Final Project Part II
MATLAB Session
ES 156 Signals and Systems
2007
SEAS
PreparedbyFrankTompkins
Outline
freqz() command
Step by step through the communication
system
Explanation of new concepts and new
MATLAB functions
High-level view of flow through the system
Next week we talk in more detail about
implementation
Eye diagrams
Start Early!
Project is complex
Not something you can do in one day
Less hand-holding than MATLAB
exercises in homework
More like real-life projects
Extra office hours possible
Email us if you have questions/problems
freqz()
Same inputs as filter()
Plots frequency response
channelFilterTaps = [1 0 -1/2 3/8 zeros(1,28)];
freqz(channelFilterTaps, 1);
Outputs frequency response
channelFilterTaps = [1 0 -1/2 3/8 zeros(1,28)];
H = freqz(channelFilterTaps, 1, 256); % evaluate H(e j) at 256
points
Overall Idea
Want to transmit a digital image from point A to
point B using radio waves, etc.
We wont actually build antennas/wires
Simulate the whole thing in MATLAB
Have to convert digital image to a wave that can
travel through the air, a wire, etc. at point A
Then convert wave back to a digital image at
point B
Well transmit DCTs instead of actual image
pixels, as is often done in real life applications
Image Pre-Processing
Break image into 8 pixel by 8 pixel blocks and
take DCT of each block
Quantize DCT coefficients into 256 levels by
representing them as 8-bit unsigned numbers
blkproc()
MATLAB command for applying a function
in blocks to a matrix
Example: apply DCT in 8 by 8 blocks
I = imread(myimage.tif');
fun = @dct2;
J = blkproc(I, [8 8], fun);
Quantization
Approximate a continuous range of values by a
set of discrete values
Quantization
In MATLAB uint8()
x = 5.7;
% x is double precision (32-bit floating point)
xq = uint8(x);
% xq is an 8-bit unsigned integer
For images, we have to scale to [0,1]
before quantizing with im2uint8()
xscaled = (1.4 - x) / (1.4 - 6.3); % x is an image matrix
xq = im2uint8(xscaled);
Conversion to a bit stream
Using reshape()and permute()
Arrange 8 x 8 DCT blocks into groups of N blocks
each
Reshape each block into a vector (length 8*8*N) to
be transmitted later
Convert each pixel in vector to a binary number
Conversion
reshape()
Takes elements columnwise
>> x = [1 2 3; 4 5 6; 7 8 9]'
x=
1
>> reshape(x,1,9)
ans =
1
permute()
>> x = rand(1,2,2)
x(:,:,1) =
0.4565
0.0185
x(:,:,2) =
0.8214
0.4447
>> permute(x,[2 1 3])
ans(:,:,1) =
0.4565
0.0185
ans(:,:,2) =
0.8214
0.4447
de2bi()
>> x = [4; 212; 19]
x=
4
212
19
>> de2bi(x)
ans =
0
Modulation
Modulate each bit by a sine wave and put into the channel
Implementation details are up to you
What is written in the PDF file is a suggestion
We will however take off points if you use more than one for
loop in your code
That one should loop over the N-sized block groups to send
each one through the channel in turn
Pulse Amplitude Modulation
PAM for short
We will use a specific simple type:
half-sine pulse
To send a bit
1
0.5
-0.5
For a 1 send sin(t)
For a 0 send sin(t)
-1
10
15
t
20
25
30
10
15
t
20
25
30
1
0.5
0
-0.5
-1
Modulated [1 0 0 1]
1
0.8
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
20
40
60
80
100
120
140
Channel
Atmosphere, telephone wire, coaxial cable
We model it as an LTI system
Impulse response h(t)
In using MATLAB, must approximate by
discrete time system h[n]
Noise
We will use zero mean AWGN
Additive White Gaussian Noise
Add an independent Gaussian random
variable to each sample passed through the
channel
AWGN
>> randn(2,3)
ans =
-0.4326
0.1253 -1.1465
-1.6656
0.2877
1.1909
y[n] = h[n] * x[n] + noise[n]
noisePower = 2;
% variance of Gaussian random variable
result = signalFromChannel + sqrt(noisePower) * randn(rows, cols);
[1 0 0 1] After Channel and Noise
3
-1
-2
-3
20
40
60
80
100
120
140
Receiver Equalization
Attempt to undo distortion of modulated signal
(sine wave) due to channel and noise
We will try two equalizing filters
Zero Forcing (ZF)
Minimum Mean Square Error (MMSE)
Zero Forcing (ZF) Filter
Just the inverse of the channel response
If H(ej) is the response of the channel, then the
ZF filter has response 1 / H(ej)
Clearly if there were no noise, ZF filter would
perfectly recover modulated signal
But since ZF doesnt take noise into account at
all, it will perform very badly if noise is strong
MMSE Filter
Takes noise into account
Derived by minimizing the average error
Well just take it on faith
If H(ej) is the response of the channel,
then the MMSE filter has response
Detection
Examine equalized signal to determine
whether a 1 or a 0 was sent over channel
Optimal detector is a thresholder
Threshold Detector
Integrate (sum) the transmitted and
equalized half-sine pulse
If integral (sum) < 0, decide a 0 was sent
Else decide a 1 was sent
Conversion to an Image
After receiving all block groups, use
reshape()and permute() to rebuild the
block DCT image
Conversion to an Image
Image Post-Processing
Use blkproc() to compute inverse block
DCT
Thats it!
Things To Play With
Noise Power
Increasing noise power will cause more distortion in
received image
ZF/MMSE Equalizers
Since MMSE handles noise, it should perform better
than ZF
Eye Diagrams
Coming up next
Different Channels
Optional
Eye Diagrams
Used to visualize how waveforms used to
send (modulate) multiple bits of data can
lead to detection errors
The more open the eye, the lower the
probability of error
Consider modulated half-sine pulses for
four subsequent transmitted bits
Modulated [1 0 0 1]
1
0.8
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8
-1
20
40
60
80
100
120
140
[1 0 0 1] After Channel and Noise
3
-1
-2
-3
20
40
60
80
100
120
140
[1 0 0 1] Eye Diagrams
Before Channel
After Channel
