Acoustic Beamforming using a TDS3230 DSK:

Final Report

Steven Bell Nathan West

Student Member, IEEE Student Member, IEEE

Electrical and Computer Engineering

Oklahoma Christian University
 i

C ONTENTS

I Introduction 1

II Theory 1
II-A Delay and Sum Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
II-B Microphone Array Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
II-B1 Microphone spacing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

III Simulation 2
III-A Source localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
III-B Spatial Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

IV System Requirements 3

V Design 4
V-A Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
V-B DSP Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
V-C Interface Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
V-C1 DSK-PC Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
V-C2 Interface GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

VI Test Methods 6
VI-A Source localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
VI-B Spatial filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

VII Results 6

VIII Discussion 6
VIII-A Sources of difficulty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

IX Conclusion 7

References 7

Appendix A: MATLAB code for beam-sweep source localization 8

Appendix B: MATLAB code for spatial filtering 11

Appendix C: C code for main processing loop 14

Appendix D: C code for summing delays 16

Appendix E: C code for calculating delays 17

Appendix F: Java Code - Main 17

Appendix G: Java Code for Beamformer Communication 18

Appendix H: Java Code for Beam Display 21

 1

Abstract—Acoustic beamforming is the use of a micro-

phone array to determine the location of an audio source
or to filter audio based on its direction of arrival. For this
project, we simulated and implemented a real-time acous-
tic beamformer using MATLAB for simulations and the
TDS3230 DSK for the real-time implementaion. Although
the final system does not meet all of our initial goals, it
does successfully demonstrate beamforming concepts.

I. I NTRODUCTION
Most broadly, beamforming is the use of an array
of antennas - or in the case of audio, microphones
- to perform signal processing based on the spatial
characteristics of a signal. We will discuss two primary
forms of beamforming in this document: source localiza-
tion and spatial filtering. Source localization attempts to
determine the location in space a signal originated from, Figure 1. Planar wavefront approaching a linear microphone array.
while spatial filtering creates an electronically-steerable
narrow-beam antenna, which has gain in one direction
and strong attenuation in others. Spatial filtering systems added. This essentially creates a spatial filter, which we
and corresponding transmission techniques can replace can point in any direction by changing the delays.
physically moving antennas, such as those used for radar. To determine the direction a signal came from, we can
Acoustic source localization is familiar to all of us: sweep our beam around the room, and record the total
we have two ears, and by using them together, our brain power of the signal recieved for each beam. The source
can tell where sounds come from. Similarly, our brains came from the direction with the highest-power signal.
are able to focus on one particular sound and tune out
the rest, even when the surrounding noise is much louder B. Microphone Array Design
than the sound we are trying to hear.
Microphone arrays can be nearly any shape: linear, cir-
cular, rectangular, or even spherical. A one-dimensional
II. T HEORY
array allows beamforming in one dimension; additional
A. Delay and Sum Method array dimensions allow for 2-dimensional beamforming.
If we have an array of microphones and sufficient Given the limited number of microphones and amount
signal-processing capability, we can measure the time of time we have, a linear array is the best choice.
delay from the time the sound strikes the first micro- 1) Microphone spacing: The spacing of the micro-
phone until it strikes the second. If we assume waves phones is driven by the intended operating frequency
originate far enough away that we can treat the edge of range.
its propagation as a plane, then the delays are simple to For spatial filtering, a narrower beam width is an
model with trigonometry as shown in Figure 1. advantage, because signals which are not directly from
Suppose we have a linear array of microphones, m1 the intended direction are attenuated. A narrow beam
through mn , each spaced Dmic meters apart. Then width is analogous to a narrow transistion band for
Ddelay , the extra distance the sound has to travel for a tranditional filter. Lower frequencies will correlate
each successive microphone, is given by better with delayed versions of themselves than high
frequencies, so the lower the frequency, the broader the
Ddelay = Dmic · cos (θ) (1) beam. Conversely, a longer array will result in a greater
At sea level, the speed of sound is 340.29 ms , which delay between the end microphones, and will thus reduce
means that the time delay is the beam width.
At the same time, the spacing between microphones
Dmic
Tdelay = · cos (θ) (2) determines the highest operating frequency. If the wave-
340.29 ms length of the incoming signal is less than the spacing
By reversing this delay for each microphone and sum- between the microphones, then spatial aliasing occurs.
ming the inputs, we can recover the original signal. If a An example is shown in Figure 2.
signal comes from a different direction, the delays will The spacing between microphones causes a maximum
be different, and as a result, the individual signals will time delay which, together with the sampling frequency,
not line up and will tend to cancel each other when limits the number of unique beams that can be made.
 2

90 90
6
120 60 120 60

150 30 150 30
2

180 0 180 0

210 330 210 330

240 300 240 300

270 270

Figure 2. Spatial aliasing. The source is 1 kHz, located at 135 degrees. Figure 3. Source localization simulation using a 400 Hz source and
An image appears at approximately 50 degrees, which means that a 4-microphone array. The source was located at 0, 45, 90, 135, and
sound coming from that direction is indistinguishable from the desired 180 degrees.
source.

90
The maximum number of beams, maxbeams , is shown 120 60

mathematically in Equation 3.
150 30
maxbeams = 2 · Fs · timespacing (3)
The variable timespacing is the maximum amount of
time it takes for sound to travel from one microphone to 180 0

an adjacent microphone, as the case when the source is Two

Four
along the line created by the array. Six
210 330

III. S IMULATION 240 300

Source localization and spatial filtering both use es- 270

sentially the same simulation code. First, we define a

coordinate system where our microphones are placed,
with the origin at the center of our linear array. We Figure 4. Comparison of beam width versus the number of micro-
defined the distance between microphones to be 25 phones. The source was 400 Hz at 90 degrees.
cm, making total array lengths of 75 cm for a four
microphone array and 25 cm for a two microphone array.
90
120 60

A. Source localization
150 30
In the source localization code, a single source is
located 10 meters away from the center of the array.
A loop creates thirty beams which cover a 180-degree 180 0
range, and the power for each beam is computed. Figure
3 shows the power sweeps for five different source
angles. Note that in each case, the maximum power is 210 330
exactly at the angle the sound originates from.
Using the beam-sweeping technique, we examined the 240 300
relative effects of the number of microphones and the 270

source frequency. Figure 4 shows a comparison of the

beamwidth as a function of the number of microphones.
Figure 5 shows a comparison of beam width versus fre- Figure 5. Comparison of beam width versus frequency using 2
quency. Note that higher frequencies produce narrower microphones. The frequencies range from 200 Hz to 700 Hz in
beams, so long as spatial aliasing does not occur. increments of 100 Hz. Higher frequencies produce narrower beams.
 3

90 FFT of the received signal

100 0
120 60
80

60 −10
150 30

Relative magnitude of signal (dB)

40

20 −20

180 0
−30

−40
210 330

−50
240 300
270
−60
0 100 200 300 400 500 600 700
Frequency (Hz)

Figure 6. Decibel-scale plot of a 4-microphone array beampattern for Figure 7. The noise source at 300Hz is attenuated by about 3dB using
a 600 Hz signal at 90 degrees. Note the relatively large sidelobes on an array of two microphones with sources separated by 60 degrees.
both sides of the main beam.

Figure 6 shows a plot of the beam in decibels, which 0

FFT of the received signal

brings out a pair of relatively large sidelobes. These

−5
can be reduced by windowing the microphone inputs, in
−10
the same way that a filter can be windowed to produce
Relative magnitude of signal (dB)

−15
lower sidelobes in exchange for a wider transition band.
However, this windowing is unlikely to be useful for our −20

small array. −25

The MATLAB source code is in Appendix A. −30

−35
B. Spatial Filtering −40

For spatial filtering, one source is placed at 10 m −45

normal to the array and 10 m tangentially from the center −50
0 100 200 300 400 500 600 700
of the array. Using this location as a reference, we can Frequency (Hz)
place a secondary source equidistant from the center of
the microphone array but some degree offset. Figure 8. An array of two microphones with sources separated by 90
For this simulation set we used two offsets, 60 degrees degrees will attenuate the noise source by about 13dB.
and 90 degrees. The formed beam was looking 45
degrees away from the array (towards the source at 10,
10). Two microphones were used in simulations because 0
FFT of the received signal

of suggestions by Dr. Waldo and four were used because

that is the maximum number of input channels on our −10
hardware. In Figure 7 the simulation has a variable
Relative magnitude of signal (dB)

number of microphones with the sources separated by −20

60 and 90 degrees. This simulation was the influence for
the specification calling for 3dB down on a 60 degree −30

source separation. The SNR is improved drastically when

the same sources are separated by 90 degrees still using −40

a 2 microphone array, as seen in Figure 8.

Figures 9 and 10 show similar but improved results −50

when the array is increased to four microphones.

−60
The MATLAB source code is in Appendix B. 0 100 200 300 400 500 600 700
Frequency (Hz)

IV. S YSTEM R EQUIREMENTS

Figure 9. An array of four microphones with sources separated by
• The system must have at least two microphones as 60 degrees results in the noise source attenuated by about 12dB.
its input.
 4

FFT of the received signal

0
Figure 11. Functional block diagram.

−5

−10
Figure 12. Flowchart for the main DSP software.
Relative magnitude of signal (dB)

−15

−20
cm apart. Theses microphones will be connected to the
−25
microphone inputs on the DSK_AUDIO4 daughtercard.
−30 The DSK will be programmed to process a 256 sample
−35 block at a time. The software delays each microphone’s
−40 input by a different number of samples then adds each
−45 input to create the output. In source localization mode,
−50
the power of the output is calculated and sent to the PC
0 100 200 300 400 500 600 700
Frequency (Hz) GUI. In spatial filtering mode, the output is passed to
one of the board’s output channels.
Figure 10. With an array of four microphones with sources separated
by 90 degrees the noise signal is about 12dB down from the signal we
attempted to isolate. A. Hardware
In addition to the standard DSK board provided to
the DSP class we will be using the Educational DSP
• The system must have a GUI running on a computer
DSK_AUDIO4 daughtercard, which has four input chan-
which communicates with the DSK board.
nels. Each of these channels will be connected to one of
• In localization mode:
the generic mono microphones provided to the class by
– The system will be able to locate the direction Dr. Waldo. The placement of the microphones introduces
of a 400 Hz tone between -90 and +90 degrees, a physical variable to the system that is important to the
with +/- 5 degrees error. This target tone will operation of hardware: according to our specifications
be significantly above the noise threshold in the and simulations the microphones should be spaced 25
room. cm apart. Ideally, they will face the direction of incoming
– The GUI must display the direction of the acoustic sources for maximum input power.
incoming sound with a latency less than 1
second.
• In directional enhancement mode: B. DSP Software
– The GUI must allow the user to select a partic- The foundation of our DSP software is the sample
ular direction to listen from, between -90 and code provided by Educational DSP and modified by Dr.
+90 degrees, and with a resolution of at least Waldo. This code initializes the DSP and the codec and
10 degrees. calls a processing function that provides samples from
– The system should play the selected directional all four input channels. Using this as a base we will
output via one of the line out channels. implement our main function inside of the processing
– A 300 Hz noise source with the same power function provided. In this main file we will also include
will be placed 60 degrees apart from the a file that defines our sum and delay functions which
desired signal source. The directional output will exist in their own files.
should give a signal-to-noise ratio of at least The main software branch opens a RTDX channel with
3 dB. the PC GUI and keeps track of the current operating
• The system will operate properly when the sound mode. The mode is a boolean value where a TRUE
sources are at least 4 meters away from the micro- means spatial filtering mode and a FALSE means source
phone array. localization mode. In the spatial aliasing mode we do
a simple delay and sum of the input data. The delay is
given by the GUI. In localization mode the current output
V. D ESIGN is also the delay and sum of the input; however, the delay
The acoustic beam forming system consists of three sweeps from a minimum possible delay to a maximum
high level subsystems: the microphone array, the DSK possible delay defined by the number of samples it takes
board, and a GUI running on a PC. A block diagram to go from −90o to 90o . After every delay and sum
showing the interconnections of these subsystems is operation the delay/power pairs will be sent to the GUI
shown in Figure 11. The microphone array should con- for display via the RTDX channel. The flow of this is
tain two to four microphones placed in a straight line 25 shown in Figure 12.
 5

Table I
Figure 13. Flowchart for the function that calculates the number of B YTE SIGNALS FOR PC→DSK COMMUNICATION
samples to delay each microphone’s input by.
Use Code Range
Figure 14. Flowchart for the DSK function that calculates the current Microphone time spacing 0-100
output sample. Number of microphones 2-4
Mode 1,2
Beam selection -100 to +100
To determine how much to delay each microphone by
we get the number of samples each microphone should
be delayed by from the GUI and multiply by the order
of that microphone in the array. The first microphone in
the array is defined as microphone 0 and has a 0 delay. Figure 16. Byte transmission order for PC → DSK communication
If the delay we get from the GUI is negative the order is
reversed so that the earliest input always has a 0 delay.
The calcDelay function will return a pointed to an array We found that we did not need any synchronization
called delays. Figure 13 shows the structure and flow for bytes, and we trimmed the communictation down to a
calcDelay. single byte for the power paired with a single byte for
The sum function accepts the current four input sam- delay. This was intended to speed up RTDX communi-
ples and delay from the GUI and passes it to calcDelays. cation, but was not successful.
Figure 14 shows a program flow for this function. It b) Settings to DSK Board: The GUI will control
will have a reverse index tracked buffer so that if several settings, summarized in Table I Physical variables
i=5 is the current input i=6 is the oldest input and such as microphone spacing and number of microphones
i=4 is the second most recent input. This requires need to be easily changed by software in real time, and
fewer rollover-error-checking comparisons relative to a the GUI provides an interface for users to the DSK to
forward-tracking buffer, which should give us an incre- change the operation as required.
mental speed improvement. The byte stream will follow the same format as the
DSK to PC stream. Data will be sent in 3 byte packets
C. Interface Software as shown in Figure 16. Packets will only be sent when
there is new data, so we expect the number of packets
1) DSK-PC Interface: The PC will communicate with per second to be very low.
the DSK board using RTDX 2.0. For speed, the DSK We did not implement this due to time constraints.
will only perform integer operations; thus, the PC will
2) Interface GUI: The interface software will be
perform the floating-point work of translating delays
written in Java. An overview of the main classes is shown
to angles and vice-versa. There will be two RTDX IO
in Figure 17.
channels, one going each direction.
The Beamformer class handles the low-level transla-
a) Power-vs-Delay to PC:
tion and communication with the DSK board. It converts
• Delay will be a 2’s complement 8 bit signed num-
from sample delays to angles and vice-versa, converts
ber, which represents the relative delay between booleans to code values, and floating-point values to
microphones. This will be calculated from the mi- integers for high-speed communication.
crophone spacing and the speed of sound. The BeamDisplay is a GUI class which displays the
• Power will be a 16-bit unsigned number, which
beam power levels and sound direction on a polar plot,
holds the relative power of the signal for a particular similar to the simulation plots. It will use the Java
beam. AWT/Swing toolkit and associated drawing classes to
In order for these two values to remain correlated, we implement a custom drawable canvas. Clicking on the
will interleave both on one channel, and signal the start display will select the direction to use when in spatial
of each new delay-power pair using the value 0xFF. This filtering mode.
will allow our system to recover from lost or delayed The Settings panel will contain a series of standard
bytes. Swing components to select the operating mode, number
This data will only be sent when the system is in of microphones, microphone spacing, the number of
source localization mode. beams per sweep. It will have a pointer to the Beam-
former object, so it can configure it directly.
The MainWindow class creates the main GUI window,
the display and control panels, and the connection to
the DSK board. Once running, it will regularly poll
Figure 15. Byte transmission order for DSK→ PC communication the beamformer class for new data, and update the
 6

MainWindow
We were unable to get any kind of beamforming
+guiComponents
-connectToDSK()
to work with four microphones. After getting the 2-
-initializeGui() microphone system working, we switched to 4 by chang-
-disconnect()
ing a #define in our code and connecting two addi-
ControlPanel
tional microphones to the DSK expansion board. When
BeamDisplay +modeSelect
we ran the GUI, we were unable to observe any kind of
+canvas: JPanel +numBeams localization.
+numMicrophones
-initialize()
+microphoneSpacing Each adjacent pair of microphones worked in a 2-
+update(newData)
-initialize() microphone beamformer, so we believe that it was not
a simple case of one microphone not working properly
Beamformer or a result of mixing up the order of the microphones.
-ControlChannel: RTDX Channel We performed a test where we merely averaged the
-DataChannel: RTDX Channel
+setMode(mode:int)
inputs from the four microphones and sent that value
+setNumBeams(newNum:int) to the output, which is equivalent to forming a beam
+setNumMicrophones(newNum:int)
+setMicrophoneSpacing(newSpacing:double) perpendicular to the array. Using one of the speakers
+setFilterAngle(newAngle:int)
+update()
as an audio source and measuring the RMS voltage of
+hasNewData() the output on the oscilloscope, we manually checked the
+getBeamPower()
directionality of the array. We were unable to measure
any significant difference between the sound source be-
Figure 17. Class Diagram for the GUI
ing directly orthogonal to the array and at a small angle
to it. This indicates that at least part of the problem is
not due to our software. There may have been additional
BeamDisplay accordingly.
problems with our software for four microphones, but
because we were unable to get this simple test working,
VI. T EST M ETHODS we did not test our software further.
A. Source localization
To test performance of the source localization method VIII. D ISCUSSION
we set up two microphones 25cm apart. Using a meter A. Sources of difficulty
stick to keep a contsant radius, an amplified speaker was
We identified several factors which contributed to our
swept radially across the plane level with the micro-
system’s failure to meet its requirements.
phones. A digital oscilloscope was used to measure the
RMS voltage of the processed output. We also observed First, there were several acoustic problems. We found
the beam displayed on the GUI. that the supplied microphones are very directional -
they are specifically labeled “unidirectional”. Because
the beamforming system assumes omnidirectional input,
B. Spatial filtering unidirectional microphones will cause oblique signals to
Because of time constraints and unexpected difficul- be attenuated. This is fine when the desired beam is per-
ties with other parts of the project, we did not explicitly pendicular to the array, but is suddenly a problem when
test spatial filtering. the goal is to amplify oblique signals over orthogonal
To test the spatial filtering mode, we would have set ones.
up two sources at for different frequencies equidistant The audio input level from the microphones seemed
from the array with a 90 degree arc length seperation. to be rather low, producing averaged output values in
The DSK would process the data with a beam formed the single-millivolt range. We are not sure if there is a
towards one of the sources and feed the output was to way to configure additional gain on the DSK board, or
a digital oscilloscope or MATLAB/Simulink GUI set to if using different microphones would help. Due to the
view the FFT. Using this method, we could compare the low input level, we had to play the signal very loudly on
relative strengths of the signals. the speakers and keep them within one or two meters of
the array.
A 1-meter source distance clearly violates the farfield
VII. R ESULTS plane-wave assumption - our initial specification as-
Our system is able to do beamforming with 2 mi- sumed that 4 meters was farfield. This may have caused
crophones, and display the result on a PC GUI. By some problems, but re-running our MATLAB simulation
observation, the beamwidth is about what we expected using a source only 1m away produced a beam virtually
for a 2-microphone array. indistinguishable from a 10-meter beam.
 7

We also experienced very strong echoes in the room, [8] R. J. Lustberg, “Acoustic Beamforming Using Microphone Ar-
because of its cement construction and solid wood rays,” Master’s thesis, Massachusetts Institute of Technology,
Cambridge, MA, June 1993.
lab benches. Working in an environment with reduced [9] N. Mitianoudis and M. E. Davies, “Using Beamforming in the
echoes would probably have yielded better results. This Audio Source Separation Problem,” in 7th Int Symp on Signal
was evident during late nights in the lab when people Processing and its Applications, no. 2, 2003.
working in the ASAP room and janitorial staff would
peek into the room to find the very loud noise echoing
throughout the HSH.
On the software side, we experienced difficulties get-
ting the bandwidth we desired over RTDX. I had also
seen poor RTDX performance before, when sending data
from a PC to the DSK during the cosine generation lab.
There are several possible causes of this: One is that
we only send a few bytes at a time, but do it dozens
of times per second. Recommendations we found online
suggested that filling the RTDX queue and sending large
packets of data is the fastest way to send data.
Another likely cause is our DSK boards. We read a
mailing list correspondence between someone who is
familiar with this family of DSKs and a hapless hobbyist
having similar issues. The “expert” claimed that there
is obfuscation hardware on the DSK boards introduced
by Spectrum Digital that slows down RTDX communi-
cation. It was mentioned that a possible workaround is
using the JTAG connector on the DSK board, but the
hardware required for this is not readily available (and
the JTAG port is blocked by the DSP_AUDIO_4). We
looked at high-speed RTDX, but it is not supported on
this platform.

IX. C ONCLUSION
Although we did not meet our initial specifications,
our project was still somewhat successful. We were able
to implement a working beamforming system, and most
of the problems we encountered were limitations of our
equiment which were not under our control.

R EFERENCES
[1] M. Brandstein and D. Ward, eds., Microphone Arrays: Signal Pro-
cessing Techniques and Applications. Berlin, Germany: Springer,
2001.
[2] J. Benesty, J. Chen, and Y. Huang, Microphone Array Signal
Processing. Berlin, Germany: Springer, 2008.
[3] J. Chen, L. Yip, J. Elson, H. Wang, D. Maniezzo, R. Hudson,
K. Yao, and D. Estrin, “Coherent Acoustic Array Processing and
Localization on Wireless Sensor Networks,” Proceedings of the
IEEE, vol. 91, no. 8, pp. 1154–1162, 2003.
[4] J. C. Chen, K. Yao, and R. E. Hudson, “Acoustic Source Localiza-
tion and Beamforming: Theory and Practice,” EURASIP Journal
on Applied Signal Processing, pp. 359–370, 2003.
[5] T. Haynes, “A primer on digital beamforming,” Spectrum Signal
Processing, 1998.
[6] W. Herbordt and W. Kellermann, “Adaptive Beamforming for Au-
dio Signal Acquisition,” Adaptive Signal Processing: Applications
to Real-World Problems, pp. 155–196, 2003.
[7] R. A. Kennedy, T. D. Abhayapala, and D. B. Ward, “Broadband
nearfield beamforming using a radial beampattern transformation,”
IEEE Transactions on Signal Processing, vol. 46, no. 8, pp. 2147–
2156, 1998.
 8

A PPENDIX A
MATLAB CODE FOR BEAM - SWEEP SOURCE LOCALIZATION

1 % General setup
2 clear;
3
4 % Speed of sound at sea level
5 speedOfSound = 340.29; % m/s
6
7 % Source setup
8 % Frequency of the audio source
9 sourceFreq = 700; % Hz
10
11 % Just so we know - since this will affect our mic placement
12 wavelength = speedOfSound / sourceFreq;
13
14 % xy location of audio source
15 sourceLocX = 0; % in meters
16 sourceLocY = 10;
17
18 % Microphone setup
19 numMicrophones = 4;
20
21 % Distance between microphones - determines the highest frequency we can
22 % sample without spatial aliasing
23 micSpacing = 0.25; % meters
24
25 % Minimum amount of time it takes for sound to travel from one mic to the next
26 timeSpacing = micSpacing/speedOfSound;
27
28 % Total width of the array - determines the lowest frequency we can
29 % accurately locate
30 arrayWidth = (numMicrophones - 1) * micSpacing;
31
32 % xy locations of the microphones
33 micLocX = linspace(-arrayWidth/2, arrayWidth/2, numMicrophones); % in meters
34 micLocY = zeros(1, numMicrophones);
35
36
37 % Sampling rate
38 Fs = 44100; % Hz
39
40
41 % Distance from the source to the mic
42 propDistance = hypot(sourceLocX - micLocX, sourceLocY - micLocY);
43
44 timeDelay = propDistance/speedOfSound;
45
46
47 % Create some of the signal
48 soundTime = 0:1/Fs:.125;
49
50 sourceSignal = sin(2 * pi * sourceFreq * soundTime);
51
 9

52 % Delay it by the propagation delay for each mic

53 for ii = 1:numMicrophones
54 received(ii,:) = delay(sourceSignal, timeDelay(ii), Fs);
55 end
56
57 % Plot the signals received by each microphone
58 % plot(received’);
59
60 % Now it’s time for some beamforming!
61 % Create an array of delays
62 numBeams = 60;
63 beamDelays = linspace(-timeSpacing, timeSpacing, numBeams);
64
65 for ii = 1:numBeams
66 summedInput = zeros(1, length(sourceSignal));
67 for jj = 1:numMicrophones
68 if(beamDelays(ii) >= 0)
69 % Delay the microphones by increased amounts as we go left to right
70 summedInput = summedInput + ...
71 delay(received(jj,:), beamDelays(ii) * (jj - 1), Fs);
72 else
73 % If the beam delay is negative, that means we want to increase the
74 % delay as we go right to left.
75 summedInput = summedInput + ...
76 delay(received(jj,:), -beamDelays(ii) * (numMicrophones - jj), Fs);
77 end
78 end
79 % Calculate the power of this summed beam
80 power(ii) = sum(summedInput .^ 2);
81 end
82
83 % directions = linspace(pi, 0, numBeams);
84 directions = acos(linspace(-1, 1, numBeams));
85
86 polar(directions, power/max(power), ’-b’);
 10

1 function [outputSignal] = delay(inputSignal, time, Fs)

2 % DELAY - Simulates a delay in a discrete time signal
3 %
4 % USAGE:
5 % outputSignal = delay(inputSignal, time, Fs)
6 %
7 % INPUT:
8 % inputSignal - DT signal to operate on
9 % time - Time delay to use
10 % Fs - Sample rate
11
12 if(time < 0)
13 error(’Time delay must be positive’);
14 end
15
16 outputSignal = [zeros(1, floor(time * Fs)), inputSignal];
17
18 % Crop the output signal to the length of the input signal
19 outputSignal = outputSignal(1:length(inputSignal));
 11

A PPENDIX B
MATLAB CODE FOR SPATIAL FILTERING

1
2
3 % General setup
4 clear;
5
6 % The frequencies we are interested in are:
7 % Right now this is optimized for sources between 100 and 500 Hz
8 % Change this to whatever range you are interested in.
9 fmin = 00; % Hz
10 fmax = 700;% Hz
11
12 % Speed of sound at sea level
13 speedOfSound = 340.29; % m/s
14
15 % Source setup
16 % Frequency of the audio sources
17 source1Freq = 300; % Hz
18 source2Freq = 400; % Hz
19
20 % Just so we know - since this will affect our mic placement
21 wavelength1 = speedOfSound / source1Freq;
22 wavelength2 = speedOfSound / source2Freq;
23
24 degreesApart=60;
25 % xy location of audio source
26 source1LocX = sqrt(10^2 + 10^2)*cosd(45+degreesApart); % in meters
27 source1LocY = sqrt((10^2 + 10^2)-source1LocX^2);
28 source2LocX = 10; % Secondary source
29 source2LocY = 10;
30
31 % Microphone setup
32 numMicrophones = 2;
33
34 % Distance between microphones - determines the highest frequency we can
35 % sample without spatial aliasing
36 micSpacing = 0.25; % meters
37
38 % Minimum amount of time it takes for sound to travel from one mic to the next
39 timeSpacing = micSpacing/speedOfSound;
40
41 % Total width of the array - determines the lowest frequency we can
42 % accurately locate
43 arrayWidth = (numMicrophones - 1) * micSpacing;
44
45 % xy locations of the microphones
46 micLocX = linspace(-arrayWidth/2, arrayWidth/2, numMicrophones); % in meters
47 micLocY = zeros(1, numMicrophones);
48
49
50 % Sampling rate
51 Fs = 96000; % Hz
 12

52
53
54 % Distance from the source to the mic
55 prop1Distance = hypot(source1LocX - micLocX, source1LocY - micLocY);
56 prop2Distance = hypot(source2LocX - micLocX, source2LocY - micLocY);
57
58 time1Delay = prop1Distance/speedOfSound;
59 time2Delay = prop2Distance/speedOfSound;
60
61 % Create some of the signal
62 soundTime = 0:1/Fs:.125;
63
64 source1Signal = sin(2 * pi * source1Freq * soundTime);
65 source2Signal = sin(2 * pi * source2Freq * soundTime);
66
67 % Delay it by the propagation delay for each mic
68 for ii = 1:numMicrophones
69 received(ii,:) = delay(source1Signal, time1Delay(ii), Fs)...
70 + delay(source2Signal, time2Delay(ii), Fs);
71 end
72
73
74 % Direct the beam towards a location of interest
75 angleWanted = 45; % Degrees (for simplicity)
76 angleToDelay = angleWanted * pi/180; % Convert to radian
77
78 % We want to take the fft of the signal only after everymicrophone is
79 % getting all of the data from all sources
80 deadAirTime = (max([time1Delay, time2Delay]));
81 deadAirSamples = deadAirTime * Fs;
82 endOfCapture = length(received(1,:));
83
84 % Start off with an empty matrix
85 formedBeam = zeros(1, max(length(received)));
86
87 % For each microphone add together the sound received
88 for jj = 1:numMicrophones
89 formedBeam = formedBeam + ...
90 delay(received(jj,:), + timeSpacing*sin(angleToDelay) * (jj-1), Fs);
91 end
92
93 % Get the PSD object using a modified covariance
94 beamPSD = psd(spectrum.mcov, formedBeam,’Fs’,44100);
95 % Get the magnitude of the PSD
96 formedSpectrum = abs(fft(formedBeam));
97 % The fft sample # needs to be scaled to get frequency
98 fftScaleFactor = Fs/numel(formedSpectrum);
99
100 % The frequencies we are interested in are:
101 % Right now this is optimized for sources between 100 and 500 Hz
102 %fmin = 0; % Hz
103 %fmax = 600;% Hz
104
105 % Plot the PSD of the received signal
 13

106 figure(1);
107 % Get the frequencies and data out of the PSD object
108 beamPSDfreqs=beamPSD.Frequencies;
109 beamPSDdata = beamPSD.Data;
110 % Get the indeces that correspond to frequencies specified above
111 indexesToPlot=find(beamPSDfreqs>fmin-1):find(beamPSDfreqs>fmax,1);
112 % Actually plot it (in a log10 scale so we have dB)
113 plot(beamPSDfreqs(indexesToPlot),20*log10(beamPSDdata(indexesToPlot)));
114 title(’PSD using a modified covariance of the received signal’);
115 ylabel(’Power/Frequency (dB)’);
116 xlabel(’Frequency (Hz)’);
117
118 % Plot the fft of our received signals
119 figure(2);
120 maxLimit=round(fmax/fftScaleFactor);
121 minLimit=round(fmin/fftScaleFactor);
122 if minLimit <= 0
123 minLimit = 1;
124 end
125 f=linspace(0,44100,44100/fftScaleFactor);
126
127 % This gets all of the fft frequencies we want to look at
128 fOfInterest = f(minLimit:maxLimit);
129 % Grab the portion of the fft we want to look at
130 spectrumOfInterest = formedSpectrum(minLimit:maxLimit);
131 % Normalize this so that the max amplitude is at 0db.
132 spectrumOfInterest = spectrumOfInterest/max(formedSpectrum);
133 % Plot it
134 plot(fOfInterest,20*log10(spectrumOfInterest));
135 title(’FFT of the received signal’);
136 ylabel(’Relative magnitude of signal (dB)’);
137 xlabel(’Frequency (Hz)’);
 14

A PPENDIX C
C CODE FOR MAIN PROCESSING LOOP

1
2 void processing(){
3 Int16 twoMicSamples[2];
4
5 Int32 totalPower; // Total power for the data block, sent to the PC
6 Int16 tempPower; // Value used to hold the value to be squared and added
7 Int16 printCount; // Used to keep track of the number of loops since we last sent data
8 unsigned char powerToSend;
9
10 Int8 firstLocalizationLoop; // Whether this is our first loop in localization mode
11 Int8 delay; // Delay amount for localization mode
12
13 QUE_McBSP_Msg tx_msg,rx_msg;
14 int i = 0; // Used to iterate through samples
15
16 RTDX_enableOutput( &dataChan );
17
18 while(1){
19
20 SEM_pend(&SEM_McBSP_RX,SYS_FOREVER);
21
22 // Get the input data array and output array
23 tx_msg = QUE_get(&QUE_McBSP_Free);
24 rx_msg = QUE_get(&QUE_McBSP_RX);
25
26 // Spatial filter mode
27
28 // Localization mode
29 if(firstLocalizationLoop){
30 delay = DELAYMIN;
31 firstLocalizationLoop = FALSE;
32 }
33 else{
34 delay++;
35 if(delay > DELAYMAX){
36 delay = DELAYMIN;
37 }
38 }
39
40 // Process the data here
41 /* MIC1R data[0]
42 * MIC1L data[1]
43 * MIC0R data[2]
44 * MIC0L data[3]
45 *
46 * OUT1 Right - data[0]
47 * OUT1 Left - data[1]
48 * OUT0 Right - data[2]
49 * OUT0 Left - data[3]
50 */
51
 15

52 totalPower = 0;
53
54 for(i=0; i<QUE_McBSP_LEN; i++){
55
56 // Put the array elements in order
57 twoMicSamples[0] = *(rx_msg->data + i*4 + 2);
58 twoMicSamples[1] = *(rx_msg->data + i*4 + 3);
59
60 tx_msg->data[i*4] = 0;
61 tx_msg->data[i*4 + 1] = 0;
62 tx_msg->data[i*4 + 2] = sum(twoMicSamples, delay);
63 tx_msg->data[i*4 + 3] = 0;
 16

A PPENDIX D
C CODE FOR SUMMING DELAYS

Listing 1. sum.h
1 #ifndef _SUM_H_BFORM_
2 #define _SUM_H_BFORM_
3 extern Int16 sum(Int16* newSamples, int delay);
4 #endif

Listing 2. sum.c
1
2 #include <std.h>
3
4 #include "definitions.h"
5 #include "calcDelay.h"
6
7 /* newSamples is an array with each of the four samples.
8 * delayIncrement is the amount to delay each microphone by, in samples.
9 * This function returns a "beamed" sample. */
10 Int16 sum(Int16* newSamples, int delayIncrement){
11 static int currInput = 0; // Buffer index of current input
12 int delays[NUMMICS]; // Amount to delay each microphone by
13 int mic = 0; // Used as we iterate through the mics
14 int rolloverIndex;
15 Int16 output = 0;
16 static Int16 sampleBuffer[NUMMICS][MAXSAMPLEDIFF];
17
18 // Calculate samples to delay for each mic
19 // TODO: Only do this once
20 calcDelay(delayIncrement, delays);
21
22 // We used to count backwards - was there a good reason?
23 currInput++; // Move one space forward in the buffer
24
25 // Don’t run off the end of the array
26 if(currInput >= MAXSAMPLEDIFF){
27 currInput = 0;
28 }
29
30 // Store new samples into sampleBuffer
31 for(mic=0; mic < NUMMICS; mic++){
32 // Divide by the number of microphones so it doesn’t overflow
33 // when we add them
34 sampleBuffer[mic][currInput] = newSamples[mic]/NUMMICS;
35 }
36
37 // For each mic add the delayed input to the current output
38 for(mic=0; mic < NUMMICS; mic++){
39 if(currInput - delays[mic] >= 0){// Index properly?
40 output += sampleBuffer[mic][currInput - delays[mic]];
41 }
42 else{
43 // The delay index is below 0, so add the length of the array
44 // to keep it in bounds
 17

45 rolloverIndex = MAXSAMPLEDIFF + (currInput - delays[mic]);

46 output += sampleBuffer[mic][rolloverIndex];
47 }
48 }
49
50 return output;
51
52 }
A PPENDIX E
C CODE FOR CALCULATING DELAYS

Listing 3. calcDelay.h
1 #ifndef _CALC_DELAY_H_
2 #define _CALC_DELAY_H_
3 extern void calcDelay(int delayInSamples, int* delays);
4 #endif

Listing 4. calcDelay.c
1 /* calcDelay
2 * Accepts delays in samples as an integer
3 * and returns a pointer to an array of delays
4 * for each microphone.
5 *
6 * Date: 9 March 2010
7 */
8
9 #include "definitions.h"
10
11 void calcDelay(int delayInSamples, int* delays){
12 int mic = 0;
13 if(delayInSamples > 0){
14 for(mic=0; mic < NUMMICS; mic++){
15 delays[mic] = delayInSamples*mic;
16 }
17 }
18 else{
19 for(mic=0; mic < NUMMICS; mic++){
20 delays[mic] = delayInSamples*(mic-(NUMMICS-1));
21 }
22 }
23 }
A PPENDIX F
JAVA C ODE - M AIN

Listing 5. MainWindow.java
1 /* MainWindow.java - Java class which constructs the main GUI window
2 * for the project, and sets up communication with the DSK. It also
3 * contains the main() method.
4 *
5 * Author: Steven Bell and Nathan West
6 * Date: 9 March 2010
7 * $LastChangedBy$
8 * $LastChangedDate$
 18

9 */
10
11 package beamformer;
12
13 import java.awt.*; // GUI Libraries
14
15 import javax.swing.*;
16
17 public class MainWindow
18 {
19 JFrame window;
20 DisplayPanel display;
21 ControlPanel controls;
22 static Beamformer beamer;
23
24 MainWindow()
25 {
26 beamer = new Beamformer();
27
28 window = new JFrame("Acoustic Beamforming GUI");
29 window.setDefaultCloseOperation(WindowConstants.EXIT_ON_CLOSE);
30 window.getContentPane().setLayout(new BoxLayout(window.getContentPane(), BoxLayout.LINE
31
32 display = new DisplayPanel();
33 display.setPreferredSize(new Dimension(500,500));
34 display.setAlignmentY(Component.TOP_ALIGNMENT);
35 window.add(display);
36
37 controls = new ControlPanel();
38 controls.setAlignmentY(Component.TOP_ALIGNMENT);
39 window.add(controls);
40
41 window.pack();
42 window.setVisible(true);
43
44 beamer.start();
45 }
46
47 public static void main(String[] args)
48 {
49 MainWindow w = new MainWindow();
50 while(true){
51 beamer.update();
52 w.display.updatePower(beamer.getPower());
53 }
54 }
55 }
A PPENDIX G
JAVA C ODE FOR B EAMFORMER C OMMUNICATION

Listing 6. Beamformer.java
1 /* Beamformer.java - Java class which interacts with the DSK board
2 * using RTDX.
3 *
 19

4 * Author: Steven Bell and Nathan West

5 * Date: 11 April 2010
6 * $LastChangedBy$
7 * $LastChangedDate$
8 */
9
10 package beamformer;
11
12 //Import the DSS packages
13 import com.ti.ccstudio.scripting.environment.*;
14 import com.ti.debug.engine.scripting.*;
15
16 public class Beamformer {
17
18 DebugServer debugServer;
19 DebugSession debugSession;
20
21 RTDXInputStream inStream;
22
23 int[] mPowerValues;
24
25 public Beamformer() {
26 mPowerValues = new int[67]; // TODO: Change to numBeams
27
28 ScriptingEnvironment env = ScriptingEnvironment.instance();
29 debugServer = null;
30 debugSession = null;
31
32 try
33 {
34 // Get the Debug Server and start a Debug Session
35 debugServer = (DebugServer) env.getServer("DebugServer.1");
36 debugServer.setConfig("Z:/2010_Spring/dsp/codecomposer_workspace/dsp_project_trunk/ds
37 debugSession = debugServer.openSession(".*");
38
39 // Connect to the target
40 debugSession.target.connect();
41 System.out.println("Connected to target.");
42
43 // Load the program
44 debugSession.memory.loadProgram("Z:/2010_Spring/dsp/codecomposer_workspace/dsp_projec
45 System.out.println("Program loaded.");
46
47 // Get the RTDX server
48 RTDXServer commServer = (RTDXServer)env.getServer("RTDXServer.1");
49 System.out.println("RTDX server opened.");
50
51 RTDXSession commSession = commServer.openSession(debugSession);
52
53 // Set up the RTDX input channel
54 inStream = new RTDXInputStream(commSession, "dataChan");
55 inStream.enable();
56 }
57 catch (Exception e)
 20

58 {
59 System.out.println(e.toString());
60 }
61 }
62
63 public void start()
64 {
65 // Start running the program on the DSK
66 try{
67 debugSession.target.restart();
68 System.out.println("Target restarted.");
69 debugSession.target.runAsynch();
70 System.out.println("Program running....");
71 Thread.currentThread().sleep(1000); // Wait a second for the program to run
72 }
73 catch (Exception e)
74 {
75 System.out.println(e.toString());
76 }
77 }
78
79 public void stop()
80 {
81 // Stop running the program on the DSK
82 try{
83 debugSession.target.halt();
84 System.out.println("Program halted.");
85 }
86 catch (Exception e)
87 {
88 System.out.println(e.toString());
89 }
90
91 }
92
93 public void setMode()
94 {
95
96 }
97
98 public void setNumBeams()
99 {
100
101 }
102
103 public void update(){
104 try{
105 // Read some bytes
106 byte[] power = new byte[1];
107 byte[] delay = new byte[1];
108
109 inStream.read(delay, 0, 1, 0); // Read one byte, wait indefinitely for it
110 inStream.read(power, 0, 1, 0); // Read one byte, wait indefinitely for it
111
 21

112 int intPower = power[0] & 0xFF; // Convert to int value from unsigned byte
113
114 mPowerValues[delay[0] + 33] = intPower; // Save them
115
116 System.out.println("D:" + (int)delay[0] + " P: " + intPower);
117 }
118 catch (Exception e)
119 {
120 System.out.println(e.toString());
121 }
122 } // END update()
123
124 public int[] getPower()
125 {
126 return mPowerValues;
127 }
128
129 /*
130 public int[] getDelays()
131 {
132 return int
133 }*/
134
135 } // END class
A PPENDIX H
JAVA C ODE FOR B EAM D ISPLAY

Listing 7. DisplayPanel.java
1 /* DisplayPanel.java - Java class which creates a canvas to draw the
2 * beamformer output on and handles the drawing.
3 *
4 * Author: Steven Bell and Nathan West
5 * Date: 9 March 2010
6 * $LastChangedBy$
7 * $LastChangedDate$
8 */
9
10 package beamformer;
11
12 import java.awt.*; // GUI Libraries
13 import javax.swing.*;
14 import java.lang.Math;
15
16 public class DisplayPanel extends JPanel
17 {
18
19 int mDelay[];// = {-3, -2, -1, 0, 1, 2, 3};
20 double mPower[];// = {0, .25, .5, 1, .5, .25, 0};
21 int mNumPoints = 67;
22
23 double mSpacing = 33;
24
25 DisplayPanel()
26 {
 22

27 mDelay = new int[mNumPoints];

28 mPower = new double[mNumPoints];
29
30 for(int i = 0; i < mNumPoints; i++){
31 mDelay[i] = i - 33;
32 mPower[i] = .5;
33 }
34 }
35
36 void updatePower(int[] newPower)
37 {
38 for(int i = 0; i < mNumPoints; i++)
39 {
40 mPower[i] = (double)newPower[i] / 255;
41 if(mPower[i] > 1){
42 mPower[i] = 1;
43 }
44 }
45 this.repaint();
46 }
47
48 static final int FIGURE_PADDING = 10;
49
50 // Overrides paintComponent from JPanel
51 protected void paintComponent(Graphics g)
52 {
53 super.paintComponent(g);
54
55 Graphics2D g2d = (Graphics2D)g; // Cast to a Graphics2D so we can do more advanced pain
56 g2d.setRenderingHint(RenderingHints.KEY_ANTIALIASING, RenderingHints.VALUE_ANTIALIAS_ON
57
58 // Determine the maximum radius we can use. It will either be half
59 // of the width, or the full height (since we do a semicircular plot).
60 int maxRadius = this.getWidth()/2;
61 if(maxRadius > this.getHeight()){
62 maxRadius = this.getHeight();
63 }
64 maxRadius = maxRadius - FIGURE_PADDING;
65
66 // Pick our center point
67 int centerX = this.getWidth() / 2;
68 int centerY = this.getHeight()-(this.getHeight() - maxRadius) / 2;
69
70 // Calculate all of the points
71 int[] px = new int[mNumPoints];
72 int[] py = new int[mNumPoints];
73
74 for(int i = 0; i < mNumPoints; i++)
75 {
76 double angle = delayToAngle(mDelay[i]);
77 px[i] = centerX - (int)(maxRadius * mPower[i] * Math.sin(angle));
78 py[i] = centerY - (int)(maxRadius * mPower[i] * Math.cos(angle));
79 }
80
 23

81 g2d.setPaint(Color.BLUE);
82 g2d.drawPolygon(px, py, mNumPoints);
83
84 // Draw the outline of the display, so we have some context
85 g2d.setPaint(Color.BLACK);
86 float[] dash = {5, 4};
87 g2d.setStroke(new BasicStroke((float).5, BasicStroke.CAP_BUTT, BasicStroke.JOIN_MITER,
88 g2d.drawLine(10, centerY, this.getWidth() - 10, centerY);
89 g2d.drawArc(10, (this.getHeight() - maxRadius) / 2, 2*maxRadius, 2*maxRadius, 0, 180);
90 }
91
92 // Takes a delay and converts it to the equivalent in radians
93 private double delayToAngle(int delay)
94 {
95 return(Math.PI/2 - Math.acos(delay/mSpacing));
96 }
97
98 // Takes an angle in radians (-pi/2 to +pi/2) and converts it to a delay
99 private int angleToDelay(double angle)
100 {
101 return(int)(mSpacing * Math.cos(angle + Math.PI/2));
102 }
103 }