Variable Bandwidth Adaptive Delta Modulation

bandwidth on the order of a factor of two relative to an initial 3-kHz bandwidth for typical speech samples. If the coder noise is white, there is an average noise power reduction by a factor proportional to the bandwidth reduction. Furthermore, the remaining portion of the noise power lies entirely within the band of the speech so that for reasonably good signal-to-noise ratios, some masking of the noise by the speech can be expected. The second case we explore is one of an adaptive sampling rate. In this case, the noise is again eliminated outside the principal speech bandwidth with a time-dependent low-pass filter. Then the average bit rate is reduced by a time-dependent decimation. Figure 1 shows a block diagram of the system implemented in software for the tests to be presented. The system includes estimation of the short-time bandwidth (discussed in Section II), time-dependent filtering to this bandwidth, sampling rate conversion via decimation/ interpolation, and a 1-bit memory ADM coder with exponential stepsize adaption. For discursive purposes, we regard each of the two bandpass filters as a cascade of independently controlled low-pass and high-pass filters. The details on the implementation of this system are given in Appendix A. 1 2