By J.D. HIRSCH
IS PHASE SHIFT AUDIBLE?
Can audio phase shift be heard? Many people are convinced that phase distortion is not only audible, but may be largely responsible for the less-than-perfect sound reproduction of some of our most highly regarded loudspeakers (or amplifiers, phono cartridges, tape recorders, etc.).
Others, with equal fervor, deny this.
Phase-shift distortion (also known as nonlinear phase response or time-delay distortion) occurs when the different frequencies embodied in a complex audio signal reach the listener's ears at different times. It is a property of most loudspeaker systems, to a greater or lesser degree. Phase-shift distortion can be produced electrically in the crossover network, mechanically by the physical construction of the drivers, and acoustically by the spacing of the drivers.
(Phase-shift distortion does not occur significantly in electronic components such as amplifiers.) The shape of a complex waveform can be drastically changed by a shift in the relative phases of its different frequencies, even if the relative strengths and amounts of the various frequencies are not altered. For example, a square wave can become a sharp spike.
The results of an extensive series of investigations in the research laboratories of the well-known Danish manufacturer Bang & Olufsen were presented in a paper delivered last year before the Audio Engineering Society in Rotterdam.
The paper's authors, Erik R. Madsen and Villy Hansen, were in charge of a research program that was designed to determine the threshold of phase-shift detection of human hearing and to obtain quantitative results if possible. Their findings are not only significant, but in some cases quite surprising.
Madsen and Hansen first conducted listening tests with a loudspeaker whose phase shift was such that it produced a roughly triangular-wave acoustic output from a square-wave electrical input.
Using an ingenious corrective method to obtain a true square-wave output from the speaker, they alternated the two signals and found that trained listeners heard a distinct difference in timbre, even though a spectrum analysis (which is insensitive to phase) showed the two signals to be identical in respect to their frequency content.
Next, they produced a transient signal with continuously variable phase shift and constant total energy- specifically, a single-cycle sine-wave burst with an adjustable zero-reference line. The out put of a high-quality electrostatic loud speaker driven by this signal was judged (in an anechoic chamber) by the listening panel. Their findings not only confirmed the original conclusions, but unexpectedly showed that absolute phase is significant. Apparently the ear favors a negative pressure impulse, so that a transient that moves a speaker cone inward will be heard with greater subjective accuracy than if the speaker leads are reversed to make the cone move out ward. At this time, no attention is being given to phase in the recording process from microphone to loudspeaker (except to avoid frequency cancellations), but Madsen suggests the possibility of a genuine improvement in reproduced sound if absolute phase is standardized.
The major effort of the B&O research team was devoted to obtaining quantitative relationships defining the audibility of phase-shift distortion. Tests were made in seven frequency ranges from 100 Hz to 10 kHz, using a number of sound-pressure levels (SPL). At each frequency, the listeners could choose from among five steps of increasing phase shift until a definite change of timbre could be heard. Wide-range electrostatic headphones were used to eliminate environmental effects.
The averaged family of curves from this test has a shape somewhat reminiscent of the familiar Fletcher-Munson equal-loudness contours. In this test, the ear was most sensitive to phase shifts at about 800 Hz, with decreasing sensitivity at lower and higher frequencies. The sensitivity to phase changes, at any frequency, increased with the SPL. It is interesting to note the magnitude of the effect. A mid-range phase shift of less than 5 degrees could be detected at an SPL of about 80 dB, while at a 61-dB SPL the shift had to reach 15 degrees before it was audible.
The tests were then repeated in a normally "live" room, using a wide-range electrostatic loudspeaker. The general shape of the resulting curves was similar to those obtained with headphones, but the region of maximum sensitivity was fairly uniform from about 200 to 1,000 Hz. The magnitudes of the minimum detectable phase shifts were about the same, but they occurred at much lower sound-pressure levels. For example, a 5-degree shift could be heard at an SPL between 65 and 70 dB, and a 22-degree shift was detected at an SPL of only 50 dB (a much lower level than would be used in listening to music at home).
Summarizing the findings of Madsen and Hansen, it appears that:
1. Phase-shift distortion can be heard.
2. Sensitivity to phase-shift distortion is greater in a reverberant (or nor mal) environment than in an anechoic (acoustical test-chamber) environment.
3. Sensitivity to phase-shift distortion increases with SPL, and is generally detected earlier at frequencies to which the ear is highly sensitive.
4. A speaker with poor transfer characteristics (nonlinearities and inadequate transient response) is more revealing of phase variations than a speaker of better quality.
5. Absolute phase is a significant factor in the realistic reproduction of transient waveforms.
Obviously, much more remains to be done in this field, especially in establishing a correlation between phase and impulse measurements and subjective listening tests. For one thing, Madsen used a rather simple, artificial test signal which had advantages in analysis but certainly was not representative of the usual program content of music or speech.
In this country, the importance of eliminating time-delay (or phase-shift) distortion has been stressed by only a few speaker designers. Intuitively, it would appear that a speaker capable of generating an acoustic transient waveform whose shape accurately reflects that of the electrical driving signal should be a superior transducer. In our experience, few speakers can approach this ideal, and we have not yet established any firm correlation between this sort of transient-response capability and the overall subjective quality of the speaker's sound.
At the conclusion of their paper, the Danish researchers indicate that their investigation ". . . shows seemingly more correlation between phase and impulse measurements and subjective quality of loudspeakers than many other kinds of objective measurements." If so, I look forward to the development of such measurements into a useful laboratory tool for speaker evaluation in the future. But my own view at this time subject to change-is that although low time-delay (or phase) distortion is certainly a desirable quality of a loudspeaker, it is not in itself sufficient to define an accurate reproducer. Those speakers we have tested that proved to have superior transient properties have been very good by any standards, but there have been many others lacking this capability that nevertheless managed to sound just about as good, if not better. It seems likely to me that in most cases, the other usual aberrations of loudspeakers, such as nonlinear distortion, modulation distortion, erratic polar characteristics, overall energy output versus frequency, and certainly many others, tend to swamp out the potentially audible effects of time-delay distortion. When the other distortions are reduced significantly, I do not doubt that the speakers with the best phase characteristics will then be able to demonstrate a clear superiority in audible performance.
----------------
Also see:
TESTED THIS MONTH: Hitachi SR-5200 Receiver Grado FTR+1 Phono Cartridge Bib Model 45 Record Cleaner Akai GX-285D Stereo Tape Deck
PERFECTING SOUND REPRODUCTION--Are discs, tapes, and FM broadcasts as good as they might be?
CONTEMPORARY MUSIC IN FRANCE--A report on the "second degree" at the "Paris Autumn" Festival, ERIC SALZMAN