640

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008

Bit-Interleaved Coded OFDM With Iterative Decoding in Impulsive Noise

Ha H. Nguyen, Senior Member, IEEE, and Trung Q. Bui

AbstractBit-interleaved coded modulation with iterative decoding (BICM-ID) and orthogonal frequency-division multiplexing (OFDM) have recently been considered separately to improve the error performance and/or spectral efciency of communication systems disturbed by impulsive noise, such as power-line communications. Given the advantages of these two techniques, this paper proposes a novel combination of them in one system. The key feature of the proposed system, termed bit-interleaved coded OFDM with iterative decoding (BI-COFDM-ID), is the iterative processing at the receiver where information exchange is carried out between the OFDM demodulator and the channel decoder. Structure and algorithms employed at the iterative receiver of BI-COFDM-ID are described in detail. The results show that BI-COFDM-ID outperforms all the previously proposed techniques at the high signal-to-noise-ratio region and, therefore, it is attractive for applications that require high spectral efciency and low bit-error rate. Index TermsBit-interleaved coded modulation (BICM), impulsive noise, iterative decoding, orthogonal frequency-division multiplexing (OFDM), power-line communications (PLC), signal mapping.

I. INTRODUCTION

ITH potential results and recent developments, the existing power-grid network is on its way from migrating from a pure energy distribution network to a multipurpose medium delivering energy, voice, data, and digital telecommunications services to every site. Due to the obvious advantages of this migration, power-line communications (PLC) are currently the focus of intense research activities. The power-grid network, however, differs considerably in structure as well as physical properties from the conventional communications systems. Varying levels of impedance and attenuation due to electrical hardware congurations are frequent. Such variations and other sources of interference lead to rather hostile properties of a PLC system. Those interferences, referred to as manmade noise, have statistical characteristics

Manuscript received August 29, 2006; revised February 27, 2007. This work was supported in part by an NSERC Discovery Grant and in part by a Graduate Scholarship from the Government of Vietnam. A part of this paper was presented at the International Conference on Communications and Electronics (ICCE), Hanoi, Vietnam, October 1011, 2006. Paper no. TPWRD-00502-2006. Ha H. Nguyen is with the Department of Electrical and Computer Engineering, University of Saskatchewan, Saskatoon, SK S7N 5A9 Canada (e-mail: ha.nguyen@usask.ca). Trung Q. Bui was with the Department of Electrical and Computer Engineering, University of Saskatchewan, Saskatoon, SK S7N 5A9 Canada. He is now with VIETPHAPMEEC, Hanoi, Vietnam (e-mail: trb610@mail.usask. ca). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TPWRD.2007.910986

that are much different from that of classical Gaussian interference and noise. Manmade noise is typically impulsive. A relatively simple model that incorporates background noise and impulsive noise is suggested in [1] and is known as Middletons class- noise. This noise model corresponds to an independent and identically distributed (i.i.d.) discrete-time random process whose probability density function is an innite weighted sum of Gaussian densities, with decreasing weights and increasing variance for the Gaussian densities [2]. Middletons classnoise model has been used widely in the performance analysis of PLC systems [3][8]. For uncoded systems (i.e., systems that do not implement error-control coding), the work in [4] considers -ary quadrature amplitude modulation ( -QAM) and designs optimum and suboptimum receivers for it. It is shown in [4] that using the receiver properly designed for impulsive noise can signicantly improve the error performance of PLC systems. More recently, the use of the orthogonal frequency-division multiplexing (OFDM) technique to combat impulsive noise has also received strong interest [7], [9], [10]. The basic principle behind OFDM is to use a properly chosen linear transform at the transmitter and its inverse transform at the receiver. The transmitted signal passes both transforms and is therefore unaffected, whereas the impulsive noise passes the receivers transform only. The energy of individual impulses is therefore dispersed (or smeared) over the increased symbol duration [11], [12]. In this way, the error oor that is typical for uncoded transmission over impulsive noise channels [3] is partially eliminated. When the number of subcarriers used in OFDM is sufciently large, the transformation at the OFDM receiver essentially converts impulsive noise to Gaussian noise (due to the central-limit theorem) and a detector designed for an additive white Gaussian noise (AWGN) channel is generally applied to demodulate the transmitted symbols over each subcarrier. Although such implementation of OFDM in the impulsive noise environment is simple, it is by no means optimal. In fact, it can perform poorer than the single-carrier system with a properly designed receiver. This is because converting impulsive noise to Gaussian noise destroys the high structure of the impulsive noise and the detector optimally designed for Gaussian noise does not make use of it. Due to the aforementioned disadvantage, a more advanced and efcient receiver based on iterative processing is recently proposed in [7] for OFDM in an impulsive noise environment. In particular, the iterative receiver in [7] consists of two information-exchanging estimators, one in the codeword domain and one in the information domain. Such iterative decoding of the OFDM is shown in [7] to perform very close to the performance of an equivalent system over an AWGN channel.

0885-8977/$25.00 2007 IEEE

NGUYEN AND BUI: BIT-INTERLEAVED CODED OFDM WITH ITERATIVE DECODING IN IMPULSIVE NOISE

641

Fig. 1. Simplied block diagram of a bit-interleaved coded OFDM system.

It is also well known that error performance of a communications system can be further improved by the use of error-control coding. Applying powerful turbo codes [13][15] to combat impulsive noise is investigated in [5] where the authors also design an appropriate turbo receiver for an additive classnoise (AWAN) channel. Motivated by the recent successes of BICM with iterative decoding (BICM-ID) over both AWGN and fading channels [16][18], the work in [8] considers BICM-ID for an AWAN channel. It is demonstrated in [8] that with the proper design of the soft-input soft-output (SISO) demodulator and appropriate constellation mapping, the use of BICM-ID can greatly improve the error performance and spectral efciency of the PLC systems. Given the effectiveness of OFDM with iterative decoding for uncoded systems in [7] and BICM-ID for coded systems in [8], [19] and [20], it is natural to combine these techniques in one system. This is precisely the main proposal of this paper and the system under consideration shall be referred to as bit-interleaved coded OFDM with iterative decoding (BI-COFDM-ID). Here, the term iterative decoding refers to the fact that information exchange is carried out between the channel decoder and the OFDM demodulator, while the latter is an iterative block itself as described in [7]. The remaining part of this paper is organized as follows. The detailed system model is described in Section II. Section III presents the novel algorithms used in the proposed iterative receiver. In Section IV, simulation and numerical results are provided to compare the error performance of BI-COFDM-ID with that of the previously known techniques. Finally, Section V concludes this paper.

fast Fourier transform (IFFT) algorithm. The IFFT yields the of OFDM symbol consisting of sequence length , where (1)

Assuming perfect synchronization and timing, the received symbols at the receiver are given by (2) Note that the impulsive noise in (2) is separated into two represents components: accounts for impulsive noise and Gaussian noise. In particular, s are i.i.d. circularly symmetric complex Gaussian random variables with variance , whose probability density function (pdf) is (3) The impulsive noise variables s are also i.i.d. with variance . Their pdf is given by Middletons class- model as [1], [3], [7]

(4) and as the Dirac delta function. The with parameter is called the impulsive index. For small (e.g., ), the noise is highly structured since only % of the samples are hit by impulses. Also for covenience, dene as the Gaussian-to-impulsive noise power ratio. (Note that the parameter is referred to later in this paper.) At the receiver, the problem of decoding the received vector arises. One can try to implement the maximum-likelihood (ML) decoder to minimize the probability of decoding error. However, such an optimum receiver is very complicated due to the presence of the random interleaver. To overcome this problem, this paper proposes a novel suboptimum iterative receiver as illustrated in Fig. 2. Similar to the BICM-ID in [8], [16], and [17], the iterative receiver implements the iterations between the SISO decoder and the SISO OFDM demodulator. Here, the OFDM demodulator is an iterative demodulator itself as proposed in [7]. Therefore, there are two iteration loopsone within the OFDM demodulator (outer loop), and the other one is between the SISO decoder and the OFDM demodulator (inner loop). The key is how

II. SYSTEM MODEL Fig. 1 highlights the transmitter side of a bit-interleaved coded OFDM system. The information bits are rst encoded by a convolutional code to produce a coded sequence . The coded sequence is then interleaved by the random interleaver . The interleaved sequence is mapped by the -ary modulator into a symbol sequence. This symbol sequence is then passed through a serial-to-parallel converter, whose output is grouped into sets of symbols each, where belongs to the -ary constellation . Here, is the number of subcarriers employed in OFDM and the symbol is transmitted over the th subcarrier. In order to generate the transmitted signal, an inverse discrete Fourier transform (IDFT) is performed on these symbols. Typically, is chosen to be a power of 2 and the IDFT can be efciently implemented using the inverse

642

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008

Fig. 2. Block diagram of the proposed iterative receiver for bit-interleaved coded OFDM.

to effectively transfer the extrinsic information from the OFDM demodulator to the SISO decoder and vice-versa. As described in [7], the outer loop exchanges the information between two estimatorsone in the codeword domain and one in the information domain. The criterion for both estimators is the minimum mean-square error (MMSE). The input information for the outer loop includes the received symbol sequence and the a priori information fed back from the inner loop. As in [7], here the capital letters denote vectors or symbols in the information domain, while the lowercase letters are used for vectors/symbols in the codeword domain. To convert information between the two domains, a transform matrix and its inverse are used as shown in Fig. 2. In OFDM, the transform matrix is the inverse Fourier transform matrix and given as in (5), shown at the bottom of the page. The inputs of the codeword-domain estimator are the received symbol sequence and the estimated symbol sequence . With these inputs, the codeword-domain MMSE estimator estimates a sequence of impulsive noise . To provide the inputs for the information-domain MMSE estimator, the estimated impulsive noise sequence together with the received sequence are converted to and by multiplying with the inverse transform matrix , respectively. The information-domain MMSE estimator in the proposed BI-COFDM-ID differs from the one in [7]. Here, not only does it use and but also takes into account the a priori probabilities to estimate the information sequence and to compute the probabilities . After being converted to the codeword domain, the estimated information sequence is then fed back as the input of the codeword-domain MMSE estimator for the next iteration. The inner loop then uses as its input to calculate the bit log-likelihood ratios (LLRs), . The LLRs of the coded bits are then deinterleaved and passed to the SISO decoder. Similar to the SISO decoder used in BICM-ID

[8], [17], [18], the SISO decoder in the receiver of the proposed BI-COFDM-ID also implements the maximum a posteriori probability (MAP) algorithm [21] to compute the hard deci. sions (if needed) as well as the extrinsic probabilities For the information exchange between the SISO decoder and the OFDM demodulator, the extrinsic probabilities are then interleaved to become the input of the outer loop. With the iterations in both the inner and outer loops, the proposed receiver for BI-COFDM-ID takes advantage of both BICM-ID and iteratively decoded OFDM to combat impulsive noise. A relevant question here is how to carry out the information exchange so that the inner and outer loops can benet from each other. The next section reviews the algorithms for the outer and inner loops and discusses different strategies of performing the information exchange between the two iteration loops. III. ALGORITHMS OF THE PROPOSED ITERATIVE RECEIVER A. Outer Iterations To obtain the MMSE estimations in the outer loop, the approximations introduced in [7] are useful. In particular, if a vector of i.i.d complex random variables with and is multiplied with the transform matrix to yield , then the components of are approximated as i.i.d complex random variables with and . Similarly, if the vector is transformed by the inverse matrix , then is also approximated as a vector of i.i.d complex random variables with and . Using these approximations, all of the marginal pdfs used in the outer loop of Fig. 2 are summarized in Table I, where denotes a Gaussian distribution with mean and variance

. . .

. . .

..

. . .

(5)

NGUYEN AND BUI: BIT-INTERLEAVED CODED OFDM WITH ITERATIVE DECODING IN IMPULSIVE NOISE

643

TABLE I MARGINAL PDFS FOR THE VECTORS USED IN THE OUTER ITERATION LOOP [7]

Furthermore, the abbreviations used in (8) are (14) (15) where the operators and evaluate the real and imaginary parts of a complex number, respectively. Similar abbreviations for and are also used in (8). Note that in order to estimate the whole vector , (8) has to be calculated for every component and its complexity grows linearly with the number of subcarriers . 2) Information-Domain MMSE Estimator: The informationdomain MMSE estimator used in the outer loop of the proposed BI-COFDM-ID differs from the one presented in [7]. Here, the and , where inputs of the estimator include (16) and (17) Basically, is the received vector transformed into the informais the transformed output of the codeword-dotion domain, main estimator, and represents the background Gaussian noise in the information domain. The a priori probabilities are the information fed back from the inner loop to increase the reliability of the outputs of the codeword-domain estimator. In are the scalar coefcients, also pro(16) and (17), is the vided by the least-square regression estimation, and error vector of the information domain estimator. To simplify notations, the superscript will also be omitted in following calculations. With these inputs and using the marginal pdfs in Table I, the MMSE estimation of each is (18) (19) (8) where is given as follows [7], [9]:

. The implementation of each estimator is described in the following: 1) Codeword-Domain MMSE Estimator: The inputs of the codeword-domain MMSE estimator are (6) and (7) is the estimated vector where is the received vector, provided by the information-domain estimator in the preceding and are the scalar coefcients iteration; provided by the least-square regression estimation; and is the error term of the information-domain MMSE estimator. will be omitted To simplify the notations, the superscript in the following computations. With the approximations mentioned earlier, the high structure of impulsive noise is the only non-Gaussian component in the codeword domain and can, therefore, be well distinguished from , and . As a consequence, the codeword-domain estimator is applied to estimate the impulsive noise vector . Since the inputs and are both approximated as vectors of i.i.d random variables, the MMSE estimation can be carried out for each and it is a function of only and . Using Table I and (6) and (7), this estimation of is given as follows [7]:

where (9) (10) (11) and the functions and are In (20), the parameters (12) and (13) (21) (22) (23) , and (20)

644

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008

Also note that in the rst iteration, the probabilities are assumed to be equiprobable and they are canceled in (18). Howare computed using the ever, from the second iteration, a priori probability information fed back from the inner loop be the number of coded bits carried by Let one symbol of the constellation . Since an ideal interleaver makes bits in one symbol independent, the a priori information can be assumed to be independent. Therefore, for each constellation symbol, the probability is calculated as (24) (25) denotes the mapping from a group of bits to a where is the value of signal point in the constellation , and the th bit in the label of symbol . The other output of the information-domain MMSE estimator , which is delivered as the input informais tion to the inner loop. The computation of is as follows:

is delivered to the SISO deAfter being deinterleaved, coder. The SISO decoder treats that information as the a priori information for the coded bits and the iteration between the OFDM demodulator and the SISO decoding keeps running. The hard decisions of the information bits can be made at the nal . The other iteration based on the extrinsic information output of the SISO decoder is the extrinsic probability , which is interleaved before being fed back as the input of the information-domain MMSE estimator. A straightforward combination of the iteratively decoded OFDM with error-control coding (such as the convolutional codes considered in this paper) is to simply compute the LLRs of the coded bits at the output of the OFDM modulator and feed them to the SISO decoder. Such a simple combination does available at the not transfer the extrinsic information output of the SISO decoder back to the information-domain MMSE estimator. For clarity in comparison, we shall refer to these systems as BI-COFDM (i.e., without iterative decoding). C. Iteration Scheduling As described before, the proposed iterative receiver for BI-COFDM-ID includes two iteration loops, namely the outer and inner loops. It is therefore necessary to investigate the strategies to schedule the iterations between the outer and the inner loops. There are many different ways that the two iteration loops can interact. This section discusses three different scheduling schemes, namely the parallel, serial, and mixed schemes whose error performances are investigated and compared in the next section. First, for parallel iteration scheduling, the iterations of the outer and the inner loops are always processed together. In particular, one iteration of the whole receiver includes one iteration in the outer loop followed by one iteration in the inner loop. In other words, the outer loop and the inner loop exchange information in every iteration of the whole receiver. Next, with serial iteration scheduling, the iterations of the outer and the inner loops are operated separately and there is no information exchange between the two loops. More speciiterations of the outer loop are processed rst. Only cally, after the last iteration of the outer loop, the iteration process of times. It can also be seen that the inner loop is carried out corresponds to the case of BI-COFDM system mentioned at the end of Subsection III-B. Finally, the mixed iteration scheduling combines parallel and serial schemes. Here, for each iteration of the outer loop, iterations are carried out for the inner loop before continuing to the next iteration of the outer loop. If the total number of iter, then the total number of iteraations of the outer loop is tions run in the inner loop is Due to more iterations implemented in the inner loop, the quality of the extrinsic information improves before it is fed back to the information-domain estimator in the next iteration. Of course, the mixed iteration scheduling becomes the parallel iteration scheme mentioned before if . At this point, it is appropriate to comment on the relative complexities of the three iteration schemes. Since the MAP decoding of convolutional codes is fairly standard and can be efciently implemented, it is reasonable to say that the complexity of the proposed iterative receiver is largely determined by the OFDM

(26) and is the subset of that where contains symbols whose th labelling bit is . The probability is given by (27) with given in (23). Under the assumption that the estimated impulsive noise components in the information domain are i.i.d and still have can be determined as the class- distribution,

(28) . Using (23), (24), (26), (27), and where of as well (28), one can obtain the estimated values as the probabilities , which is indicated as in Fig. 2. B. Inner Iterations One of the outputs of the outer iteration loop, namely the set of probabilities , is transferred to the inner loop as the input of the bit LLR calculator. Using these probabilities, the bit LLR calculator computes the log likelihood ratio for each bit of the received symbol as follows: (29)

NGUYEN AND BUI: BIT-INTERLEAVED CODED OFDM WITH ITERATIVE DECODING IN IMPULSIVE NOISE

645

Fig. 3. 8PSK with different mappings.

demodulator (this is especially true when the number of subcar). Given the same number of riers is large, such as outer iterations that involve the OFDM demodulator, it can be seen that the serial iteration scheme is simpler than the other two schemes. This is because with serial iteration, the information needs to be computed only once at the last iteration of the outer loop. Moreover, since the mixed iteration scheme includes a parallel iteration as the simplest case (when ), its complexity is the highest for the same number of outer iterations. IV. RESULTS AND DISCUSSION This section investigates the error performance of BI-COFDM-ID systems with subcarriers. A four-state, rateconvolutional code with generator polynomials is employed throughout. Each information block has a length of 4096 b. Two modulation schemes considered are quadrature phase-shift keying (QPSK) and eight phase-shift keying (8PSK). For 8PSK, three different mappings shown in Fig. 3 are considered. They are gray, set partitioning (SP), and semi set partitioning (SSP) mappings. Since QPSK is only used with uncoded OFDM for performance comparison, Gray mapping shall be used. All of the simulation results are obtained for a typical impulsive noise environment with parameters and . A. Performance Improvement with Iterations in BI-COFDM-ID Fig. 4 shows the bit-error rate (BER) performance of BI-COFDM-ID when a parallel iteration scheme is carried out for SSP, SP, and Gray mappings of 8PSK. For the SNR displayed on the horizontal axis, is the average energy per information bit and is one-sided power spectral density of the total white noise, which includes both Gaussian and impulsive noise components in (2) (i.e., ). First, observe that error performance improvement due to iterations is very signicant for all mapping schemes considered. Second, different from BICM-ID with a Gray mapping in [8] (where iterations are useless with Gray mapping), here iterations can signicantly improve the error performance even when Gray mapping is used for BI-COFDM-ID. This is expected because the OFDM demodulator is an iterative module itself and the iteration process inside the OFDM demodulator works independently with the mapping. The outer iterations therefore help to improve the overall performance of the proposed BI-COFDM-ID, regardless of the mapping schemes.

Fig. 4. BER performance of BI-COFDM-ID over an AWAN channel with 8PSK and different mappings.

Furthermore, a close investigation of Fig. 4 reveals that all mapping schemes of 8PSK used in BI-COFDM-ID perform almost identically up to ve iterations. The superiority of SSP mapping over the other two mappings is only clearly visible at high SNR and after seven iterations. However, different from BICM-ID in [8], here performance improvement by SSP mapping is not very signicant compared to both SP and Gray mapand after seven iterations, pings. For example, at a BER of SSP mapping achieves SNR gains of only 0.25 dB and 0.5 dB over the SP and Gray mappings, respectively. These SNR gains for BI-COFDM-ID are very small compared to the SNR gains observed in BICM-ID systems in [8]. This is again because, unlike BICM-ID, the error performance of BI-COFDM-ID can be improved signicantly by the outer iteration loop, regardless of the mapping scheme used. B. Comparison of BI-COFDM-ID and BI-COFDM To see the effect of the inner-loop iterations, Figs. 5 and 6 compare the error performance between the proposed BI-COFDM-ID and the BI-COFDM, where there is no feedback from the SISO decoder to the OFDM demodulator and the SISO decoder is run only one time at the end of the last outer-loop iteration. The comparisons are investigated for SSP and Gray mappings of 8PSK in Figs. 5 and 6, respectively. It is obvious from both Figs. 5 and 6 that the inner-loop

646

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008

Fig. 5. Performance comparison of BI-COFDM-ID and BI-COFDM with 8PSK/SSP mapping.

Fig. 7. Comparison of different iteration strategies for BI-COFDM-ID with 8PSK/SSP mapping.

Fig. 6. Performance comparison of BI-COFDM-ID and BI-COFDM: Rate-2=3 convolutional code, 8PSK/Gray mapping over an AWAN channel with A = 0:1 and 0 = 10 .

iterations help to improve the overall error performance of BI-COFDM-ID greatly. As can be seen from Figs. 5 and 6, performance improvement in BI-COFDM saturates after ve iterations, while a signicant gain can still be achieved by parallely running more than ve iterations of outer and inner loops in BI-COFDM-ID. In particular, after seven iterations, the SNR gains due to the inner loop are 2.5 and 3.5 dB at the BER level as can be seen from Fig. 5 and predicted from Fig. 6, of respectively. The aforementioned observations clearly show that it is necessary and benecial to run the inner-loop iterations in the proposed receiver of the BI-COFDM-ID system. C. Effects of Iteration Schemes to the Performance of BI-COFDM-ID The effects of different iteration schemes on the error performance of BI-COFDM-ID with 8PSK/SSP mapping are demonstrated in Fig. 7. Here, the mixed scheme is designed with three inner-loop iterations for one outer-loop iteration and the results

are shown for six iterations of the outer loop. The results with the serial iteration scheduling are obtained by processing six iterations in the outer loop rst, then carrying out ve iterations in the inner loop. Finally, the error performance with the parallel iteration scheme is obtained by running nine iterations over both the outer and the inner loops. For comparison, the gure also displays the performance of BI-COFDM with nine iterations of the OFDM demodulator, but only one iteration of the SISO decoder (i.e., there is no information feedback to the demodulator). As can be observed from Fig. 7, all three iteration schemes help to improve the error performance of BI-COFDM-ID compared to BI-COFDM. It is also seen that the mixed scheme offers the best error performance. In particular, at the BER level of , the mixed iteration scheme provides SNR gains of 0.7 dB and 0.4 dB compared to the parallel and the serial schemes, respectively. Also note that BI-COFDM-ID with a serial iteration scheme also performs well while its receiver complexity is the lowest. Nevertheless, given the very close performance of all iteration schemes and due to the simplicity in adjusting the number of iterations, the parallel iteration scheduling is used in the rest of this section when comparing BI-COFDM-ID with other techniques. It should also be pointed out that the comparison of different iteration schemes in this section is merely to show that all the schemes work well and outperform BI-COFDM as long as the appropriate numbers of iterations for the inner and the outer loops are chosen. For a given BER level, nding the optimal choice of the iteration scheme as well as the numbers of inner/ outer iterations for it requires detailed complexity analysis and it is beyond the scope of this paper. D. Comparison of BI-COFDM-ID and BICM-ID As demonstrated in [8], the BICM-ID designed for impulsive noise shows impressive performance. However, the excellent performance of this BICM-ID over the impulsive noise environment degrades when the channel is less impulsive or close to Gaussian noise. Since iteratively decoded OFDM can still perform very well in such an impulsive noise environment, it is anticipated that BI-COFDM-ID, which combines coding and

NGUYEN AND BUI: BIT-INTERLEAVED CODED OFDM WITH ITERATIVE DECODING IN IMPULSIVE NOISE

647

Fig. 8. Performance comparison of BI-COFDM-ID and BICM-ID with 8PSK/SSP mapping.

Fig. 9. Performance comparison of BI-COFDM-ID and iteratively-decoded OFDM.

OFDM, should perform well even when the channel is less impulsive. This prediction is clearly conrmed in Fig. 8, where the error performance of BI-COFDM-ID is compared against BICM-ID when 8PSK/SSP mapping is used. The gure shows that while BICM-ID outperforms BI-COFDM-ID at the low SNR region (i.e., high BER levels), its error performance quickly approaches a very high error oor. On the other hand, the performance of BI-COFDM-ID quickly improves with both the number of iterations as well as the SNR. The superiority of BI-COFDM-ID over BICM-ID is clearly observed at practical BER levels between and . Though now shown here in the interest of space, similar observations apply for different convolutional codes. However, the performance advantage of BI-COFDM-ID over BICM-ID degrades and is only prominent at a lower BER level if a more powerful convolutional code is used. This is because a more powerful convolutional code can help to improve the error performance of BICM-ID better and lower its error oor. Also with a more powerful code, the role of the OFDM demodulator becomes less signicant. E. Comparison of BI-COFDM-ID and Iteratively Decoded OFDM As discussed in Section I, OFDM is also a common technique to combat impulsive noise [7], [12], [22]. As shown in [7], viewing OFDM as a code over complex numbers and using iterative decoding, impulsive noise can almost be completely removed. Specically, after four iterations, the error performance of iteratively decoded OFDM can converge to the Gaussian bound. It is therefore of interest to compare the error performance of BI-COFDM-ID and iteratively decoded OFDM. To make a fair comparison in terms of spectral efciency, here, the BI-COFDM-ID is investigated with 8PSK/SSP mapping and a rateconvolutional code. Sine no channel coding is used in iteratively decoded OFDM, QPSK modulation with Gray mapping is employed.

Fig. 10. Performance comparison of BI-COFDM-ID and other techniques.

Fig. 9 shows the error performances of both BI-COFDM-ID and iteratively decoded OFDM with one, three, and seven iterations. It can be seen that most of the performance improvement can be achieved after three iterations in the case of iteratively decoded OFDM. On the other hand, performance of BI-COFDM-ID improves more signicantly with iterations and there is a big improvement when moving from three iterations to seven iterations. Although the performance of BI-COFDM-ID , it improves much is poorer in the high BER levels faster with SNR, making it superior at the lower BER levels . This is, of course, due to the time diversity provided by the convolutional code employed in BI-COFDM-ID. Finally, Fig. 10 summarizes and compares the BER performance of the relevant techniques in combating impulsive noise. All techniques yield the same spectral efciency of 2 b/s/Hz. As can be seen from this gure, BICM-ID never provides the best performance at any SNR value. Instead, BI-COFDM performs the best at low SNR, while iteratively decoded OFDM gives the best performance at medium SNR. The superiority of

648

IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 23, NO. 2, APRIL 2008

BI-COFDM-ID is seen at a high SNR region, where the BER curves decay very fast with SNR. In particular, at the BER level , the SNR gain provided by BI-COFDM-ID over iteraof tively decoded OFDM is 2 dB. It should be pointed out, however, that although having the poorest error performance, the complexity of BICM-ID is the lowest since it does not involve the OFDM modulator and demodulator.

Denote squared error

and for training vector with length

. The is

(32) (33) (34) denotes the Hermitian transpose of a where the superscript can be obtained matrix. The derivation of the squared error as

V. CONCLUSION A novel combination of BICM with OFDM was introduced to combat impulsive noise. The key feature of the proposed technique, called BI-COFDM-ID, lies at the receiver, where iterations are performed within the OFDM demodulator (outer loop) and between the OFDM demodulator and the channel decoder (inner loop). Various iteration schemes were also discussed and investigated for the proposed iterative receiver. The error performance of BI-COFDM-ID was thoroughly compared with the performance of previously proposed techniques. It was demonstrated that BI-COFDM-ID outperforms all other techniques at the high SNR region and, hence, is very attractive for applications that require high spectral efciency and low BER. The theoretical development and extensive computer simulation results presented in this paper clearly demonstrated the advantage and usefulness of the BI-COFDM-ID technique. It is of great interest to carry out experimental measurements on actual power lines to validate our proposed framework in a real-life PLC environment.

(35) (36) (37) Thus, the scalar coefcients are given by (38) Substituting this value of into (32), one can obtain the min. Finally, the variance of the imum of the squared error error vector can be calculated as follows: (39) where is the average squared error for each training OFDM . symbol and given by REFERENCES
[1] D. Middleton, Statistical-physical models of electromagnetic interference, in Proc. 2nd Symp. Technical Exhibition Electromagnetic Compatibility, Jun. 1977, vol. EMC-19, pp. 331340. [2] E. Biglieri and P. de Torino, Coding and modulation for a horrible channel, IEEE Commun. Mag., vol. 41, no. 5, pp. 9298, May 2003. [3] A. D. Spaulding and D. Middleton, Optimun reception in an impulsive interference environmentPart I: Coherent detection, IEEE Trans. Commun., vol. COM-25, no. 9, pp. 910922, Sep. 1977. [4] S. Miyamoto, M. Katayama, and N. Morinaga, Performance analysis of QAM systems under class-A impulsive noise environment, IEEE Trans. Electromagn. Compat., vol. 37, no. 2, pp. 260267, May 1995. [5] D. Umehara, H. Yamaguchi, and Y. Morihiro, Turbo decoding in impulsive noise environment, in Proc. IEEE Global Telecommun. Conf., Jul. 2004, pp. 194198. [6] H. Dai and H. V. Poor, Turbo multiuser detection for coded DMT VDSL systems, IEEE J. Select. Areas Commun., vol. 20, no. 2, pp. 351362, Feb. 2002. [7] J. Haring and A. H. Vinck, Iterative decoding of codes over complex numbers for impulsive noise channels, IEEE Trans. Inf. Theory, vol. 49, no. 5, pp. 12511260, May 2003. [8] H. H. Nguyen and T. Q. Bui, Bit-interleaved coded modulation with iterative decoding in impulsive noise, IEEE Trans. Power Del., vol. 22, no. 1, pp. 151160, Jan. 2007. [9] J. Haring, Error Tolerant Communication Over the Compound Channel. Aachen, Germany: Shaker Verlag, 2001. [10] Y. H. Ma, P. L. So, and E. Gunawan, Performance analysis of OFDM system for broadband power line communications under impulsive noise and multipath effects, IEEE Trans. Power Del., vol. 20, no. 2, pt. 1, pp. 674682, Apr. 2005. [11] G. R. Lang, Rotational transformation of signals, IEEE Trans. Inf. Theory, vol. IT-9, no. 3, pp. 191198, Jul. 1963.

APPENDIX LEAST-SQUARE ESTIMATION OF THE REGRESSION COEFFICIENTS To implement the MMSE estimators in both the codeword and information domains, one needs to obtain the scalar coefcients with the least-square regression estimation. This Appendix derives the least square estimations of coefcients for the codeword-domain MMSE estimator. Estimation of the coefcients for the information-domain MMSE estimator can be carried out similarly. In the codeword domain estimator, the impulsive noise can be modeled by

(30) where the vectors , and are randomly generated and used as the training data to estimate at the th iteration. The vector is the output of the codeword-domain estimator. The error vector is unknown and its variance needs to be minimized. For notational simplicity, the superscript is omitted in the following computations. First, (30) can be rewritten as

(31)

NGUYEN AND BUI: BIT-INTERLEAVED CODED OFDM WITH ITERATIVE DECODING IN IMPULSIVE NOISE

649

[12] J. Bingham, Multicarrier modulation for data transmission: An idea whose time has come, IEEE Commun. Mag., vol. 28, no. 5, pp. 514, May 1990. [13] C. Berrou, A. Glavieux, and P. Thitimajshima, Near shannon limit error-correction coding and decoding: Turbo codes, in Proc. IEEE Int. Conf. Commun., 1993, pp. 10641070. [14] P. Robertson, E. Villebrun, and P. Hoeher, A comparison of optimal and sub-optimal map decoding algorithms operation in the log domain, in Proc. IEEE Int. Conf. Commun., 1995, pp. 10091013. [15] S. Lin and D. J. Costello, Error Control Coding, 2nd ed. Upper Saddle River, NJ: Prentice-Hall, 2004. [16] X. Li and J. A. Ritcey, Bit-interleaved coded modulation with iterative decoding, IEEE Commun. Lett., vol. 1, no. 2, pp. 169171, Mar. 1997. [17] X. Li, A. Chidapol, and J. A. Ritcey, Bit-interleaved coded modulation with iterative decoding and 8-PSK, IEEE Trans. Commun., vol. 50, no. 8, pp. 12501256, Aug. 2002. [18] N. H. Tran and H. H. Nguyen, Signal mappings of 8-ary constellations for bit-interleaved coded modulation with iterative decoding, IEEE Trans. Broadcast., vol. 52, no. 1, pp. 9299, Jan. 2006. [19] T. Q. Bui and H. H. Nguyen, Bit-interleaved coded modulation with iterative decoding in impulsive noise, in Proc. IEEE ISPLC Conf., Mar. 2006, pp. 98103. [20] T. Q. Bui and H. H. Nguyen, Error performance of BICM-ID in impulsive noise, in Proc. IEEE Canadian Conf. Electrical Computer Engineering, May 2006, pp. 201204. [21] S. Benedetto, D. Divsalar, G. Montorsi, and F. Pollara, Soft-input softoutput maximum a posteriori (MAP) module to decode parallel and serial concatenated codes, in Proc. IEEE Int. Conf. Commun., 1995, pp. 10091013. [22] J. Haring and A. H. Vinck, OFDM transmission corrupted by impulsive noise, in Proc. Int. Symp. Powerline Communication , Apr. 2000, pp. 914.

Ha H. Nguyen (M01SM05) received the B.Eng. degree from Hanoi University of Technology, Hanoi, Vietnam, in 1995, the M.Eng. degree from Asian Institute of Technology, Bangkok, Thailand, in 1997, and the Ph.D. degree from the University of Manitoba, Winnipeg, MB, Canada, in 2001. He joined the Department of Electrical and Computer Engineering, University of Saskatchewan, Saskatoon, SK, Canada, in 2001, and became a Full Professor in 2007. His research interests include diversity techniques in wireless communications, spread-spectrum systems, and error-control coding. He is an Associate Editor for IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, and International Journal on Information and Coding Theory. He is also an Adjunct Professor of the Department of Electrical and Computer Engineering, University of Manitoba and a Funded Scientist with TRLabs, Saskatoon. Dr. Nguyen is a registered member of the Association of Professional Engineers and Geoscientists of Saskatchewan.

Trung Q. Bui received the B.Eng. degree in power engineering from Hanoi University of Technology, Hanoi, Vietnam, in 2001, and the M.Sc. degree in electrical engineering from the University of Saskatchewan, Saskatoon, SK, Canada, in 2006. He was the recipient of a Vietnamese Governments Graduate Scholarship for his M.Sc. studies in Canada. His research interests are in the areas of coded modulation and iterative decoding with a special emphasis on power-line communications. Currently, he is a Project Manager for VIETPHAPMEEC, Hanoi, Vietnam.