So far, we’ve been talking about a monaural phantom image, that is derived from two identical signals coming from each speaker of the stereo pair. The phantom becomes more interesting when we start thinking of it as a stereo phantom image. It becomes stereo as a function of short time delays between the left and right speakers. These delays have a very powerful impact on the phantom’s location, and we can exert some rather powerful control (much more so than with a pan-pot) over where the phantom is localized and the stability of that localization (the phantom doesn’t even have to be between the speakers - a fact which has led to the development of systems like Q-Sound and the Roland RSS space simulator!).

Back in about 1950, Helmut Haas published a paper about the audibility of early echoes that lead to the term: “The Haas effect” (also known as “the precedence effect” and “the law of first wavefront”). Basically, the effect is that we localize a sound based on the angle of arrival of the direct sound, and we ignore the angle of arrival information of all delayed versions of that sound (i.e. room reflections). This means that we don’t hear room reflections as such, so long as they arrive quickly enough after the direct sound. The window of time for this is from about 1 millisecond to somewhere between 10 and 50 milliseconds, depending on the kind of sound. In nature, you never have the exact same sound coming at you from two directions at exactly the same time. One of the versions is always a reflection. The invention of the loudspeaker changed this.

What this means for stereo is: a monaural signal that is undelayed to either speaker is perceived to originate at a point in space compatible with the idea that the sounds from both loudspeakers are reflections (the non-existent direct sound is inferred from these reflections - this is why it is a “phantom” image); a monaural signal that is delayed by less than seven tenths of a millisecond to one of the two loudspeakers implies reflection paths from a different location in the room. It will as a result cause the phantom image to shift toward the earlier speaker. The amount of shift in localization is related to the amount of time delay. I’ve noticed (fooling around with an SPX 900, for instance) that the phantom shifts quite reliably in linear increments from the center to the earlier speaker as I step through .1 millisecond delays from .1 ms to 1 millisecond. This is consistent with what Haas (and many others) found.

	Approximate locations of phantom image for various delay times. With a 1 millisecond delay, the phantom appears to emanate from the earlier speaker. Longer delays yield erratic results, with the phantom emanating from various places, depending on the acoustics of the room and the spectra of the signal, among other things. By 10 milliseconds, the late signal begins to be heard as an echo, depending on the envelope of the signal.

These stereo phantom images are quite realistic and powerful, and when coupled with some extra early delays (less than 50 milliseconds) sent to left and right speakers, give a sense of liveness to the recording that is quite remarkable.

Delay time between speakers in milliseconds	localization effect
.1 to .7 ms.	- causes phantom image to migrate toward earlier speaker
.8 to 10 ms.	- causes phantom to appear erratically around the sound stage. often ambiguous
10 ms.	- short sounds begin to be heard as two distinct sounds, original and echo, while long sounds are still ambiguous, hard to localize
50 ms.	- long sounds begin to be heard as two sounds, original and echo

Figure 3. Impact of delays between left and right speakers for otherwise identical signals.