Transform domain energy modeling of natural images for wireless SoftCast optimization

Zhihai Song,Ruiqin Xiong,Xiaopeng Fan,Siwei Ma,Wen Gao

DOI: https://doi.org/10.1109/ISCAS.2014.6865335

2014-01-01

Abstract:The SoftCast scheme recently proposed for wireless visual communication avoids the threshold effect that traditional communication systems usually suffer from. It provides graceful quality transition by sending images in a sequence of whitened transform coefficients using dense-constellation modulation and analog-like transmission. A key point in SoftCast is that it allocates transmission power among coefficients unequally, according to the expected energy of each coefficient. Importantly, the energy diversity utilized by power allocation should be shared with the receiver for correct decoding. Signaling the energy for each coefficient individually is prohibitive since it requires a large set of meta data. Grouping coefficients into a few chunks and signaling the energy at chunk level, on the other hand, may compromise the efficiency of SoftCast remarkably. In this paper, we investigate the energy distribution of natural images in transform domain and propose a model to approximate this distribution. We apply the model to guide the power allocation in SoftCast. Experimental results show that the proposed method outperforms the equal-chunk approach in the original SoftCast by 2~5 dB, and reduces the number of required meta data significantly at the same time.