Data compression algorithm using multi-pulse adaptive sub-band coding (MASC)
Abstract (summary)
A system of data compression is proposed using an efficient parametric coding algorithm and based upon the psychoacoustic characteristics of the human auditory system, where the technique provides accurate coding of audible signals but does not waste channel capacity encoding redundant signals of low perceptual significance. This research has investigated three fundamental waveform coding techniques (APCM, ADPCM and ADM) and their performances compared against the LPC parametric coding technique, to determine their suitability with sub-band coding. Proposed as the Multi-pulse Adaptive Sub-band Coding (MASC) technique in this thesis, the system combines the application of multi-pulse linear predictive coding (LPC) and sub-band coding (SBC).
The perceptual trade-off between the number of excitation pulses allocated versus the number of sub-bands required in the MASC system is examined. An adaptive perceptual error weighting filter is also investigated to psychoacoustically motivate the pulse locations. Proposed as the Psychoacoustic Optimization Algorithm (POA), it is based upon the F-weighting function and adapts according to the audio signals in each frame, thus coding information which matches the human hearing characteristic.
By introducing an additional strategy that allocates the number of excitation pulses within the sub-band set based upon their short-term near-instantaneous frame average power, an additional coding advantage is obtained. Using the Dynamic Excitation Pulse Allocation Algorithm (DEPAA) derived, further data rate reduction can thus be obtained.
In the course of this research, a data rate reduction from 705.6 kbits/s (16-bits/44.1 kHz) to 94.4 kbits/s per channel, or a compression ratio of approximately 7:1, is achieved using the MASC system with promising results. This is equivalent to linear PCM coding of audio signals with approximately 2-bits at 44.1 kHz sampling rate.