IT ] 17 S ep 2 00 9 Constructing Linear Codes with Good Spectra
Shengtian Yang,Thomas Honold,Yan Chen,Zhaoyang Zhang,Peiliang Qiu
2009-01-01
Abstract:In the authors’ former work, it was proved by the code-spectr um approach that linear codes with good joint spectra serve as good candidates for lossless joint source channel coding (JSCC). However, it is still unknown how to construct such co des in practice. This paper aims to present some methods for constr ucting linear codes with good joint spectra, and moreover, i t intends to provide an extensive investigation of the general proble m of constructing linear codes with good spectra in a unified f ramework. Firstly, for preparation, a series of new methods and result s of the code-spectrum approach are established. Important ones among them are spectrum generating functions, spectra with coordinate partitions, and the MacWilliams identities. Secondly, the general criteria of good linear codes are revi ewed and discussed for three cases: lossless source coding, channel coding, and lossless JSCC, respectively. In the framework o f code-spectrum approach, it is shown that all these issues j st correspond to the problems of constructing linear codes wit h good kernel spectra, good image spectra, and good joint spe ctra, respectively. It is further shown that, under some conditio ns, good joint spectra imply good kernel spectra and good ima ge spectra. This thus indicates the fundamental role of linear codes with good joint spectra and then allows us to concentra t on the construction of these special codes. Next, the general principles for constructing linear codes with good joint spectra are derived. A family of random linea r codes derived from Maximum Rank Distance (MRD) codes is pres ented and proved to have good joint spectra. When the output (resp., input) alphabet is an elementary abelian group, it i s proved that linear codes with good joint spectra can be cons tructed based on good linear source (resp., channel) codes. Further mor , a general scheme based on serial concatenation is prop osed for constructing linear codes with good joint spectra. The i nt resting role of nonsingular rate1 codes in such a scheme is also discussed. Finally, to address the practical concern, an explicit cons truction based on a serial concatenation scheme is proposed , where the inner code is a regular low-density generator matrix (LD GM) code and the outer code is a regular low-density parity-c heck (LDPC) code. The joint spectra of regular LDGM codes and the d istance property of regular LDPC codes are both analyzed. It is proved that this construction can yield linear codes with good joint spectra. Index Terms Code spectrum, linear codes, low-density generator matrix (LDGM) codes, low-density parity-check (LDPC) codes, MacWilliams identities, weight. The material in this paper was presented in part at the 2008 In ternational Conference on Communications and Networking i n China (CHINACOM’08), Hangzhou, China, August 2008. This work was supported in part by the Natural Science Founda tion of China under Grants 60772093, 60802014, and 60872063 , the Chinese Specialized Research Fund for the Doctoral Program of Higher Education u nder Grants 200803351023 and 200803351027, the National Hi gh Technology Research and Development Program of China under Grant 2007AA0lZ257. S. Yang, T. Honold, Y. Chen, Z. Zhang, and P. Qiu are with the De partment of Information Science and Electronic Engineerin g, Zhejiang University, Hangzhou 310027, China and also with Zhejiang Provincial Ke y Laboratory of Information Network Technology, Zhejiang U niversity, Hangzhou 310027, China (email: yangshengtian@zju.edu.cn; honold@zju.edu .cn; qiupl418@zju.edu.cn; ningming@zju.edu.cn; qiupl@zju.edu.cn). SUBMITTED TO IEEE TRANSACTIONS ON INFORMATION THEORY (VERS ION: SEPTEMBER 17, 2009) 1 Constructing Linear Codes with Good Spectra