`in Audio Systems*
`
`PAPERS
`
`STANLEY P. LIPSHITZ, MARK POCOCK, AND JOHN VANDERKOOY
`
`University of Waterloo, Waterloo, Ontario, Canada N2L 3G1
`
`The current state of our knowledge regarding the audible consequences of phase
`nonlinearities in the audio chain is surveyed, a series of experiments is described which
`the authors have conducted using a flexible system ofall-pass networks carefully con-
`structed for this purpose, and some conclusions are drawn regarding the audible effects
`of midrange phase distortions.
`It is known that the inner ear possesses nonlinearity (akin to an acoustic half-wave
`rectifier) in its mechanical-to-electrical transduction, and this would be expected to
`modify the signal on the acoustic nerve in a manner which depends upon the acoustic
`signal waveform, and so uponthe relative phase relationships of the frequency compo-
`nents of this signal. Some of these effects have been known for over 30 years, and are
`quite audible on even very simple signals. Simple experiments are outlined to enable the
`readers to demonstrate these effects for themselves.
`Having satisfied ourselves that phase distortions can be audible, the types of phase
`distortions contributed by the various links in the audio chain are surveyed, and it is
`- concluded that only the loudspeaker contributes significant midrange phase nonlineari-
`ties. Confining the investigation to the audibility of such phase nonlinearities in the
`midrange}.circuitry.is:described which enables such effects to be assessed objectively for
`their audible consequences. The experiments conducted so far lead to a number ofcon-
`clusions:
`1) Even quite small midrange phase nonlinearities can be audible on suitably chosen
`signals.
`2) Audibility is far greater on headphones than on loudspeakers.
`3) Simple acoustic signals generated anechoically display clear phase audibility on
`headphones.
`4) On normal music or speech signals phase distortion appears not to be generally
`audible, although it was heard with 99% confidence on some recorded vocal material.
`It is clear that more work needs to be doneto ascertain acceptable limits for the phase
`linearity of audio components—limits which might become morestringent as improved
`recording/reproduction systems become available. It is stressed that none ofthese exper-
`iments ‘hus far has indicated a present requirement for phase linearity in loudspeakers
`for the reproduction of music and specch.
`
`0 INTRODUCTION
`
`Is phase distortion audible? This innocent-sounding
`question continues to remain a source of controversy
`140 years after Ohm formulated his celebrated acoustic
`‘‘phase law” [1], [2],! according to which only the power
`spectrum (and not the relative phases of the compo-
`nents) of a sound determines its character. Certainly
`phase-etfects are normally sufficiently subtle that it
`is
`not surprising that the early investigators, with the rela-
`tively insensitive apparatus of the preelectronic age,
`were for the most part absorbed by the much more
`,
`:
`
`* Presented at the 67th Convention of the Audio Engineer-
`ing Society, New York, 1980 October 31-November3; title
`revised 1982 June 29.
`
`blatant audible effects caused by changes in the power
`spectrum of a sound. Nevertheless, exceptions to the
`phase law were not entirely unknown, and Helmholtz
`[4] was cautiousto.restrict its applicability to the ‘““mu-
`sical” (that is, continuous or steady-state) portion of a
`sound, as opposed to its buildup and decay (thatis,
`transient) sections. Indeed, even Rayleigh [5] in 1896
`cast serious doubt onthe validity of Helmholtz’s con-
`clusions regarding phase inaudibility in simple conso-
`
`.
`;
`An excellent survey article on the current state of our
`knowledge of the human hearing system will be found in
`Schroeder [3]. This reference also contains a substantial bib-
`liography. We shall cite only a limited selection of the rele-
`vant research papers in this publication.
`
`580
`
`© 1982 Audio Engineering Society, inc.
`
`0004-7554/82/090580-16$00.75
`
`J. Audio Eng. Soc., Vol. 30, No. 9, 1982 September
`
` Sony Exhibit 1031
`
`Sony Exhibit 1031
`Sony v. MZ Audio
`Sony v. MZ Audio
`
`
`
`PAPERS
`
`AUDIBILITY OF MIDRANGE PHASE DISTORTION IN AUDIO SYSTEMS
`
`nances—this very. early reference is worth consulting.
`It is clear that phase distortion of sufficient magni-
`tude must be audible. For example, if a transmission
`channel were sufficiently dispersive to delay one por-
`tion of the audible spectrum by, say,
`1 second, with
`respect to the remainder, it is obvious that severe dis-
`tortion of any normal musical or speech signals would
`occur. The question thus reducesto one of ascertaining
`the magnitude. of those. phase distortions that can be
`allowed without audible degradation, that is, thresholds
`for phase distortion must be determined. Stimulated by
`the obvious need for phase correction of long-distance
`telephone lines (see, for example, [6]), a considerable
`amountof careful research has been conducted to ob-
`
`tain audibility thresholds, usually expressed in terms of
`the allowable group (or envelope) delay as a function of
`frequency. The group delay distortions introduced by
`most links in the modern audio chain, including the
`transducers, are generally regarded as being sufficiently
`small as to be inaudible, on the basis of the above-
`determined thresholds.
`And yet, there exists a substantial body of evidence
`that even more subtle phase effects are audible. Part of
`the difficulty lies in the fact that the group delay is
`frequently not a physically meaningful measure of the
`“time delay” of a frequency component of a signal
`through a system, as pointed out by Heyser [7]. For
`example, a minimum-phase system can have a group
`delay which is negative over certain frequency bands,
`but this cannot be taken to mean that it behaves acaus-
`ally in these frequency bands, giving an output before
`its signal
`input. Formulating a useful definition of
`phase distortion requires consideration of what phase
`characteristic guarantees the absence of phase distor-
`tion. If H(s) is the transfer function of a linear system
`with magnitude A(w) and phase $(w), that 1s,
`
`Hijo) = A( weit)
`
`(D
`
`then for distortionless transmission over a given fre-
`guency band it is necessary and sufficient that A(@) be
`constant and #(w) be linear in w, that is, é(@) = —Tw.
`Thusthe frequency response mustbeflat, and the phase
`response mustbea straight line through the origin, with
`slope 7 representing the constant time delay to which
`this system subjects the signal. When A(w) and/or ¢(w)
`do notsatisfy these conditions, linear distortion 1s pres-
`ent. (For a detailed discussion, see Preis [8].) The two
`common measures of phase nonlinearity are the phase
`delay
`:
`
`A,=-%
`
`and the group delay
`
`A
`do
`T= 7 Go
`
`(2)
`
`(3)
`
`both in general being functions of frequency. As al-
`ready intimated, neither the group delay nor the phase
`
`J. Audio Eng. Soc., Vol. 30, No. 9, 1982 September
`
`delay byitself is a very meaningful measure of the phase
`distortion. In seeking a better indicator of the deviation
`from phaselinearity, one is led to define what Leach [9]
`has called the differential time delay distortion Ar of the
`system as the difference
`
`(4)
`
`which is.a first-order measure of this deviation. The
`smaller Ar, the smaller the first-order phase distortion,
`and indeed, Ar = 0 over the relevant bandwidth is a
`necessary and sufficient condition for phase linearity.
`The related phase angle
`
`A@ A ‘wAr
`
`(5)
`
`is likewise a measure of the deviation from perfect
`phase response, and has been called the phase intercept
`distortion [8] or the differential phase shift distortion
`[9]. It should be clear thatit is not the actual phase shift
`in a system, as a function of frequency, which matters,
`but rather the amount by which this phase shift differs
`from a pure time delay, and Ar is a good measure of
`this. It seems not to be generally appreciated how good
`the phase linearity of most low-passfilters is, and this
`has often led to ridiculous and unnecessary demands
`being made on component high-frequency bandwidth
`in a mistaken belief that such bandwidths are necessary
`for good time delay performance (see [9]). High-pass
`filters are another matter. In any event, any purely fre-
`quency-domain quantity, whether group delay ordif-
`ferential time delay distortion, must be interpreted with
`considerable caution when assessing its audible signifi-
`cance, since the human hearing system does not oper-
`ate purely in one domain but behaves as a mixed time-
`and frequency-domain processor of daunting complexity.
`The last few decades and, as regards transducers,
`especially the last few years, have seen a considerable
`improvementin the amplitude flatness of audio com-
`ponents. Some loudspeakers, pickup cartridges, disk-
`cutting systems and magnetic tape recorders now have
`impressively flat frequency responses over most of the
`audio band, and nonlinear distortions which are rea-
`sonably low at moderate signal levels. The stage has
`thus been reached when the only major respect in which
`technical improvement can be madein their transfer
`functions is as regards their phase performance. It is for
`this reason that it is now timely that an assessment be
`made of whether improvements to their phase responses
`would provide any audible benefits, even for the most
`sensitive listeners.
`
`Before examining in more detail some of the phasc
`aberrations exhibited by the various links in the audio
`chain, an important point must be made. Systems can
`be classified into two distinct classes which differ in a
`significant respect. A minimum-phase(lag) system is
`one in which there exists an intimate relation between
`the system’s logarithmic amplitude /n A(w) and phase
`¢(w). In fact, in mathematical terms these two func-
`tions are Hilbert transforms of each other(see, for ex-
`581
`
`
`
`LIPSHITZ ET AL.
`
`PAPERS
`
`our loudspeakerlistening tests we have therefore used
`the Quad ESL. Still more phase-accurate loudspeakers
`like the new Quad ESL-63 [10] are now appearing.
`Similarly, electrostatic headphonesare a suitable choice
`for experimental purposes.
`
`1 THE STATE OF OUR KNOWLEDGE
`REGARDING PHASE AUDIBILITY
`
`As intimated above, phase effects are controversial.
`Nevertheless, there are some well-documented results
`of careful experiments which demonstrate the clear
`audibility of certain kinds of phase distortion on suita-
`bly chosen signals. For example, Mathes and Miller
`[11}in 1947 and Craig and Jeffress [12] in 1962 showed
`that a simple two-component tone, consisting of a fun-
`damental and a second harmonic only, changed in tim-
`bre as the phase of the second harmonic was varied
`relative to the fundamental. (See also [5].) This is a very
`basic and most musical type of signal, and can hardly
`be called “highly specialized.” The effect is so startling-
`ly audible that it is well worth the trouble of demon-
`strating it for oneself, and this can easily be done. Two
`sine-waveoscillators of good purity should have their
`outputs summed and. fed in-phase to both ears via a
`pair of good-quality headphones. The fundamental fre-
`quencyshould be set at 200-300 Hz and at a comfort-
`able sound pressure level. The combined oscillatorsig- °
`nal should be viewed on an oscilloscope, and the second
`oscillator tuned to the second harmonic, and adjusted
`so that the relative phase between the oscillators slips
`360° (that is, 1 cycle) every few seconds. The amplitude
`of second harmonic should be varied. It will be found
`that the tone changes timbre cyclically at the slip rate.
`If the two oscillators are phase locked, or adjusted to
`
`
`
`
` -90"
`
`ample, [7]), so that they are not independent.” Most
`audio electronics fall into this category, as do many of
`the remaining components, at least over parts of the
`frequency range. The reason that a minimum-phase com-
`ponent is so desirable from the technical point of view
`is that
`this correspondence between amplitude and
`phase guarantees that equalization (by any minimum-
`phase means) will improve the phase response as the
`frequency response is flattened, so that in the limit as
`perfcct frequency responseis achieved, the correspond-
`ing phase response becomespertectly linear (because of
`the fact that a perfect system has flat amplitude and
`linear phase response). To emphasize this point were-
`mark that it follows that even mechanical systems with
`resonanccs like loudspeakers or pickup cartridges are
`guaranteed to be improved by (minimum-phase) equal-
`ization, provided only that they themselves are min-
`imum phaseto begin with. It is therefore wrong, as is
`sometimes done, to object to an equalizer on the grounds
`that it ‘rings,’ when this very ringing must of necessity
`occur as a consequenceof its frequency response, if it is
`to accurately correct a complementary frequencyre-
`sponse aberration in some other component, and result
`in a flatter system free from such ringing.
`Those systems that are not minimum phasc may have
`their phase response degraded as their frequencyre-
`sponse is improved by (minimum-phase) equalization.
`It is for this reason important to know whether or nota
`system is of the minimum-phase type before one pro-
`ceeds to manipulate its frequency response. (This state-
`mentis of particular relevance for loudspeaker systems,
`as we shall see.) Non-minimum-phase systems can be
`represented as the cascade of a minimum-phase system,
`a pure time delay, and an all-pass system [that is, one
`having a flat magnitude A(w) = constant, but nonlin-
`ear phase (w)]. The effect of normal minimum-phase
`equalization on sucha systemis to make its minimum-
`phase part perfect, so that we are left with a system
`whose overall characteristic is that of an all-pass net-
`work. A non-minimum-phase equalizer is required to
`phase linearize such a system. Magnetic tape recorders
`and loudspeaker systems constitute non-minimum-
`phase systems in general, and great care must be taken
`in using such components when trying to assess the
`audible significance ofadditional phase distortions con-
`tributed by some other component. Weshall have more
`to say about this in the sequel. There do, however,exist
`a few minimum-phase loudspeaker systems with ade-
`(P= ESE ER EE
`quately flat frequency response (and hence approximate-
`wwe
`
`20 10000=20006"400u030
`
`ly linear phase response). The amplitude and phase
`FREQUENCY
`Hz
`curves of Fig.
`| show that the original Quad full-range
`electrostatic loudspeaker falls into this category, and so
`is suitable for conducting meaningful tests of the audi-
`bility of phase distortion elsewhere in the chain. For
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
` amplitude
`
`
`
`
`
`
`? The name ‘‘minimum-phase” derives fromthe fact that of
`all the possible systems with the same amplitude response, the
`minimum2phase system is the one having the least possible
`phase lag. Every other system differs from it by an all-pass
`function, which provides additional phase lag without chang-
`ing the frequency response.
`582
`
`J. Audio Eng. Soc., Vol. 30, No. 9, 1982 September
`
`1. Free-field amplitude and phasc responsesofthe orig-
`Fig.
`inal Quad electrostatic loudspeaker at
`1 m on axis (solid
`lines), shown together with
`the computed minimum-phase
`portion of its phase response (brokenline). Note that, apart
`from a small linear-phase (i.c., time delay) difference in the
`computed curve, the system is minimum phase to above 15
`kHz. The response irregularities are a consequence of the
`close microphone distance, necessitated by the measuringset-
`up used. At distances greater than 2 m the loudspeakerdis-
`plays the flat frequency response for which it
`is renowned.
`These curves demonstrate that a loudspeaker can be both
`minimum phase and phaselinear.
`
`
`
`PAPERS
`
`AUDIBILITY OF MIDRANGE PHASE DISTORTION IN AUDIO SYSTEMS
`
`producea stationarytrace, it can be verified quite easi-
`ly that the timbre is a function of the relative phase
`between second harmonic and fundamental. (This is
`most easily accomplished if one has access to a Fourier
`synthesizer which permits independent adjustment of
`harmonic amplitudes and phases.) The waveform seen
`on the oscilloscope will be asymmetrical positive to nega-
`tive due to the presence of even (second) harmonics
`only. This duplicates the results of [11], [12].
`But let us proceed further. Adjust the phase differ-
`ence to produce a stationary pattern which is markedly
`asymmetrical positive to negative, and compare the
`sound quality when the acoustic polarity of this signal
`‘is reversed, by reversing the polarity of the connections
`to both earpieces simultancously. This is best accom-
`plished by a switch which allows instantaneous change-
`over. Again a changewill be heard, and it will be found
`to correspond precisely to the timbre change caused by
`a 180° shift of the second harmonic, for this is equiva-
`lent to a polarity reversal in the case of this composite
`signal. This experiment can be repeated on different
`headphones, and also on loudspeakers, and the effect
`will still be found to be audible, although notas clearly
`so on loudspeakers unless conducted in an anechoic
`chamber, due to standing wavesin the room. An objec-
`tion can be raised that any transducer nonlincarity
`could account for the effect. This can be resolved in a
`
`number of ways. First, two transducers could be used,
`one for each oscillator signal. This would prevent any
`possible intermodulation, but does not remove the ef-
`fects of harmonic distortion in each separately. It also
`introduces path-length difference effects which mean
`that the listener must not move. Second, one can verify
`by measurement of the distortion in the transducer’s
`acoustic output that, at the levels and frequencies in-
`volved, they are adequately linear. Most available trans-
`ducerssatisfy this requirement easily. Third, if the trans-
`ducer is of symmetrical planar construction, and so can
`be auditioned from front and back (such as some elec-
`trostatic [10] and planar dynamic designs),
`it can be
`absolved totally as a contributory factor by the simple
`expedient of comparing the sound fromits front with
`that from its back when feeding it with an electrical
`signal of reversed polarity. For, by listening from the
`back, the acoustic polarity of any transducer asymme-
`tries is reversed (as is the signal); hence simultaneously
`reversing the electrical signal while listening from the
`back results in an acoustic signal of the same polarity
`as originally, but with all transducer asymmetry (that
`is, nonlinearity) contributions reversed in polarity. If
`no difference can be heard, it can be deduced that the
`transducer’s asymmetries are below audibility, and so
`not a contributory factor. We have in fact performed
`this verification for ourselves on a planar loudspeaker.
`Indeed, these phase effects have proved audible on all
`reasonable transducers tried. In this connection it
`is
`interesting to note that not all transducers are of the
`same acoustic polarity, that ts, not all produce an acous-
`tic compression in response to a positive going electri-
`cal signal. (The original Quad ESL, for example,
`is
`
`J. Audio Eng. Soc., Vol. 30, No. 9, 1982 September
`
`polarity-inverting.) Transducer polarity can actually be
`assessed solely on the basis of the timbre heard when
`listening to the simple two-tone signal discussed above.
`This experiment also suggests that an acoustic polarity
`inversion may be audible on music and speech, and this
`is indeed true, although the effect is much more subtle
`than those described above.
`The authors have demonstrated the two-tone exper-
`iment described above to numerous people on different
`systems. No one has ever failed to hear the timbral
`change with phase, and discern the polarity reversal on
`this signal with unvarying accuracy. Indeed, in a dou-
`ble-blind demonstration to eleven members of the
`SMWTMS audio group [13], the accuracy score was
`100% on the summed 200-Hz and 400-Hz tones over
`loudspeakers, and overall, including musical excerpts,
`the results on the audibility of the polarity inversion of
`both loudspeaker channels were 84 correct responses
`out of 137, this representing confidence of more than
`99% in the thesis that acoustic polarity reversal is audi-
`ble. (See also [14], [15].) Some designers [16] neverthe-
`less still believe this effect to be inaudible.
`Let us pursue the polarity reversal question a bit
`further. The inversion of the polarity of a time signal
`[f@) — —f(d)] is equivalent to a constant phase shift
`of 7 radians in its complex Fourier transform. This is a
`nonlinear phase distortion (in fact, phase intercept dis-
`tortion), even though the group delay is zero (that is,
`no dispersion), for the phase curveis not a straight line
`through the origin. It leads to severe waveform distor-
`tion—in fact, the interchange ofpositive and negative
`polarities in the time domain, of course. Now, many
`musical and speech sounds are markedly asymmetrical
`[17]. Furthermore, it is now well established that the
`inner ear behaves largely like a half-waverectifier (see,
`for example, [3]) with neural output from the acoustic
`nerve occurring predominantly during the rarefaction
`half of the acoustic waveform, for signal frequencies
`below about 1 kHz. Similar results have been found to
`
`applyto other animals, and interestingly enough, a sim-
`ple creature like the sprat has been found to have hair
`cells of two types, one group responding specifically to
`compressions, while the other responds to rarefactions
`[18]. The important point is that there is a well-estab-
`lished mechanism in the inner ear for detecting wave-
`form asymmetries and hencepolarity reversal of asym-
`metrical signals.? What is perhaps surprising is how
`subtle this effect generally appears to be on music and
`speech. As the above-mentioned experiment [13] indi-
`cates, however, it is an audible factor, and should be
`taken into account when performing comparisons of
`
`? This is all too often overlooked, and numerous null experi-
`ments have been reported on the audibility of phase and
`polarity effects, based on listening tests with continuoussig-
`nals like square waves, which are symmetrical positive to
`negative (that is, which contain only odd harmonics), and
`remain symmetrical irrespective of the type of phase shifting
`to which they are subjected. Such experiments are not justi-
`fied in drawing any conclusions about asymmetrical wave-
`forms, and so cannot contribute to a resolution of the issuc.
`
`583
`
`
`
`LIPSHITZ ET AL.
`
`PAPERS
`
`audio components [15], [19], [20]—acoustic polarity
`should be maintained. In fact, for this very reason,
`Stodolsky [21] recommended in 1970 that polarity stan-
`dardization be adoptedfor all audio components, some-
`thing whichis easy to implement, but has still not been
`done. There exist polarity standards only for micro-
`phones (of necessity, due to potential phase problems
`in multimiking), and unofficial standards for pickup
`cartridges (necessitated by CD-4 carrier disks) and loud-
`speakers. Even these are not uniformly followed. For
`audio electronics,
`tape recorders, disk records, and
`other components no standards have been adopted. We
`strongly advise the standardization of magnetic (analog
`and digital) tape recorder and of phonograph disk po-
`larities. A recent proposal [22], [23] for tape recorder
`polarity unfortunately only exacerbates the situation
`by recommending precisely the opposite to the stan-
`.dard which Stodolsky recommended 10 years ago.‘
`Anothervery simple experiment can be used to dem-
`onstrate conclusively that the inner ear responds asym-
`metrically. It is common experience that reversing the
`polarity of only one channel of a pair of headphones
`produces a markedly oppressive and very audible effect
`on both monauralandcoincident stereophonic material.
`Theeffect is predominantly oneaffecting frequency com-
`ponents below I kHz, as can beverified by listening to
`filtered music or noise signals restricted to frequency
`bands below and above | kHz. Since the reversal of the
`polarity does not introduce any time-delay or disper-
`sive effects into the acoustic signal, but merely changes
`compressions into rarefactions and vice versa, the au-
`dible effects are due solely to the constant 180° phase
`shift which polarity reversal entails. Since no interaural
`cross correlations occur before the olivary complexes
`to which the acoustic nerve bundles connect, it must
`follow that the acoustic nerve output from the cochlea
`is changed by the polarity reversal. This change is due
`to two factors: the cochlea is now responding to the
`opposite polarity half of the waveform, and this wave-
`form hasa shifted time relationship relative to the sig-
`nal heard by the other ear. This is merely a very simple
`confirmation of the asymmetry(thatis, half-waverecti-
`fying nature) of the inner ear. This experiment does nor
`demonstrate that polarity reversal of both channels (or
`monaural polarity reversal if the signal is applied to
`one ear only) is audible. What it does show is that the
`neural output signal from the earis phase sensitive, and
`this suggests that, when further processed by the brain,
`it would be surprising if no monaural, that is, single
`ear, phase effects were audible. The earlier demonstra-
`tions mentioned above serve to confirm this audibility,
`although the effect is more subtle than one would antici-
`pate. The mechanism responsible for this pronounced
`transduction asymmetry appears to be the hair cells,
`which respond unilaterally to motion of the cilia in-
`
`4 We have therefore suggested [24] that Stodolsky’s stan-
`dard be adopted, and have prepared a polarity test tape [25]
`which permits determiningthe polarity of a record/reproduce
`system relative to Stodoisky’s convention. Wewill gladly sup-
`ply a short segment ofthis test tape on request.
`584
`
`duced by fluid and basilar membranevibrations in the
`cochlea. This is well documented [3]. Individual neu-
`rons in the acoustic nerve can, for brief periods, fire at
`up to 1000 times per second, although for continuous
`stimulation, rates of a few hundred pulses per second
`are the norm. Since these firings occur predominantly
`during acoustic rarefactions,it follows that the ear has,
`at frequencies below | kHz, the ability to ‘follow’ the
`negative half of the waveform of an acoustic stimulus
`by modulating the neuron firing rate in sympathy with
`the signal waveform. This feature can explain many, if
`not most, of the phase and polarity effects discussed
`above.
`
`To conclude this section, we would like to give a few ©
`specific references to the literature to guide the further
`reading of those who maybeinterested in pursuing the
`subject in greater depth. The paper by Hansen and
`Madsen [26] may fairly be classed as one of the most
`influential in recent years. The authors discuss the ef-
`fect on single sinusoids of an adjustable dc offset and
`starting phase, and find pronounced phaseeffects. In
`[27], Berkovitz and Edvardsen surveythe field and com-
`ment further on the results reached in [26]. The research
`of Plomp and Steeneken [28] enables a comparison of
`phase sensitivity cffects with those due to amplitude
`(that is, frequency response) changes, and is most inter-
`esting. In [29] Cabot ef ai. use as their test signal a
`400-Hz fundamental with third harmonic added, this
`producing only symmetrical waveforms, and conse-
`quently leading to much less pronounced phaseeffects
`than those displayed by asymmetrical signals. Blauert
`and Laws[30] use specially constructed all-pass filters
`in-their phase distortion tests, but are forced to band-
`limit their test signals rather severely in order to stay
`within the flat response region oftheir all-passfilters.
`Consequently their pulse test signals were already quite
`badly dispersed before being subjected to further phase
`shifting. The paperis nevertheless interesting and re-
`flects many of the ideas which we adopted in the exper-
`iments to be outlined below. Most recently, Suzuki e¢
`al. [31] have performed careful phase audibility exper-
`iments reinforcing many of our conclusions in the
`sequel.
`An argument frequently put forward to justify why
`phase distortion cannot be significant for material re-
`corded and/or reproduced in reverberant surroundings
`is that the reflections cause gross phase distortions them-
`selves, which are very position sensitive. This is true,
`but in both cases the first-arrival direct sound is not
`subject to these distortions, and very important direc-
`tional and other analyses are conducted duringthe first
`few milliseconds after its arrival, and before the pre-
`dominant reverberation arrives. We do not accept that
`the presence of reverberation renders phase linearity
`irrelevant, and for confirmatory evidence refer to a re-
`cent paper by Bridges [32]. Some fascinating new work
`on the overriding importance of phase preservation in
`at least the long-term spectrum of acoustic signals is
`given in [33], [34], and leads one to speculate about
`possible experiments which could be conducted to re-
`
`J. Audio Eng. Soc., Vol. 30, No. 9, 1982 September
`
`
`
`PAPERS
`
`AUDIBILITY OF MIDRANGE PHASE DISTORTIONIN AUDIO SYSTEMS
`
`solve some aspects of the short-term phase controversy.
`Finally we would refer to papers by Bauer [35], Moir
`[36], Bank and Hathaway [37] and Lee [38] for some
`of the opposing views on phase matters. Although
`we do not believe that they have proven their cases for
`phase inaudibility, their arguments should be examined.
`Oneinteresting point made by Bauer [35] is that phase
`distortion is frequently used in commercial, broadcast-
`ing processors because of its waveform-symmetrizing
`properties andfor its ability to reduce signal crest fac-
`tors (that is, peak-to-rms ratios) and hence facilitate
`greater avcrage modulation levels without overmodula-
`tion. One wonders whether all these “‘benefits” can real-
`ly be achieved without audible consequences. Regard-
`ing [37], [38], we believe that the deviations from phase
`linearity still present in the loudspeaker systems usedin
`the experiments weresufficient to classify them as all-
`pass rather than as minimum-phase, and hence the con-
`clusions reached are suspect in our view.
`
`2 PHASE DISTORTION IN
`AUDIO COMPONENTS
`
`The preceding discussion would be irrelevant were
`audio components free from phase nonlinearities, but
`they are not. To a greateror lesser extent they all con-
`tribute some phasedistortion. As already indicated, as
`long as the phase distortion is of the minimum-phase
`kind, and so goes hand in hand with corresponding
`frequency response deviations, it is relatively innocu-
`ous, since any (minimum-phase) equalization of the one
`will automatically improve the other, and the flatter of
`two components will be the more nearly phase linear. It
`is the excess or all-pass distortion which concerns us
`more, for it remains after equalization.
`Since all audio components are of necessity band-lim-
`ited, their frequency response rolloffs will be accom-
`panied by phase nonlinearities which encroach into the
`audio band for a few octaves at both the low- and
`high-frequency extremes. The high-frequencyrolloff is
`generally almost phase linear over the passband, where-
`as the low-frequencyrolloff is accompanied by consid-
`erable phase distortions (see [9]). Much available
`research indicates that the high-frequency phase nonlin-
`earity is innocuous, but moreresearch needs to be done
`in this area, as well as regarding the effects of the ex-
`treme low-frequency phase distortion. What concerns
`us more is phase distortion occurring between these
`frequency extremes, say, from 100 Hz to 3 kHz—what
`we shall refer to as midrange phase distortion in this
`paper. It is in this area that most researchers have found
`that the ear’s sensitivity to phase distortion is greatest.
`Let us therefore briefly examine the components which
`make up the audio chain for the types of midrange
`phase aberrations which they introduce. Among the
`papers already cited, [7], [8], [24], [30] contain useful
`results in this area, and Preis [39] is a useful general
`discussion.
`
`Since virtually all electronic components by them-
`selves are minimum phase, their only significant phase
`
`J. Audio Eng. Soc., Vol. 30, No. 9, 1982 September
`
`distortions are those that inevitably accompanytheir
`bandwidth limitations, and so will not concern us here.
`Disk cutter heads [40] and modern wide-band pickup
`cartridges [41], [42] have beenfound to exhibit simple
`minimum-phase behavior to beyond 20 kHz, even in-
`cluding whatever resonances and mechanical breakup
`modes may occur below this upper limit. So again, no
`«midrange phase distortions of consequence are to be
`found here. Studio.microphones scem to remain largely
`an open question, and not much work appears to have
`been done regarding their phase response. Professional
`measuring microphones, like