Information theory Flashcards

1
Q

What is the relationship between a probability of an event and the amount of information it contains?

A

A certain event contains zero information.
An event with zero probability contains infinite information
IA = -log PA
IA is the information in event A, PA is the probability

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

How many bits do you need for 4, 8 equiprobable messages?

A

4 - requires 4 distinct binary pulse patterns (2 bits)
8 - requires 8 distinct binary pulse patterns (3 bits)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

How can you calculate the information contained by message of different symbols with different independent probabilities

A

Ii = log_2(1/Pi) bits
Say you had message X = ABC, sum log_2(1/Pi) of A B and C

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What is the entropy?

A

Entropy H is the average information. H = sum(i=1 to n) of Pi*log_2(1/Pi)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What is the rate of information transmission?

A

R = rH bits per second
r is the symbol rate
H is the entropy (average information per symbol)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What is Shannon’s theorem?

A

Says that R<=C, ie. rate of information transmission is less than channel capacity.
C is the max rate of reliable transmission through the channel without error.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What is the Hartley-Shannon theorem?

A

C = Blog_2(1+S/N) bits per second
C = channel capacity
B = channel bandwidth
S/N = mean square signal to noise ratio. Needs to be not in dB. dB = 10 log_10(x)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

What is the maximum error rate for a binary system?

A

50%

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

How does an extended source work?

A

If you consider a message of individual symbols to be divided into blocks of n symbols, you can consider this a source of K^n alphabet symbols where K is the number of symbols in the original alphabet. Then Hnew = nH

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

How can data be source encoded?

A

Source encoding means representing information efficiently. This can be done by representing frequent symbols with short codewords and rare symbols with longer codewords

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What are the two requirements of source encoders?

A

The codewords are in binary form
The source code is uniquely decodable.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

How can we calculate the coding efficiency of a source encoder?

A

ƞ = Lmin/Lavg
L is codeword length
Lavg is calculated by the sum of (probabilities times the number of bits in each codeword)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

What is a prefix code?

A

A code in which no codeword is the prefix of any other codeword. It is always uniquely decodable and has a decision tree as you receive more and more bits. The end of the codeword is always recognisable. H <= Lavg < H+1

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

What is Huffman coding?

A

It is a procedure for creating prefix codes that assigns longer codewords to symbols with more information. It however requires statistical knowledge of the source.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

What is Lempel-Zev coding?

A

Takes care of the fact that you might not know the statistics of the data you are encoding.
Uses fixed length codes to represent a variable number of source symbols.
Only has true advantage with long data strings.

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What is channel coding?

A

Aims to increase the resistance of the communication to channel noise by introducing redundancy.
Understood by block codes - message is divided into blocks k bits long mapped onto an n bit block where n>k

16
Q

What is repetition coding?

A

Each bit in a message is repeated several (n) times where n is an odd integer
Improved reliability is at cost of decreased code rate

17
Q

What is error control coding?

A

If a channel has capacity C, there exists a coding technique to transmit over the channel with arbitrarily probability of error.
This is done through the addition of redundancy which adds complexity and increases bandwidth.

18
Q

What is a linear block code?

A

A code is linear if any two codewords can be added in mod-2 to produce a third codeword in the code.
The block consists of a message vector m and a parity vector b and a code vector c.
c = [b|m] or [m|b] (could go either way around)

19
Q

What are repetition codes?

A

The simplest type of linear block codes where all the parity bits are repetitions of the single message bit per block.

20
Q

In Hamming coding, how long is each part of the block? How many distinct message words are there?

A

A family of linear block codes where:
block length n = 2^m -1
message bits k = 2^m - m - 1
parity bits n-k=m
m>= 3
There are 2^k distinct message words
They are single error correcting binary perfect codes.

21
Q

What is forward error correction?

A

A form of error control that provides coding gain but adds complexity and requires greater bandwidth for the same message rate. Two types include block coding and convolutional coding.

22
Q

Draw the binary symmetric channel

A
23
Q

How can you calculate the code rate?

A

Code rate = rate of message data / rate of codeword = m/c

24
Q

Given the parity check matrix or generator matrix, how can you find the other one?

A

Parity check H = [ I : P’ ]
Generator G = [ P : I ]
Look for I in one of them to identify P. Then make the other.

25
Q

How does Hamming coding work? Sketch G. What is the relationship between G and c?

A

1

26
Q

How can you know if a linear block code received is a valid codeword?

A

Multiply the received codeword by the transpose of the parity check matrix. Equals 0 if it is a valid codeword. Equals something else means there is an error.
cH’=0

27
Q

How can you work out the block error probability for a repetition code?

A

Say a block has 5 bits. Probability of getting the block wrong = probability of getting 3 bits wrong + probability of getting 4 bits wrong. + probability of getting 5 bits wrong.
Eg. probability of 3 bits wrong = (Perror^3)(Pnoerror^2)(5 choose 3)

28
Q

What is the relationship between H and G in linear block coding?

A

HG’ = 0
GH’ = 0

29
Q

What are syndrome errors and how can you write down all the syndrome errors for a Hamming code?

A

A received signal r = c + e where e is an error vector.
The syndrome is the result of multiplying the parity check transpose with the received message. By looking at the list of syndromes that correspond to codewords, we can work out which bit had a likely error.
s = rH’, s is the syndrome vector
s = [0 0 0] if no error.

See 2

30
Q

What is SNR and SIR?

A

Signal to Interference Ratio - interference from other reflected signals travelling different paths
Signal to Noise Ratio - background noise (thermal, ratio)

31
Q

How can you calculate the signal to noise ratio, linear or dB?

A

SNR (dB) = 10 log_10 (mean s^2/ mean n^2)
S/N (linear) = mean s^2 / mean n^2

note we are squaring the voltages

32
Q

How can you calculate the rms voltage arising from thermal noise at a given temperature across a resistor R?

A

mean v^2 = 4kTRB
T = temp in K
R = resistance
B = bandwidth
k = Boltzmann’s constant

Don’t forget to square root

33
Q

How can you calculate the power of noise across a certain bandwidth?

A

Integrate the power spectral density across the bandwidth. (multiply B by n^2 for AWGN)

34
Q

What is noise figure?

A

The ratio of input S/N ratio to output S/N ratio, based on the idea that a device like an amplifier will add noise to the amplified signal.

F = (S/N)o/ (S/N)i

34
Q

What is noise temperature?

A

The effective temperature of a white thermal noise source at the input of a device that represents all the noise in the device.

35
Q

What does the cascaded noise figure equation show?

A

The overall noise figure of a cascade of amplifiers can be calculated using the equation, with the noise figures and gains of the first, second amplifiers etc. being F1 G1, F2 G2….
Says you should put your best amplifier first.

36
Q

What are the key features of convolution coding?

A

Code rate = 1/2 (one bit in, two bits out)
Constraint is 3 (how many steps each bit is in the system for)