Hybrid technique for adaptive delta modulators.

Abstract

The simplicity and excellent noise immunity of Delta Modulation (DM) for digitally encoding speech waveforms has inspired many refinements to the basic invention. The most important refinement is in the form of Adaptive Delta Modulation (ADM), which employs variable step?size quantization. A large variety of such adaptive quantization schemes has been proposed, and most of these are feedback?type adaptations. Depending on the adaptive scheme used, ADM systems are broadly classified into three categories: syllabic (long?term), instantaneous (short?term), and hybrid. The continuous DM (CDM) and the Continuously Variable Slope Delta modulation (CVSD) are examples of syllabic adaptation, whereas Jayant’s one?bit memory systems such as CFDM?1, the constant factor delta modulator (CFDM?2), and Song’s Voice ADM (SVADM) are representatives of instantaneous adaptations. The hybrid combining DM (HCDM) is a combination of CFDM and syllabic adaptation. The variation of step size in instantaneously adapted DM can be either multiplicative or incremental. The CFDM?1 and CFDM?2 are of the multiplicative type, while SVADM is of the incremental type. Among ADM algorithms, the SVADM and CVSD coders are the least complex and provide the best performance in the presence of random and independent channel errors, and are less prone to quantizing noise. However, these coders have limited dynamic range due to slow attack/decay times. Therefore, there is a need to develop new adaptation algorithms that enhance the dynamic range of these systems without degrading other performance factors. The present investigations have resulted in some versatile ADM systems, viz., Hybrid Constant Factor Incremental Delta Modulators (HCFIDM) and Modified Continuously Variable Slope Delta Modulators (MCVSDM). The HCFIDM system combines the incremental type of instantaneous adaptation and the envelope type of syllabic adaptation. In the MCVSDM system, the syllabic filter exciting signal is made to vary as a function of the input signal level to provide small step?size changes for small input levels and fast changes for large input levels. The HCFIDM and MCVSDM systems have been studied on a digital computer using Gaussian signals, sine waves, and digitized speech as coder inputs, and their performances compared with similar systems. The figure of merit used is the signal?to?noise ratio (SNR). With realistic upper and lower limits of step sizes, a 55?dB dynamic range was obtained, as against 45?dB and 40?dB in the case of SVADM and CVSD coders respectively. Effects of channel errors and tandem encoding were also studied. The performance in each case was found to be comparable to that of the CVSD system.

The simulation results have been verified by building the necessary laboratory models. Both objective and subjective testing have been carried out. Subjectively, it has been observed that the HCFIDM and MCVSDM systems yield about 50?dB dynamic range at 32?kb/s, as against 40?dB and 30?dB in the case of SVADM and CVDM systems respectively. At the same bit rate, the mean opinion score (MOS) of the HCFIDM and MCVSDM systems is approximately 4 on a five?point rating scale.

The thesis is organised as follows: • Chapter?1, the introductory chapter, presents an overview of the investigations carried out. • Chapter?2 gives a brief review of speech digitization methods and the approach to the present problem. • The proposed idea of the HCFIDM and MCVSDM systems and their analyses are presented in Chapters?3 and 4 respectively. • The experimental verification is presented in Chapter?5. • Chapter?6, which is the concluding chapter, presents the summary of results and scope for further work. Some portions of this work have already appeared in IEEE Proceedings, Electronics Letters, International Journal of Electronics, and Journal of I.I.Sc.