We're Speaker Art, and as the name suggests, we build loudspeakers. We've been doing so for some two decades now, quietly refining the expression of our art while satisfying a small but highly discriminating clientele. Here we'd like to share with you some of the hard won insights we've gleaned in the course of our extended efforts to elevate loudspeaker performance beyond the norm.
Loudspeaker drivers--even the best of them, and we use the very best--make music reluctantly, constantly trammeled by their own inertia, stiffness, and internal resonances. On the other hand, loudspeaker cabinets make music all too exuberantly, resonating haphazardly like a chorus of tipplers in a Karaoke bar. The normal behavior of both subsystems, drivers and enclosures is obviously undesirable.
If speakers in a box are to do anything more than going through the motions of reproducing a music signal, both the drivers and the enclosures have to be placed under severe operating constraints. The drivers themselves must be restricted to the operating range where their behavior is linear, while the enclosure must do nothing more than enclose an appropriate volume of air, and must exhibit no audible mechanical resonances of its own. If both conditions can be met-a very big if-a significant increment in performance can be realized over that of conventional design, including-sad to say-almost all audiophile products.
Achieving these two performance objectives, namely restricting drivers to their most linear frequency ranges, and quelling cabinet resonances, is extremely difficult, and normally imposes unpalatable trade-offs upon the designer. The normal response of the design community has been to settle on various half measures, but the Speaker Art approach has always been otherwise. We have sought to eliminate compromise rather than to embrace it, and that objective is best met by avoiding the design conundrums requiring compromise. And that in turn demands nothing less than reformulating the terms in which the basic engineering issues are framed as well changing the context in which they occur.
Such a radical rejection of conventional design principles has thrust us as a company into areas outside the normal range of engineering disciplines and standard construction techniques utilized by loudspeaker system designers, and has necessarily required a far greater expenditure of time and effort than textbook approaches. But we think the results have been amply justified.
A Fundamental New Design of Crossover Network
clef Loudspeaker designers of conventional opinions play a depressing zero sum game when they set out to implement a crossover network. Ideally one would want the crossover to maintain constant acoustical power and flat on-axis response, preserve absolute phase linearity through the audio band, minimize vertical and horizontal lobing, and attenuate driver output very sharply beyond the designated passband, unfortunately these goals cannot be met in their entirety by a high level crossover, and normally any optimization in one or two areas will result in considerably degraded performance in some other area.
The majority of loudspeakers made today utilize classic Butterworth filter types, which provide for maximally flat on-axis response, but, which except for the slow slope, first order variety, suffer from nonlinear phase response. An alternative, which is rapidly growing in popularity, is the fourth order Linkwitz-Riley type, which offers steep filter skirts and well-controlled, highly symmetrical off-axis output, but which exhibits very poor phase behavior and some nonlinearity in the amplitude domain as well. When utilized skillfully, both types of filters will permit reasonably smooth driver integration but to entail what we believe to be unacceptable tradeoffs, particularly in the time domain-tradeoffs that are commonly manifested in blunted transient attacks, compromised imaging, and an unacceptably narrow listening window.
What can be done to extricate the designer and ultimately his customer from this dismal predicament?
No matter what anyone tells you, the fundamental tradeoffs engendered by this basic behavior of electrical filters cannot be entirely eliminated in any design. Such tradeoffs can be rendered benign, however, and that's where unconventional filter topologies must be sought.
The filter topology that we utilize in all of our speaker systems is know as the Kaminsky fourth order series network. Named after the mathematician who described it, this filter type, provides for both an unusual and highly advantageous set of trade-offs, as well as an unprecedented degree of practicality. Yet in spite of its manifest advantage, it has never been successfully employed by any other manufacturer-we at Speaker Art have spent literally years of engineering time in achieving practical realizations of the design, and we currently use the topology across 80% of our speaker line. We are the only speaker manufacturer in the world to do so.
Since series networks are admittedly exotic, here we offer a brief explanation:
In a series network, all filter elements are placed in a single electrical path between hot and ground-a path that includes the drivers themselves. Because every element in the filter loads every other element, the number of variables can be enormous, and thus series networks of any complexity become extremely difficult to design. Consequently, in virtually every previous instance where a series network has been used in a commercial product, the filter has been comprised of a capacitor and a coil only.
Such simple "quasi-second-order" series filters are hardly ideal, but they exhibit a number of uniquely desirable properties-which led us to explore the large subject of series networks in the first place.
The quasi-second order crossover maintains full amplitude linearity through the crossover region, and exhibits a rising rate of attenuation beyond the crossover point reaching a maximum of 12dB per octave. Moreover, phase linearity is better than for any other crossover type excepting the first order Butterworth.
The combination of good on-axis phase and amplitude response, and moderate driver attenuation has recommended the quasi-second order to a number of noted designers, but the industry in general has found the filter slope steepness insufficient, and most designers have experienced severe difficulties in utilizing the topology when the impedances of the drivers themselves have not been substantially identical. Thus the quasi-second order crossover has earned at best only a small cult following.
Still, when we launched our crossover research program some fifteen years ago, we were sufficiently intrigued by the virtues of the quasi-second order filter type to investigate higher order series networks. Unfortunately, at the time such high order filters existed only as theoretical possibilities. No one had ever troubled to characterize them mathematically, and no one, to our knowledge, had ever attempted to build one. We were venturing into completely unknown territory.
Exploring that territory was a time consuming and often frustrating enterprise, but when, after extensive experimentation, we succeeded in modeling the behavior of a cascaded series of quasi-second order filters, we found that we could achieve frequency division in a more nearly ideal manner than we had ever believed possible.
Indeed, the design exhibits a well nigh incredible array of virtues. The fourth order series filter-perhaps quasi-fourth order is the more accurate term-markedly improved upon the quasi-second order variety by providing steep filter skirts (exceeding 30dB) with resultant enhancements in driver behavior and in off-axis system response, but it does so with approximately half the component count of a parallel Linkwitz-Riley filter.
Amplitude response is nearly flat through the crossover region with just a slight ripple, and phase shift is extremely linear over the audio band approximating a Bessel filter function. As with any fourth order crossover system, a 360 degree phase rotation ultimately occurs, but the rotation is gradual, exhibiting none of the violent and audible phase discontinuities that characterize the behavior of the common fourth order Linkwitz-Riley type.
Asymmetrical Guassian parallel filters can in theory achieve similar results, but practical implementations have proven illusive. The Kaminsky filter, on the other hand, lends itself to real world applications with a wide range of drivers, and, very interestingly, provide the invaluable side benefit of virtually foolproof tweeter protection without the inclusion of sonically deleterious fuses. The advantage of the Kaminsky filter are real and not simply theoretical, as is the case with a number of textbook minimum phase filters. And they make for simply better sound.
All of our speakers are individually hand assembled and all subassemblies are fully tested and evaluated prior to final inspection. Each speaker is subjected to a thorough listening evaluation where it is compared to a laboratory reference as well as being closely matched to its own partner. Because we build in small quantities we can provide a degree of customization within our line, including not only custom finishes, but in some cases special voicing.