Parallel concatenated joint source-channel coding - Electronics Letters

Introduction: Joint source-channel coding (JSCC) approaches have recently emerged as a good alternative to the strict application of Shannon’s source-channel separation principle for delay-constrained transmission scenarios. A subset of these techniques is given by joint source-channel decoding, where residual source redundancy increases the error-correction capability of the decoder [1, 2]. Especially, the serial concatenation of implicit source redundancy and explicit redundancy from a channel code can be decoded in an iterative fashion analogous to the decoding of serially concatenated channel codes [3, 4]. However, parallel concatenated JSCC (PCJSCC) schemes could be an alternative to serially concatenated approaches when one aims to increase the decoding performance especially for high bit error rates [5]. In this Letter a novel joint source-channel coding scheme for the reliable transmission of variable-length encoded waveform signals over additive white Gaussian noise (AWGN) channels is proposed. Instead of using a standard serial concatenation of source and channel encoder we employ a parallel concatenation of variable-length codes (VLCs) and channel encoding. Simulation results show that this new scheme outperforms a previously published serially concatenated VLC JSCC approach [6] in terms of reconstruction signal-to-noise ratio (RSNR). The transmission model is shown in Fig. 1. The vector U1⁄4 [U1, . . . , Uk, . . . , UK] represents a packet of K correlated source symbols Uk, where after quantisation with M bits we obtain the resulting index vector I1⁄4 [I1, I2, . . . , IK]1⁄4 [i1,1, i1,2, . . . , iK,M] with Ik2I , I 1⁄4 {0, 1, . . . , 2 1}, and ik,‘2 {0, 1}, ‘1⁄4 1, . . . , M, denoting the ‘th bit of Ik. Owing to the assumed source correlation the indices Ik show dependencies, which may, for example, be modelled as a first-order stationary Gauss-Markov process. The VLC encoder in Fig. 1 maps each fixed-length index Ik to a variable-length bit vector c(l)1⁄4C(Ik1⁄4 l) for l2I and the variable-length codetable C, leading to the binary sequence w1⁄4 [w1, . . . , wn, . . . , wNS], wn2 {0, 1}, of length NS. Also, a bit-interleaved version of I is applied to a terminated rate-Rc recursive systematic convolutional (RSC) channel code. As for turbo codes we only consider the parity bits vpm2 {0, 1} of each codeword resulting in a length-NC binary sequence vp1⁄4 [vp1, . . . , vpm, . . . , vpNC], which leads to a channel code rate of RCp. The sequences w and vp are then multiplexed and transmitted over a BPSK-modulated AWGN channel with noise variance se 1⁄4N0=2Es.