Vous êtes sur la page 1sur 5

Coded M-PSK Modulation using Convolutional Self-Doubly Orthogonal Codes

Christian Cardinal and Bakitanga-Florian Mvutu


Department of Electrical Engineering, Ecole Polytechnique de Montr al e P.O. Box 6079 Station Centre-Ville, Montr al, QC, Canada H3C 3A7 e Tel.: 1-514-340-4711 ext. 5082, Fax: 1-514-340-4078 Email: christian.cardinal@polymtl.ca Finally, Section VII concludes the paper. II. DEFINITIONS AND NOTATIONS Convolutional Self-Doubly Orthogonal Codes (CSO2 C) are a new class of convolutional codes that can be decoded by using an iterative threshold decoding technique [7]. These codes have specic orthogonal properties that ensure independence of the observables over the rst two iterations. Two types of CSO2 C codes can be dened: wide-sense convolutional self-doubly orthogonal codes (namely, CSO2 C-WS) and strictsense convolutional self-doubly orthogonal codes (namely, CSO2 C-SS). The CSO2 C-WS codes have some of their orthogonal properties that are not completely satised, due to undesirable repetitions of the observables at the second decoding iteration. Consequently, the threshold iterative decoding technique will not converge to its best error performance [7]. On the other hand, CSO2 C-SS codes do not present any repetition at the second iteration. They have better error performance than those obtained with CSO2 C-WS codes, but with larger decoding complexity and latency. Since, only the wide-sense codes are considered in this paper, the abbreviation CSO2 C will then be used without any confusion. To make the paper self contained, we recall the denition of CSO2 C [7]. Let us note that the code-generator vector of a 1 2 rate CSO2 C systematic encoder is specied by a set {j , j = 1, 2, . . . , J : j1 < j } of the J connection positions. Denition: For a single shift-register encoder, a systematic convolutional code is said to be Self-Doubly Orthogonal in the Wide-Sense if the set of its connection positions satises the following conditions: 1) the differences (j k ), j = k are distinct; 2) the differences of differences (j k ) (m l ) are distinct for all, except for the unavoidable repetitions, with j = k, m = l, l = k, m = j; 3) the differences of differences are distinct from the differences. Figure 1 illustrates an example of a CSO2 C systematic encoder of rate R = 1 with J = 4 connection positions 2 {1 = 0, 2 = 4, 3 = 14, 4 = 15}. Its memory length (or span) is 4 = 15.

AbstractIn this paper, coded modulation technique based on the use of convolutional self-doubly orthogonal codes (CSO2 C) and threshold iterative decoding is presented. Two different methods of coded modulation are considered. The rst approach is based on a conventional pragmatic scheme. In the second approach, all information bits are coded and a simpler Gray coding is used for the bit-to-symbol mapping. Simulation results indicate that for a bit error rate of 105 , the pragmatic 8-PSK coded modulation scheme with R = 2 coding rate provides a coding 3 gain of 3 dB as compared with the error performances of the uncoded QPSK system. The simulation results also demonstrate that by using the second approach, a larger coding gain of 4.5 dB is obtained when compared with uncoded QPSK modulation.

I. INTRODUCTION Several works on CSO2 C codes have been conducted and published in the literature [5]- [8], where signicant contributions of these codes were demonstrated. Particularly, these works investigated puncturing technique for CSO2 C codes as an approach to circumvent the difculties of the search and determination of CSO2 C codes with coding rates R > 1 . 2 Those works motivated us to investigate coded modulation techniques using these new class of convolutional codes. In addition, by combining the CSO2 C codes with M-ary modulation, power and bandwidth efciencies are improved simultaneously as it was shown in [2], [3], [4] for the usual convolutional codes. In this paper, coded modulation techniques using convolutional self-doubly orthogonal codes is investigated. Two different methods are considered. The rst approach is based on a pragmatic coded modulation, which is similar to the one proposed by V iterbi et al. [3]. The second approach is based on the use of Gray coding coded modulation where all information bits are coded. Considering coded 8-PSK modulation, several simulation results were found and are presented for both schemes. The paper is organized as follows. The denition of convolutional self-doubly orthogonal codes [5]- [8] is presented in Section II and the decoding algorithm of these codes is reviewed in Section III. In Section IV, a brief description of pragmatic coded modulation using CSO2 C codes is explained. Section V deals with coded modulation based on Gray coding for coded symbols to constellation points mapping. Simulations results for both approaches are presented in Section VI.

978-1-4244-2517-4/09/$20.00 2009 IEEE

UI UI

(v) i

u yi

+
j=1

p (yi+j

j1 k=1

i+j k

(v1)

PI

J k=j+1 u = yi +

i+j k )
J

(v)

Fig. 1. Example of a R = 1 CSO2 C systematic encoder with J = 4 2 connection positions {1 = 0, 2 = 4, 3 = 14, 4 = 15}.

j=1

j,i

(v)

(4)

III. THRESHOLD ITERATIVE DECODING ALGORITHM A. Review of Binary modulation Case Let us consider a systematic convolutional self orthogonal encoder [1]. As explained in Section II, the J connection positions {1 , 2 , . . . , J } specify the CSO2 C code. Each parity symbol is generated as a modulo-2 sum by a set of J information bits:
J

For the rst two iterations, a recursive application of (4) yields the following approximative MAP value at iteration v = 2 [5]- [8]:
J j=1 j1 J k=j+1 J n=1,n=k u yi+(j k )(l n ) (1)

(2)

u = yi +

p {yi+j

i+j k
p (yi+j (k n )

(2)

k=1 n1

u [yi+j k +

pi =
j=1

uij , i = 0, 1, 2, ...

(1)

l=1 J

Assuming binary modulation and transmission over an Additive White Gaussian Noise channel (AWGN) [8], let the u received information symbol at time i be denoted yi , and the p received parity symbol at time i be yi . A threshold decoder will use a specic selection of the harddecision values on the received symbols to generate (J+1) equations which are orthogonal on the information bit ui being decoded [7]. A weighed combination of theses equations provides the maximum a posteriori (MAP) value for the current decoded symbol [5]- [8]. Let us dene the add-min operator as follow [7]:

l=n+1

i+(j k )(l n ) )]}

(5)

From (5), some conditions must be satised in order to obtain a function of independent observables [5]- [8]. The main condition to be satised is that all values produced by the differences and the differences of differences of the integer values j must be distinct as dened in Section II. At the last decoding iteration v = L, a hard decision is taken for the (L) current information bit ui : if i > 0 then ui = 1, else ui = 0. 1 2 = sign(1 )sign(2 ) min(|1 |, |2 |) (2) Since the CSO2 C codes considered in this paper are dened By using this operator, the approximate MAP value i , i = in the wide sense, according to the denition given in Section II, correlation between observables at iterations v 2 may 0, 1, . . ., is given by: occur due to the presence of unavoidable repetitions of some differences of differences. One way to compensate for the j1 J J lack of independence between observables at decoding is to p u u i = yi + (yi+j yi+j k i+j k ) introduce a weighting factor a in (4) as follow [7]:
j=1 J k=1 k=j+1 u = yi + J

j,i
j=1

(3)

(v)

u = a(yi +

J j=1

j,i ), v = 1, 2, ..., L

(v)

(6)

where j=1 j,i represents the extrinsic information. Thus, the threshold iterative decoding process is the repetitive application of the usual threshold decoding algorithm [1]. At iteration (v 1), the soft output of the decoder, denoted by (v1) i , is combined with the received parity symbols to yield the extrinsic value which is used by the decoder at iteration u v, in conjunction with the received information symbol yi , to (v) produce the soft output i [5]- [8]:

The weighting factor can be obtained by computer simulation. For a given Eb /N0 , the error performance is minimized in respect of the weighting factor a varying between 0 and 1. Note that another way to obtain a strict independence of J observables is to use coding rates of the form 2J , leading to 2 strict-sense CSO C codes [7]. However, in this paper, only convolutional codes dened in wide sense are considered.

B. Application to M-PSK modulation Case From the Section III-A, for a binary modulation such as BPSK modulation, the decoder provides an approximative MAP value of the current decoded information bit. However, with M-PSK modulation, a coded symbol represents more than one bit, so that the coded symbols may be split into binary symbols before decoding. Thus, the log-likelihood criterion can be used as described in [5]. b Considering a CSO2 C of coding rate R = b+1 , let the transmitted channel symbol be Sq = f (sq,1 ; sq,2 ; ...; sq,m ) where f (.) represents the mapping function and sq,r ; r = 1, ..., m, is the rth bit in the symbol Sq and m = b + 1 = log2 (M ). The symbol Sq are transmitted on an AWGN channel. Let the received channel symbol be y. The receiver must determine the reliability value of each of the m bits, yr ; r = 1, ..., m, contained in y. For 1 r m, yr = 1 if and only if: P (yr = 1/y) )0 (7) L(yr ) = log( P (yr = 0/y) Let S0,r ( respectively S1,r ) be the set of indices q of signals Sq , 1 q M of the M-PSK constellation for which sq,r = 0 ( respectively sq,r = 1 ). The sets S0,r (respectively S1,r ) are dened as follow, for r = 1, ..., m: Sb,r = {q {1, 2, . . . , M } | sq,r = b}, b {0, 1} (8)

U1

B1 = U 1 S T = (S T I ;S TQ ) B2 = U2
BITS-TO-SIGNAL POINT MAPPING AWGN CHANNEL

(S T I + N I ;S TQ + N Q )

U2

1/2-RATE CONVOLUTIONAL ENCODER

B3 = P

Fig. 2.

Block diagram of the pragmatic encoder/modulator for 8-PSK


Y
S3 010 S4 100 S5 101 S6 111 S7 110 S 2 011 S1 001 S0 B1B2 B3 = 000

X d 0 = 0.765 E

Coset = 00 Y
S4 100 S0 000

Coset = 01 Y

S1 001

Coset =11 Y S2 011

Coset = 10 Y
S3 010

X d =2 E 2
S5 101

X
S6 111

X
S7 110

Fig. 3. Pragmatic approach mapping of three coded symbols to 8-PSK constellation signals

Assuming that the bits 0 and 1 are equally likely, that is, P (yr = 0) = P (yr = 1) = 1 , (7) becomes: 2 L(yr ) = = P (yr = 1)P (y/yr = 1) ) P (yr = 0)P (y/yr = 0) P (y/yr = 1) ) log( P (y/yr = 0) log(
No 2 ,
2

(9)

in [4] are replaced respectively by the CSO2 C codes and the threshold iterative decoding algorithm [7]. Secondly, due to the large memory of CSO2 C codes and the fact that threshold decoding is a symbol-by-symbol decoder, the analysis of its coded modulation by pragmatic the approach or by the Ungerboeck method cannot be practically carried out by using the trellis representation, as conducted in [3] and [4]. A. TRANSMITTER BLOCK DIAGRAM As stated in [3] and [4], the pragmatic approach of coded modulation is another version of the Ungerboeck method [2]. That is, for b input bits, only the lowest order bit is coded by the binary convolutional encoder while the (b 1) remaining bits are not. For example, consider a rate 1 - convolutional 2 encoder and 8-PSK modulation as shown in Figures 2 and 3 where b = 2. The two output bits of the convolutional encoder will select one subset of the M-PSK constellation, which is characterized by one of four phases, also called coset, according to the following Gray code [3]: 00 01 11 10 {0, }.

For an AWGN channel with noise variance


qS1,r qS0,r

(9) yields

L(yr ) =

log(

exp( exp(

ysq
No No

) )

ysq

(10)

Therefore, we just have to determine whether the nearest neighbor of the received signal has a bit 0 or 1 in the rth position to make a decision on the current decoded bit. The metrics evaluated by the expression (10) which are given at the inputs of the iterative threshold decoder represents the reliability values of the received information and parity symbols. IV. PRAGMATIC CODED MODULATION Based on the works of Viterbi et al. [3] and [4], the pragmatic approach of coded M-PSK modulation using CSO2 C codes [7] can be conceptually represented by Figure 2 and Figure 4, respectively for the transmitter and the receiver. However, to apply this approach of coded modulation using this novel class of convolutional codes (CSO2 C), some conceptual aspects need to be mentioned. Firstly, the usual convolutional codes and the Viterbi decoding technique presented

{/4, 5/4} {/2, 3/2} {3/4, 7/4} (11)

The uncoded (b 1) bits will choose the constellation point within the above selected subset. The mapping constructed in this manner leads to pragmatic coded M-PSK modulation.

SR

S RI
DEMODULATOR METRIC S RQ COMPUTATION

~ u2 ~ p
2

THRESHOLD ITERATIVE DECODER

u2

1/2-RATE CONVOLUTIO NAL CSO2C ENCODER

u2 p

u1
DEMAPPING

u2

DELAY

Fig. 4.

Block diagram of the pragmatic demodulator/decoder for 8-PSK

We can observe that for the pragmatic approach, there are always a total number of four subsets for any M-PSK constellation, where M = 2b+1 , and 2b1 constellation points within each subset. That is, for an 8-PSK constellation, i.e. b + 1 = 3, there are four subsets and two constellation points within each subset. Likewise, for 16-PSK, i.e. b + 1 = 4, there will also be four subsets, but with four constellation points per subset. Thus, in the block diagram of the pragmatic encoder/modulator represented in Figure 2, the (b 1) uncoded bits and the two output bits of the encoder form a total of (b + 1) bits. These (b + 1) bits are mapped into the constellation points or modulated signals by following the pragmatic mapping illustrated by Figure 3, [4] for an 8PSK constellation1 . Therefore, these modulated signals will be transmitted over an AWGN channel. B. RECEIVER BLOCK DIAGRAM Again, as stated in [3] and [4], Figure 4 illustrates the modied two-stage decoding technique of pragmatic coded modulation to decode the CSO2 C codes. At the rst stage, only one information bit coded at the transmission side is decoded by the threshold iterative decoding algorithm which is well documented in [7]- [8] and its brief description is given in Section III. At the second stage, the uncoded information bit at the transmission side is decoded in two steps. In the rst step, the decoded information bit obtained at the output of the threshold decoder is re-encoded, providing an estimated parity bit. This estimated parity bit and the decoded information bit dene the coset, which along with the position of the received channel symbol, helps nally in the estimation of the uncoded bit at the transmission side. V. GRAY-CODING CODED MODULATION Although the error performances for the pragmatic approach are good, they can still be improved. One way to accomplish this improvement for the coded 8-PSK modulation system using CSO2 C codes, with a low complexity scheme, is to simply decode in one stage all the received channel symbols by the usual iterative threshold decoder. Based on simulation results, it was observed that: (1) All the input bits at the transmitter should be coded by a CSO2 C systematic encoder of coding rate R = 2 . 3
1 The

(2) Coded symbols to constellation points mapping should be done by following Gray coding. (3) The re-encoding, distance minimization, and delay function which are part of the second stage in pragmatic coded modulation, should be removed in the block diagram of the receiver, thus, reducing its complexity. Therefore, the resulting coded M-PSK modulation scheme dened as a Gray-coding coded modulation is illustrated in Figure 5 for the transmitter and Figure 7 for the receiver. Graycoding is applied to constellation points as depicted in Figure 6.
U1 B1 = U 1 S T = (S TI ; S TQ )
2/3-RATE B2 = U2 SYSTEMATIC CONVOLUTIONAL CSO2C ENCODER B = P
3

U2

BITS-TO-SIGNAL POINT MAPPING

AWGN CHANNEL

(S T I + N I ;S TQ + N Q )

Fig. 5.

Block diagram of the Gray-coding encoder/modulator for 8-PSK

Y
S3 010 S4 110

S2 011 S1 001 S0 B1B2 B3 = 000

X
S5 111 S7 100 S6 101

Fig. 6. Gray-coding mapping of three coded symbols to 8-PSK constellation signals.


b A simple way for constructing a rate b+1 CSO2 C is to 2 consider a set of J connections of a CSO C of coding rate R = 1 and to split it into b sets of connections having the same 2 cardinality equal to J . Since the set of connections of the 1 b 2 rate CSO2 C code satises the double orthogonality conditions, the b sets of connections thus obtained also satisfy the double orthogonality conditions. For example, starting with a 1 -rate 2 CSO2 C code, J = 18, specied by the set of connections = {0, 199, 375, 405, 1284, 2038, 2560, 2901, 2916, 4495, 5467, 5762, 5766, 5831, 5872, 5893, 5905, 5918}, we can construct a CSO2 C of coding rate R = 2 by splitting the 3 set in two sets of J = 9 connections, that is, 1 = {0, 1284, 2038, 2560, 2916, 5762, 5766, 5872, 5918} and 2 = {199, 375, 405, 2901, 4495, 5467, 5831, 5893, 5905}.

SR

S RI
DEMODULATOR METRIC S RQ COMPUTATION

~ u2 ~ p
2

THRESHOLD ITERATIVE DECODER

u1 u2

Fig. 7.

Block diagram of the Gray-coding demodulator/decoder for 8-PSK

VI. SIMULATION RESULTS Some error performances obtained by computer simulations for the two coded M-PSK modulation schemes using CSO2 C codes are given in Figures 8 and 9. It can be observed that

extension for 16-PSK is straightforward [3].

a better error performance is obtained with the Gray coding modulation scheme as compared to that of the pragmatic coded modulation approach. For each scheme, the best weighing factor a introduced in (6), which minimizes the error performance at the last iteration step L = 8 was determined by computer simulations. As an illustration, pragmatic coded 8-PSK modulation with a CSO2 C code of rate R = 1 having the set of connection positions {0, 2 9, 21, 395, 584, 767, 871, 899, 912} was used. For the Gray coding approach, a 8-PSK modulation in conjunction with a CSO2 C code of rate R = 2 having the sets of connection 3 {0, 1284, 2038, 2560, 2916, 5762, 5766, 5872, 5918} and {199, 375, 405, 2901, 4495, 5467, 5831, 5893, 5905} were considered. Comparing the results of both schemes with those of uncoded QPSK modulation, we observe that at the 8th iteration, E coding gains of 3 dB (at Nb = 6.6dB) for the pragmatic 0 Eb approach and 4.5 dB (at N0 = 5.1dB) for the Gray-coding scheme are achieved, both at bit error rate of 105 . The improvement in coding gain for the 8-PSK coded modulation using the gray coding approach is due to the fact that all the input information bits are coded. Moreover, as mentioned in Section V, the scheme using the Gray coding approach is much simpler than the one using the pragmatic approach. On the other hand, even though the memory length of these codes are not minimized with respect to the coded modulation, the 2 -rate CSO2 C code with J = 9 used in the 3 Gray coding scheme introduced a large decoding latency [6] - [8].
10
1

10

Weighting factor a=0.20, J=09, R=2/3, 8 iterations 1st iteration 3rd iteration 4th iteration 8th iteration QPSK (Uncoded)

10

10 Bit Error Rate

10

10

10

10

6 7 Eb/No(dB)

10

Fig. 9. BER vs. Eb /N0 for Gray coded 8-PSK modulation, R = 2 , 3 CSO2 C code positions: {0, 1284, 2038, 2560, 2916, 5762, 5766, 5872, 5918} and {199, 375, 405, 2901, 4495, 5467, 5831, 5893, 5905} [6], dmin = 10.

Therefore, the search and determination of suitable CSO2 C codes are among future works to further improve the error performance and reduce the decoding latency of coded modulation using this new class of convolutional codes. Based on the results presented in this paper, the coded modulation technique using CSO2 C codes with iterative threshold decoder have a good potential for applications in reliable high speed wireless communication systems where high bandwidth efciency is required. R EFERENCES

Weighting factor a=0.20, J=9, R=2/3, 8 iterations 1st iteration 3rd iteration 4th iteration 8th iteration QPSK (Uncoded)

10

10 Bit Error Rate

10

10

10

10

7 Eb/No(dB)

10

Fig. 8. BER vs. Eb /N0 for pragmatic coded 8-PSK modulation, R = 1 , 2 CSO2 C code positions: {0, 9, 21, 395, 584, 767, 871, 899, 912} [7], dmin = 10.

[1] J. L. Massey, Threshold Decoding, Cambridge, MA, MIT Press 1963. [2] G. Ungerboeck, Treillis-Coded Modulation with Redudant Signal Sets II: State of the Art, IEEE Comm. Mag., vol. 25, issue 2, pp. 12-21, Feb. 1987. [3] A. J. Viterbi, E. Zehavi, and R. Padovani, A Pragmatic Approach to Treillis-Coded Modulation, IEEE Comm. Mag., pp. 11-19, Jul., 1989. [4] R. H. Morelos-Zaragoza, The Art of Error Correcting Coding, Wiley, New York, 2002. [5] Dru, F., Haccoun, D., Iterative threshold decoding of punctured convolutional self doubly orthogonal codes and its application to M-ary modulation, IEEE Pacic Rim Conference on Communications, Computers and signal Processing. PACRIM, vol. 1, pp. 124-127, 26-28 Aug., 2001. [6] C. Cardinal, E. Roy, D. Haccoun, Simplied Convolutional Self-Doubly Orthogonal Codes: Search Algorithms and Codes Determination, Accepted for publication in IEEE Trans. on comm., June 2008. [7] C. Cardinal, D. Haccoun, and F. Gagnon, Iterative threshold decoding without interleaving for convolutional self-doubly orthogonal codes, IEEE Trans. Comm., vol. 51, issue 8, pp. 1274-1282, Aug. 2003. [8] D. Haccoun, C. Cardinal, High-Rate Punctured Convolutional SelfDoubly Orthogonal Codes for Iterative Threshold Decoding, IEEE Trans. Comm., vol. 53, issue 1, pp. 55-63, Jan. 2005.

VII. CONCLUSION In this paper, coded M-PSK modulation techniques using the CSO2 C codes was performed by applying two approaches: the pragmatic approach and Gray-coding. Although a coding gain as large as 4.5 dB was obtained, we can note that the CSO2 C codes used in our work were generated to be optimum only for the puncturing technique of CSO2 C [8].

Vous aimerez peut-être aussi