MIT 6.02 introduction to EECS II(digital communication) video notes

//////////////////////////////////////////////////////////////
22:16 2014-6-17
MIT 6.02 introduction to digital communication


video I, encoding information


22:19 2014-6-17
communication channel


22:19 2014-6-17
mathematical model


wired networks, wireless network


22:21 2014-6-17
CAN == Controller Area Network


automative network


22:22 2014-6-17
digital communications


22:23 2014-6-17
communication across space   // flash disk


communication across time


22:23 2014-6-17
python network stack protocol


22:26 2014-6-17
pointer-to-pointer communication channels:


transmitter => receiver


22:27 2014-6-17
untangle of the some bad things happend  // error correction


22:28 2014-6-17
corruption


22:29 2014-6-17
signal processing, error detection, error correction


22:29 2014-6-17
real world


22:29 2014-6-17
sharing a channel


22:32 2014-6-17
* pointer-to-pointer networks


* multi-hop networks


packet switching,


efficient routing


22:32 2014-6-17
engineering goals for networks


* efficient


* secure


* scalable


* cost


* reliable


22:38 2014-6-17
robust: survive even with failures


22:38 2014-6-17
information resolves uncertaity


22:39 2014-6-17
likely events carries less information than unlikely events


22:40 2014-6-17
encoding


22:47 2014-6-17
measuring information content


22:47 2014-6-17
binary encoding


22:48 2014-6-17
// coin toss


1 encoding head


0 encoding tail


22:48 2014-6-17
data rate <=> information rate


22:49 2014-6-17
discrete random variable


22:51 2014-6-17
expected information content: entropy


22:54 2014-6-17
Why care about entropy?


* entropy tells us the average amount of information that must


  be delivered in order to resolve all uncertainty


* data rate <=> information rate  // match?


* entropy gives us a metric to measure encoding effectiveness


22:56 2014-6-17
capacity of the communication channel


22:56 2014-6-17
average amout of information


22:57 2014-6-17
more fancy encoding scheme


23:14 2014-6-17


* fixed-length encoding


* varialbe-length encoding  // David Huffman, MIT 1950


23:18 2014-6-17
Huffman tree


23:21 2014-6-17
optimal encoding


23:23 2014-6-17
a moment of disruption  // MOOC edx


a moment of reflection 


a moment of transformation/evolution


digital evolution


23:24 2014-6-17
distributed power


0:03 2014-6-18
talking to dead people // Rome


0:10 2014-6-18
justice


///////////////////////////////////////////////////////////////////////////


17:21 2014-6-18 Wednesday


information rate <=> data rate


17:23 2014-6-18
let's get closer to the information rate


17:23 2014-6-18
adaptive variable-length codes


LZW algorithm


17:29 2014-6-18
transmit table indices


17:35 2014-6-18
the decoder can reconstruct the table 


17:51 2014-6-18
symbol table // string table


17:56 2014-6-18
encoder notes:


* encoder algorithm is greedy


* no waste


* compression increases as encoding progresses 


17:57 2014-6-18
video compression


17:57 2014-6-18
about the decoder:


you building table the same way as the encoder does


17:59 2014-6-18
lossless compression,


lossy compression


18:00 2014-6-18
Huffman & LZW are lossless encoding


18:00 2014-6-18
lossy compression: target at audio & video


18:03 2014-6-18
perceptual coding


18:06 2014-6-18
high spatial frequencies


18:06 2014-6-18
bandwidth constraints


18:07 2014-6-18
perceptual coding example:


spend more bits on luminance than color


// encode separately


18:08 2014-6-18
RGB to YCbCr conversion


18:14 2014-6-18
energy levels at different spatial frequencies


18:16 2014-6-18
YCrCb representation


18:17 2014-6-18
basis function


18:19 2014-6-18
basis expansion


18:19 2014-6-18
2D DCT Basis Functions


18:19 2014-6-18
spatial frequencies


18:22 2014-6-18
DC component


18:22 2014-6-18
JPEG compression


18:24 2014-6-18
JPEG == Joint Photographic Expert Group


18:26 2014-6-18
DCT coefficient


18:26 2014-6-18
quantization  // lossy part


18:26 2014-6-18
this is the size of the bucket


////////////////////////////////////////////////////////////


8:58 2014-6-19 Thursday
bit stream


8:58 2014-6-19
ADC == Analog Digital Converter


8:59 2014-6-19
what is the point of digital communication?


9:02 2014-6-19
anlog woe


9:04 2014-6-19
analog errors accumulate


9:15 2014-6-19
digital signaling


9:17 2014-6-19
binary signaling scheme


9:17 2014-6-19
we need a "forbidden zone"


9:34 2014-6-19
NM == Noise Margin


9:53 2014-6-19
continuous time, discrete time


10:05 2014-6-19
sampling rate, sampling period


10:05 2014-6-19
samples per second


10:06 2014-6-19
sample clock, 


transmit clock


10:14 2014-6-19
clock recovery @receiver


10:20 2014-6-19
infered clock edges


10:20 2014-6-19
receiver can infer presence of clock edges everytime


there is a transition in the received samples


10:21 2014-6-19
then using sample period


10:21 2014-6-19
using middle sample to determine message bit


10:22 2014-6-19
recoding


10:25 2014-6-19
ensure transition every so often


10:25 2014-6-19
I need to know where the group are?


10:29 2014-6-19
8 bit, 10 bit code


10:33 2014-6-19
clock recovery, 8 bit => 10 bit code


--------------------------------------------------------
10:35 2014-6-19
start introduction to digital communication video 4,


LTI systems


10:46 2014-6-19
building mathematical models for the transmission channel


10:47 2014-6-19
system input => response


10:48 2014-6-19
input =>[transmission channel]=>  response


10:49 2014-6-19
input sequence, response sequence


10:51 2014-6-19
unit step, unit step response


10:51 2014-6-19
u[n]: unit step 


s[n]: unit step response


10:57 2014-6-19
unit sample


11:00 2014-6-19
unit sample, unit sample response


11:03 2014-6-19
unit sample decomposition


11:04 2014-6-19
unit step decomposition


11:09 2014-6-19
the invariant system


11:11 2014-6-19
linear system


11:13 2014-6-19
for LTI systems, we have convolution


x[n] * h[n]  // convolution sum


11:17 2014-6-19
convolution sum, convolution integral


11:18 2014-6-19
the channel behavior


11:19 2014-6-19
the unit sample response sequence h[n]


completely characterize the LTI system


11:20 2014-6-19
convolution operator


11:21 2014-6-19
properties of convolution:


1. cummutative


2. associative


3. distributive


11:22 2014-6-19
parallel interconnection of LTI system


11:28 2014-6-19
series interconnection of LTI system


11:33 2014-6-19
channel as LTI systems


11:33 2014-6-19
causal == nonantipitative


11:37 2014-6-19
real-world channels are causal


11:37 2014-6-19
relationship between h[n] & s[n]? // unit sample response & unit step response


h[n] = s[n] - s[n-1]


11:47 2014-6-19
if I put this into a channel, what comes out?


11:48 2014-6-19
8 samples per bit


11:59 2014-6-19
real signal going in, and real signal going out


11:59 2014-6-19
digitization threshold


12:00 2014-6-19
why the "middle sample"? // 8 samples per bit


12:02 2014-6-19
Faster transmission: 4 samplers per bit


12:04 2014-6-19
convolution sum, convolution integral
-----------------------------------------------------------


start Harvard happiness, video 14


////////////////////////////////////////////////////////////


12:52 2014-6-20 Friday
start Harvard happiness, video 15


perfectionism


12:57 2014-6-20
---------------------------------------------------------
start MIT introduction to digital communication


video 5, ISI, deconvolution


14:45 2014-6-20
ISI == InterSymbol Inference


14:45 2014-6-20
transmission over a channel


15:55 2014-6-20
convolution sum: "flip & slide"


15:55 2014-6-20
8 samples/bit


16:22 2014-6-20
bit cell  // 4 samples/bit, 8 samples/bit


=> ISI(InterSymbol Interference)


16:24 2014-6-20
in doing the convolution sum, the impulse response h[n]


might cover more bit cell, thus came the concept of "ISI"


16:25 2014-6-20
neighboring cell


16:27 2014-6-20
neighboring bits


16:27 2014-6-20
I'm getting interference from lots of neighboring bits


16:27 2014-6-20
frequency domain


16:30 2014-6-20
the bottom line is ....


16:30 2014-6-20
I have to be concerned with contribution from neighboring bits


16:31 2014-6-20
Given h[n], how bad is ISI?


16:32 2014-6-20
compute B, how many bits "covered" by h[n]?


B = length of active portion of h[n] / N + 2


16:34 2014-6-20
Generate a test pattern


16:36 2014-6-20
transmit the test pattern over the channel


16:36 2014-6-20
instead of plot of y[n], plot the response as an "eye diagram"


1. break the plot up into short segments each containing 2N+1 samples


   start at samples 0, N, 2N,...


2. plot all the short segments on top of each other


16:38 2014-6-20
generate all possible combinations of B bits(not samples)


// for 4 samples/bit, B == 6


16:49 2014-6-20
"eye diagram"


16:49 2014-6-20
Is the eye big enough, can we transmit the information


we want?


16:50 2014-6-20
"eye diagram" and h[n] both have the same information


16:50 2014-6-20
width of eye


16:50 2014-6-20
worst case signaling


16:51 2014-6-20
"width" of eye   // vertical distance


16:52 2014-6-20
welcome to the wonderful world of jargon


16:52 2014-6-20
the maximize the NM(noise margin)


16:53 2014-6-20
digitization threshold


16:53 2014-6-20
How to maximize noise margin using "eye diagram"?


* pick the best sample point => widest point in the eye


* pick the best digitization threshold => half-way across width


16:56 2014-6-20
What's the point of "eye diagram"?


* choosing the best "sample point" & "digitization threshold"


* choosing the best "Samples/Bit"


17:00 2014-6-20
no eye


17:02 2014-6-20
noise margin: "the digitization threshold" to "the widest point at the eye"


17:02 2014-6-20
If I give you a noise margin, you can determine how 


many "samples per bit" can use?


17:03 2014-6-20
fast channel


17:40 2014-6-20
slow channel


17:43 2014-6-20
ringing channel


17:44 2014-6-20
bit cell


17:45 2014-6-20
multi-hop networks


17:46 2014-6-20
can we recover from ISI?


solution: deconvolution


because transmit over a channel can be modeled as y[n] = x[n] * h[n]


we can recover x[n] using deconvolution


17:48 2014-6-20
if I have perfect deconvolver


17:55 2014-6-20
if w[n] is a perfect estimate of x[n]


17:56 2014-6-20
we can compute w[n] from y[n] & w[n] using "plug & chug"


18:01 2014-6-20
sensitivity to noise


18:11 2014-6-20
amplifying noise is never a good thing


18:15 2014-6-20
noisy deconvolution example


18:16 2014-6-20
stability criterion


//////////////////////////////////////////////////////
16:42 2014-6-21 Saturday
start MIT introduction to digital communication, 


video 6, noise bit errors


16:42 2014-6-21
mathematical model for the noise


16:43 2014-6-21
bad things happen to good signal


16:44 2014-6-21
noise-free signals


16:44 2014-6-21
random process


16:46 2014-6-21
additive noise


16:47 2014-6-21
multipath interference


16:47 2014-6-21
unit sample response of the channel: h[n]


16:49 2014-6-21
mean, power, energy


17:00 2014-6-21
SNR == Signal Noise Ratio


17:06 2014-6-21
SNR is often measured in decibels(dB)


17:07 2014-6-21
power amplifier


17:13 2014-6-21
amplification factor


17:14 2014-6-21
signal quality degrades with lower SNR


17:14 2014-6-21
What's a "random process"?


random processes, such as noise, take on different 


sequences for different trials


17:17 2014-6-21
a whole collection of trials


17:18 2014-6-21
stationary random process,


ergodic random process


17:18 2014-6-21
What is a "stationary random process"?


statistical behavior is independent of shifts 


in time in a given trial.


// time invariant


17:20 2014-6-21
statistical sampling can be performed at one sample time


across different trials, 


or across different sample times of the same trials with no


change in the measured result.


17:22 2014-6-21
assumption of stationarity implies histogram should converge


to a shape known as a PDF


17:24 2014-6-21
statistical distribution


17:24 2014-6-21
PDF == Probability Distribution Function


17:25 2014-6-21
by the stationary assumption, we know that


it's going to converge to some statistical distribution


// PDF


17:28 2014-6-21
this histogram converge to a shape


17:29 2014-6-21
PDF == Probability Distribution Function


17:30 2014-6-21
minus infinity, plus infinity


17:34 2014-6-21
uniform PDF


17:39 2014-6-21
What's the meaning of "ergodicity"?


Assumption of ergodicity implies the value occuring at a given


time sample, noise[k], across many different trials has the same


PDF as estimated in our previous experiemnt of many time samples


and one trial.


17:49 2014-6-21
sample value distribution


17:52 2014-6-21
mean & variance


17:52 2014-6-21
visualizing mean & variance


17:54 2014-6-21
noise on a communication channel


17:59 2014-6-21
CLT == Central Limit Theorem


18:01 2014-6-21
what's the significance of CLT(Central Limit Theorem)?


1000 trials of summing 100 random samples draw from [-1, 1]


using two different distributions.


// one is triangular PDF, the other is uniform PDF


the sum both takes on a "normal PDF"


18:03 2014-6-21
the Central Limit Theorem tells me that the sum


is normally distributed


18:05 2014-6-21
Gaussian PDF == normal PDF


18:06 2014-6-21
standard deviation:


square root of variance


18:07 2014-6-21
CDF == Cumulative Distribution Function


18:10 2014-6-21
symmetric distribution


18:13 2014-6-21
unit normal


18:16 2014-6-21
BER == Bit Error Rate


18:18 2014-6-21
corrupted by noise


18:18 2014-6-21
bit == symbol


samples/bit


--------------------------------------------------
18:28 2014-6-21
start MIT 6.02 introduction to digital communication


video 7


18:28 2014-6-21
ISI == InterSymbol Interference


BER == Bit Error Rate


18:30 2014-6-21
digital communication channel


18:30 2014-6-21
noise-free version


18:30 2014-6-21
sequence generated by a random process


18:31 2014-6-21
sigma is the standard deviation


18:34 2014-6-21
BER == Bit Error Ratio


18:47 2014-6-21
noise-free channel with no ISI


18:47 2014-6-21
noise-free signal


18:48 2014-6-21
original digital sequence


18:48 2014-6-21
mean, power


18:51 2014-6-21
p(bit error)


18:52 2014-6-21
Gaussian noise


18:52 2014-6-21
digitization threshold


18:52 2014-6-21
what is p(bit error)?


assume the channel has Gaussian noise,


p(bit error) = p(transmit "0") * p(error|transmit "0")


    + p(transmit "1") * p(error|transmit "1")


18:55 2014-6-21
the voltage is corrupted by noise


18:58 2014-6-21
BER == Bit Error Rates


    == Bit Error Ratio


19:01 2014-6-21
BER(no ISI) vs. SNR


19:03 2014-6-21
so if you give me an SNR, I can compute BER


19:03 2014-6-21
BER == p(bit error)


19:05 2014-6-21
convolution sum, ISI,


unit sample response of communication channel: h[n]


samples per bit


19:22 2014-6-21
all possible transmission


19:22 2014-6-21
test sequence to generate the eye diagram


19:23 2014-6-21
eye diagram


19:25 2014-6-21
width of the eye


19:25 2014-6-21
p(bit error) with ISI


19:29 2014-6-21
the error rate because so large that the channel


becomes unusable


19:32 2014-6-21
you're much sensitive to noise if you have ISI


19:39 2014-6-21
digitization threshold


19:40 2014-6-21
choosing Vth


19:44 2014-6-21
minimizing BER


19:49 2014-6-21
minimizing BER when p(0) != p(1)


19:49 2014-6-21
triangular distribution


19:51 2014-6-21
noise PDF


19:54 2014-6-21
random noise


19:54 2014-6-21
AWGN == Additive White Gaussian Noise


19:58 2014-6-21
AWGN channel


19:58 2014-6-21
white: frequency distribution of the noise


noise has equal energy at all frequencies


19:58 2014-6-21
error corruption & detection


19:59 2014-6-21
probability distribution function for the noise


21:27 2014-6-21
Cumulative Distribution Function


21:29 2014-6-21
BER == Bit Error Rate(Bit Error Ratio)


21:30 2014-6-21
bit error


21:31 2014-6-21
What fraction of the sample we get are going to corrupted?


21:31 2014-6-21
that get our a sense of how bad the channel is


21:32 2014-6-21
we want to compare the power of signal versus the power of noise


21:49 2014-6-21
p(bit error)  // without ISI


21:50 2014-6-21
corrupted by noise


21:56 2014-6-21
use PDF in anger


21:59 2014-6-21
BER(no ISI) vs. SNR


21:59 2014-6-21
p(bit error) == BER


22:02 2014-6-21
given a particular SNR, we can compute BER


using the previous formula


22:02 2014-6-21
SNR & BER
ISI & BER


22:06 2014-6-21
test sequence to generate eye diagram


22:14 2014-6-21
a channel with ISI


a channel without ISI


22:23 2014-6-21
bit error rate // BER


22:24 2014-6-21
you're much more sensitive to noise if you have a 


lot of ISI


22:25 2014-6-21
signaling protocol


22:33 2014-6-21
Gaussian noise distribution


triangular noise distribution


22:38 2014-6-21
error correction & detection


//////////////////////////////////////////////////////////


13:49 2014-6-22 Sunday
start MIT 6.02 introduction to digital communication


video 8, ECC


13:49 2014-6-22
* coping with errors using packets


* checksums, CRC


* hamming distance & single error correction


* (n,k) block codes


13:50 2014-6-22
digital signaling scheme


13:52 2014-6-22
recover original signal despite small amplitude errors


introduced by ISI & noise


13:53 2014-6-22
digital data stream


13:53 2014-6-22 
bit errors


13:53 2014-6-22
Gaussian PDF


13:54 2014-6-22
probability density function


13:54 2014-6-22
Bit Errors


13:56 2014-6-22
BER == Bit Error Rate // Bit Error Ratio


13:56 2014-6-22
dealing with errors: packets


13:57 2014-6-22
digitizaton of your voice


13:58 2014-6-22
divide into chunks


13:58 2014-6-22
fixed-sized packets


13:58 2014-6-22
sequence number + message + checksum


13:59 2014-6-22
request a retransmission of the packets


14:02 2014-6-22
reliable transport protocol


14:03 2014-6-22
throw things that are apparent bad, and request a retransmission


14:03 2014-6-22
check bits


14:04 2014-6-22
packet size


14:04 2014-6-22
choose a proper packet size


14:04 2014-6-22
overhead + payload


14:05 2014-6-22
engineering tradeoff


14:05 2014-6-22
check bits: hash function


14:06 2014-6-22
detecting errors


14:08 2014-6-22
likely errors


14:09 2014-6-22
calling good packets bad


14:09 2014-6-22
checksum, check bits


14:09 2014-6-22
likely errors:


* random bits(BER)


* error bursts


14:25 2014-6-22
CRC == Cyclical Redundancy Check


14:46 2014-6-22
shift register


14:49 2014-6-22
generating the CRC


14:49 2014-6-22
it will be caught by the CRC, that packet will


be flaged as bad


14:51 2014-6-22
wireless channel,


wire channel


14:52 2014-6-22
BER == Bit Error Rate


14:52 2014-6-22
packet retransmission


14:53 2014-6-22
SEC == Single Error Correction


15:05 2014-6-22
retransmission rate


15:05 2014-6-22
single-bit errors


15:05 2014-6-22
Single Error Correction


15:06 2014-6-22
ECC == Error Correction Code


SECC == Single Error Correction Code


15:07 2014-6-22
digital transmission using SECC


15:07 2014-6-22
the meaning of ECC?


reduce the number of retransmission


15:08 2014-6-22
digital channel


15:09 2014-6-22
error correction


15:09 2014-6-22
error detection & correction


15:10 2014-6-22
bit stream


15:10 2014-6-22
single-bit error


15:10 2014-6-22
Hamming distance


15:13 2014-6-22
What are the Hamming distance between these two sequence?


15:14 2014-6-22
Error Detection


15:18 2014-6-22
valid code word,


invalid code word


15:19 2014-6-22
single-bit error


15:19 2014-6-22
parity bit


15:20 2014-6-22
even parity


15:20 2014-6-22
what we need is an encoding


15:39 2014-6-22
ECC == Error Correction Code


SECC == Single Error Correction Code


15:40 2014-6-22
if D is the minimum Hamming distance between code words,


we can detect up to (D-1) bit errors


15:40 2014-6-22
valid code word, 


invalid code word


15:40 2014-6-22
parity check


15:41 2014-6-22
parity bit


15:41 2014-6-22
parity(even parity)


15:42 2014-6-22
if more than 1 bit is flipped, that in technical word


is screwed


15:43 2014-6-22
this is the take-home message


15:45 2014-6-22
error detection,


error correction


15:46 2014-6-22
SECC == Single Error Correcting Codes


* use multiple parity bits


* syndrome bits


----------------------------------------------------


15:55 2014-6-22
start MIT 6.02 introduction to digital communication


video 9, 


15:55 2014-6-22
* How many parity bits?


* Dealing with burst errors


* Reed-Solomon codes


15:56 2014-6-22
systematic block code


15:59 2014-6-22
to deduce which bit is bad?


16:01 2014-6-22
parity check


16:02 2014-6-22
compute a syndrome bit // error bit


16:02 2014-6-22
parity bit(P)


syndrome bit(E)


data bits(B)


16:17 2014-6-22
code word //(n, k) code


message bits + parity bits


16:17 2014-6-22
systematic block codes


16:17 2014-6-22
What is (n, k, d)?


k: message bits


n-k: parity bits


d: the minimum Hamming distance between code words


16:18 2014-6-22
code rate: k/n


16:18 2014-6-22
(n, k, d)


16:18 2014-6-22
row parity bits,


column parity bits


16:27 2014-6-22
my even parity is violated


16:31 2014-6-22
parity check


16:34 2014-6-22
parity bits


16:36 2014-6-22
How many parity bits do I need?


16:36 2014-6-22
SECC == Single Error Correcting Code


16:36 2014-6-22
syndrome bit


16:38 2014-6-22
ECC == Error Correcting Code


16:42 2014-6-22
rectangular codes with row/column parity


16:47 2014-6-22
noise model:


* Gaussian noise


* impulse noise


16:48 2014-6-22
Dealing with burst errors


16:50 2014-6-22
Can we think of a way to turn a B-bit error burst into B


single-bit errors?


16:50 2014-6-22
interleaving


16:52 2014-6-22
interleave bits


16:52 2014-6-22
at the receiver, do "deinterleaving"


16:52 2014-6-22
interleaving:


compute CRC => partition => apply ECC => interleave => transmit


16:53 2014-6-22
interleaving & deinterleaving


16:54 2014-6-22
error correction scheme


16:55 2014-6-22
almost all modern code use interleave at some level


16:56 2014-6-22
pretty nift idea


16:56 2014-6-22
interleaved block,


ECC block


16:58 2014-6-22
8b10b encoding


16:59 2014-6-22
mark the begining of these interleaved blocks


16:59 2014-6-22
How to transmit?


* packetize: {#, data, chk}   // packet


* SECC: (n, k, d) // split {#, data, chk} into k-bit block, add parity bits 


 // code word


* 8b10b   // synchronization info  // frame


* bit => samples per bit voltage sample


17:06 2014-6-22


Framed bit stream: Reed-Solomon codes


unframed bit stream: convolutional codes


17:09 2014-6-22
oversampled polynomial


17:14 2014-6-22
packet => code word => frame


checksum, parity, sync


17:33 2014-6-22
finite field // Galois field


17:33 2014-6-22
R-S code


17:39 2014-6-22
CD error correction


* de-interleave  // turn burst erros into single bit errors


17:41 2014-6-22
concatenated R-S codes


//////////////////////////////////////////////////////////////////


7:46 2014-6-23 Monday
start MIT introduction to digital communication


video 6, convolutional codes


7:52 2014-6-23
ECC == Error Control Codes


7:53 2014-6-23
linear block code


7:53 2014-6-23
codeword


7:54 2014-6-23
minimum Hamming distance


7:54 2014-6-23
Bi-orthogonal codes


7:55 2014-6-23
picture transmission errors


7:55 2014-6-23
Green machine


7:56 2014-6-23
data rate


7:56 2014-6-23
efficient decoding algorithm


7:56 2014-6-23
FFT == Fast Fourier Transform


7:57 2014-6-23
command system


7:57 2014-6-23
quantitative command


control command


7:59 2014-6-23
convolutional codes with Viterbi decoding


8:01 2014-6-23
combination of convolutional code:


turbo code


8:10 2014-6-23
convolutional code


8:15 2014-6-23
data rate


8:15 2014-6-23
bps == bit per second


8:15 2014-6-23
block code


8:16 2014-6-23
message bits,


parity bits


8:16 2014-6-23
parity bits are computed from blocks of message bits


8:16 2014-6-23
rate 1/r codes == r parity bits/message bits


8:18 2014-6-23
sliding window


8:18 2014-6-23
constraint length K


8:18 2014-6-23
so it's a linear combination of some "message bits"


8:21 2014-6-23
What does "convolution code" do?


message bits => parity bits // generate using convolution operation


8:25 2014-6-23
K: the "constraint length" of the code


=> greater redundancy


=> better error correction possibilities


8:34 2014-6-23
we'll transmit the parity sequence, not the message itself


// convolutional code


8:37 2014-6-23
generator


8:38 2014-6-23
shift register


8:41 2014-6-23
shift register view of convolutional code


8:42 2014-6-23
state of the register


8:51 2014-6-23
state-machine view


9:01 2014-6-23
state diagram


9:06 2014-6-23
state machine view


shift-register view


9:19 2014-6-23
unfolding things in time // state machine diagram


9:27 2014-6-23
it's the same story all over again


9:30 2014-6-23
minimum Hamming distance


9:43 2014-6-23
Trellis view


10:13 2014-6-23
constraint length


code rate


coefficient of the generator


No. of states in state machine of this code


----------------------------------------------------
10:22 2014-6-23
start MIT introduction to digital communication,


video 7, viterbi decoding


10:25 2014-6-23
key concept of decoding: Trellis


10:25 2014-6-23
parity check bits


10:26 2014-6-23
message bits, parity bits


10:28 2014-6-23
Viterbi decoding


10:36 2014-6-23
Trellis diagram vs state diagram


10:38 2014-6-23
Trellis View at Transmitter


10:58 2014-6-23
decoding: finding the ML path


11:09 2014-6-23
ML == Maximum Likelihood


11:09 2014-6-23
hard decision decoding


soft decision decoding


11:17 2014-6-23
most likely


11:22 2014-6-23
Viterbi algorithm:


want: most likely message sequence


Have: (possibly corrupted)received parity sequence


11:23 2014-6-23
intermediate stage


11:25 2014-6-23
finding the Maximum Likelihood path


11:32 2014-6-23
branch metric, path metric


11:39 2014-6-23
BM == Branch Metric


PM == Path Metric


11:41 2014-6-23
BM == Branch Metric


Hamming distance between expected parity bits &


received parity bits


11:42 2014-6-23
Hard-Decision Viterbi Decoding:


a walk through the trellis


11:48 2014-6-23
path metric: min of all paths leading to state


11:49 2014-6-23
soft branch metric:


the square of the Euclidian distance between received voltages


& expected voltages


11:51 2014-6-23
navigating through the trellis is exactly the same


11:54 2014-6-23
post-decoding BER


12:02 2014-6-23
Hamming code, convolutional code


12:03 2014-6-23
constraint length


12:10 2014-6-23
Hamming code, convolutional code


12:10 2014-6-23
block code,  // Hamming code?


convolutional code


---------------------------------------------------
12:24 2014-6-23
start MIT introduction to digital communication


video 12, sharing MAC protocols


12:24 2014-6-23
shared medium => media access control // MAC


TDMA  // Time Division Multiple Access


contention protocols, Alohanet


12:25 2014-6-23
shared communication channel


12:26 2014-6-23
avoid collision,


communication protocol


12:27 2014-6-23
measure good // metric


12:29 2014-6-23
good performance


12:30 2014-6-23
channel capacity is a limited resource


12:30 2014-6-23
high utilization


12:30 2014-6-23
protocol overhead


12:32 2014-6-23
MAC == Media Access Control


12:34 2014-6-23
divide resource equally along all requesters


12:34 2014-6-23
bounded wait


12:38 2014-6-23
isochronous communication // voice, video


12:39 2014-6-23
good performance:


* High utilization


* fairness


* bounded wait


* scalability


12:39 2014-6-23
dynamically accomodate more users


12:40 2014-6-23
sharing protocols:


MAC == Media Access Control


12:41 2014-6-23
time division


12:41 2014-6-23
time slot


12:41 2014-6-23
requester


12:41 2014-6-23
Time Division:


* prearranged: TDMA


* Not prearranged: contention protocol(e.g. Alohanet)


12:42 2014-6-23
contention protocols


12:43 2014-6-23
frequency division


12:43 2014-6-23
FDMA == Frequency Division Multiple Access


12:43 2014-6-23
CDMA == Code Division Multiple Access


12:45 2014-6-23
* orthogonal pseudorandom code


* dot product to select


12:50 2014-6-23
utilization:


U = total throughput over all nodes / maximum data rate of channel


12:50 2014-6-23
* backlogged


* offered load


12:51 2014-6-23
Fairness


12:56 2014-6-23
contention protocol


12:57 2014-6-23
time slot


12:57 2014-6-23
shared medium


12:57 2014-6-23
queue of packets


13:01 2014-6-23
round-robin scheme


13:04 2014-6-23
centralized resource allocator


13:06 2014-6-23
time synchronization


13:06 2014-6-23
TDMA is very good from a fairness point of view


13:08 2014-6-23
TDMA for GSM phones


13:08 2014-6-23
RAC == Random Access Channel


13:09 2014-6-23
contention protocol: Alohanet


13:13 2014-6-23
allocation is not predetermined


13:14 2014-6-23
burst data patterns


13:14 2014-6-23
Alohanet:


Alohanet was a satellite-based data networking 


connecting computers on the Hawaiian islands.


13:15 2014-6-23
time slot:


* success


* idleness


* collisions


13:22 2014-6-23
slotted Aloha:


if a node is backlogged, it sends a packet in 


the next time slot with probability p.


13:24 2014-6-23
maximining utilization


13:28 2014-6-23
slotted Aloha


13:29 2014-6-23
Taylor'series expansion


13:32 2014-6-23
backlogged nodes


13:36 2014-6-23
stabilization


13:38 2014-6-23
stabilizing Aloha:


finding a p that maximizes utilization as loading changes


13:38 2014-6-23
binary exponential Back-off


13:40 2014-6-23
exponentially decreasing p


13:40 2014-6-23
increasing p on success


13:41 2014-6-23
stabilized Aloha


13:44 2014-6-23
utilization is good, but fairness is awful


13:44 2014-6-23
one node "captures" the network for a period of time


13:48 2014-6-23
limiting the capture effect


--------------------------------------------------------------
MIT introduction to digital communication, 


video 13, more MAC protocols


13:50 2014-6-23
unslotted Aloha


carrier sense, contention windows  // CSMA/CD, CAN


code division glimpse


////////////////////////////////////////////////////////////
8:02 2014-6-24 Tuesday
contention protocols


* different nodes have different amount


* traffic is very bursty


8:03 2014-6-24
slotted Aloha


8:10 2014-6-24
maximize the utilization of the network


8:12 2014-6-24
capture effect


8:13 2014-6-24
we put an upper bound


8:14 2014-6-24
trying to transmit according to probability p,


and then adjust p up & down a little bit


8:16 2014-6-24
stabilized Aloha


8:21 2014-6-24
stabilization protocol


8:22 2014-6-24
unslotted Aloha


* packets take T time slots to transmit


* collisions are no longer "perfect"


8:23 2014-6-24
both packets are deemed corrupted


8:23 2014-6-24
it doesn't take a rocket science


8:25 2014-6-24
window of vulnerability


8:25 2014-6-24
unslotted Aloha


slotted Aloha


8:31 2014-6-24
self interference


8:31 2014-6-24
what's the maximum rate the channel can hold


-----------------------------------------------------------
19:21 2014-6-24
resume MIT introduction to digital communication,


video 13, 


19:21 2014-6-24
unslotted Aloha


19:21 2014-6-24
self interference


19:23 2014-6-24
* slotted Aloha


* unslotted Aloha


19:25 2014-6-24
utilization & fairness


19:32 2014-6-24
carrier sense


19:33 2014-6-24
CSMA/CD == Carrier Sense Multiple Access with Collision Detection


19:34 2014-6-24
Carrier Sense:


reduce collisions with on-going transmissions by


transmitting only if channel appears not to be busy


19:35 2014-6-24
on-going transmission


19:37 2014-6-24
window of vunerability


19:44 2014-6-24
carrier sense => 1 slot window of vulnerability


19:44 2014-6-24
simulation of carrier sense


19:46 2014-6-24
carrier sense => increased utilization


19:48 2014-6-24
collision avoiding


19:48 2014-6-24
detection time // idle detectioin


19:48 2014-6-24
collision avoidance


19:49 2014-6-24
carrier sense + collision avoidance


19:50 2014-6-24
CW == Contention Window


19:50 2014-6-24
Carrier Sense + Collision Avoidance == Contention Window


19:52 2014-6-24
MAC == Multiple Access Control


    == Media Access Control


19:55 2014-6-24
MAC protocols


19:55 2014-6-24
Goal of MAC protocols:


to maximize utilization & fairness


19:55 2014-6-24
TDMA == Time Division Multiple Access


* Round-robin sharing


* contention protocol   // slotted Aloha, unslotted Aloha


   distributed protocol


   parameter (p, CW) adjusted


19:58 2014-6-24
centralized controller


19:58 2014-6-24
transmission experience


19:59 2014-6-24
backlogged nodes


20:00 2014-6-24
orthogonality principle


20:01 2014-6-24
FDMA == Frequency Division Multiple Access


20:01 2014-6-24
orthogonal vectors


20:02 2014-6-24
CDMA == Code Division Multiple Access


20:02 2014-6-24
chip code


20:05 2014-6-24
channel will sum the transmitted vectors


20:05 2014-6-24
How can I get back the original 4 messages?


20:06 2014-6-24
orthognality principle


dot product


20:07 2014-6-24
CDMA receiver


20:07 2014-6-24
Asynchronous CDMA


synchronous CDMA


20:11 2014-6-24
PN == Pseudo-Noise


20:13 2014-6-24
FDMA == Frequency Division Multiple Access


-----------------------------------------------------------------------
20:44 2014-6-24
complex exponential


DTFS


spectral coeffient,


band-limited signal


20:45 2014-6-24
frequency division multiplexing


20:51 2014-6-24
LTI channel


20:52 2014-6-24
your energy is still in the same frequency


20:52 2014-6-24
complex exponential


20:53 2014-6-24
periodic sequence


20:53 2014-6-24
fundamental frequency


20:54 2014-6-24
radians/sample


21:01 2014-6-24
negative frequency


21:02 2014-6-24
complex exponential


21:08 2014-6-24
sine & cosine both can be represented as 


sum of complex exponential using Euler's formula


21:12 2014-6-24
DFS == Discrete-time Fourier Series


21:25 2014-6-24
spectral coefficient


21:25 2014-6-24
orthogonal basis


21:26 2014-6-24
wavelet, etc...


21:26 2014-6-24
basis expansion, projection


basis * coefficient == component


21:26 2014-6-24
synthesis equation


analysis equation


21:36 2014-6-24
time domain <=> frequency domain


21:37 2014-6-24
time domain sequence => frequency domain coefficient


the coefficient Ak are also complex numbers!


21:43 2014-6-24
spectrum of digital communication


21:51 2014-6-24
band-limited signal


21:52 2014-6-24
effect of band-limiting a transmission


21:53 2014-6-24
the eye is starting to close


------------------------------------------------


22:03 2014-6-24 
start MIT introduction to digital communication


video 15, Frequency Response


22:04 2014-6-24
fundamental frequency, harmonics


22:09 2014-6-24
complex conjugate


22:12 2014-6-24
complex exponential is the eigenfunction of LTI system


frequency response is the eigenvalue


22:14 2014-6-24
sine & cosine are just sums of complex exponential


22:15 2014-6-24


LTI => convolution sum


22:15 2014-6-24
from convolution sum: x[n] * h[n], we have


frequency response


22:16 2014-6-24
characterize LTI system:


h[n] // impulse, in time domain


H    // frequency response, in frequency domain


22:22 2014-6-24
the frequency response tells us how the system will


affect each of the spectral coefficient that determine 


the input...


22:25 2014-6-24
unit sample, unit sample response


22:26 2014-6-24
unit sample response: h[n]


frequency response: H


22:28 2014-6-24
MA == Moving Average


22:33 2014-6-24
moving average is a way of throwing away high frequencies


22:35 2014-6-24
DTFT == Discrete-Time Fourier Transform


22:43 2014-6-24
building something let a particular frequency to go away


22:46 2014-6-24
series interconnection of LTI system


h1[n] * h2[n] // time domain           impulse response
 
H1[n] . H2[n] // frequency domain      frequency response


22:50 2014-6-24
convolution in the time domain <=> multiplication in the frequency domain


// convolution theorem


22:52 2014-6-24
LPF == Low Pass Filter


22:53 2014-6-24
unit sample <=> frequency response


// DFT, IDFT


23:08 2014-6-24
zero padding


23:11 2014-6-24
non-causal h[n] // non-causal impulse response


23:11 2014-6-24
if you are doing real-time signal processing, then


causality of h[n] is important, so


adding an N/2 sample delay


23:22 2014-6-24
BPF == Band Pass Filter


23:26 2014-6-24
FFT == Fast Fourier Transform


23:28 2014-6-24
spectral coefficient


23:28 2014-6-24
frequency response of channels


23:31 2014-6-24
fast channel, slow channel, ringing channel


////////////////////////////////////////////////////////


18:55 2014-6-25 Wednesday


start MIT introduction to digital communication,


video 16, modulation


* sharing the frequency


* modulation


* demodulation


18:56 2014-6-25
k // the spectral coefficient index


18:57 2014-6-25
band-limited signal


19:06 2014-6-25
baseband


19:06 2014-6-25
spectral coefficient


19:18 2014-6-25
modulation


19:18 2014-6-25
digital transmission waveform


19:32 2014-6-25
burst of carrier frequency


19:34 2014-6-25
demodulation + LPF


19:55 2014-6-25
this is the energy I want


19:57 2014-6-25
demodulation carrier


20:03 2014-6-25
multiple transmitters


20:09 2014-6-25
start MIT introductiont to digital communication,


video 17, more modulation


20:18 2014-6-25
* mismatch of receiver's frequency & pahse


* quadrature demodulation


* BPSK, QPSK, DQPSK, QAM


20:18 2014-6-25
standard digital transmission sequence


20:27 2014-6-25
frequency error in demodulator


20:42 2014-6-25
pilot tone


20:45 2014-6-25
phase error in demodulator


// frequency is the same, phase differs


20:46 2014-6-25
channel delay


21:02 2014-6-25
the reason you cannot hear me is that the phase


error has caused a scaling


21:08 2014-6-25
scaling factor:


* phase error


* channel delay


21:08 2014-6-25
fixing phase problems in the receiver:


quadrature demodulation


21:10 2014-6-25
Inphase, Quadrature


21:11 2014-6-25
quadrature modulation


21:23 2014-6-25
3 things:


* magnitude


* frequency 


* phase


21:30 2014-6-25
AM == Amplitude Modulation


FM == Frequency Modulation


PM == Phase Modulation


21:31 2014-6-25
Phase Modulation == PSK // Phase Shift Keying


21:34 2014-6-25
BPSK == Binary Phase-Shift Keying


21:35 2014-6-25
the message bit selects one of the two pahses for 


the carrier: pi/2, -pi/2


21:36 2014-6-25
BPSK: dealing with phase ambiguity


think the phase encoding as differential


21:43 2014-6-25
change in phase // encode my information


21:44 2014-6-25
differential encoding


21:44 2014-6-25
differential phase encoding


21:45 2014-6-25
* encode bits


* encode transitions


21:46 2014-6-25
DBPSK == Differential Binary Phase-Shift Keying


DQPSK == Differential Quadrature Phase-Shift Keying


21:47 2014-6-25
Differential PSK


21:55 2014-6-25
the idea of putting more points in the constellation


is a good idea!


21:55 2014-6-25
QAM == Quadrature Amplitude Modulation


21:59 2014-6-25
// QAM


using more message bits => generate larger constellations


21:59 2014-6-25
constellation points


21:59 2014-6-25
QPSK // QAM-4


22:01 2014-6-25
using QAM-16, 4 message bits are encoded into


one of 16 constellations points.


22:04 2014-6-25
QAM receiver


////////////////////////////////////////////////////
7:44 2014-6-26
start MIT introduction to digital communication,


video 18, switching


7:44 2014-6-26
point-to-point communication


7:44 2014-6-26
multi-hop communication


7:44 2014-6-26
switch


7:48 2014-6-26
packet switching <=> circuit switching


7:48 2014-6-26
MTBF == Mean Time Between Failure


7:51 2014-6-26
network reliability


7:52 2014-6-26
Redundancy:


* no single point of failure


* fail-safe, fail-safe, fail-hard


* automated adaption to component failure


8:00 2014-6-26
network scalability


8:03 2014-6-26
IPv4, IPv6


8:06 2014-6-26
incremental build-out


8:10 2014-6-26
NRE(non-recurring expenses, one-time-costs)


8:13 2014-6-26
dealing with system complexity:


manage complexity with abstraction layer


8:14 2014-6-26
easier to design if details have been abstract away


8:15 2014-6-26
layed abstraction


8:15 2014-6-26
API == Application Programming Interface


8:24 2014-6-26
abstraction layer


8:24 2014-6-26
network topologies


* point-to-point channels // simplex, half-duplex, full-duplex


* multiple-access channels


* LAN & WAN


8:48 2014-6-26
full-connected graph


----------------------------------------------
13:43 2014-6-26
fully connected network


13:43 2014-6-26
the disadvantage of fully-connected network is


it's very expensive


13:46 2014-6-26
the othe end of the world is "start connectivity"


13:47 2014-6-26
network architecture


13:57 2014-6-26
centralized network => decentralized network => distributed network


13:57 2014-6-26
modern networks: LANs + WANs


14:01 2014-6-26
sharing the internetwork links:


* circuit switching // isochronous


* packet switching  // asynchronous


14:04 2014-6-26
circuit switching:


* first establish a circuit between end points


* tear down(close) circuit


14:08 2014-6-26
multiplexing/demultiplexing


14:09 2014-6-26
packet switching:


* used in the internet


* data is sent in packets 


// header contains control info(source, dest addr)


* per-packet routing


14:17 2014-6-26
router, routing table


14:19 2014-6-26
packets are organized using a queue


14:19 2014-6-26
best efforts delivery


14:21 2014-6-26
acknowledgement/retransmission protocol


14:22 2014-6-26
end-to-end argument


14:24 2014-6-26
queues are essential


14:24 2014-6-26
the longer the queue is, the longer the delay


pro: absorb bursts


con: add delay


14:26 2014-6-26
Little's law


14:27 2014-6-26
FIFO delivering mechanism


14:29 2014-6-26
average rate of delivery


14:29 2014-6-26
mean delay for packets


-----------------------------------------------
17:49 2014-6-26
packet switching network


17:49 2014-6-26
Little's law


17:50 2014-6-26
packet find their way through the networks


17:50 2014-6-26
finding the shortest path


17:56 2014-6-26
routing: who's linked to whom?


17:59 2014-6-26
routing: what path the packet should follow?


17:59 2014-6-26
What am I supposed to do with this packet?


18:01 2014-6-26
IP filter


18:01 2014-6-26
TCP/IP   // TCP is reliable network protocol


18:01 2014-6-26
VoIP use data diagrams


18:02 2014-6-26
header checksum


18:03 2014-6-26
information in the header:


destination address,


CRC,


TTL // Time-To-Live


18:06 2014-6-26
routing table


18:06 2014-6-26
a quick table lookup using the destination 


as a key!


18:08 2014-6-26
routing table is the key to everything, everything


else is relatively straightforward


18:12 2014-6-26
every individual switch has its own routing table


18:20 2014-6-26
just think of this as a giant dictionary in python


18:20 2014-6-26
shortest path routing


18:21 2014-6-26
mean cost routing


18:21 2014-6-26
distributed approach:


each switch build its own routing table based on the


information it receives, then making its own routing decisions


18:23 2014-6-26
distributing the overhead over the whole network


18:24 2014-6-26
no centralized approach


18:24 2014-6-26
link-state protocol


18:25 2014-6-26
distance-vector routing


18:25 2014-6-26
there is cheaper path using the other links


18:26 2014-6-26
B only makes only a local decision


18:28 2014-6-26
DV protocol // Distance-Vector protocol


18:28 2014-6-26
"Hello" packet


18:34 2014-6-26
If I don't hear from John for a while... // no "hello" packet,


so I delete John from my neighbour list,


who my live neighbour is?


18:34 2014-6-26
distance-vector routing


18:36 2014-6-26
* distance-vector routing


* link-state routing


18:42 2014-6-26
so I considering using you as my way to 


get to MIT


18:43 2014-6-26
Bellman-Ford algorithm


18:45 2014-6-26
routing table


18:53 2014-6-26
I can get to myself with zero cost


18:54 2014-6-26
DV advertisement


18:54 2014-6-26
take multiple generation of these advertisement


18:57 2014-6-26
B & C will also send advertisement about A


19:02 2014-6-26
happily, they send advertisement to D & E


19:02 2014-6-26
sending advertisement


19:03 2014-6-26
after 2nd generation of advertisement


19:06 2014-6-26
I find that I can get to A more cheaply,


so I "update routing table"


19:07 2014-6-26
update routing table


19:07 2014-6-26
this is the steady-state


19:07 2014-6-26
Distance-Vector protocol


19:11 2014-6-26
DV routing  // Distance-Vector Routing


19:11 2014-6-26
gateway node // don't give you what they know


19:13 2014-6-26
gateway to the outside world


19:13 2014-6-26
at the next generation of advertisement


19:21 2014-6-26
update routing table


19:22 2014-6-26
partition the network


-----------------------------------------------
19:33 2014-6-26
Routing:


* distance-vector routing


* link-state routing


19:33 2014-6-26
DV == Distance Vector


19:34 2014-6-26
we want to equip each node with a "routing table"


19:34 2014-6-26
shortest path


19:46 2014-6-26
update routing table


19:46 2014-6-26
distance-vector routing


19:47 2014-6-26
What is a DV(Distance-Vector) routing?


* advertisement


at each ADVERT interval, nodes tell neighbors (dest, cost) for


all routers in their routing tabel


* update


nodes add link cost to neighbour's routing costs and keep


their routing table up-to-date with shortest-path route


19:52 2014-6-26
link is down


19:55 2014-6-26
down link


19:55 2014-6-26
An unfortunate combination of down links


might partition the network


19:58 2014-6-26
count for infinity


20:01 2014-6-26
infinite cost, which means we don't have a route


20:04 2014-6-26
packet A just bounce back & forth the network


20:05 2014-6-26
How does we solve routing loop problem?


we have the TTL(Time-To-Live) field in the packet header


20:06 2014-6-26
routing loop


20:09 2014-6-26
DV == Distance Vector


PV == Path Vector


20:10 2014-6-26
PV(Path-Vector) Routing


20:10 2014-6-26
advertisement both (path, cost)


20:10 2014-6-26
distance vector routing


path vector routing


20:11 2014-6-26
send advertisement, building routing tables


20:11 2014-6-26
pros & cons of PV routing:


pros: simple, works well for small network


cons: ony works for small network


20:12 2014-6-26
unreachable nodes are quickly removed from tables


20:14 2014-6-26
Path-Vector routing only works for small networks


20:15 2014-6-26
hierarchical structured local routing


20:17 2014-6-26
Link-State Routing


20:17 2014-6-26
LSA == Link-State Advertisement


20:18 2014-6-26
which neighbour is currently live?


20:18 2014-6-26
continually update on people who I can get to?


20:21 2014-6-26
Link-State Routing


* Advertisement step


* Integration


20:22 2014-6-26
LSA flooding


20:28 2014-6-26
LSA == Link-State Advertisement


20:30 2014-6-26
* LSA travels each link in each direction


* termination: each node rebroadcasts LSA exactly once


* all reachable nodes eventually hear every LSA


20:32 2014-6-26
rebroadcast LSA


20:32 2014-6-26
LSA flooding


20:32 2014-6-26
link-state table


20:33 2014-6-26
LSA table // Link-State Advertisement table


20:33 2014-6-26
how straightforward routing is!


20:34 2014-6-26
What eventually will be in your routing table?


20:35 2014-6-26
gateway node


20:36 2014-6-26
Dijkstra's Shortest Path Algorithm


20:37 2014-6-26
nodeset


spcost == shortest path cost


20:42 2014-6-26
link-statement announcement


20:44 2014-6-26
heap queue, priority queue


20:48 2014-6-26
Why is network routing hard?


20:53 2014-6-26
inside domain:  interior router


between domain: boarder router


20:57 2014-6-26
Hierarchical Routing


20:58 2014-6-26
DV protocol: Distance-Vector protocol


LS protocol: Link-State protocol


-------------------------------------------------
21:54 2014-6-26
review link-state routing


21:54 2014-6-26
DV routing // Distance-Vector Routing


21:55 2014-6-26
DV advertisement: (dest, cost)


each node tells their neighbour (dest, cost)


at the ADVERT interval


21:59 2014-6-26
update routing table


22:00 2014-6-26
routing advertisement


22:01 2014-6-26
2nd generation of routing advertisement


22:01 2014-6-26
break a link(link is down)


22:08 2014-6-26
down link


22:09 2014-6-26
miss a "hello" packet


22:09 2014-6-26
loss packet


22:09 2014-6-26
chop the network into pieces! // more down link


22:10 2014-6-26
partition the network


22:11 2014-6-26
Bellman-Ford algorithm


22:12 2014-6-26
DV == Distance Vector


PV == Path Vector


22:15 2014-6-26
packet A just bound around the network,


this is just a "routing loop"


22:17 2014-6-26
routing loop


22:17 2014-6-26
we have the "Time-To-Live" field in


the packet header


22:17 2014-6-26
TTL == Time-To-Live


22:17 2014-6-26
imperfect information about the available routes


22:18 2014-6-26
bad engineers build things that work in the perfect world,


22:20 2014-6-26
defend against a lot of possibilities


22:21 2014-6-26
What is a PV routing?


Path-Vector Routing is an improvement of Distance-Vector Routing,


instead of advertise (distance, cost), advertise (path, cost)!


so it can detect "routing loop"


22:22 2014-6-26
sending a lot of advertisement &  building our routing tables


22:24 2014-6-26
by using Path-Vector Routing, unreachable nodes


are quickly removed from tables


22:26 2014-6-26
it really only works for small networks


22:26 2014-6-26
Hierarchical structured local routes


22:29 2014-6-26
LS Routing // Link-State Routing


22:29 2014-6-26
Bellman-Ford algorithm only gets information from neighbours


22:29 2014-6-26
Can we devise an algorithm in which


all nodes get topological information about the whole network?


then using "Shortest Path" algorithm locally


22:30 2014-6-26
LSA == Link-State Advertisement


22:31 2014-6-26
LSA: 


* send information about its "links" to its neighbours


// instead of its own (distance, cost)


* do ti periodically


22:33 2014-6-26
integration:


if #seq in incoming LSA > seq# in saved LSA,


I knew it's a new anouncement!


I then update LSA,


rebroadcast to my neighbours  // flooding


22:37 2014-6-26
I rebroadcast to my neighbours


22:38 2014-6-26
source code


22:40 2014-6-26
eventually each node discovers current map of the network


22:40 2014-6-26
building routing table based on this LSA(Link-State Advertisement)


22:41 2014-6-26
LSA flooding 


22:42 2014-6-26
LSA == Link-State Anouncement


22:42 2014-6-26
each node eventually hears every LSA


22:45 2014-6-26
verification is aided by simplicity


22:45 2014-6-26
routing strategy


22:45 2014-6-26
LS Routing


22:46 2014-6-26
Bellman-Ford based on only gets information about your neighbours


22:46 2014-6-26
LS routing: all the nodes know about the topology of the network,


then they run their shortest path algorithm locally 


=> Link-State Routing


22:46 2014-6-26
LS Routing only talks about


advertisement I sent out instead of just talking about my


routing table, only talk about the links:


* which neighbours I have?


* How much cost to reach each neighbour?


22:49 2014-6-26
the only information about the advertisement is which


links is current live from that node?


22:50 2014-6-26
everytime I just increment the sequence number, so every node


knows this is a new advertisement


22:51 2014-6-26
continually updating people who I can get to


22:52 2014-6-26
If the incoming anouncement from George is a newer version


of advertisement(the sequence number is bigger than the sequence


number I have), I knew it is a new Link-State Anouncement(LSA)


22:54 2014-6-26
flooding


22:54 2014-6-26
I rebroadcast to my neighbours


22:54 2014-6-26
send LSA


22:55 2014-6-26
LSA flooding the network


22:55 2014-6-26
eventually the LSA from everybody will reach everybody else


22:55 2014-6-26
we keep the same source node


22:56 2014-6-26
If for a long time, we don't hear LSA from a node(seq# is too far out-of-date),


then we remove saved LSAs!


22:57 2014-6-26
result: each node discovers current map of the network


22:58 2014-6-26
building routing table:


* periodically each node runs the same "shortest path algorithm" over its map


* if each node implements computation correctly & each node has the same map,
 
  then routing table will be correct


22:59 2014-6-26
LSA flooding // Link-State Anoucement Flooding


23:00 2014-6-26
each node rebroadcast LSA exactly once


23:00 2014-6-26
LSA table


23:02 2014-6-26
neighbours & costs


23:02 2014-6-26
How straightforward routing is !


23:03 2014-6-26
gateway node in my network


23:04 2014-6-26
Dijkstra's Shortest Path Algorithm


23:16 2014-6-26
go to bed


//////////////////////////////////////////////////
6:53 2014-6-27 Friday


video 21, reliable data transportation


6:53 2014-6-27
* redundancy via careful retransmission


* sequence numbers & acks


* RTT estimation & timeouts


* stop-and-wait protocol


6:55 2014-6-27
best-effort network


7:07 2014-6-27
communicate reliably


7:15 2014-6-27
the problem:


* packets may be lost arbitrarily


* packets may be reordered arbitrarily


* packets delays are variable


* packets may even be duplicated


7:17 2014-6-27
Send S & Reciever R want to communicate reliably


7:17 2014-6-27
reliable transport protocol


7:18 2014-6-27
application "layed above" transport protocol


7:18 2014-6-27
transmitter: 


each packet includes a sequentially increasing sequence number


7:27 2014-6-27
(xmit time, packet)


7:29 2014-6-27
un-ACKed list


7:30 2014-6-27
ACK == acknowledgement


7:30 2014-6-27
unacknowledge list


7:35 2014-6-27
stop & wait protocol


7:36 2014-6-27
timeout:


some time constant which I can tell whether I 


should retransmit


7:38 2014-6-27
stop & wait protocol


7:39 2014-6-27
RTT == Round-Trip Time


7:39 2014-6-27
data loss + retransmission


7:41 2014-6-27
lose acknowledgement


7:42 2014-6-27
Receiver deliver packet payload to application 


in sequence number order


7:44 2014-6-27
retransmission


7:48 2014-6-27
packet buffer @transmitter @receiver


7:49 2014-6-27
RTT == Round-Trip Time


7:50 2014-6-27
choose appropriate timeout value


7:51 2014-6-27
change my estimate of what the RTT(Round-Trip Time) is?


7:53 2014-6-27
RTT measurement


7:55 2014-6-27
CDF of RTT


8:01 2014-6-27
CDF == Cumulative Distribution Function


determine timeout value from the "CDF of RTT" diagram


8:05 2014-6-27
RTT can be highly variable


8:10 2014-6-27
estimating RTT from data


8:11 2014-6-27
Chebyshev's Inequality


8:15 2014-6-27
standard deviation


8:16 2014-6-27
we need some more incremental method


8:25 2014-6-27
EWMA == Exponential Weighted Moving Average


8:25 2014-6-27
the RTT is changed dramatically each time I sample it,


so filter it with EWMA!


8:28 2014-6-27
srtt == smoothed RTT


8:36 2014-6-27
EWMA for smoothed RTT // srtt


8:36 2014-6-27
use another EWMA for smoothed RTT deviation(srttdev)


8:39 2014-6-27
timeout = srtt + k * srttdev


TCP use k = 4


8:40 2014-6-27
1 packet every T seconds


throughput = 1 / T


8:41 2014-6-27
we cannot just assume T = RTT since packets get lost


8:49 2014-6-27
RTT == Round-Trip Time


8:54 2014-6-27
sliding window protocol 


provided by TCP


------------------------------------------------
9:14 2014-6-27
start MIT introduction to digital communication,


video 22, sliding window protocol


9:14 2014-6-27
stop & wait protocol too slow


9:16 2014-6-27
small identifier: sequence number


9:17 2014-6-27
1 packet per RTT


9:19 2014-6-27
with packet loss & timeout, throughput even lower


9:22 2014-6-27
in order to improve performance:


solution: use a window


* allow W packets outstanding in the network at once // W == Window Size


* overlap transmissions with ACKs


9:23 2014-6-27
sliding window


9:25 2014-6-27
pipelining


9:25 2014-6-27
the window slides, packet 2 ~ 6 now standing out


9:28 2014-6-27
sliding widnow implementation


9:28 2014-6-27
sliding window <=> stop & wait


9:31 2014-6-27
un-ACKed list


9:33 2014-6-27
undelivered packet queue


9:38 2014-6-27
duplicate packet


9:38 2014-6-27
acknowledgement


9:38 2014-6-27
sliding window protocol


9:42 2014-6-27
list of undelivered packet


9:42 2014-6-27
packet buffer


9:46 2014-6-27
Sliding Window Size


9:58 2014-6-27
Setting the Window Size:


Applying Little's Law


10:00 2014-6-27
W == #packets in window


B == rate of slowest(bottleneck) link


RTT == avg delay


10:04 2014-6-27
if W = B * RTT, path will be fully utilized


10:04 2014-6-27
bandwidht-delay product  // key concept in transport protocols


10:04 2014-6-27
bottleneck link


10:08 2014-6-27
propogation delay


10:22 2014-6-27
Round Trip propogation delay


10:23 2014-6-27
sliding window transport protocol


---------------------------------------------------------------------
10:57 2014-6-27
review:


stop & wait protocol too slow =>


sliding window protocol


10:59 2014-6-27
RTT == Round-Trip Time


11:01 2014-6-27
solution: use a window


* allow W packets outstanding in the network at once // W == Window Size


* overlap transmissions with ACKs


11:03 2014-6-27
sliding window


11:06 2014-6-27
sliding window in action


11:07 2014-6-27
fixed-size sliding window


11:09 2014-6-27
unacknowldged packets


11:18 2014-6-27
duplicate receptiion


------------------------------------------------
12:39 2014-6-27
stop watching OCW videos!!!