I Have Heard the Future...(DSP Room Acoustics Correction) Page 2
Fig.1 Impulse response of Snell B-Minor at the 1992 Summer CES (23.2ms time window). Note the multiple reflections of the speaker's direct sound from the room boundaries.
Fig.2 Snell B-Minor, in-room frequency response at the listening position. The major irregularities are due to the speaker's direct sound being interfered with by the boundary reflections.
Fig.3 Snell B-Minor, uncorrected in-room cumulative spectral-decay plot. The impulse decays rapidly at high frequencies, but note the deep valley near 50Hz and the persistent peak near 100Hz.
Then Voecks switched a SigTech AEC-1000 Acoustic Environment Correction filter into the signal path (footnote 2). This device, produced by Cambridge Signal Technologies, is a programmable digital filter. It uses a Motorola 56001 DSP chip and a fistful of FIR filters that synthesize an "inverse room" in the digital domain. During installation the impulse response of the speaker/room system is measured, and a separate computer calculates the parameters of the required inverse-room filter. The parameters are stored in the AEC-1000 unit, which synthesizes a 2500-tap digital filter that can adjust the response at more than 1000 individual frequency points.
What really matters, of course, is not the frequency resolution but the fact that the filtering is a time-dependent process. Since the heart of the filter is a multi-tap digital time-delay circuit, the processor causes the speaker to emit a succession of wave launches that have the same time-delays as the room reflections. Each delayed signal arrives at the listener's chair simultaneously with a room reflection, but with an inverted pattern of peaks and valleys (fig.4). Voecks displayed the waterfall plot of the correction filter on the computer screen (fig.5): its first-arrival response was nearly flat, but later curves (at intervals of about 5ms) exhibited increasingly deep notches at about 100Hzthe inverse of the 100Hz standing-wave peak seen in the initial speaker/room waterfall plot. The later curves also contained some low-bass boost to compensate for the 32Hz standing-wave null that occurs near the middle of the room.
Fig.4 Frequency response of the correction filter calculated by the SigTech digital signal processor.
Fig.5 Cumulative spectral-decay plot of the DSP correction filter.
By now you may have guessed the result. When the digital filter was switched into the signal path, the valleys and peaks in its delayed responses canceled out the peaks and valleys in the room's delayed reflections (fig.6). I should stress this point, to clear up any potential misunderstanding: the DSP filter does not try to entirely cancel the room's reflections, simulating an anechoic environment. (That would not sound very good, as you can prove to yourself by setting up your system outdoors.) The digital filter simply cancels the peaks and valleys due to the reflections. The room acoustics are still there, but the reflections and standing waves no longer color the sound. Voecks showed the waterfall plot of the entire system with the filter in-circuit (fig.7): the first-arrival and all of the delayed arrivals exhibited nearly flat response. In effect, the DSP preconditions the musical signal going to the speakers so the signal arriving at your head mimics the sound of an ideal speaker in an ideal room.
Fig.6 Snell B-Minor, DSP-corrected in-room frequency response at the listening position. The rise in the low bass may indicate over-correction.
Fig.7 Snell B-Minor, DSP-corrected in-room cumulative spectral-decay plot. While the correction is not flawless, the system's transient response can be seen to be dramatically improved. After the initial impulse, the sound decays evenly, with only a slight hangover in the bass and low-midrange.
This may seem like a purely technical feat, but with music, the subjective effect of this correction was mind-boggling. Pipe-organ pedals that were barely perceptible without the filter were felt and heard clearly (even in the middle of the room) when the Acoustic Environment Correction was switched in. More important, the elimination of the 100Hz standing-wave peak removed a dense, turgid quality that affected virtually every recording. The entire soundstage opened up: the psychoacoustic "masking" caused by excess midbass was taken away, exposing formerly covered instrumental lines, midrange details, ambience, and depth. Recordings of male vocalists (Joe Williams, Mel Tormé), which I had always thought were made with mediocre cardioid microphones, turned out to be far more enjoyable than I believed possible. Bass textures (organ, bass fiddle, cello, baritone), which are thickly colored in many playback systems because of the rooms' midbass peaks, were rendered in a far more relaxed, open, and lifelike way. The sound had a freedom from coloration that I usually hear only when I set up "near-field" monitors at a recording session.
There is a small population of very lucky audiophiles for whom this process might not produce a dramatic benefit. If you have a very large listening room with irregular walls, if you have managed to tame your room's standing waves with active bass traps and a lot of experimentation, or if you sit very close to your speakers in order to hear mainly the direct wavefront with little contamination from the room, perhaps you don't need this correction. But boundary reflections and standing waves have compromised the performance of every audio system I've ever owned, in every place I've lived, and most other systems I've heard elsewhere. In my judgment, acoustic environment correction is the single most important advance in audio since the CDperhaps the most important advance since the advent of stereo.
The basic idea of using adaptive digital filtering to compensate for speaker/room interaction was devised ten years ago by Robert Berkovitz and Ron Genereux at Acoustic Research in Massachusetts (footnote 3). Articles were published describing the concept, but digital signal-processing hardware with sufficient speed and power did not exist at that time in a practical and affordable form, so the method was patented and shelved. Three years ago it was revived, and a spinoff company was formed to create a working product. Ex-AR engineer Ron Genereux headed the development team, and the SigTech AEC-1000 is the result. But while it can be used in a living room, it was designed to be a pro-audio product, especially for correcting acoustical problems in recording-studio control rooms. The price is $10,000 plus an IBM-compatible PC. A more affordable consumer version of the SigTech may be developed next year.
When Kevin Voecks demonstrated the SigTech processor at CES, his purpose was not to sell the product but to provide a preview of the remarkable transformation in sound quality that DSP makes possible. DSP products for home use are being developed on several continents. For example, major investigations of speaker/room interaction were launched three years ago in Europe and at Canada's National Research Council, with the expectation that the resulting knowledge will be used both to design more room-compatible speakers and to design DSP equipment to correct for the room. Engineers at Philips in Holland, Matsushita in Japan, and elsewhere are currently working on several DSP applications that will reach the market during the next two years.
In the US a consortium of developers joined forces under the AudioSoft banner. They include Snell Acoustics (whose loudspeakers have always been designed with particular attention to speaker/room interaction), engineer Doug Goldberg of Audio Alchemy (an authority on digital hardware design and efficient manufacturing using surface-mount circuitry), software developer John Solar, and a fourth member whose identity has not been announced. While other developers are trying to develop DSP systems from the ground up, AudioSoft has leapfrogged rapidly to designs that are nearly ready for production, thanks to John Solar's experience with computer-based signal analysis for military applications. Digital processing to correct for room coloration is based on principles that already are widely used in military signal-analysis computers to cancel spurious echoes in radars and undersea sonar-mapping systems. (Many of the technical papers in this field, at least those that are not classified secret, have been published in the Journal of Underwater Acoustics.) Practical DSP correction for room acoustics may turn out to be yet another beneficial consumer spinoff from our military tax dollars, along with personal computers and Teflon frying pans.
I have focused on using DSP to correct for coloration caused by room acoustics because that is the most dramatic benefit of this technology. It can also be used to correct some of a speaker's built-in imperfections (footnote 4). For example, all woofers roll off below some low-frequency limit. The steeper that rolloff, the greater the associated group-delay (phase shift) in the octave above the cutoff frequency. As a result, transients from a bass-reflex speaker tend to sound soggy next to the taut, solid bass produced by a low-Q acoustic-suspension woofer. Sharp infrasonic filtering in preamps and amplifiers can have a similar unwanted side-effect. A few years ago designer Laurie Fincham at KEF demonstrated that the quality of bass reproduction could be improved by equalizing a woofer's response all the way down to 10Hz, because the EQ flattened its group-delay. Amplitude and phase are tied together in analog signals, but a DSP system can correct them separately, removing group-delay without boosting the bass to excess. DSP can also fix crossover phase problems and equalize a speaker's overall response, if it isn't already flat.
According to Voecks, the first DSP speaker/room processors for the consumer market will be introduced by Snell Acoustics within a few months (footnote 5). The "black box" from Snell Digital will accept plug-in software cartridges that provide specific phase and amplitude corrections for Snell loudspeakers (including older models no longer on the market). When you buy the processor, the dealer will bring a measuring microphone and a portable computer to your home and measure the speaker/room transfer function at your chair. At your choice, the measurements can be averaged over a broad listening area or can be concentrated at one chair. A program will calculate the required filter parameters and load them into your processor, after which the dealer can take the computer away. This calibration process can be repeated whenever you rearrange your room, buy new speakers, or move to a different house. Eventually, of course, the AudioSoft programs will be marketed so that you can use them with your own computer.
Next year Audio Alchemy will market a general-purpose version of the processor, usable with all loudspeakers (footnote 6). Audio Alchemy's involvement is likely to mean a dramatic reduction in the price of this complex processing. And as DSP speaker/room correction becomes widespread, loudspeaker designers may have to rethink how they should tailor a speaker's on-axis and off-axis responses so as to make the best use of the processing. For example, built-in crossovers may become optional, with separate input terminals for woofers and tweeters; once you have a DSP computer for room correction, it can easily be programmed to do a phase-perfect crossover too.
I have heard the future, and the sound is wonderful.Peter W. Mitchell
Footnote 2: This brand name, with a capital letter in the middle of the word, betrays the product's parentage in the computer industry, where names like WordStar, SideKick, and CompuPro are widespread.Pete W. Mitchell
Footnote 3: I heard the AR ADSP (Adaptive Digital Signal Processor) successfully demonstrated in the spring of 1982 (see "Music by Numbers, the AR ADSP," HFN/RR May 1982), but, due to the speed limitations of DSP technology a decade ago, it could only apply a correction for room problems below 500Hz. The problem of recombining the corrected low-frequency signal with the uncorrected higher-frequency one was, I believe, never satisfactorily solved in the ADSP, though it did indicate the future path audio reproduction was eventually going to take.John Atkinson
Footnote 4: Though a Dallas company, Audile, showed a prototype loudspeaker at the 1991 SCES which used DSP to optimize its sound quality, the first digital equalizer commercially available in quantity was Celestion's DLP-600, which I reviewed in August 1992 (p.145). This black box adjusts the phase and frequency response of the Celestion SL600Si loudspeaker above 500Hz to be optimal when the listener sits on the intended axis, but doesn't do anything about the speaker's low-frequency performance. As a rule of thumb, the lower in frequency you want to extend your full-range correction with DSP, the more coefficients the digital filter has to have, meaning that there has to be more computing power, with a concomitant increase in cost.John Atkinson
Footnote 5: The Snell correction box metamorphosed into a device from NAD, then made its ultimate appearance as a ready-for-primetime product from Tact.John Atkinson
Footnote 6: The Audio Alchemy processor wasn't made available befire the company went belly-up. However, it resurfaced as the Perpetual Technologies P-3A, though the room correction software never really got out of beta testing.John Atkinson