Transcript 1-p

Human InformationProcessing
Li Liu
Human Computer Interaction
”Human
Information
Processing”
How much
information can
she receive?
Case 1
Personal Area Network
Principle
Speech – Control Technology
Name Dialing
Hand Free
Speech Recognizer
Speech
Recognizer
”repeat”
Speech Signal Waveform
Effect of Background Noises
Performance
Words Correct Rate (%)
100
0
-30
Speech-to-Noise Ratio (dB)
0
Lip Tracker
Performance of Speech Recognition
Words Correct Rate (%)
100
Auditory and
visual cues
Auditory
cues only
0
-30
Speech-to-Noise Ratio (dB)
0
How to Design A Headset?
Tip ?
Mobile Phones
Braille Code for English
Tactile Display
Problem: Usable?
Efficiency?
Reliability?
Braille Code for English
How Many Letters Can be
Represented?
Combination
Tactile Pins
Braille Code for English
W=2.4093
The average number
of active tactile pins.
Wopt=1.2118
Letters in written English
New Code
W=1.2316
Information
Information Theory
If all eight horses are equally good,
then the chance for a horse to win
the race is 1/8!
What is the probability?
The probability for a horse to win is
Pi=1/8
Information Theory
If a horse wins the game, how much
information is it given?
How to measure information?
Measure of information
1/8
1/8
1/8
1/8
• A source A={a1,a2…aN} the probability
of each event P={P1,P2…PN}
• More information if Pi is low, less
information is Pi is high
1/2
1/4
1/8
1/16
• I(ai)=log(1/Pi) is called self
information
Entropy
1/8
1/8
1/8
1/8
Taking the mean value over all
symbols with the alphabet
H(x)   Pi log2 Pi (bits / symbol)
1/2
1/4
1/8
1/16
Communication Theory I
H
R
source
If R < H
error-free representation impossible!
Coding
1/8
1/8
1/8
1/8
000
001
010
111
1/2
1/4
1/8
1/16
0
10
110
1110
111100
111101
111110
111111
How much information does a
picture contain?
255
254
0
How much information does a
picture contain?
Suppose the size of a picture is 256 x256,
And each pixel is represented by 8 bits, then
Total bits to represent such a picture is
8x256x256= 524288 bits/picture
How many pictures can be represented
by a half Mbits?
0100110101000011101110101010..............
Total number of pictures
2524288 = 10158000
How many pictures have
human beings been perceived
Up to now, the number of all pictures which
have been perceived by human beings is only
1019.
1019 <<< 10158000
Assume that 30 frames of pictures per second
are perceived by human eyes. If the average
life-span is 70 years and 8 hours per day are
spent on sleep, then about 108 frames are
received by one person in all his/her life.
Furthermore, if the total number of human
beings who lived in the earth is 1011, then, the
total number of all pictures which
have been perceived by human beings is only
1019.
Entropy of an image
probability
intensity
45
123
208
001
00010
00011
H=6.9
Entropy of human face images
A face image = 50 bytes (400 bits) !
Communication
Brain Wave
H
C
source
channel
Communication Channel
Abstract Channel Model
Ex. Noiseless Binary Channel
Channel Capacity
C= 1 bit
0
0
1
1
Communication Channel
Abstract Channel Model
Ex. Noisy Binary Channel
0
Channel Capacity
1
C= 1 + p log2 p + (1-p) log2 (1-p) bits
1-p
p
p
1-p
0
1
Communication & Design
receiver
transmitter
Different Decoders
”When a group of people look at an object, none of
them sees exactly the same thing as anyone else.
Even if they receive approximately the same image
on their retina and interpret the image in basically
the same way, this image is always revised by the
observer’s personality and situation.”
Rune Monö
Real-World Channel Model
Ex. Band-limited Gaussian Channel
Y(t)
X(t)
Z(t)
Gaussian noise
C = Wlog2(1+P/N) bits/s
W: channel bandwidth
P/N: signal to noise ratio
Real-World Channel Model
Telephone signals are band-limited to
3300Hz and have a SNR of 20 dB
(P/N=100),
C=3,300 log (1+100)= 22,000 bits/s
Practical modems achieve transmission
rates up to 19,200 bits/s
Communication Theory II
H
Source
R
C
Channel
If H <= C lossless transmission possible
If H > C
lossless transmission impossible!
Brain Wave
Information Rate
Sensory
Bandwidth
sensation
Perceptual
Bandwidth
data
perception
information
Cognitive
Bandwidth
cognition
knowledge
Information Rates
Sense
Information Stream
Bits/s
Vision
10.000.000
Hearing
100.000
Touch
10.000
Smell
1.000
Taste
1.000
Sense
Bandwidth of
consciousness bits/s
vision
40
hearing
30
touch
5
Taste and smell
1
Channel=”Bandpass Filter”
Effect of Bandpass
Effect of degrading the speech signal by
spectral filtering
The original speech signal contains
significant energy up to 7 kHz
a) signal bandpass filtered (0.25 - 0.75 kHz)
b) signal bandpass filtered (1.0 – 3.0 kHz)
c) original signal
Bandpass System
Most long-haul transmission systems have a bandpass
Frequency response
The transfer function can be written as
R
=
L
H ( f )  1 jQ(1f  fo )
fo
C
f
where the resonant frequency fo and quality factor Q are
fo  2 1 LC
QR
C
L
The 3dB bandwidth between the lower and upper cutoff
1.0
0.707
frequencies is
B
fl
fo
fu
f
B  fu  fl 
fo
Q
Bandpass System
Since practical tuned circuits usually have 10 < Q <100,
the fractional bandwidth B/fc should be kept within the
range
B
0.01 
 0.1
fc
As a rough rue of thumb, the carrier frequencies
and corresponding nominal bandwidth can be
B  0.02fc
Bandpass System
Selected carrier frequencies and nominal bandwidth
Frequency band
Longwave radio
Carrier
frequency
Bandwidth
100 kHz
2 kHz
5 MHz
100 kHz
100 MHz
2 MHz
5 GHz
100 MHz
Millimeterwave
1 00 GHz
2 GHz
Optical
5x1014 Hz
1013 Hz
Shortwave radio
VHF
Microwave