A Correlated Kalman Filtering Model for Subband Coding over Noisy Channels.

In this paper, we extend a framework for optimal reconstruction from noisy subband samples based on Kalman filtering to the case where some of the signals at hand are correlated, for an application to the reconstruction of a signal from quantized subband samples that have been subject to transmission noise, modeled here as bit errors. The application space envisioned for this research is the reconstruction of multimedia signals after compression and transmission on a noisy channel: decomposition of a signal in subbands, followed by quantization, is a common tool for compression, and it is thus of high interest to be able to reconstruct the original signal from its noisy, quantized subband samples. Our results show a significant improvement in PSNR on the signal samples thanks to our method, as compared to traditional reconstruction through synthesis filters.