FIG.1: In crosstalk cancellation, sounds from the right speaker are alsosent to the left speaker (1) and delayed just enough to be out of phasewith the right-speaker sounds at the listener's left ear (2), whichmeans that only the right ear hears the sounds from the right speaker(3).
In EM's March 2001“Tech Page,” I discussed a technology called DolbyHeadphone, which convincingly simulates 5.1-channel surround sound inconventional stereo headphones. Now, the company is applying the sameidea to audio systems with two speakers. Dubbed Dolby Virtual Speaker,the system accepts a 5.1-channel input and creates the illusion ofhaving five speakers around the listening position.
Like Dolby Headphone, Dolby Virtual Speaker uses head-relatedtransfer functions (HRTFs) derived from acoustic studies with actualfive-speaker setups to re-create the sonic signature of each speaker,which depends in part on its location with respect to the listener andwithin the room. The Dolby HRTFs take into account both direct andreflected sound, and they allow the virtual speakers to be placed withpinpoint accuracy at 0°, ±30°, and ±105° at areasonable distance from the listener. By contrast, most othersurround-simulation systems use much simpler HRTFs based on the directsound only, which create only a general sense of envelopment withoutprecise speaker locations.
But Dolby Virtual Speaker must do even more, since the listener'sears are not isolated, as they are with headphones. To address thisproblem, the system uses a technique called crosstalkcancellation: sounds from the right speaker are also sent to theleft speaker, filtered to mimic the response rolloff caused by the headpartially blocking the higher frequencies, and delayed just enough tobe out of phase with the right-speaker sounds at the listener's leftear
(see Fig. 1). As a result, the left ear doesn't hear soundsfrom the right speaker, and vice versa. Of course, using two speakersinstead of headphones requires the listener's head to be in the sweetspot, equidistant from the speakers, for the simulation to beeffective. However, the Dolby algorithms allow much more latitude inthe listener's position along the centerline than other systems, thanksto the fact that the complex HRTFs allow lower-ordercrosstalk-cancellation filters to be used, which reduces audible phaseartifacts.
All this takes some serious DSP power to accomplish in real time.While most surround-simulation schemes assume there will be littleprocessing bandwidth available (constraining them to using simpleHRTFs), Dolby looked at dedicating lots of processing to the system.For example, the acoustic path from each virtual speaker to each ear issimulated by a 7,000-tap, finite impulse-response (FIR) filter. Thatmany large filters would normally require excessive processing powerand introduce pronounced latency, but Dolby's convolution algorithmsand the decreasing cost of DSP bandwidth eliminate these problems.
The Dolby Virtual Speaker algorithms have been implemented by allmajor IC manufacturers, including Analog Devices, Motorola, TexasInstruments, Cirrus Logic, and Sanyo. In addition, the algorithms canrun as software on a PC with an Intel Pentium II or III or an AMD K6-IICPU running at a clock speed of at least 400 MHz. The specific DSPrequirements vary from platform to platform, but in general, DolbyVirtual Speaker requires about 20 MIPS (million instructions persecond) and 32 to 64 KB of coefficient storage to operate in realtime.
The most obvious musical application is to use Dolby Virtual Speakerwith computer audio systems, which would allow composers in crampedquarters to monitor multichannel mixes without having to set up anactual 5.1 system. In the consumer-electronics realm, Denon hasannounced it will include Dolby Virtual Speaker in some of its A/Vreceivers, including the new D-M71DVXP. Clearly, this technology has alot of potential to bring the surround-sound experience to those whowouldn't otherwise be able to enjoy it.