Optimization of irregular mapping for error floor removed bitinterleaved coded modulation with iterative decoding and 8PSK
 Chen Cheng^{1}Email author,
 Guofang Tu^{1}Email author,
 Can Zhang^{1} and
 Jing Dai^{1}
https://doi.org/10.1186/16871499201431
© Cheng et al.; licensee Springer. 2014
Received: 4 August 2013
Accepted: 27 January 2014
Published: 25 February 2014
Abstract
Bitinterleaved coded modulation with iterative decoding (BICMID) is investigated for bandwidth efficient transmissions, where the error performance can be improved by employing a suitable symbol mapping. In this paper, we introduce a lowcomplexity irregular mapping optimization for BICMID with irregular doping and 8ary phaseshift keying (8PSK) modulation over both additive white Gaussian noise (AWGN) and Rayleigh fading channels, for the purpose of achieving nearcapacity performances. The Euclidean distance spectrum and the extrinsic information transfer (EXIT) chart analysis are aided for the proposed optimization to provide design guidelines of mappings. The bit error rate (BER) results demonstrate that the BICMID system with the proposed optimal irregular mapping and doping outperforms the other typical symbol mappings, and yields a gain of about 0.1 and 0.5 dB for AWGN and Rayleigh fading channels, respectively. Moreover, it is only about 0.5 and 0.73 dB away from the discreteinput continuousoutput memoryless channel (DCMC) capacity limits of AWGN and Rayleigh fading channels, respectively, at the BER of 10^{−4} and for the spectral efficiency of 2 bits/channel use.
Keywords
Bitinterleaved coded modulation with iterative decoding (BICMID) Irregular mapping Doping Distance spectrum Extrinsic information transfer (EXIT) chart1 Introduction
Bitinterleaved coded modulation (BICM) [1, 2] is the serial concatenation of a channel encoder, a bitwise interleaver, and a symbol mapper. It is a bandwidthefficient approach primarily considered for fading channels, which increases the time diversity of coded modulation and yields a better coding gain over Rayleigh fading channels than trelliscoded modulation (TCM). The performance of BICM can be greatly improved through iterative decoding (ID), which is an effective technique to improve decoding performances. The BICM with iterative decoding (BICMID) [3–5] takes advantages of iterative information exchanges between the demapper and the channel decoder and provides excellent performances over both additive white Gaussian noise (AWGN) and Rayleigh fading channels.
An additional unityrate recursive inner encoder named doping [6, 7] is implemented for BICMID, which is referred to as unityrate precoding in [8]. The doping module adds no redundancy yet introduces dependencies between adjacent bits. The dependent bits have a significant influence on the mutual information exchanges and bring about an arbitrary low error rate with an infinite interleaver length. It has been demonstrated in [6] that the error floor can be reduced or even removed by doping techniques in iterative schemes based on BICM.
Since the symbol mapper is a basic constituent part of BICMID, the optimization of symbol mappings is crucial for the error performance of BICMID. The Euclidean distance spectrum for mappings [9] is defined to characterize mappings and derive precise error bounds. To overcome the complexity problems of the exhaustive search for mappings of higher order constellations, the binary switching algorithm (BSA) [9] is implemented to find the optimal mapping, by means of searching for the best cost function (CF) based on the characteristics of the distance spectrum. Furthermore, adaptive BSA (ABSA) is proposed in [10], which adaptively changes the CFs of the BSA with the aid of the extrinsic information transfer (EXIT) chart analysis [11–13]. On the other hand, all mappings of a given modulation are classified into a number of classes of unique mappings according to bitwise distance spectra [14], which brings significant complexity reduction of the search for suitable mappings.
Instead of using the same signal constellation and symbol mapping, which are named as the regular modulation and regular mapping, different signal constellations or symbol mappings are employed for the modulation of BICMID [7, 8, 15–17], which are referred to as irregular modulations and irregular mappings, respectively. The design of irregular modulations is focused in [15], while that of [16] is based on the design of irregular mappings. Jointly considering the irregular modulations and mappings with doping, a bitinterleaved coded irregular modulation (BICIM) scheme is proposed in [7], which assigns different signal constellations and mappings with doping to maximize the average bandwidth efficiency for a constant channel quality. Furthermore, with the combination of the irregular convolutional codes, the irregular unityrate codes, and the irregular mappers, an optimal error performance is achieved in [8]. In addition, a simplified irregular mapping scheme is provided in [17], and a nearcapacity performance is obtained with the ABSA for optimal irregular mapping search.
According to the above backgrounds and discussions, as a goal of this paper, we propose a lower complexity irregular mapping optimization for BICMID with 8ary phase shift keying (8PSK) modulation compared with the existing algorithms. The proposed optimization algorithm is a curvefitting approach with the aid of the EXIT chart analysis, where there is no need to calculate the CFs of every pair of constellation points as presented in [9, 10]. Apart from the curvefitting approach, we also use the classification method for mappings proposed in [14] to further decrease the complexity of the optimization algorithm. By that means, we only focus on a small number of unique mappings out of the total mappings of the constellation, according to the classification with the Euclidean distance spectrum. In the meantime, a nearcapacity performance is achieved by irregular doping. The irregular doping is carefully designed in the proposed optimization algorithm for error floor removal, which leads to a further improvement of the error performance. On the other hand, a performance tradeoff between AWGN and Rayleigh fading channels is presented in this study. With the aid of signal space diversity (SSD) [18, 19], we preliminarily investigate the simultaneous optimization over both AWGN and Rayleigh fading channels.
Since the scope of this paper is focused on the optimization of irregular mapping and doping, the following discussion will be restricted to a rate of 2/3 convolutional coded (CC) [20] system with 8PSK, and the spectral efficiency is 2 bits/channel use.
The rest of this paper is organized as follows. Section 2 briefly reviews the scheme of BICMID with doping. Section 3 analyzes the characteristics of mappings by the EXIT chart and the Euclidean distance spectrum, thus provides design guidelines for the mappings. Section 4 proposes the optimization of irregular mapping with irregular doping, where the EXIT curves of BICMID with the proposed optimal irregular mapping and doping are provided to reveal the performance improvement. The performance tradeoff between AWGN and Rayleigh channels is presented in Section 4 as well. Section 5 provides the simulation results of BER performances and decoding trajectories with EXIT charts and demonstrates the advantages of the proposed mappings over both AWGN and Rayleigh channels. Finally, Section 6 concludes the paper.
2 Review of BICMID with doping
3 Characteristics of mappings
As mentioned, symbol mapping is crucial for the error performance of BICMID. Hence, the analysis of the characteristics of mappings is important. In this section, we use the analysis tools of the EXIT chart and the Euclidean distance spectrum to characterize mappings both intuitionally and theoretically; thus, we obtain a design guideline of irregular mapping optimization. Note that in this section, we only discuss about the characteristics of mappings over AWGN channels. In regard to Rayleigh fading channels, the characteristics of mappings are similar as those for AWGN channels; thus, they are omitted here.
3.1 EXIT chart analysis
The EXIT chart is based on mutual information to describe the flow of extrinsic information through the SISO constituent decoders, which is originally proposed in [11]. It is proved to be a powerful analysis tool to provide design guidelines for mappings of an iterative demapping and decoding scheme.
In general, an EXIT chart plots the mutual information I_{ E } as a function of I_{ A }, where I_{ A } is the average a priori information going into the decoder and I_{ E } is the average extrinsic information coming out of the decoder. The EXIT chart usually has two EXIT curves for a serially concatenated system, that is, the EXIT curve of the inner decoder and the inverted EXIT curve of the outer decoder. As the EXIT chart is an efficient tool for the convergence prediction of the error performance, the behavior of the decoding algorithm could be approximately predicted before actually running the algorithm, by the following properties [12, 13]:

If there is a crossing point between the two EXIT curves, the value of this crossing point indicates the error floor of the decoding algorithm.

If there is an open EXIT tunnel between the two EXIT curves, an infinitesimally low BER could be achieved, when large interleavers and iterations are provided.

The open tunnel’s area is proportional to how closely can the scheme operate to the channel capacity.
The SSP mapping is found to be the best mapping of 8PSK for a general BICMID system [5], which can easily be revealed by the EXIT chart in Figure 3 as well. The EXIT curve of SSP mapping reaches the highest value of ${I}_{{E}_{1}}$, thus yields the lowest error floor. However, even the best mapping of 8PSK cannot create an open EXIT tunnel since the demapper’s EXIT curve cannot reach the point (1,1). As a result, the error performance of the general BICMID with 8PSK is limited.
As mentioned, additional doping has a significant influence on mutual information exchanges. With the help of doping, the open EXIT tunnel is conceivable if a proper symbol mapping is provided. As shown in Figure 4, all the three doped demapper’s EXIT curves reach the point (1,1) with P=50. Besides, the SP and SSP curves are both above the inverted SISO decoder’s EXIT curve, that is, open EXIT tunnels are obtained at E_{ b }/N_{0}=4 dB over AWGN channels. It is thus clear that better error performances can be achieved with the help of doping.
3.2 Distance spectrum
for Rayleigh fading channels. χ denotes the signal set of the Mary PSK constellation, and ${\chi}_{b}^{i}\phantom{\rule{1em}{0ex}}\text{and}\phantom{\rule{1em}{0ex}}{\chi}_{\stackrel{\u0304}{b}}^{i}$ denote the two complementary signal subsets, where $\stackrel{\u0304}{b}=1b\in \left\{0,1\right\}$; ${s}_{k}\phantom{\rule{1em}{0ex}}\text{and}\phantom{\rule{1em}{0ex}}{\u015d}_{k}$ are two coded symbols, where ${s}_{k}\in {\chi}_{b}^{i}$ and ${\u015d}_{k}\in {\chi}_{\stackrel{\u0304}{b}}^{i}$. For an Mary PSK modulation, m= log2M.
Except for determining corresponding CFs, the Euclidean distance spectrum can also determine the corresponding bitwise mutual information [14], thus determining the corresponding EXIT curves. The bitwise distance spectra are used to classify mappings of the Mary PSK constellation. A given set of bitwise distance spectra [W_{ 0 },W_{ 1 }] characterizes an entire class of equivalent mappings, where W_{ 0 } is the bitwise distance spectrum given no a priori information and W_{ 1 } is the bitwise distance spectrum given full a priori information. W_{ i } is expressed as
denotes the total Hamming distance for the j th bit between one symbol and all other symbols at Euclidean distance d_{ k }, averaged over all symbols of the constellation, while ${w}_{1}^{\phantom{\rule{0.3em}{0ex}}j}\left(k\right)$ is denoted similarly for the j th bit when the other bits of a symbol are all known in the demapping process. Therefore, the mappings can easily be classified through an exhaustive search of unique sets of distance spectra [W_{ 0 },W_{ 1 }].
It is remarkable that the classification is based only on the distance spectra for the two extreme cases of either no a priori or full a priori information; thus, mappings with the same [W_{ 0 },W_{ 1 }] can still exhibit different distance properties for arbitrary a priori information in between the two extreme cases. Hence, the classification provides a lower bound of the number of unique 8PSK mappings, which leads the optimization algorithm into a suboptimal approach rather than a global optimal one. As a result, there is a tradeoff between the complexity and the performance for the design of the optimization algorithm. A significant reduction of the complexity is obtained by the classification, while the nearcapacity performance is slightly influenced, which is acceptable for overall consideration.
4 Optimization of irregular mapping
To obtain the discreteinput continuousoutput memoryless channel (DCMC) capacity associated with 8PSK of 2 bits/channel use, the AWGN channel needs a minimum E_{ b }/N_{0} of 2.75 dB, while the Rayleigh fading channel needs a minimum E_{ b }/N_{0} of 5.37 dB. For a closer approach to the DCMC capacity, the regular modulations and mappings are not effective; thus, the irregular mapping optimization is under consideration. Taking advantages of the characteristics of mappings discussed in Section 3, we investigate a curvefitting optimization algorithm of finding an EXIT curve of the irregular doped demapper at a minimum E_{ b }/N_{0}, while ensuring that this curve is all above the SISO decoder’s inverted EXIT curve and the area within the open EXIT tunnel is as small as possible. Instead of calculating CFs as in the BSA, this optimization is a lowcomplexity algorithm with the aid of EXIT chart analysis. In addition, the proposed algorithm only focuses on the 86 unique mappings of 8PSK constellation out of a total of 5,040 mappings, which is much more efficient than the exhaustive search algorithms. Furthermore, since the doping rate P can be used to adjust the EXIT curve shape of BICMID with a given mapping, a carefully designed irregular doping rate is taken into account for irregular mapping optimization. Note that the optimization algorithm is designed for AWGN and Rayleigh fading channels separately; we have made further efforts to the simultaneous optimization over both AWGN and Rayleigh channels in Section 4.3.
4.1 Irregular construction
It is worth mentioning that doping is able to ‘bend up’ the tail of the curve for high ${I}_{{A}_{1}}$ while lowering the small ${I}_{{A}_{1}}$ with the decreasing doping rate P. On the other hand, dependent values between the adjacent bits are introduced by doping to reduce the error floor; thus, the randomness of the coded bits after interleaving is affected, which will result in performance degradation if the doping rate is too small. As a result, we set the doping rate P∈{10,50,100} over all considerations.
Denote that $\left({I}_{{A}_{1}}^{{P}_{1}},{I}_{{E}_{1}}^{{P}_{1}}\right),\left({I}_{{A}_{2}}^{{P}_{2}},{I}_{{E}_{2}}^{{P}_{2}}\right)\in \left\{\left({I}_{A}^{P},{I}_{E}^{P}\right)\right\}$ are the doped demapper’s EXIT curves of mapping T_{1} with doping rate P_{1} and mapping T_{2} with doping rate P_{2}, respectively, where $\left\{\left({I}_{A}^{P},{I}_{E}^{P}\right)\right\}$ is the set of the classified unique EXIT curves with doping rate P∈{10,50,100}. Similarly, denote $\left({I}_{{A}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}},{I}_{{E}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}}\right)$ as the mutual information of the system with irregular mapping and doping. The irregular mapping is denoted as T_{ I r }=[T_{1},T_{2}], and the corresponding doping rate is P_{ I r }=[P_{1},P_{2}]. Taking advantages of the average approximate of the EXIT chart analysis [13], we have
where H_{ b }(·) is the binary entropy function and p_{ l } is the bit error probability. According to the construction of irregular codeblock shown in Figure 7, ${I}_{{E}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}}$ is formulated as
From Equations 4 and 5, we arrive at
Therefore, the doped demapper’s EXIT curve with the irregular mapping T_{ I r }=[T_{1},T_{2}] and doping rate P_{ I r }=[P_{1},P_{2}] is equivalent to the linear combination of the two EXIT curves of mapping T_{1} with doping rate P_{1} and mapping T_{2} with doping rate P_{2} directly. We obtain that irregular mapping optimization can be transformed to a curvefitting problem of the classified unique EXIT curves, which calculates the irregular doped demapper’s EXIT curve $\left({I}_{{A}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}},{I}_{{E}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}}\right)$ at a minimum E_{ b }/N_{0} according to Equation 6, while ensuring that this curve is all above the SISO decoder’s inverted EXIT curve, and the area A_{ I r } within the open EXIT tunnel is as small as possible. Hence, the optimal ${T}_{{\mathit{\text{Ir}}}_{\mathit{\text{opt}}}}=\left[{T}_{1\mathit{\text{opt}}},{T}_{2\mathit{\text{opt}}}\right]$, ${P}_{{\mathit{\text{Ir}}}_{\mathit{\text{opt}}}}=\left[{P}_{1\mathit{\text{opt}}},{P}_{2\mathit{\text{opt}}}\right]$, and α_{ o p t } are obtained.
The irregular mapping of the considered two mappings with linear combination of the corresponding EXIT curves can easily be generalized as the condition of N(N>2) mappings, according to [7]:
where the coefficient α_{ i }∈[0,1] satisfies the condition
It is conceivable that the irregular mapping optimization of N(N>2) mappings may achieve a better error performance than the irregular mapping optimization of two mappings, which is not discussed in this paper and is currently under investigation.
Algorithm 1 Irregular mapping optimization
4.2 Irregular mapping optimization algorithm
The pseudocode of the irregular mapping optimization algorithm is shown in Algorithm 1, where the number of elements in the set of the unique mappings {T} is Q=86, according to the discussion in Section 3.2. The doping rate is P^{ k }∈{P} with k=1,2,…,K, where we set K=3 according to Section 4.2, that is, {P}={10,50,100}. The EXIT curve of the i th regular mapping with doping rate P^{ k } is denoted as $\left({I}_{{A}_{i}}^{{P}_{i}^{k}},{I}_{{E}_{i}}^{{P}_{i}^{k}}\right),i=1,2,\dots ,Q$, where ${P}_{i}^{k}\in \left\{P\right\}$ with k=1,2,…,K. The SISO decoder’s inverted EXIT curve is denoted as $\left({I}_{A}^{\text{CC}},{I}_{E}^{\text{CC}}\right)$.
For the sake of achieving a nearcapacity performance, the minimum E_{ b }/N_{0} will be determined before searching for the optimal irregular mapping. We set an experience factor δ=(δ_{1},δ_{2},δ_{3}),δ_{1}>δ_{2}>δ_{3}>0 to adjust the initial value and the searching step size of E_{ b }/N_{0}. First, E_{ b }/N_{0} is initialized as E_{ b }/N_{0}=C_{ l i m }+δ_{1}, where C_{ l i m } is the DCMC capacity limit valued at 2.75 dB for AWGN channels and 5.37 dB for Rayleigh fading channels. δ_{1} is large enough to ensure that at least one open EXIT tunnel is existing at the initial E_{ b }/N_{0} between the demapper’s EXIT curve of the regular mapping $\left({I}_{A}^{P},{I}_{E}^{P}\right)$ and the SISO decoder’s inverted EXIT curve. Second, the value of E_{ b }/N_{0} is slightly decreasing with δ_{2} until no open EXIT tunnel is obtained with the regular mappings. Then, the searching for the curve $\left({I}_{{A}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}},{I}_{{E}_{\mathit{\text{Ir}}}}^{{P}_{\mathit{\text{Ir}}}}\right)$ of the optimal irregular mapping is activated, the optimization algorithm will find a minimum E_{ b }/N_{0} while ensuring that the EXIT curve is all above the SISO decoder’s inverted EXIT curve, and the area A_{ I r } within the open EXIT tunnel is as small as possible. If no irregular mapping is found, E_{ b }/N_{0} will slightly increase with δ_{3}. In this study, the experience factor is set as δ=(δ_{1},δ_{2},δ_{3})=(0.5,0.05,0.01), according to the practical experiments.
4.3 Optimization tradeoff between AWGN and Rayleigh fading channels
The proposed optimization algorithm in Section 4.2 is designed for AWGN and Rayleigh fading channels separately; thus, the optimal irregular mapping suited for AWGN channels can obviously not achieve the optimal performance over Rayleigh fading channels. As a result, the proposed optimization algorithm is infeasible for the case of a simultaneous optimization over both AWGN and Rayleigh fading channels. The technique of SSD [10, 18, 19] provides a significant diversity gain over Rayleigh fading channels while it leads to nondegradation over AWGN channels, which is suitable for the case of the simultaneous optimization. With the aid of SSD, we give a preliminary investigation of the optimization tradeoff between AWGN and Rayleigh fading channels.
where $f\left({s}_{k}\right)=\underset{{\u015d}_{k}\in {\chi}_{\stackrel{\u0304}{b}}^{i}\left(\theta \right)}{max}\left\{\frac{1}{{\left\mathit{\text{Re}}\left({s}_{k}{\u015d}_{k}\right)\right}^{2}}+\frac{1}{{\left\mathit{\text{Im}}\left({s}_{k}{\u015d}_{k}\right)\right}^{2}}\right\}$. The symbolic representations in Equation 9 are the same as those of the CF for Rayleigh fading channels in Equation 2. The difference between Equations 9 and 2 is that value D in Equation 9 depends on the rotation angle θ, and the optimal θ is chosen by maximizing D.
Since we have obtained the optimal ${T}_{{\mathit{\text{Ir}}}_{\mathit{\text{opt}}}}^{a}$, ${P}_{{\mathit{\text{Ir}}}_{\mathit{\text{opt}}}}^{r}$, and ${\alpha}_{\mathit{\text{opt}}}^{a}$ over AWGN channels in Section 4.2, we use exactly the same parameters for the simultaneous optimization between AWGN and Rayleigh fading channels, in order to firstly maintain the nearcapacity performance over AWGN channels. Then, the optimal parameters for AWGN channel is used for the BICMIDSSD with doping scheme over Rayleigh fading channels. With the aid of SSD, the design criterion in Equation 9 is used for choosing an optimal rotation angle θ; thus, a better performance is achieved over Rayleigh fading channels. For the reason that the SSD has no influence on transmissions over AWGN channels, the proposed BICMIDSSD with doping scheme with the optimal rotation angle θ will achieve a nearcapacity performance for both AWGN and Rayleigh fading channels.
As shown in Figure 13, the demapper’s EXIT curve with the optimal rotation angle θ=9° becomes steeper than the curve without rotation, which is the same as mentioned in [19]. With the aid of SSD, the open EXIT tunnel between the demapper’s EXIT curve and the SISO decoder’s inverted EXIT curve is obtained over Rayleigh fading channels, while using the optimal ${T}_{{\mathit{\text{Ir}}}_{\mathit{\text{opt}}}}^{a}$, ${P}_{{\mathit{\text{Ir}}}_{\mathit{\text{opt}}}}^{r}$, and ${\alpha}_{\mathit{\text{opt}}}^{a}$ over AWGN channels. Since no performance degradation is caused by SSD for transmissions over AWGN channels, the proposed BICMIDSSD with doping for 8PSK with the rotation angle θ=9° performs the same as the separate optimization over AWGN channels. In other words, the simultaneous optimization over both AWGN and Rayleigh fading channels reaches the same performance as the separate optimization algorithm over AWGN and Rayleigh fading channels. However, the optimization criterion is based on the harmonic mean of the square of Euclidean distance [18], which is not effective enough as mentioned in [19]. Further study is needed for an effective optimization, which is now under investigation.
5 Simulation results
This section provides the simulation results to confirm the advantages of the proposed optimal irregular mappings of 8PSK with BICMID with irregular doping. We also provide the simulation of the general BICMID for comparison. The 2/3rate, fourstate CC is implemented as the encoder, and the generator sequences are g_{ 1 }=(6,2,6), g_{ 2 }=(2,4,4) [20]. The spectral efficiency is 2 bits/channel use, which yields a DCMC capacity limit of 2.75 dB for AWGN channels and 5.37 dB for Rayleigh fading channel. We set the block length of the proposed system to 200,000 bits and the iterations to 50.
6 Conclusions
In this paper, we have proposed a lowcomplexity irregular mapping optimization for BICMID with irregular doping and 8PSK. The Euclidean distance spectrum and the EXIT chart are aided for the proposed optimization to provide design guidelines of mappings, and the computational complexity is largely decreased by the classification of mappings. The simulation results demonstrate that the proposed optimal irregular mapping and doping yield a better performance than the typical symbol mappings of 8PSK, that is, a 0.1dB gain over AWGN channels and a 0.5dB gain over Rayleigh fading channels at BER= 10^{−4}. In addition, the optimal irregular mapping performs about 0.5 and 0.73 dB away from the DCMC capacity limit over AWGN and Rayleigh channels, respectively, at BER= 10^{−4} and for the spectral efficiency of 2 bits/channel use.
Declarations
Acknowledgements
This research work was supported by the State Key Program of National Natural Science Foundation of China (grant no. 61032006), the National Science Foundation of China (grant no. 61271282), and the Award Foundation of Chinese Academy of Sciences (grant no. 2069901).
Authors’ Affiliations
References
 Zehavi E: 8PSK trellis codes for a Rayleigh channel. IEEE Trans. Commun 1992, 40(5):873884. 10.1109/26.141453View ArticleGoogle Scholar
 Caire G, Taricco G, Biglieri E: Bitinterleaved coded modulation. IEEE Trans. Inf. Theory 1998, 44(3):927946. 10.1109/18.669123MathSciNetView ArticleGoogle Scholar
 Li X, Ritcey J: Bitinterleaved coded modulation with iterative decoding using soft feedback. Electron. Lett 1998, 34(10):942943. 10.1049/el:19980651View ArticleGoogle Scholar
 ten Brink S, Speidel J, Yan RH: Iterative demapping and decoding for multilevel modulation. In IEEE Global Telecommunications Conference, 1998. GLOBECOM 1998. The Bridge to Global Integration. Piscataway: IEEE; 1998:579584.Google Scholar
 Li X, Chindapol A, Ritcey JA: Bitinterleaved coded modulation with iterative decoding and 8 PSK signaling. IEEE Trans. Commun 2002, 50(8):12501257. 10.1109/TCOMM.2002.801524View ArticleGoogle Scholar
 Pfletschinger S, Sanzi F: Error floor removal for bitinterleaved coded modulation with iterative detection. IEEE Trans. Wireless Commun 2006, 5(11):31743181.View ArticleGoogle Scholar
 Schreckenbach F, Bauch G: Bitinterleaved coded irregular modulation. Eur. Trans. Telecommunications 2006, 17(2):269282. 10.1002/ett.1107View ArticleGoogle Scholar
 Tee R, Maunder RG, Hanzo L: Exitchart aided nearcapacity irregular bitinterleaved coded modulation design. IEEE Trans. Wireless Commun 2009, 8: 3237.View ArticleGoogle Scholar
 Schreckenbach F, Gortz N, Hagenauer J, Bauch G: Optimized symbol mappings for bitinterleaved coded modulation with iterative decoding. In IEEE Global Telecommunications Conference, 2003 (GLOBECOM’03). Piscataway: IEEE; 2003:33163320.Google Scholar
 Xie Q, Yang Z, Song J, Hanzo L: EXITchartmatchingaided nearcapacity coded modulation design and a BICMID design example for both Gaussian and Rayleigh channels. IEEE Trans. Vehicular Technol 2013, 62(3):12161227.View ArticleGoogle Scholar
 ten Brink S: Convergence behavior of iteratively decoded parallel concatenated codes. IEEE Trans. Commun 2001, 49(10):17271737. 10.1109/26.957394View ArticleGoogle Scholar
 Ashikhmin A, Kramer G, ten Brink S: Extrinsic information transfer functions: model and erasure channel properties. IEEE Trans. Inf. Theory 2004, 50(11):26572673. 10.1109/TIT.2004.836693MathSciNetView ArticleGoogle Scholar
 Hagenauer J: The EXIT chartintroduction to extrinsic information transfer in iterative processing. Proceedings of the 12th European Signal Processing Conference (EUSIPCO), Vienna, 6–10 Sept 2004 15411548.Google Scholar
 Brannstrom F, Rasmussen LK: Classification of unique mappings for 8PSK based on bitwise distance spectra. IEEE Trans. Inf. Theory 2009, 55(3):11311145.MathSciNetView ArticleGoogle Scholar
 Schreckenbach F, Bauch G: Irregular signal constellations, mappings and precoder. International Symposium on Information Theory and its Applications (ISITA), Parma, 10–13 Oct 2004 13321336.Google Scholar
 Szczecinski L, Chafnaji H, Hermosilla C: Modulation doping for iterative demapping of bitinterleaved coded modulation. Commun. Lett. IEEE 2005, 9(12):10311033. 10.1109/LCOMM.2005.1576578View ArticleGoogle Scholar
 Liu Z, Peng K, Cheng T, Wang Z: Irregular mapping and its application in bitinterleaved LDPC coded modulation with iterative demapping and decoding. IEEE Trans. Broadcasting 2011, 57(3):707712.View ArticleGoogle Scholar
 Chindapol A, Ritcey JA: Bitinterleaved coded modulation with signal space diversity in Rayleigh fading. In Conference Record of the ThirtyThird Asilomar Conference on Signals, Systems, and Computers 1999. Piscataway: IEEE; 1999:10031007.Google Scholar
 Xie Q, Song J, Peng K, Yang F, Wang Z: Coded modulation with signal space diversity. IEEE Trans. Wireless Commun 2011, 10(2):660669.View ArticleGoogle Scholar
 Lin S, Costello DJ: Encoding of convolutional codes. In Error Control Coding. Englewood Cliffs: PrenticeHall; 2004:287295.Google Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.