Spreading Codes and Characteristics. Error Correction Codes

Transcription

1 Spreading Codes and Characteristics and Error Correction Codes Global Navigational Satellite Systems (GNSS-6) Short course, NERTU Prasad Krishnan International Institute of Information Technology, Hyderabad December, 206

2 Spreading Codes

3 Talk Outline - Spreading Codes Spreading codes idea Need for spreading codes in GNSS Generating spreading codes for GNSS

4 What are Spreading Codes? Example Consider that you have a message bit b (can be zero or one). Instead of transmitting b we transmit - b[ ]. The vector [ ] is like a carrier - we call it the code. To decode, multiply received vector by [ ] T.

5 What are Spreading Codes? Example 4 different bits with 4 orthogonal codes transmitted at the same time. b 0 [ 0 0 0]+ b [0 0 0] + b 2 [0 0 0] + b 3 [0 0 0 ] =b 0 x0 + b x + b 2 x2 + b 3 x3

6 What are Spreading Codes? Example 4 different bits with 4 orthogonal codes transmitted at the same time. b 0 [ 0 0 0]+ b [0 0 0] + b 2 [0 0 0] + b 3 [0 0 0 ] =b 0 x0 + b x + b 2 x2 + b 3 x3 Note that xi xj T = { if i = j 0 if i j Can get the bit b i by multiplying with xi T. (i represents the shift in time).

7 What are Spreading Codes? - Finding Delays Example Consider that a coded bit b comes with an arbitrary unknown delay (0 j 3). Received vector = bxj (0 j 3). Then can we find out the delay j and bit b? Multiplying with xi (0 i 3) does the trick.

8 More generally - Pseudo-Random-Noise Sequences Consider two or more binary sequences ({+, }) sequence which have good autocorrelation properties and good cross correlation properties. Such sequences are called Pseudo-Random-Noise sequences.

9 More generally - Pseudo-Random-Noise Sequences Consider two or more binary sequences ({+, }) sequence which have good autocorrelation properties and good cross correlation properties. Such sequences are called Pseudo-Random-Noise sequences. Good autocorrelation property of a sequence x x i x T j has a high value if i = j and low value if i j. Good cross-correlation property of sequences x and y x i y T j has a low value for any i, j.

10 How are PNR sequences useful in GNSS? Each satellite has its own unique PRN sequence, and uses it to modulate data transmitted to receivers. Ranging Good autocorrelation properties Find Delay due to separation between Rx and Satellite Tx. Delay (from multiple satellites) User Location.

11 How are PNR sequences useful in GNSS? Each satellite has its own unique PRN sequence, and uses it to modulate data transmitted to receivers. Ranging Good autocorrelation properties Find Delay due to separation between Rx and Satellite Tx. Delay (from multiple satellites) User Location. Saving spectrum Good cross-correlation properties Decode info from different satellites. Multiple satellites can transmit over the same frequency.

12 How are PNR sequences useful in GNSS? Each satellite has its own unique PRN sequence, and uses it to modulate data transmitted to receivers. Ranging Good autocorrelation properties Find Delay due to separation between Rx and Satellite Tx. Delay (from multiple satellites) User Location. Saving spectrum Good cross-correlation properties Decode info from different satellites. Multiple satellites can transmit over the same frequency. Gain in SNR Same bit is encoded in a long PRN sequence redundancy. Redundancy provides SNR gain (and hence lowers prob. of error).

13 Generating Pseudo-Random-Noise PN Sequences deterministically generated, yet possess properties of randomly generated sequences PN sequences generated using linear feedback shift registers.

14 Linear Feedback Shift Registers i - Output of the shift-register circuit is transformed to if it is 0, and if it is. Output sequence is given by 2 c i+m = g m c i+m + g m 2 c i+m g c i+ + c i (mod 2)

15 Characteristic Polynomial of LFSR Since the operation is binary addition, the above output equation can be rewritten as where g 0 = g m = m g l c i+l = 0, l=0 The characteristic polynomial of the LFSR is given by g(x) = m g l x l We are interested in special kinds of LFSRs which can output PRN sequences. l=0

16 Primitive Polynomial Every polynomial g(x) with coefficients in binary field having g(0) = divides x N + for some N. The smallest N for which this is true is called the period of g(x). An irreducible polynomial of degree m whose period is 2 m is called a primitive polynomial

17 MLS Generator m-sequences : Examples of PRN sequences DANISH GPS CENT Random sequences can be generated using a maximum-length sequence (MLS) generators Clock MLS Output n is the length of the shift register (0 for GPS) An LFSR produces an m-sequence (maximum length) if and Length of the MLS is : only if its characteristic polynomial is a primitive polynomial In the above example, N the polynomial 2 n MLS is g(x) = + x 3 + x Danish GPS Center 20

18 Delay and Add Property of m-sequences The cyclic shift of an m-sequence is also an m-sequence The sum of an m-sequence and a cyclic shift of itself is also an m-sequence

19 Autocorrelation Function of m-sequences Let (s t ) be an m-sequence of period N = 2 n Then the autocorrelation of the m-sequence is { 2 n if τ = 0(mod2 n ) θ s,s (τ) = if τ 0(mod2 n )

20 The normalized periodic autocorrelation function of an m- Plot of Autocorrelation Function N j i j sequence, defined asρ( i ) = ( ) is equal to for N j= 0 Normalized i = 0(mod autocorrelation N) and /N function for i 0(mod N) c c + - i i ρ ( i ) = (# of 0 s in c T c - # of s in c T c ) N - proved easily by shift and add property

21 Gold Sequences m-sequences have good auto-correlation properties but poor cross-correlation properties (cross-correlation can be high, which we don t want). Two m-sequence generators are used to generate a Gold Sequence which has good cross-correlation properties

22 Generating Gold Sequences C/A Code Generator DANISH GPS CENTER 0.23 MHz clock 0 G generator G code Reset circuit Phase selector S S2 + + C/A code (Gold code) G2 generator + Epoch detector 20 G2 code 50Hz clock for navigation data 2009 Danish GPS Center 22

23 ration: nomials Gold Sequences in GPS DANISH GPS CENTER x 8 x 9 x

24 Forward Error Correction

25 Talk Outline - Error Correcting Codes Need for Error Correction Error Correcting Codes (general principles and examples) Encoding and Decoding of a Block Code (Hamming Code) Types of Error Correcting Codes (in GNSS) Encoding and Decoding of Convolutional Codes

26 Channel Coding Sender 0 (- p) p p (- p) Receiver 0 p : cross-over probability, say 0. The bit cross-over probability p (< 0.5) is a property of the channel. Free to manipulate the input and output to the channel. Encode messages to codewords (add redundancy cleverly) : Reduce effective Prob(error).

27 A trivial code example - Repetition code Repeat the same bit three times Message 0 [0 0 0] (codeword), Message [ ]. Decode by Majority logic. For the above channel, probability of error comes down (Check!).

28 General ideas behind FEC for the binary symmetric channel Messages Channel Code Encoder Codeword Channel Noisy codewords Channel Code Decoder Message estimate Decoder decides the Tx codeword c based on received Noisy Codeword y. Decoding rule : Decoding for the most likely codeword. (Choose that c which maximizes p(y c)).

29 General ideas behind FEC for the binary symmetric channel Messages Channel Code Encoder Codeword Channel Noisy codewords Channel Code Decoder Message estimate Decoder decides the Tx codeword c based on received Noisy Codeword y. Decoding rule : Decoding for the most likely codeword. (Choose that c which maximizes p(y c)). Probability that every transmitted bit is flipped is p < 0.5 If you don t code at all, the rule decodes to the bit that was received as it is.

30 General ideas behind FEC for the binary symmetric channel For a code of length n: Choose the codeword which is closest to Rx Vector y in terms of number of flipped bits. Minimum Hamming Distance Rule.

31 General ideas behind FEC for the binary symmetric channel For a code of length n: Choose the codeword which is closest to Rx Vector y in terms of number of flipped bits. Minimum Hamming Distance Rule. Example Repetition Code The Majority Decoder is infact the Minimum Hamming Distance Decoder. The Repetition Code can correct any single-bit error.

32 General ideas behind FEC for the binary symmetric channel For a code of length n: Choose the codeword which is closest to Rx Vector y in terms of number of flipped bits. Minimum Hamming Distance Rule. Example Repetition Code The Majority Decoder is infact the Minimum Hamming Distance Decoder. The Repetition Code can correct any single-bit error. Increase n to decrease P(error). Rate n : For bit of message, we need to send n coded bits. (Pretty bad!)

33 Can we do better? - Linear Block Codes k message bits Code word of length n bits Each set of k message bits maps to a unique codeword Each of the n bits is a linear combina9on of k message bits n = length of the code, k = dimension of the code

34 Linear Block Codes Let u be the message vector and c is the corresponding codeword. How to get c from u? Linear Block Codes used a Linear Map c = ug G is a full-rank matrix of size k n (k n). The code is a (n, k) linear block code.

35 Linear Block Codes Let u be the message vector and c is the corresponding codeword. How to get c from u? Linear Block Codes used a Linear Map c = ug G is a full-rank matrix of size k n (k n). The code is a (n, k) linear block code. Repetition Code G = ( ). Message u {, 0}. Codewords are [ ] and [0 0 0].

36 Linear Block Codes - Examples Hamming Codes - a class of single error correcting codes. Reed Solomon Codes. Example The (7, 4) Hamming Code G =

37 Error Correcting Capability of a Block Code

38 Error Correcting Capability of a Block Code

39 Error Correcting Capability of a Block Code A block code can correct t errors if and only if Hamming balls of size t around codewords don t intersect. Minimum distance of the code must be at least 2t +. Linearity = Minimum weight of the code is at least 2t +.

40 Decoding - The Parity Check Matrix The Parity Check Matrix : A full-rank n k n matrix such that GH T = 0. For the Hamming Code : H =

41 Decoding Received vector y = c + e. Compute s = yh T = ch T + eh T = xgh T + eh T = eh T. Corresponding to any error vector of weight upto t there is an unique syndrome.

42 Syndrome Decoding Syndrome decoding for errors of weight upto t.. Find the syndrome s 2. Find e corresponding to s. 3. Find c = r e. Map it back to x.

43 Types of Codes used in GNSS n and Application Table 2. Channel coding comparison Coding NAV CNAV Galileo CNAV-2 Hamming Yes No No No Convolution No Yes Yes No CRC No Yes Yes Yes Interleaving No No Yes Yes LDPC No No No Yes BCH No No No Yes ame The LC signal provides the first navigation signal using modern advanced FEC-LDPC (Low Density Parity

44 Convolutional Encoding Convolutional codes are used in applications that require good performance with low encoding complexity. Convolution codes have memory that utilises previous bits to encode or decode following bits (block codes are memoryless)

45 Convolutional Encoding + y [n] x[n] D x[n- ] D x[n- 2] Encoder + y 2 [n] y [n] = x[n] x[n ] x[n 2] y 2 [n] = x[n] x[n 2] Rate 2 convolutional encoder Constraint length for each input is 2

46 State Diagram 0 / 0/ /0 0/ 0 /0 /00 0/0 /0 State diagram State transitions are given by input/output

47 Example Encoding 0 / 0/ /0 0/ 0 /0 /00 0/0 /0 State diagram Input: Output:

48 Brute Force Approach Going through the list of possible transmit sequences and comparing Hamming distance is highly complex A transmit sequence of N bits has 2 N possible strings, exponential complexity Low Complexity Decoder: Viterbi Decoder - decoding on trellis

49 Branch Metric State Time: i Received 00 i+ 00 0/00 0 / 2 0 0/0 /0 0 0/ 2 /00 0 0/0 /0 The branch metric for hard decision decoding. In this example, the receiver gets the parity bits 00 Two of the branch metrics are 0, corresponding to the only states and transitions where the corresponding Hamming distance is 0 Other non-zero branch metrics correspond to cases where there are bit errors

50 Computing Path Metric Value of PM[s, i] - total number of bit errors detected when comparing the received parity bits to the most likely transmitted message, considering all messages that could have been sent by the transmitter until time step i If the transmitter is at state s at time step i +, then it must have been in only one of two possible states at time step i, say α and β Path Metric update is given by PM[s, i+] = min(pm[α, i]+bm[α s], PM[β, i]+bm[β s])

51 Viterbi Decoding: Step Rcvd:

52 Viterbi Decoding: Step 2 Rcvd:

53 Viterbi Decoding: Step 3 Rcvd: Showing only survivor paths

54 Viterbi Decoding: Step 4 Rcvd:

55 Viterbi Decoding: Step 5 Rcvd: To produce the message, start from final state with smallest path metric and word backwards and then reverse the bits

56 Hard Decision Decoding Hard decision decoding digitizes the received voltagee signals by comparing it to a threshold, before passing it to the decoder Loss of Information and are both treated as by the decoder even it is more likely that is a Hamming distance as branch metric

57 Soft Decision Decoding Soft Decision Decoding does not digitise the incoming samples prior to decoding If the convolutional code produces p parity bits and p corresponding analog samples are v = v, v 2,..., v p, a soft decision branch metric is given by BM soft [u, v] = p (u i v i ) 2 i=

58 Thanks!