[ Pobierz całość w formacie PDF ]
.M/S stereo coding:The masking behavior of stereo signals is improved if a two-channel signal can beswitched between left/right and sum/difference representation.Both broadbandand critical band-wise switching has been proposed [Johnston, 1989a].Intensity stereo coding:For high frequencies, phase information can be discarded if the energy envelopeis reproduced faithfully at each frequency, This is used in intensity stereo coding[van der Waal and Veldhuis, 1991, Herre et al., 1992].Coupling channel:In multichannel systems, a coupling channel is used as the equivalent to an n-channel intensity system.This system is also known under the names dynamiccrosstalk or generalized intensity coding.Instead of n different channels, for partof the spectrum only one channel with added intensity information is transmitted[Fielder et al., 1996, Johnston et al., 1996].Stereo prediction:In addition to the intra-channel version, prediction from past samples of onechannel to other channels has been proposed [Fuchs, 1995]. APPLICATIONS OF DSP TO AUDIO AND ACOUSTICS50Spectrum flattening:As a special version to enhance the efficiency of the quantization and codingmodule, an LPC analysis has been proposed to normalize the spectral values[Iwakami et al., 1995].2.3.3 Perceptual EntropyThe term Perceptual Entropy (PE, see [Johnston, 1988]) is used to define the lowestdata rate which is needed to encode some audio signal without any perceptual differenceto the original.An estimate of the PE (there is not enough theory yet to calculate a real PE) can be used to determine how easy or how difficult it is to encode a givenmusic item using a perceptual coder.In practice, the calculation of the PE requires an analysis filter bank and a perceptualmodel.The PE is defined as(2.1)where N is the number of frequency components between f and fu l is the lowerl , ffrequency limit (e.g.fl = 0 Hz), f is the upper frequency limit (e.g.f = 20000u uHz), signal(f) is the amplitude of the frequency component f and threshold( f ) is theestimated threshold level at the frequency f.This definition of the PE of course needsthe existence of a concept of audibility resp.an auditory threshold.Examples for thisare given later in this chapter.2.4 DESCRIPTION OF CODING TOOLS2.4.1 Filter banksThe filter bank is the deciding factor for the basic structure of a perceptual coding sys-tem.Figure 2.6 shows the basic block diagram of an static n-channel analysis/synthesisfilter bank with downsampling by k.If k = n, it is called a filter bank with criticalsampling.A number of basic parameters can be used to describe filter banks used foraudio coding:Frequency resolution:Over the past years, two main types of filter banks have been used for highquality audio coding: Low resolution filter banks (e.g.32 subbands), normally combined with aquantization and coding module which works on blocks in time direction.These are frequently called subband coders. PERCEPTUAL CODING OF HIGH QUALITY DIGITAL AUDIO51 High frequency resolution filter banks (e.g.512 subbands), normally com-bined with a quantization and coding module which works by combiningadjacent frequency lines.These have traditionally been called transformcoders.Figure 2.6 Basic block diagram of an n-channel analysis/synthesis filter bank withdownsampling by k (Reprinted from [Herre, 1995] © 1995, courtesy of the author)Mathematically, all transforms used in today s audio coding systems can be seenas filter banks.All uniform subband filter banks can be seen as transforms ofL input samples into N spectral components as derived for example in [Edler,1995, Temerinac and Edler, 1993].There is no basic difference between bothapproaches, so any attempt to distinguish between subband coders and transformcoders is against current scientific knowledge.Therefore, in the following wewill use the term filter bank synonymously to subband filter bank and/or transform .A higher resolution filter bank for most signals exhibits a larger transform gain.For this reason, high frequency resolution filter banks are the tool of choice foraudio coding systems built for maximum coding efficiency at low bit-rates.Perfect reconstruction:Perfect reconstruction filter banks allow the lossless reconstruction of the inputsignal in an analysis synthesis system without quantization.While not a neces-sary feature, the use of a perfect reconstruction filter bank simplifies the designof a coding system, While at some point other filter banks have been proposedfor use in perceptual coders (e.g.wave digital filters, see [Sauvagerd, 1988]),all currently used filter banks are either perfect reconstruction or near perfect APPLICATIONS OF DSP TO AUDIO AND ACOUSTICS52reconstruction (very small reconstruction error in the absence of quantization ofthe spectral components).Prototype window:Especially in the case of low bit-rates (implying that a lot of quantization noiseis introduced), the filter characteristics of the analysis and synthesis filters asdetermined by the prototype window / windowing function are a key factor forthe performance of a coding system.Uniform or non-uniform frequency resolution:Both uniform or non-uniform frequency resolution filter banks have been pro-posed since the first work on high quality audio coding.While a non-uniformfrequency resolution is closer to the characteristics of the human auditory sys-tem, in practical terms uniform frequency resolution filter banks have been moresuccessful.This may be due to the fact that even at high frequencies for somesignals the larger coding gain of a high frequency resolution is needed [Johnston,1996] [ Pobierz caÅ‚ość w formacie PDF ]