Features

One way, an initial compression at the microphone caused the speaker to deliver an initial compression (absolute signal phase maintained) and the other way, the initial compression at the microphone caused the speaker to deliver an initial rarefaction (absolute phase inverted). Extending this to hi-fi equipment, he found, when applying a rigorous methodology, developed with his co-worker John Vanderkooy, to two items which still sounded different when all the above-mentioned differences had been removed, that the matching of their absolute phase characteristic then removed that audible difference. Occam's razor—don't multiply entities unnecessarily—therefore must ascribe the audible difference between the two designs to the absolute phase difference and nothing else!

All the magical properties, supposedly unmeasurable by normal test methods which were supposed to exist between two different amplifiers, for example, could be boiled down to the effect on asymmetric signals of whether a design was phase inverting or non-inverting if—and this "if" is itself a significant source of real audible differences—all frequency response differences are eliminated and neither amp is being driven into even momentary overload with the speaker in use. If you're convinced that your Quad 405 doesn't sound as good as a highly touted audiophile amp, you can reverse the phase of both speaker leads and start to enjoy your record collection again—unless you use Quad Electrostatics, of course, which are themselves arbitrarily connected to be inverting (footnote 11), in which case you'll start enjoying your records less!

The problem is that there is no one standard governing signal polarity (apart from that for unbalanced mikes, which states that a positive-going pressure change should produce a positive-going voltage). After that, all amplifiers, mixing desks, tape machines and recording lathes used to produce the record are designed totally arbitrarily when it comes to preservation of signal phase polarity. The listener has no idea how many inverting or non-inverting stages the signal has gone through to get on to the record. And even if he did know, hi-fi amp design is again random. Some power amps invert; most do not. Some preamps are inverting with the tone controls switched "in" but non-inverting when switched to "cancel." Some are non-inverting except when the signal goes through a tape loop. With some the disc input is inverting but not the line level inputs. Some have an inverting headamp but a non-inverting moving-magnet input. Some tape recorders put an inverted signal onto tape but the replay amp is non-inverting. Peter Moncrieff has even reported (footnote 12) that the circuit tracking the audio signal in the Dolby-B circuitry is inverting on record but not on playback. As it incorporates a half-wave rectifier to provide the necessary DC control signal, asymmetric waveforms—such as speech and music—can give up to a 3% frequency response error due to compandor mistracking.

Now there still remains the major aspect that, despite all this undoubtedly being true, its importance does depend on the absolute phase polarity being audible on real music program. It seems that all the psycho-acoustic researchers into audibility of phase change that I have been able to find in the HFN/RR library, right the way from Helmholtz (footnote 13), who stated in 1862 that "the quality of the musical portion of a compound tone depends solely on the number and relative strength of its partial simple tones and in no respect of their differences of phase," up to Ronken (footnote 14) in 1970, Madsen/Hansen of B&O (footnote 15) in 1972 and Cabot et al (footnote 16) in 1976, have used artificial test signals—clicks, single or dual sinusoids, sinewave pulses with shifted zero points etc.—-as a (necessarily repeatable) asymmetric program source. Most of the work since Helmholtz suggests that on such signals, the human ear can detect differences, although all workers indicate that the effect is subtle.

Obviously, it is not necessarily incumbent upon the discoverers of an audible effect to show that the necessary mechanism within the ear to detect absolute phase does exist, but the necessary physiological evidence would be useful in validating such work. Mark Davis, in the Boston Audio Society Speaker (footnote 17), summarises work done on hearing mechanisms and how a change in phase on simple test signals could be detected. Although he doesn't deny the audibility of either phase reversal or the group delay of upper harmonics with respect to the fundamental on special test signals, he states that in connection with a proposed ear non-linearity phase polarity detection system, "The dynamic character of music is sufficient to render this effect inaudible. . . it seems fairly safe to conclude that distortion-dependent phase sensitivity has no effect on the reproduction of music, and can be completely discounted as a source of audible aberration." The finite resolving power of the ear's hair cell filters could also lead to a mechanism whereby phase shifting with respect to the fundamental of high order (10th) harmonics can be audible. Davis concludes, however, that "the audibility of this effect is about on a par with the phase/distortion mechanism, which is to say just barely audible under worst-case, laboratory conditions. Substitution of music . . . renders the effect inaudible by virtue of the masking ability of the music abetted again by its dynamic nature, and the fact that few musical instruments produce adequate energy at harmonics of sufficiently high order."

However, Davis here, in straying onto the audible effect of group delay has left the absolute phase argument and so this last piece of evidence, though interesting, is perhaps irrelevant (except to designers of linear phase loudspeakers). Certainly other workers have reported more positively on the effect of absolute phase reversal on music program. Ralph West (footnote 18) recalls an experiment using recorded bass drum beats (shown to have an asymmetric waveform on an oscilloscope) which had slightly positive results, while Lipshitz and Moncrieff both quote more easily audible effects using music as a source. However, whereas Lipshitz is very careful in extrapolating to the possible effects on music program, Moncrieff has no doubts at all, stating that (footnote 19) "experiments...have demonstrated that absolute phase polarity is far more audible than most people have suspected," and that "The characteristic sonic quality caused by phase inversion is so striking that it can be perceived through many other changed and changing sonic properties."

As indicated, only asymmetric waveforms should produce an audible difference upon having their polarity reversed, speech being the most obvious of examples. However, as all continuous waveforms can be shown by Fourier analysis to be equivalent to a combination of symmetrical—by definition—sinewaves, one might reasonably suspect that audible differences would most possibly be detected using program material consisting mainly of transients, which can be markedly asymmetric—Ralph West's bass drum, for instance, or handclaps. Similarly, plucked and struck instruments should all produce waveforms possessing such asymmetry, and yet the work carried out by Moncrieff gave inconclusive results when using such program material. Perhaps the transient in these cases is over too quickly for the ear/brain to be aware of whether it was positive-going or negative-going.

However, Stan Lipshitz did find that the handclaps on a DG Archiv disc, Golden Dance Hits of 1600 (2533 184), were particularly revealing. Richard Cabot, then with Tektronix (footnote 20) noted that music signals off-record that he was using for slew rate analysis had a distinct waveform asymmetry with, in general, the greater degree of asymmetry being associated with music played on a smaller number of instruments. This asymmetry polarity varied from disc to disc, presumably reflecting the arbitrary final phase condition of the record. I would have thought that this asymmetry was connected with the polarity of transients rather than of continuous signals, and this is what would be audible—if at all—but Moncrieff notes (footnote 21) that "the most dramatic difference occurs instead during sustained notes by a massed ensemble of human voices, strings, or wind instruments. . . The sonic effect of inverting the absolute phase is that of both a tonal balance depression and suck-out through the midrange and upper midrange. The altered tonal balance in turn provides psycho-acoustic illusions of an apparent loss of presence, solidity, and midrange clarity. There may also seem to be less midrange definition, less depth within the ensemble, and more depth between the listener and ensemble location."

Listening Tests
We have never been that positive ourselves about the audible effects of polarity reversal, but since then, in true pragmatic spirit, HFN/RR has been careful, whenever possible, to preserve absolute phase polarity in equipment reviews, apart from a lapse in November 1978 when we inadvertently carried out a listening test between the Quad 33 (inverting) and Hafler DH101 (non-inverting) preamps without maintaining the same phase polarity. A sharp reminder from Lipshitz (footnote 22) forced us to redo the tests, thankfully without having to modify our original conclusions. And after all, only on truly phase-coherent recordings will it matter, anyway. On a normal multimike recording, the absolute phase of even different sections of the orchestra might be opposite, so changing the absolute phase won't then introduce any significant overall improvement at all. However, compulsive hi-fi neurosis led to nagging thoughts that it might be important on enough records to do some further investigation. My colleague Ivor Humphreys designed a suitable line-level black box and we did some serious listening.

The results were interesting, to say the least. Following precedent, we decided first to see if any difference in phase polarity could be heard using an artificial, "worst case" signal, in this case a regular series (roughly 5Hz) of clicks derived from a simple CMOS astable circuit. Initially thinking it best to use a "linear-phase" loudspeaker, we decided on the trusty Quad Electrostatics, as a "time-smeared" design would perhaps introduce competing sonic effects. With the pulse width set to about 50ms, blind A/B-ing over a long period between the 5Hz pulse-streams producing compressions and a rarefactions showed that a difference could be reliably detected, but that the incidence of correct detection varied widely from listener to listener, being from 50% (no detection) to 100% (complete detection).

Ralph West had pondered whether the ability to detect differences in absolute phase was similar to that of "perfect pitch" and it was instructive that one listener detecting the change in absolute phase 100% of the time did have perfect pitch. He reported the subjective difference between the two phase states as being that of approximately a semitone change of pitch, and this would tie in with the B&O study, which also indicated that the change in absolute phase was interpreted by the listener as a pitch change.

Of course, it might be argued that it is not absolute pitch, per se, that is being detected, but a secondary effect; maybe the room itself reacts differently to positive-going pulses than to negative-going ones, or that the loudspeaker cabinet has different resonances excited in the two cases. However, repeating the pulse train experiment in another room with moving-coil speakers (B&W 801s) produced much the same results, as did repeating it in the open air and with headphones. And in any case, whether a primary effect of absolute phase, or a secondary effect, the fact remains that an audible difference is produced on the pulse train by reversing its absolute phase polarity.

An important point to note is that we didn't have any idea which way round was which, so the "lower-pitched" pulse-train could have been either inverted or non-inverted: this knowledge isn't relevant when trying to detect differences on such an artificial signal. On music program, though, one way round is "correct," as then the phase polarity will be that existing at the original microphone. However, if a difference can be heard when reversing the absolute phase of a record, then there is still no easy way of knowing which position would be correct, apart from making a cynical assumption that it must be that which makes the record more enjoyable. In the absence of evidence to the contrary, "reality" is the condition leading to increased happiness!

Following the click experiments, we carried out a number of blind A/B tests using both homebrewed speech recordings and commercial crossed-pair music recordings, as well as live applause conveniently provided by the BBC for the Prom broadcasts. This time, the results were not nearly so conclusive within the limited time permitted by a rapidly approaching copy date; Martin Colloms indicating that any audible difference was of the order of that between nominally identical drive-units, for instance. I wouldn't go as far as to say that the effect of maintaining absolute phase polarity is not that important, but more that it is up to the individual listener to determine the importance with his record collection, on his equipment, for himself, using something like the the HFN/RR phase box.

If you can then reliably detect a difference, then all that is necessary to do is to mark the relevant sleeve either "+" or "–," depending on which way round you prefer, remembering to press the appropriate button when you put the record on. To quote Lipshitz in his original BAS Speaker article: "How many of your records would sound better when reproduced with reversed polarity?" Only you can decide what the answer is.

---

Footnote 11: Letter to the Editor, HFN/RR, July 1978 p.83

Footnote 12: "Phase Inversion," J. Peter Moncrieff IAR 4, 1979 p.13.

Footnote 13: "On the sensations of tone," H. L. F. Helmholtz, Dover 1954.

Footnote 14: "Monaural detection of a phase difference between clicks," D. A. Ronken, J. Acoust Soc. Am Vol.47 p.109.

Footnote 15: "On aural phase detection," V. Hansen & E.R. Madsen, JAES Vol.22 No.1.

Footnote 16: "Detection of phase shifts in harmonically related tones", Cabot et al, JAES Vol.24 No.1.

Footnote 17: Comment on Lipshitz article, Mark Davis, BAS Speaker March 1978.

Footnote 18: Letter to the Editor, HFN/RR March 1978 p.65.

Footnote 19: Moncrieff IAR 4.

Footnote 20: "Measurement of audio signal slew rate," Richard Cabot, preprint 1414 61st AES Convention 1978.

Footnote 21: Moncrieff IAR 4

Footnote 22: Letter to the Editor, HFN/RR April 1979 p.95.