Exploiting 2-dimensional source correlation in channel decoding with parameter estimation for unknown source

Abstract

Source redundancy does not contain any significant information for transmission in a communication system and therefore, one of the approaches to overcome this issue is by exploiting the source redundancy for error-correction via Joint Source Channel Coding (JSCC) design. Existing JSCC systems were developed to exploit 1-Dimensional (1D) correlation exhibited by a source and later, a JSCC system exploiting 2-Dimensional (2D) source correlation known as the 2D JSCC system was introduced and had been proven to outperform the 1D JSCC system in terms of Bit Error rate (BER). However, the source correlation knowledge in the 2D JSCC system has been assumed to be perfectly known at the receiver. In a real communication system, source correlation knowledge may not always be available and thus, this research aims to develop a high performance 2D JSCC system for the unknown source correlation knowledge. A parameter estimation technique had been developed based on the Baum-Welsh algorithm and employed jointly with iterative channel decoding. Simulation results revealed that the proposed 2D JSCC system with parameter estimation (2D-JSCC-PET1) for an unknown source correlation knowledge can achieve performance very close to the ideal 2D JSCC system with a known source correlation knowledge by a difference of 0.05 dB. Furthermore, the proposed 2D-JSCC-PET1 system outperformed the benchmark 2D JSCC system using a different estimation technique (2D-JSCC-PET2) by 0.84 dB at a source correlation of p = 0.7 and the performance difference became larger with the increase of source correlation strength. The effectiveness of the proposed 2D-JSCC-PET1 system is demonstrated through image transmission simulations and the simulation results reveal that despite the correlation knowledge is unknown, the proposed 2D-JSCC-PET1 system can perform very close to the ideal 2D JSCC system with only 0.26 % and 0.06 % difference in Pixel-error percentage for an image exhibiting strong and weak correlation, respectively.