Transcript Document
Communication Systems IK1500
Anders Västberg [email protected]
08-790 44 55 IK1500 1
IK1500 Communication Systems
• TEN1: 7,5 hec. • Seminars – Active participation in the seminars gives the grade E. For higher grades or if you missed the seminars then you can write the exam.
• Required reading: – Kumar, Manjunath, & Kuri,
Communication Networking
, Elsevier, 2004.
– G. Blom, et.al.,
Sannolikhetsteori och statistikteori med tillämpningar
, Studentlitteratur, 2005 • Course Webpage: – http://www.kth.se/student/program kurser/kurshemsidor/ict/cos/IK1500/HT09-1 HT08/P1 IK1500 2
Supplementary rules for examination
• Rule 1: All group members are responsible for group assignments • Rule 2: Document
any
help received and
all
sources used • Rule 3: Do
not
copy the solutions of others • Rule 4: Be prepared to present your solution • Rule 5: Use the attendance list correctly HT08/P1 IK1500 3
Mathematica
• Download the program from: – http://progdist.ug.kth.se/public/ • General introduction to Mathematica – http://www.cos.ict.kth.se/~goeran/archives/Ma thematica/Notebooks/General/ HT08/P1 IK1500 4
Course Overview
HT08/P1 IK1500 5
HT08/P1 IK1500 6
HT08/P1 IK1500 7
Course Aim
• Gain
insight
into how communication systems work (building a mental
model
) • Develop your
intuition
about
when
to model and
what
to model • Use
mathematical modelling
to analyse models of communication networks • Learning how to use
power tools
HT08/P1 IK1500 8
Modelling
• Find/built/invent a model of some specific system • Why? – We want to answer questions about the system’s characteristics and behaviour.
• Alternative: Do measurements!
– However, this may be: • too expensive: in money, time, people, … • too dangerous: physically, economically, … – or the system may not exist yet (a very common cause) • Often because you are trying to consider
which
system to build!
HT08/P1 IK1500 9
Modelling
• Models have
limited
areas of
validity
• The
assumptions
about
input parameters
and the system must be valid for the model to give
reliable
results.
• Models can be
verified
by comparing the model to the
real
system • Models help you not only with design, but give
insight
about
what to measure
HT08/P1 IK1500 10
Use of models
• Essential as input to
simulations
• Use models to detect and analyse errors – Is the system acting as expected?
– Where do I expect the limits to be?
• Model-based control systems HT08/P1 IK1500 11
Example: Efficient Transport of Packet Voice Calls
Voice coder and packetizer Voice coder and packetizer Depacketizer voice decoder Depacketizer voice decoder Router Communication link C bits/s Router Voice coder and packetizer Depacketizer voice decoder
Problem
: Given a link speed of C, maximize the number of simultaneous calls subject to a constraint on voice quality. HT08/P1 IK1500 12 [Kumar, et. al., 2004]
Voice Quality
• Distortion – The voice is sampled and encoded by, for example, 4 bits.
– At least a fraction a of the coded bits must be received for an acceptable voice quality.
Example: If a=0.95, then at least 3.8 bits per sample must be delivered. • Delay – Packets arrive at the link at random, only one packet can be transmitted at a time, this will cause queuing of packets, which will lead to variable delays.
HT08/P1 IK1500 13
Queuing Model
B C • B bits: The level of the multiplexer buffer that should seldom be exceeded.
• C bits/s: Speed of the link Leads to the delay bound B/C (s) to be rarely exceeded HT08/P1 IK1500 14
Design alternatives
• Bit-dropping at the multiplexer – If the buffer level would exceed B, then drop excess bits – Same as buffer adaptive coding (the queue length controls the source encoder)
Closed loop control
• Lower bit-rate coding at the source coder – Lower the source encoder bit rate – The probability of exceeding buffer level B is less than a small number (e.g. 0.001).
Open loop control
HT08/P1 IK1500 15
B HT08/P1 0
Multiplexer Buffer Level
bits dropped IK1500 time 16
Results
1.2
1 0.8
0.6
0.4
0.2
0 0 HT08/P1 bit-dropping low-bit-rate coding 5 10 15 20 delay bound (in packet transmission times) IK1500 25 30 17
Achievable Throughput in an Input-Queuing Packet Switch
• N
input
ports and N
output
ports • More than one cell with the same output destination can arrive at the inputs • This will cause
destination conflicts
.
• Two solutions: – Input-queued (IQ) switch – Output –queued (OQ) switch HT08/P1 IK1500 18 [kumar, et. al., 2004]
HT08/P1
Input-queued (IQ) switch
c4 b3 a1 1 f1 e1 d1 2 h2 g2 3 j3 i2 4 4 X 4 Switch 1 2 3 4 c time f a e d i h g b j IK1500 19
Output – queued (OQ) switch
• All of the input cells (fixed size small packets) in one time slot must be able to be switched to the
same
output port.
• Can provide 100% throughput • If N is large, then this is difficult to implement technically (speed of memory).
HT08/P1 IK1500 20
0.5
Markov chain representation N=2
0.25
1, 1 1, 2 0.25
0.25
0.25
0.25
0.25
Number of states =
N N
0.25
0.25
0.25
0.25
0.25
2, 1 2, 2 0.5
0.25
HT08/P1 IK1500 21
Saturation throughput
N Saturation throughput
1 2 3 4 5 6 1.0000
0.7500
0.6825
0.6553
0.6399
0.6302
Capacity of a switch is the maximum rate at which packets can arrive and be served with a bounded delay.
The insight gained: capacity ≈ saturation throughput 7 8 0.6234
0.6184
Converges to: 2 2 0 .
586 HT08/P1 IK1500 22
Virtual Output Queuing
• A virtual output queue at input
i
for output and is denoted by
VOQ ij
• Maximum-weight matching algorithm
j
VOQ 11 VOQ 12 VOQ 21 VOQ 22 2 x 2 switch 1 Q 11 Q 12 2 Q 21 Q 22 1 2 HT08/P1 IK1500 23