 5.7. All other phones veered inconsistently around different volume levels, except for the LG K20. The LG K20’s FFT analysis proposed some interesting results. A spike in volume around 12000 HZ and 20000 HZ can be seen in figures 5.18 and 5.19. Figure 5.18 was from a noise sample that was recorded in the lab testing. Where figure 5.19 was from a noise sample that was recorded in the live testing. No other phones created these unique spikes.
In addition, spectrogram analysis of the LG Stylo3 in figures 5.3 and 5.4 clearly demonstrated a visual representation of the adaptive noise filter in action. We can also see this in the LG K20 in figures 5.6 and 5.7. The noise-only samples from the test recording further demonstrated that the method was successful in indicating whether the adaptive noise filter was
32


on or off. The SNR averages showed a clear increase in dB when the filter was switched off, as seen in tables 5.1 and 5.2.
It should be noted that the LG Stylo3’s SNR performance was poor and inconsistent at times. It seemed that the higher the simulated background noise the worse the filter performed. The SNR average values in figures 5.4 and 5.3 were drastically different than the other six phones. This may be due to the fact that the LG Stylo3 does not possess a second microphone to attain the noise signal. Intense noise reduction is needed in a single microphone system, which can result in lower speech quality [23],
The method discussed in the live scenario tests yielded mixed results. Live recordings were tested on an iPhone 7, LG Stylo3 and LG K20. With the live testing of the LG K20, very little differences in SNR averages were recorded. Unfortunately, spectrogram analysis was unsuccessful in detecting differences in adaptive noise reduction filter being turned on or off.
The iPhone 7’s noise sample generated similar results in the FFT analysis, however, the FFT analysis of the Live Scenarios iPhone 7 noise sample generated similar results to the lab test’s iPhones. The Live Test scenario of the iPhone 7 demonstrated a frequency dip at for 4000 HZ and a consistent volume level of around 120 dB.
After finalizing all testing, it would appear that in order to fully detect whether adaptive noise canceling is on or off, the dB level would have to be at a high level. The higher the volume threshold, the more active the noise filter. Test recordings in the lab test with volume levels of 70 dB were concluded with very little results. However, when volume levels were increased to 80 dB, the effects of the adaptive noise filter could be seen and recorded.
33


Since the proposed method of capturing the adaptive noise cancellation filter in action is obtainable, further research into categorizing cell phone SNR values by manufacturer and model is possible for forensic use.
34


CHAPTER VII
FURTHER RESEARCH
Since adaptive noise cancellation filters are prevalent in most cell phone brand and models. It would be neglectful of the forensic scientific community to ignore such a key attribute. The proposed methods from chapter 4 could be utilized to test other cell phone manufacturers. For instance, cell phones from manufacturers like Sony, Google and HTC, Asus and Huawei were not tested for this thesis. All of these brands offer a noise reduction feature and need to be analyzed in similar manners listed in chapter Five. It seems that SNR values of the noise reduction filters exhibits unique qualities in relation to cell phone models and brands. FFT analysis of the noise signal also provided some unique characteristics. Further forensic research of this subject is needed. A creation of a database that recorded SNR and FFT results could lead to advancements in cell phone forensic audio analysis. A subject in forensic science that is vastly overlooked.
35


REFERENCES
[1] Bagnell, Juan Carlos. “Creating Content: the State of Smartphone Noise Reduction in 2016.” Pocketnow, 4 Feb. 2016, pocketnow.com/smartphone-noise-reduction.
[2] Triggs, Robert. “Why Noise Cancellation Is the Most Important Tech in Mobile Right Now.” Android Authority, Android Authority, 22 Dec. 2017, www.androidauthoritv.com/noise-cancellation-explained-820149/.
[3] Williams, Martyn. “Apple Sued over Noise Reduction Technology.” Macworld, Macworld, 9 July 2012,
www.macworld.com/article/1167599/apple sued over noise reduction technology.html
[4] Haniltji, Cemal, and Tomi Kinnunen. “Source Cell-Phone Recognition from Recorded Speech Using Non-Speech Segments.” Digital Signal Processing, vol. 35, 2014, pp. 75-85., doi:10.1016/j.dsp.2014.08.008. Accessed20Nov. 2018.
[5] Oinonen, Mika. Development of Active Noise Control and Voice Communication Systems for Personal Hearing Protectors . TAMPEREEN TEKNILLINEN YLIOPISTO, TAMPERE UNIVERSITY OF TECHNOLOGY, 22 Sept. 2006, tutcris.tut.fi/portal/files/1517363/oinonen.pdf.
[6] Watts, Lloyd, et al. Voice Processor Based on the Human Hearing System . Audience, 25 Aug. 2008, www.hotchips.org/wp-content/uploads/hc archives/hc20/2 Mon/HC20.25.320.pdf.
[7] Eassa, Ashraf. “Cirrus Logic Wins Spot in Next-Generation Apple AirPods.” The Motley Fool, The Motley Fool, 29 June 2018, www.fool.com/investing/2018/06/29/cirrus-logic-wins-spot-in-next-generation-apple-ai.aspx..
[8] “Device Finder.” Qualcomm, 19 July 2018, www.qualcomm.com/snapdragon/devices/all.
[9] Jesdanun, Anick. “First Look: Sony's Xperia Z2 Noise-Cancelling Tech Doesn't Impress-Technology News, Firstpost.” Firstpost, Firstpost, 27 Feb. 2014, www.firstpost.com/tech/news-analvsis/first-look-sonys-xperia-z2-noise-cancelling-tech-doesnt-impress-3648399.html.
[10] Rose, Brent. “What's the Point of Android Skins?” Gizmodo, Gizmodo.com, 17 June 2013, gizmodo.com/5963773/whats-the-point-of-android-skins.
[11] R. W. Hamming, Digital Filters. Englewood Cliffs, N.J: Prentice-Hall, 1977
[12] Gang, Ren, et al. “Audio Phase Singularity Detection for Room Acoustics Parameter Estimation.” Ieee, 2012 IEEE International Conference on Consumer Electronics (ICCE), 2012, pp. 77-78, doi: 10.1109/ICCE.2012.6161748. Accessed 25 Oct. 2018.
[13] Singh, Aarti. “Adaptive Noise Cancellation ” Dept, of Electronics & Communication Netaji Subhas Institute of Technology, 2001.
36


[14] V.p.chandrasekharyadav, G., and B. Ananda Krishna. “Study of Different Adaptive Filter Algorithms for Noise Cancellation in Real-Time Environment.” International Journal of Computer Applications, vol. 96, no. 10, 2014, pp. 20-25., doi: 10.5120/16829-6589.
[15] “ B. Chandrashaker Reddy, D. Vamsi Krishna, Shrinath Raygond, N. Ravi Teja. Analysis of Adaptive Filter Approach for Speech Enhancement Using Simulink, International Journal of Advance Research, Ideas and Innovations in Technology, www.IJARnD.com.
[16] Dhiman, J. et al. “Comparison between Adaptive filter Algorithms ( LMS , NLMS and RLS ).” (2013).
[17] “Apple Unlikely to Use Audience's Noise Suppression Technology in IPhone 5.” IPhone Hacks | #1 IPhone, IPad, IOS Blog, 7 Sept. 2012, www.iphonehacks.com/2012/09/audience-noise-suppression-technology-unlikelv-in-iphone-5.html.
[18] Z. Fu, F. Fan and J. Huang, "Dual-microphone noise reduction for mobile phone application," in 2013, . DOI: 10.1109/ICASSP.2013.6639068.
[19] Y. Chen et al, "Background Noise Reduction Design for Dual Microphone Cellular Phones: Robust Approach," IEEE ACM Transactions on Audio, Speech and Language Processing (TASLP), vol. 25, (4), pp. 852-862, 2017.
[20] “Comparative Examples of Noise Levels.” Industrial Noise Control, www.industrialnoisecontrol.com/comparative-noise-examples.htm
[21] Hansen, Colin. FUNDAMENTALS OF ACOUSTICS. Department of Mechanical Engineering University of Adelaide, 2001,
www.who.int/occupational health/publications/noisel.pdf.
[22] Heckbert, Paul. Fourier Transforms and the Fast Fourier Transform (FFT) Algorithm. Jan. 1995, www.cs.cmu.edu/afs/andrew/scs/cs/15-463/2001/pub/www/notes/fourier/fourier.pdf.
[23] “Leveraging Advanced Noise Cancellation Technologies in Snapdragon Processors.” Technical Note, Qualcomm & Inforce Computing, Apr. 2016,
www.inforcecomputing.com/public docs/WhitePapers/Inforce support for Fluence Audio 04-2016-1,pdf.
37


Full Text

PAGE 1

P ROPOSED METHOD FOR R ECORDING AND ANALYZING C ELL PHONE A DAPTIVE N OISE R EDUCTION F ILTERS b y ANTHONY NELSON B.S., University of Colorado Denver, 201 8 A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfilment of the requirements for the degree of Master of Science Recording Arts Program 2018

PAGE 2

i This thesis for the Master of Science degre e by Anthony Nelson Has been approved for the Recording Arts Program b y Catalin Grigoras, Chair Jeff Smith Cole Whitecotton Date: December 15 , 2018

PAGE 3

ii Nelson, Anthony (M.S., Recording Arts Program) Proposed Method for Recording and Analyzing Cell Phone Adaptive Noise Reduction Filters Thesis directed by Associate Professor Catalin Grigoras ABSTRACT During cell phone conversations, the noises from the world around us can interfere with call clarity ; noise being an unwanted sound that ruins the quality of the signal. One solution to minimize the noise is to utilize adaptive noise cancellation filters. The majority of cell phone manufacturers have implemented this technology [1][2] into their products , aiming to increase caller intelligibility by reducing ambient background noise. Many aspects of a cell phone's noise filtering features are proprietary, [3] which is unfortunate from a forensic standpoint. The algorithms used in many cell phone adaptive noise cancellation systems are kept private from the public. There is a void in forensic audio analysis of cell phones [4]. Very little forensic research has been devoted to the experimentation of adaptive noise filters, especially in regards to their use in cell phones. Since a large number of cell phones utilize such filtering, it is important that the forensic community considers this key attribute offered in cell phones today. The adaptive to forensic sciences. This thesis provides in depth research into the performance of adaptive noise cancellation filters in cell phones and proposes methods to test and analyze them. A method to capture the adaptive noise filters via audio recording was te sted and proven to be successful. In addition to this, signal to noise ratio (SNR) values were recorded and categorized by cell phone maker and model. The form and content of this abstract are approved. I recommend its publication. Approved: Ca talin Grigoras

PAGE 4

iii DEDICATION I would like to dedicate this thesis to my friends and family.

PAGE 5

iv A CKNOWLEDGEMENTS I would like to thank Catalin Grigor a s, Jeff Smith, Cole Whitecotton and the NCMF program for the amazing help throughout this program. I would also like to acknowledge my friends and family for volunteering their time and cell phones for testing purposes. Especially the one and only P fiancée Nicole Pacheco, who supported me through this entire process. Without the involvement from these people my drive and ability to finish this thesis would have been altered.

PAGE 6

v TABLE OF CONTENTS CHAPTER I. INTRODUCTION ................................ ................................ ................................ ................................ ................. 1 Background ................................ ................................ ................................ ................................ ................................ 2 Scope and Limitations ................................ ................................ ................................ ................................ ........... 3 II. ADAPTIVE NOISE REDUCTION FILTER ................................ ................................ ............................ 4 Introduction Into Filters ................................ ................................ ................................ ................................ ........ 4 How Adaptive Filters Work ................................ ................................ ................................ ............................... 4 Adaptive Noise Cancellation Filters In Cell P hones ................................ ................................ .............. 5 Adaptive F ilter Algorithms ................................ ................................ ................................ ................................ . 6 III. CELL PHONE LAYOUT ................................ ................................ ................................ ................................ ... 8 General Cell Phone Microphone L ayout ................................ ................................ ................................ ...... 8 Apple iPhone ................................ ................................ ................................ ................................ ............................. 9 LG ................................ ................................ ................................ ................................ ................................ ................ 11 Samsung ................................ ................................ ................................ ................................ ................................ .... 12 IV . TESTING ................................ ................................ ................................ ................................ ................................ . 14 Lab Testing ................................ ................................ ................................ ................................ .............................. 15 Live Scenario Testing ................................ ................................ ................................ ................................ ......... 16 V. ANALYSIS AND RESU LTS ................................ ................................ ................................ ....................... 18 Spectrogram ................................ ................................ ................................ ................................ ............................ 20

PAGE 7

vi SNR Results ................................ ................................ ................................ ................................ ............................ 22 FFT Analysis ................................ ................................ ................................ ................................ ........................ 28 VI. CONCLUSION ................................ ................................ ................................ ................................ ..................... 32 VII. FURTHER RESEARCH ................................ ................................ ................................ ................................ .. 35 REFERENCES ................................ ................................ ................................ ................................ ................................ ...... 36

PAGE 8

vii LIST OF TABLES TABLE 5.1 Average SNR Values at 70 dB ................................ ................................ ................................ .............................. 26 5 .2. Average SNR Values at 80 dB ................................ ................................ ................................ ............................ 26 5.3 Average SNR Values Live ................................ ................................ ................................ ................................ ...... 26 5.4 Average SNR Values Noise of 70 dB ................................ ................................ ................................ ............... 27 5.5 Average SNR Values Noise of 80 dB ................................ ................................ ................................ ............... 27 5.6 Average SNR Values Live Noise ................................ ................................ ................................ ......................... 27

PAGE 9

viii LIST OF FIGURES FIGURE 2.1 Adaptive Filter Signal Flo w ................................ ................................ ................................ ................................ ....... 5 2.2 ANC Signal Flow ................................ ................................ ................................ ................................ ............................ 6 3.1 General Cell Phone Microphone Array ................................ ................................ ................................ ................ 8 3.2 iPhone 6 Microphone Array ................................ ................................ ................................ ................................ ....... 9 3.3 iPhone 7 Microphone Array ................................ ................................ ................................ ................................ .... 10 3.4 iPhone 8 Microphone Array ................................ ................................ ................................ ................................ .... 10 3.5 LG Stylo3 Microphone Array ................................ ................................ ................................ ................................ 11 3.6 LGK20 Microphone Array ................................ ................................ ................................ ................................ ...... 11 3.7 Samsung Galaxy S6 Microphone Array ................................ ................................ ................................ ............ 12 3.8 Samsung Galaxy S9 Microphone Array ................................ ................................ ................................ ............ 13 4.1 Master Signal ................................ ................................ ................................ ................................ ................................ .. 16 5.1 Sample Recording Before Speech Removal ................................ ................................ ................................ .... 19 5.2 Sample Recording After Speech Removal ................................ ................................ ................................ ....... 19 5.3 LG Stylo3 With Noise Suppression ................................ ................................ ................................ ..................... 20 5.4 LG Stylo3 No Noise Suppression ................................ ................................ ................................ ......................... 21 5.5 LG K20 With Noise Suppression ................................ ................................ ................................ ......................... 21 5.6 LG K20 No Noise Suppression ................................ ................................ ................................ ............................. 22 5.7 SNR Averages at 70 dB ................................ ................................ ................................ ................................ ............ 23

PAGE 10

ix 5.8. SNR Averages at 80 dB ................................ ................................ ................................ ................................ ........... 23 5.9 SNR Averages Live ................................ ................................ ................................ ................................ ..................... 24 5.10 SNR Averages of Noise at 70 dB ................................ ................................ ................................ ....................... 24 5.11 SNR Averages of Noise at 80 dB ................................ ................................ ................................ ....................... 25 5.12 SNR Averages of Live Noise ................................ ................................ ................................ ............................... 25 5.13. FFT iPhone 6 80 dB ANC On vs Off ................................ ................................ ................................ ............. 28 5.14 FFT iPhone 7 80 dB ANC On vs Off ................................ ................................ ................................ ............... 29 5.15 FFT iPhone 8 80 dB ANC On vs Off ................................ ................................ ................................ ............... 29 5.16 FFT iPhone 7 Live ANC On vs Off ................................ ................................ ................................ .................. 30 5.17 FFT LG K20 80 dB ANC On vs Off ................................ ................................ ................................ ................ 30 5.18 FFT LG K20 Live ANC On vs Off ................................ ................................ ................................ ................... 3 1

PAGE 11

x LIST OF TERMS TERM SNR Signal to Noise Ratio ANC Adaptive Noise Cancellation FFT Fast Fourier Transform Hz Hertz PCM Pulse Code Modulation SPL Sound Pressure Level

PAGE 12

1 CHAPTER I INTRODUCTION The first active noise control system was patented in 1936 by developer Paul Lueg. active noise control systems have made their way into a variety of devices, most notably headphones, automobiles, and cell phones. In 2008, Audience introduced a development in cell phones that enabled them to utilize adaptive noise control in voice calls. The feature was aimed at improving caller intelligibility and reducing background noise [6]. Soon after the market began to change and adopted the feature for their products. Noise cancellation is a feature that can be found in most cell phones. Depending on the qualities of noise cancellation for forensic researc h purposes. This thesis considers seven different cell phones, as follows: iPhone 6 iPhone 7 iPhone8 LG Stylo3 LG K20 Samsung S6 Samsung S9

PAGE 13

2 Background The purpose of noise cancellation in a cell phone conversation is to increase caller clarity and reduce background noise. Cell phone conversations occur in a multitude of acoustic noise environments and as a result, cell phone conversation can be plagued w ith unwanted background noise that can deteriorate the call quality. The solution is to use adaptive noise cancellation. For noise cancellation to be possible a cell phone needs to have multiple microphones to access the incoming signals. The first cell p hones to offer adaptive noise cancellation did not have an additional microphone. Headphones were needed to activate the feature by providing a second microphone. Later designs in the early ell phone manufacturer will license through a third party to develop the active noise canceling application. Like how Apple used the company Cirrus Logic to develop audio chips for their product [7]. Since cell phone manufacturers, such as Apple and Samsun g, use a proprietary algorithm [6] for their products noise cancellation feature, this could indicate that different cell phone manufacturers adaptive noise cancellation filters would hinder different results in caller quality. An analysis of the speech qu ality could help identify cell phone brand or manufacture. SNR can be used as a mathematical measurement of filter quality. This could offer a way to identify a cell phone by brand or model. Recording SNR values in how they pertain to cell phone brand and model is beneficial for forensic research, since the data could later be used in a database for further forensic research. Some notable third party companies that produce adaptive noise cancellation technologies for cell phone manufacturers and their chi psets include: Qualcomm Snapdragon

PAGE 14

3 Audience A1010 Voice Processor Cirrus Logic noise cancellation feature [8]. However, it appears that the cell phones user interface might be responsible for the features and quality of the adaptive noise filter. For instance, Sony offers an adaptive noise cancellation filter that allows the user to change the setting based on the type of noisy environment they are in [9]. Thi s feature seems to be unique to the Sony Xperia user noise cancellation filter. Cell phone manufacturers develop their own user interfaces to control and chang e how the Operating System performs certain tasks [ 10]. Scope and Limitations The scope of the proposed methods has been limited to three individual cell phone manufacturers. LG, Apple, and Samsung. The limitation of this study is that it has not yet categ orized cell phone brand by the chipsets. In addition, this thesis does not actively assume which algorithm for adaptive filter a manufacturer uses. It does attempt to interpret and record cancellation filters.

PAGE 15

4 CHAPTER II ADAPTIVE NOISE REDUCTION FILTER Introduction Into Filters performs a transformation of an electrical signal into something different [11]. The point of using a filter is to remove unwanted parts of a signal. The unwanted signal could be noise or a certain frequency range. For this paper, we can separate filters into two different categories known as fixed and adaptive filters. A fixed filter estimates the input signal and removes the unwanted noise and is dependent on prior knowl edge of the input signals. Since prior knowledge of the noise characteristics is vital in the design of a fixed filter, it limits the filters used in a real life scenario. For instance, a fixed filter would yield minimal results in filtering out the noise of a busy city, since the noisy environment of a city generally has constantly changing noise. characteristics. This makes adaptive filters perfect for noise cancel lation applications. [ 12] How Adaptive Filters Work Figure 1 shows a general layout for an adaptive filter, with x(k) being the input signal, d(k) being the desired response, h(k) being the impulse response of the filter, and y(k) being the filtered outpu t. If filter x(k) filtered we get y(k), we subtract y(k) from the desired input d(k). This would result in e(k)[13]. From there, we pass the signal into a coefficient adjustment hrough the adaptive algorithm, it is fed back into the adaptive filter to make an adjustment to the signal coefficients. In essence, it changes how the filter is filtering the signal. The coefficients in an adaptive filter

PAGE 16

5 are the various impulse responses from the signal, the most common algorithms being least mean square, recursive least square and normalized least mean square. Figure 2.1 Adaptive Filter Signal Flo w Adaptive Noise Cancellation Filters In Cell Phones Adaptive noise cancellation filters acclimate to the noise of the changing environment and filter out unwanted sounds. Adaptive noise cancellation works by utilizing two inputs, which are the primary and references inputs [ 13]. An adaptive noise cancellati on filter application on a cell phone can be demonstrated in figure 2 ANC Signal Flow. The primary signal in a cell be comprised of a noisy city street. The back ground noise is decreasing the intelligibility of the associated with the noisy city street. The noise source is passed through an adaptive filter where

PAGE 17

6 it produces a filtered output noise. The filtered output noise is then subtracted from the primary signal and creates a desired output signal. Figure 2.2 ANC Signal Flow Adaptive F ilter Algorithms There are many different types of algorithms used for adaptive filters. The most common types consist of LMS (least mean square), NLMS (normalized least mean square), and RLS (recursive least square) [14]. The implementation of the algorithm is used to gen erate desired results. Different algorithms poses both wanted and unwanted outcomes. In the study Analysis of Adaptive Filter Approach for Speech Enhancement Using Simulink, International Journal of Advance Research, Ideas and Innovations in Technology [ 15 ], the authors found that the RLS algorithm provides a faster and smaller error rate when processing coefficients in an unknown setting. The RLS algorithm was recommended for adaptive filtering. The study Comparison between Adaptive filter Algorithms (LMS , NLMS and RLS ) [16], also found the same results involving the RLS algorithm, and recommended this algorithm for speech enhancement adaptive

PAGE 18

7 filters. It is important to understand the attributes of these algorithms due to the fact that they play a crucial role in the adaptive filter design.

PAGE 19

8 CHAPTER III CELL PHONE LAYOUT General Cell Phone Microphone Layout Most cell phones contain an array of microphones in order to record the multiple signals involved in a phone call and active noise reduction. Figure 3 demonstrates a general layout for a cell phone's microphone array. The microphone on the bottom is used t o record the voice signal while the microphone at the top of the phone near the camera is used to record the noise signal source. In figures 4 10 we can see the general microphone array layout for iPhones, LG, and Samsung cell phones. This appears to be th e optimal layout for a cell phone microphone array. Figure 3.1 General Cell Phone Microphone Array

PAGE 20

9 Apple iPhone Apple first introduced the noise cancellation feature on the iPhone 4. Since then, various brands have implemented the noise cancellation or reduction feature in their devices. Apple licensed the company Audience A1010 Voice Processor [17]. However, as of 2010, Apple has employed the company Cirrus Logic to develop their noise reduction chips. This thesis examines the adaptive noise cancella tion filter for the iPhone 6, iPhone 7 and iPhone 8. The microphone layout in figures 4, 5 and 6 shows the basic layout used. One microphone is located at the bottom of the phone and the other two microphones are located on the top of the phone. Fig ure 3.2 iPhone 6 Microphone Array. By https://www.att.com/devicehowto/index.html#!/devicediagram?make=Apple&model=iPhone6

PAGE 21

10 Figure 3.3 iPhone 7 Microphone Array. by https://www.apple.com/iphone 8/specs/ Figure 3.4 iPhone 8 Microphone Array. by https://www.apple.com/iphone 8/specs/

PAGE 22

11 LG LG cell phones utilize Qualcomm Snapdragon processor chips for active noise cancellation filtering. For this study, an LG Stylo 3 and K20 phone were examined. Their mi crophone arrays are shown in figures 7 and 8. In LG cell phones, the noise cancellation filter that the LG Stylo 3 only has one microphone. This could indic ate that only one microphone records both the voice signal and the background noise signal. Figure 3.5 LG Stylo 3 Microphone Array by LG Stylo 3 manual Figure 3.6 LG K20 Microphone Array by LG K20 manual

PAGE 23

12 Samsung Samsung deploys Qualcomm Snapdragon chips into their devices, is responsible for the adaptive noise cancellation. Previous Samsung Galaxy phones used to allow the user to specify if they would like to use noise suppression or not. This was provided by a to ggle switch in the call setting menu. With the introduction of the Samsung Galaxy S6, this option is no longer available. All Samsung cellular products now have the noise suppression turned on by default. This information was provided by Samsung customer s upport. The noise suppression features were tested on Samsung Galaxy S9 and Samsung Galaxy S9. Figure 9 demonstrates the microphone layout for Samsung Galaxy S6, and figure 10 demonstrates the microphone layout for Samsung Galaxy S9. Both phones utilize a microphone on the bottom and the top. Figure 3.7 Samsung Galaxy S6 Microphone Array by Samsung Galaxy S6 manual

PAGE 24

13 Figure 3.8 Samsung Galaxy S9 Microphone Array by Samsung Galaxy S6 manual

PAGE 25

14 CHAPTER IV TESTING different environments. Two tests were conducted for the purpose of this thesis. Both Test 1 and Test 2 were replicated three different times with the adaptive no ise cancellation feature on and off. This was done for comparison reasons. Two cell phones were needed to conduct the test. Cell Phone A represents the caller's cell phone. Cell phone B represents the receivers signal cell phone. To acquire the recordings the application an named Another Call Recorder by the company NLL APPS was installed on the receiver signal cell phone. To mitigate, bias both tests were deployed in different locations and studios. It is important to note that cell phone noise cancellatio n feature needs to signal to have both sources active at the same time. The first signal was the caller's voice and the second signal was the noise. In addition, it is important to note that Samsung cell phones models from Galaxy S4 to Galaxy S5 allow the consumer to turn the active noise cancellation on and off. However, turning adaptive noise cancellation off for Samsung Galaxy S to S9 was not an option due to the fact that the noise suppression feature is built into the phone. Without the option to turn it off. This was confirmed by Samsung Technical Support. Both tests were modeled after the tests found in the studies Dual microphone noise reduction for mobile phone application [18], and Background Noise Reduction Design for Dual Microphone Cellular Ph ones: Robust Approach [ 19]. Both studies aimed to test the adaptive noise filter of a cell phone and its enhancement quality of a caller s speech. In Dual microphone noise reduction for mobile phone application , the approach to testing was to record and c ompare the adaptive noise filter of a cell phone in three different live scenarios. The first scenario was recorded in an office space, the second was recorded in a cafeteria, and the third scenario was

PAGE 26

15 recorded in a bus. In Background Noise Reduction Desi gn for Dual Microphone Cellular Phones: Robust Approach adaptive filters in a lab scenario. Noise generators were introduced in different locations while the source cell phone recorded the noi se and filtered it out. These two tests provided a fantastic approach to produce and examine a cell phone's adaptive noise filter. The live and lab testing influenced the approach in the following test; Lab testing and Live Scenario Testing. Lab Testing T loudspeakers, while a human spoke into cell phone A. Cell phone A transmitted the signals to Cell phone B, where the application Another Call Recorder recorded the signals. Each sample recording consisted of 20 seconds. Seven different cell phones were used in Test 1. Each cell phone was recorded 3 times wit h the noise cancellation filter and 3 times without. In addition, the source cell phone was recorded at two different volume levels. The first volume level for the Master Waveform was set at 70 dBa. 70 dBa fs would be the average volume level for a noisy e nvironment [20], such as traffic and that of everyday life. The second volume level for Master Waveform was set at 80 dBa. 80 dBa fs would be the above average volume level for a noisy environment like a vacuum cleaner or garbage disposal. Please note that a volume threshold is needed to trigger the noise cancelation filter. All volume dB limits were measured with an SPL meter at A weighted scaled. The Master Waveform was needed to simulate a controlled noisy signal. White noise was used as the noise due to the fact that it disperses noise evenly across the frequency spectrum [21.] In addition to this, impulses were added at the 1 second and 19 second mark to help create a reference for future synchronization.

PAGE 27

16 Figure 4.1 Master Signal ARC recorded all test recordings at 16bit, 44100 kHz, WAV PCM. Cell phone communication only transmits at the frequency range of 8bit 8 kHz. This reaction then caused all sample recordings to be downsampled to 8bit, Wav PCM, and a ll redundant data above 8 kHz to be removed. Comparison of the difference between the noise cancelation filter being turned on or off and by manufacture model can be seen in Chapter 5. To mitigate bias, Test 1 was performed at different labs in different p arts of the country. The main location for lab testing was performed in Denver, CO Other locations included Colorado Springs, CO, Davis, CA and Berkeley, CA. The replication of Test 1 in different environments hindered similar results. Live Scenario Tes ting The implementation of the live scenario testing was performed to replicate the methods of the lab test in a real life scenario. The caller's cell phone was operated by an individual providing the speech signal. The noise signal is the natural acousti cal environment of the marketplace. Implementation of a real life scenario was performed at a marketplace during rush hour, which can result in a volume level of 70 dBa. The same method of recording between the caller's cell phone and the receiving cell ph one from the lab test was replicated in the live scenario. Each cell phone recorded three 20 second test recordings with the noise cancellation filter on and 3 test recordings with the noise cancellation filter off. Since playing broadband noise is not pos sible in a public location, a twenty second sample was recorded as an exemplary

PAGE 28

17 reference noise sample. In addition to this, clicks by a finger snap were implemented into the background at the 1 second and 19 second mark of test sample recordings. This was done to mimic the impulses in the Master Waveform from the lab test for synchronization in later analysis. Results for the real life scenario can be seen in chapter 5. .

PAGE 29

18 CHAPTER V ANALYSIS AND RESULTS All sample recordings from the lab test were imported into Adobe Audition, where each sample recording was synced in the timeline by utilizing the impulses at the 1 second and 19 second mark. After syncing, all test recordin gs were trimmed to 20 second increments. All recordings from the Live Test were also imported into Adobe Audition. The Live Test recordings do not have impulses at the 1 and 19 seconds marks. They were synchronized by the snaps at the beginning of the li ve test recordings. In addition to trimming, all test recordings underwent a speech signal removal process. Figure 12 demonstrates the test recording in its full form. Figure 13 represents the test recording after the speech signal was removed. Removing the speech signal from the sample recording allowed further analysis of the effects of adaptive filter on the noise signal in the original sample recordings.

PAGE 30

19 Figure 5.1 Sample Recording Before Speech Removal Figure 5.2 Sample Recording After Speech Removal

PAGE 31

20 Spectrogram A spectrogram is a great way to capture and visualize the adaptive noise filter at work. The spectrograms in figures 5.3 and 5.4 are from two recordings made by an LG Stylo by utilizing test/method 1. The recording in figure 5.3 is with noise cancellation and the recording in figure 5.4 is without. In figure 14, we can see that the noise is evenly reduced as indicated by the breaks in the red color. This tells us that the noise is filtering out the noise signal and allowing the voice signal to pass through. We can also see these results for the test recordings in figures 5.5 and 5.6. Figure 5.3 LG Stylo 3 With Noise Suppression

PAGE 32

21 Figure 5.4 LG Stylo 3 No Noise Suppression Figure 5.5 LG K20 With Noise Suppression

PAGE 33

22 Figure 5.6 LG K20 No Noise Suppression SNR Results SNR or Signal to Noise Ratio was used to plot every test cell phone with and without the adaptive noise filter on and off. Similarly, in the paper Dual microphone noise reduction for mobile phone application [18], the authors used SNR to analyze the SNR relationship between compare an SNR relationship between a cell phones noise c anceling feature turned on and off. In addition, the SNR was compared between different cell phone brands. The SNR helps to apply a calculated representation of the ANC filter in use over time. Figures 16 to 45 plots signal to noise over time against the originally recorded sample. SNR averages between cell phones and volume threshold, in relation to the adaptive noise filter on vs off can be seen in figures 18 through 23. In addition to this SN R average values of the

PAGE 34

23 sample, recording can be seen in tables 5.1 through 5.6. The error bar in the following figures 5.1 through 5.6 represents the standard deviation. Figure 5.7 SNR Averages at 70 dB Figure 5.8 SNR Averages at 80 dB

PAGE 35

24 Figure 5.9 S NR Averages Live Figure 5.10 SNR Averages of Noise at 70 dB

PAGE 36

25 Figure 5.11 SNR Averages of Noise at 80 dB Figure 5.12 SNR Averages of Live Noise

PAGE 37

26 Tables Table 5.1 Average SNR Values at 70 dB Table 5.2 Average SNR Values at 80 dB Table 5.3 Average SNR Values Live

PAGE 38

27 Table 5.4 Average SNR Values Noise of 70 dB Table 5.5 Average SNR Values Noise of 80 dB Table 5.6 Average SNR Values Live Noise

PAGE 39

28 FFT Analysis The Fast Fourier transform (FFT) is a mathematical expression that can be used to convert an audio signal from the time domain to the frequency domain [22]. This allows us to visualize an audio signal through frequency (Hz) over gain (dB). Where frequency utilizes the x axis and gain(dB) utilizes the Y axis, as displayed in figure 5.14. All background noise samples discussed previously in this chapter w ere processed through FFT analysis. Each recording was classed into categories by phone, adaptive noise filter setting and volume threshold of the noise. For instance, iPhone test samples that were recordings at 70 dB were compared between the adaptive noi se filter being on and off. The blue frequency in figures 5.14 through 5.19 represents the FFT frequency analysis with the adaptive noise filter on, where all red frequencies in figures 5.14 through 5.19 represent the FFT frequency analysis with the adapti ve filter off.

PAGE 40

29 Figure 5.13 FFT iPhone 6 80 dB ANC On vs Off Figure 5.14 FFT iPhone 7 80 dB ANC On vs Off Figure 5.15 FFT iPhone 8 80 dB ANC On vs Off

PAGE 41

30 Figure 5.16 FFT iPhone 7 Live ANC On vs Off Figure 5.17 FFT LG K20 80 dB ANC On vs Off

PAGE 42

31 Figure 5.18 FFT LG K20 Live ANC On vs Off

PAGE 43

32 CHAPTER V I CONCLUSION With the lab test, it is apparent that after analyzing the SNR average values and viewing a spectrogram that one could detect if a cell phone has the adaptive noise filter or noise cancellation on or off. For instance, the LG Stylo 3 , at 80 dB, has an increase in its SNR when noise reduction is off. This would indicate that the noise is bleeding into the desired signal of the sample recording. One can see similar results with the plotted SNR in figures and tables listed in Chapter 5. The lab test hindered similar SNR average differences between filter being on and off across all test cell phones. The increase in SNR average when the adaptive noise filters off makes sense, due to the fact that more of the noise signal is not being suppressed. The FFT analysis of the noise signals also yielded some interesting results. All iPhone samples performed a consistent dip at 4,000 Hz and remained at a volume level of around 120 dB. We can see this consistency between all iPhones in figures 5.15, 5.16, 5.17 and 5.7. All other phones veered inconsistently around different volume levels, exc ept for around 12000 HZ and 20000 HZ can be seen in figures 5.18 and 5.19. Figure 5.18 was from a noise sample that was recorded in the lab testing. Where figure 5. 19 was from a noise sample that was recorded in the live testing. No other phones created these unique spikes. In addition, spectrogram analysis of the LG Stylo 3 in figures 5.3 and 5.4 clearly demonstrated a visual representation of the adaptive noise f ilter in action. We can also see this in the LG K20 in figures 5.6 and 5.7. The noise only samples from the test recording further demonstrated that the method was successful in indicating whether the adaptive noise filter was

PAGE 44

33 on or off. The SNR averages showed a clear increase in dB when the filter was switched off, as seen in tables 5.1 and 5.2. It should be noted that the LG Stylo 3 times. It seemed that the higher the simulated background noise the worse the filter performed. The SNR average values in figures 5.4 and 5.3 were drastically different than the other six phones. This may be due to the fact that the LG Stylo 3 does not possess a second microphone to attain the noise signal. Intense noise reductio n is needed in a single microphone system, which can result in lower speech quality [ 23]. The method discussed in the live scenario tests yielded mixed results. Live recordings were tested on an iPhone 7, LG Stylo 3 and LG K20. With the live testing of the LG K20, very little differences in SNR averages were recorded. Unfortunately, spectrogram analysis was unsuccessful in detecting differences in adaptive noise reduction filter being turned on or off. in the FFT analysis, however, the FFT analysis of the Live Scenarios iPhone 7 noise sample generated similar results to the lab HZ and a consistent volume level of around 120 dB. After finalizing all testing, it would appear that in order to fully detect whether adaptive noise canceling is on or off, the dB level would have to be at a high level. The higher the volume threshold, the more active the noise filter. Test recordings in the lab test with volume levels of 70 dB were conclud ed with very little results. However, when volume levels were increased to 80 dB, the effects of the adaptive noise filter could be seen and recorded.

PAGE 45

34 Since the proposed method of capturing the adaptive noise cancellation filter in action is obtainable, further research into categorizing cell phone SNR values by manufacturer and model is possible for forensic use.

PAGE 46

35 C HA PTER VI I FURTHER RESEARCH Since adaptive noise cancellation filters are prevalent in most cell phone brand and models. It would be neglectful of the forensic scientific community to ignore such a key attribute. The proposed methods from chapter 4 could be utilized to test other ce ll phone manufacturers. For instance, cell phones from manufacturers like Sony, Google and HTC, Asus and Huawei were not tested for this thesis. All of these brands offer a noise reduction feature and need to be analyzed in similar manners listed in chapte r Five. It seems that SNR values of the noise reduction filters exhibits unique qualities in relation to cell phone models and brands. FFT analysis of the noise signal also provided some unique characteristics. Further forensic research of this subject is needed. A creation of a database that recorded SNR and FFT results could lead to advancements in cell phone forensic audio analysis. A subject in forensic science that is vastly overlooked.

PAGE 47

36 REFERENCES [1] Bagnell Pocketnow , 4 Feb. 2016, pocketnow.com/smartphone noise reduction. Android Auth ority , Android Authority, 22 Dec. 2017, www.androidauthority.com/noise cancellation explained 820149/ . Macworld , Macworld, 9 July 2012, www.macworld.com/article/1167599/apple_sued_over_noise_reduction_technology.html [4] Hanilçi, Cemal, and Tomi K Phone Recognition from Recorded Speech Using Non Digital Signal Processing , vol. 35, 2014, pp. 75 85., doi:10.1016/j.dsp.2014.08.008. Accessed 20 Nov. 2018. [5] Oinonen, Mika. Development of Active Noise Control and Voice Communication Systems for Personal Hearing Protectors . TAMPEREEN TEKNILLINEN YLIOPISTO, TAMPERE UNIVERSITY OF TECHNOLOGY, 22 Sept. 2006, tutcris.tut.fi/portal/files/1517363/oinonen.pdf. [6] Watts, Lloyd, et a l. Voice Processor Based on the Human Hearing System . Audience, 25 Aug. 2008, www.hotchips.org/wp content/uploads/hc_archives/hc20/2_Mon/HC20.25.320.pdf . [7] The Motley Fool , The Motley Fool, 29 June 2018, www.fool.com/ investing/2018/06/29/cirrus logic wins spot in next generation apple ai.aspx . . [8] Qualcomm , 19 July 2018, www.qualcomm.com/snapdragon/devices/all . [9] Jesdanun, Cancelling Tech Doesn't Impress Firstpost , Firstpost, 27 Feb. 2014, www.firstpost.com/tech/news analysis/first look sonys xperia z2 noise cancelling tech doesnt impress 3648399.html . [10] Gizmodo , Gizmodo.com, 17 June 2013, gizmodo.com/5963773/what s the point of android skins. [11] R. W. Hamming, Digital Filters. Englewood Cliffs, N.J: Prentice Hall, 1977 [12] 2012 IEEE International Conference on Consumer Electronics (ICCE) , 2012, pp. 77 78, doi:10.1109/ICCE.2012.6161748. Accessed 25 Oct. 2018. [13] Dept. of Electronics & Communication Netaji Subhas Institute of Technology , 2001.

PAGE 48

37 [14] V.p Algorithms for Noise Cancellation in Real International Journal of Computer Applications , vol. 96, no. 10, 2014, pp. 20 25., doi:10.5120/16829 6589. [15] B. Chandrashaker Reddy, D. Vamsi Krishna, Shrinath Raygond, N. Ravi Teja. Analysis of Adaptive Filter Approach for Speech Enhancement Using Simulink , International Journal of Advance Research, Ideas and Innovations in Technology, www.IJARnD.com . [16] IPhone Hacks | #1 IPhone, IPad, IOS Blog , 7 Sept. 2012, www.iphonehacks.com/2012/09/audience noise suppression technology unlikely in iphone 5.html . [18] Z. Fu, F. Fan and J. Huang, "Dual microphone noise reduction for mobile phone application," in 2013, . DOI: 10.1109/ICASSP.2013.6639068. [19] Y. Chen et al , "Background Noise Reduction Design for Dual Microphone Cellular Phones: Robust Approach," IEEE/ACM Transactions on Audio, Speech and Language Processing (TASLP), vol. 25, (4), pp. 852 862, 2017. [20] Industrial No ise Control , www.industrialnoisecontrol.com/comparative noise examples.htm [21] Hansen, Colin. FUNDAMENTALS OF ACOUSTICS . Department of Mechanical Engineering Universi ty of Adelaide, 2001, www.who.int/occupational_health/publications/noise1.pdf . [22] Heckbert, Paul. Fourier Transforms and the Fast Fourier Transform (FFT) Algorithm . Jan. 1995, www.cs.cmu.edu/afs/andrew/scs/cs/15 463/2001/pub/www/notes/fourier/fourier.pdf . Technical Note , Qualcomm & Inforce Computing, Apr. 2016, www.inforcecomputing.com/public_docs/Wh itePapers/Inforce_support_for_Fluence_Audio_04 2016 1.pdf .