Introduction to Source Coding

Transcription

1 Comm. 52: Communication Theory Lecture 7 Introduction to Source Coding - Requirements of source codes - Huffman Code

2 Length Fixed Length Variable Length Source Code Properties Uniquely Decodable allow user to invert mapping to the original. Prefix No codeword can be a beginning of any other codeword. A prefix-free encoding is useful because it is self-defining: if we have any string of symbols, there cannot be two ways to decode it.

3 Examples: Not uniquely decodable Codes Code : symbols A and C are assigned the same binary sequence. Thus, the first requirement of a useful code is that each symbol be assigned a unique binary sequence. Code 2: is also not uniquely decodable because it is confusing in decoding. Source Code Code 2 Codeword Codeword A B C D

4 Prefix Coding A prefix code is defined as a code in which no codeword is the beginning of another codeword. A prefix code is uniquely decodable but the converse is not true. Code C is uniquely decodable since bit indicates the beginning of each code word in the code. Source Prefix Code Code A Code B Code C Codeword Codeword Codeword s s s 2 s 3 Not Uniquely Decodable Nor Prefix Code Uniquely Decodable Codes

5 Example: Prefix-Uniquely decodable Prob. Code Code 2 Code 3 Code 4 A P[A]=/2 B P[B]=/4 C P[C]=/8 D P[D]=/6 E P[E]=/6 Average Length 3/6 3/6 3/6 33/6 HW: Uniquely decodable : Code, 2, 3. No confusion in decoding Determine Instantaneous the prefix decodable: codes and Code the, uniquely 3. No need decodable to look ahead. codes from Prefix codes are uniquely the decodable above codes??? but the converse is not true. *** Code 4????

6 Example: Decoding of a Prefix Code Decision Tree for Code B s Source s k Code B Codeword c k Initial State s s 2 s s s 2 s 3 s 3 Example : Decode Answer : s s 3 s 2 s s A prefix-free encoding is useful because it is self-defining: if we have any string of symbols, there cannot be two ways to decode it.

7 Example Using Prefix decoding tree, decode the following code Decoding Tree symbol code a b a b c d c r d r abracadabra =

8 Kraft-McMillan Inequality K k 2 v k A prefix code must satisfy the Kraft McMillan s inequality. If a code satisfies this inequality, it doesn t mean that the code is a prefix. For code D = 9/8 This means that Code D IS NOT A PREFIX CODE Source s k Codeword C k Code D Codeword Length v k s s 2 s 2 3 s 3 2 Proof in the BOOK (Not required)

9 Use of Kraft-McMillan Inequality We may use it if the number of symbols are large such that we cannot simply by inspection judge whether a given code is a prefix code or not. WHAT Kraft-McMillan Inequality Can Do: It can determine that a given code IS NOT A PREFIX CODE WHAT Kraft-McMillan Inequality Cannot Do: It cannot guarantee that a given code is indeed a prefix code

10 Example Source s k Codeword Code E Codeword Length For code E = Is code E a PREFIX code? NO WHY? s 3 is a beginning of s 2 c k v k s s 3 s 2 3 s 3 2

11 For a Prefix Code H H Shannon s First Theorem S L S L H S if P k 2 v k k What is the Efficiency η? η= if P 2 vk for some k η< k

12 Huffman Coding Huffman code is a prefix, variable length code that can achieve the shortest average code length for a given input alphabet. Huffman code constructs a binary tree starting with the probabilities of each symbol in the alphabet. The tree is built in a bottom-up manner The tree is then used to find the codeword for each symbol. An algorithm for finding the Huffman code for a given alphabet with associated probabilities is given in the following slide.

13 Huffman Encoding Algorithm. Source symbols are listed in order of decreasing probability. The two source symbols of lowest probability are assigned a and. This part of the step is referred to as a splitting stage. 2. These two source symbols are regarded as being combined into a new source symbol with probability equal to the sum of the two original probabilities. (The list of source symbols and therefore source statistics is thereby reduced in size by one). The probability of the new symbol is placed in the list in accordance with its value. 3. The steps are repeated until we are left with a final list of source statistics of only two for which a and a are assigned.

14 Huffman encoding: Example Use Probabilities to order coding priorities of letters Low probability get codes first (more bits) This smoothes out the information per bit Letter X X2 X3 X4 X5 X6 X7 Probability

15 Huffman encoding Use a code tree to make the code Combine the symbols with lowest probability to make a new block symbol Assign a to one of the old symbols code word and to the other symbol Now reorder and combine the two lowest probability symbols of the new set Each time the symbol has lowest probability the code words get shorter x x2 x3 x4 x5 x6 x7 D D D 2 D 3 D 4

16 Result: Huffman encoding Entropy: H(X) = 2. (The best possible average number of bits) Average code word length: _ 7 L v k P( k x k v k = number of bits per symbol ) 2.2 Compression ratio: CR=3/2.2=.357 Note: The average code word length approaches the entropy (fundamental limit) The average code word length does satisfy: S L S H H So the efficiency = H ( X _ L ) 95.5%

17 Huffman Coding: Example 2 Compute the Huffman Code for the source shown H S 4. log log log L Source s k Probability p k s. s.2 s 2.4 s 3.2 s 4.

18 Solution A Source s k Stage I s 2.4 s.2 s 3.2 s. s 4.

19 Solution A Source s k Stage I Stage II s s.2.2 s s..2 s 4.

20 Solution A Source s k Stage I Stage II Stage III s s s s..2 s 4.

21 Solution A Source s k Stage I Stage II Stage III Stage IV s s s s..2 s 4.

22 Solution A Source s k Stage I Stage II Stage III Stage IV s s s s..2 s 4.

23 As high as Possiblee Source s k Solution A Stage I Stage II Stage III Stage IV Code s s s s..2 s 4.

24 Solution A Cont d Source s k Probability p k Code word c k s. s.2 s 2.4 s 3.2 s 4. H S L S L S H H THIS IS NOT THE ONLY SOLUTION! CR. 364

25 As low as Possible Source s k Alternate Solution B Stage I Stage II Stage III Stage IV Code s s s s..2 s 4.

26 Alternative Solution B Cont d Source s k Probability p k Code word c k s. s.2 s 2.4 s 3.2 s 4. H S L S L S H H CR

27 What is the difference between the two solutions? They have the same average code length They differ in the variance of the code length 2 K k P k ( v k L) 2 Solution A σ 2 =.6 Solution B σ 2 =.36 both have the same CR: CR

28 Compute Entropy (H) Exercise Build Huffman tree Compute average code length Compression ratio CR (S) P (S) A. B.2 C.4 D.2 E. Code BCCADE

29 Solution Compute Entropy (H) H = 2. bits Build Huffman tree Compute code length - L = 2.2 bits CR=3/2.2=.364 P(S) Code A. B.2 C.4 D.2 E. Code BCCADE =>

30 Properties of Huffman Codes The Huffman coding technique is optimal because: - It satisfies prefix condition (so decoding is unambiguous) - It has small average code length (approach minimum). In Huffman encoding, symbols that occur more frequently have shorter Huffman codes (but we must know the probabilities of each symbol for this to be true) Huffman encoding process (Huffman tree) is not unique. The code with the lowest code length variance is the better one.