From Albert Von Schweikert – 3/1/2008 6:47 AM THE ENGINEERING PROBLEM: DISTORTION AND COLORATION IN SPEAKERS For the past year, I’ve been working at the lab at Cal Tech - revisiting an engineering problem I discovered years ago. Greg, you and I discussed my work on driver integration and distortion reduction back in 2003 or 2004, when you reviewed the VR-4 Gen III. As you may recall, I had discovered a type of distortion that is not discussed in textbooks: waveform distortion caused by non-linearities in the cone caused by bending wave modes. In addition, there are additional distortions generated by the voice coil moving in the magnetic gap as the driver attempts to translate the electrical signal into air pressure. On top of all of these transducer-generated distortions, there is always the problem of integrating the “transient speed” and “timbre” of different sized drivers that have different tonal qualities due to completely different cone types and motor systems. To compound these problems, the sources of sound waves at different frequencies emanate from different parts of the baffle, creating a very difficult engineering feat to integrate all of these different waves into a coherent “picture” of the original sound field. By using a novel method to take pictures of the actual sound waves generated by the transducers, I have discovered that the reproduced waveforms are "warped" by non-linearities in the drive units, improper crossover design, and baffle reflections as the speaker system attempts to replicate the original wave form. The sum of these various "wave form" distortions are audible but extremely difficult to measure, since they are additive and create the colorations that are often taken for granted as a necessary evil of transduction (the conversion of electrical energy into mechanical motion). In addition, the distortion caused by multiple transducers that are fed “chopped up” signals at different frequency ranges compounds to the problems of making the speaker behave as a true point source that has the transient speed and pickup pattern of a recording microphone. "Conventional" distortion measurements such as those used to measure amplifier distortion don't show these colorations, since these distortions are not related to harmonic or transient intermodulation effects. We are talking about the warping of the actual waveforms of air pressure created by the musical instruments. PASSIVE SERVO CONTROL About a year ago, I developed a form of "passive servo control" applied to the circuit design to compensate for a portion of the waveform distortion that is amenable to correction via time correction equalization. Since I know what the distortion looks like in the actual air motion behavior (from the photos we took of the sound waves in the lab), I was able to "precondition" the signal via equalization in the crossover circuit. The circuit design required an intense study of the transient response and cumulative spectral decay of the entire speaker system, often called the “waterfall plot” by laymen. MEASUREMENT OF DISTORTION Below is an independent “waterfall plot” of the original VR-4 design, as measured by the Swedish magazine, HiFi & Musik. At this time (1996), this was (and still is) the best cumulative spectral decay measurement recorded, of any speaker system at any price (according to the measurements seen in magazines then or since). What is amazing is that this form of measurement does not completely describe the actual “sound” of a speaker system, even though this is a very important measurement. In fact, the new VR-5 Anniversary model sounds 100% better than the original VR-4 design, even though the Cumulative Spectral Decay, distortion curves, and phase measurements look very similar to this VR-4 waterfall plot, taken 12 years ago! This goes to show that measurements alone cannot begin to describe the sound of any speaker, so what is the ultimate test, anyway? In my opinion, our “live versus the speaker” A/B testing is the only way to determine if the speaker actually sounds like the original musical input, whether you are using a voice, acoustic guitar, harmonica, or just keys that you shake. We use a variety of different brands and types of microphones to average out the sound, so that we are not building a speaker that mimics any certain brand or model of microphone, even if that microphone is a $25,000 hand-built Dick Brauner mic made in CUMULATIVE SPECTRAL DECAY PLOT SHOWING TRANSIENT RESPONSE SPEED
CONE SELECTION FOR DISTORTION REDUCTION The next thing I worked on after discovering that the wave forms of actual air pressure were being “warped” by the speaker system was to design some method of reducing this distortion, and controlling the cone motion was the “key” to success in this area. I chose drivers that had extremely rigid and light cones, but was careful not to use cones that had high Q resonances that would not sound good due to metallic or ceramic resonances that would need to be reduced by filter circuits. That is why I rejected 99% of the drivers used by other manufacturers who like to use the metal tweeters and ceramic midrange cones: they ring! The resonances cannot be controlled by filters, since the filters only reduce the input signal at the offending frequency peak, the “suck out filters” don’t actually “cure” the ringing effects. My choice of woofer cone is the magnesium alloy manufactured by Excel of Norway, since this cone has no bending mode distortion and circumvents the common problem of motor distortion with its novel voice coil, top plate, and pole piece, which are exclusive to this motor. For the all-important midrange driver, A.A.C. of For the treble range, I have found no better driver for low distortion and coloration than the good old silk dome, and the one I like best is the Dual Ring Revelator by Scanspeak of Denmark. This tweeter has the cleanest treble response I have heard, cost-no-object and puts the diamond tweeters to shame! Although I love ribbon tweeters, they do not have wide dispersion and cannot be operated down to the lower midrange frequency range, making them “supertweeters” by default, and not a good choice with a large diameter midrange as used in the VR-5 Anniversary. INTEGRATION OF TRANSIENT RESPONSE AND REDUCTION OF ACOUSTIC WAVE FORM DISTORTION CREATED BY THE TRANSDUCERS AND BAFFLE LAYOUT In order to reduce the distortion and create a “synthesized” single point source speaker system that could mimic an omni-directional microphone, I found it necessary to design a specialized circuit that would reduce the wave form distortion, along with compensating for imbalances in the different transducer types used for bass-to-treble, as well as "harmonically” blending the different frequency bands together without creating additional forms of distortion from an overly complex circuit. Aha, the solution is never quite that simple, as we will find later in this discussion! My novel solution was to use a form of “servo control” in the crossover circuits that could be pre-programmed in advance, knowing what causes the distortion and what it looks like after it occurs. Note that I am not claiming to have invented this technique: an unknown Japanese scientist leading a team of researchers at a Japanese amplifier company invented the "look forward" circuit I adapted when they designed an active circuit some 20 years ago to reduce amplifier distortion. This circuit would “know” what the distortion was "in advance" so the error correction would be more efficient in time coherence. Essentially this is another method to implement servo control, but in advance of the occurrence of distortion, not "after-the-fact.” For instance, when you install a motion detection sensor on the voice coil, then use a feedback loop to attempt to stop the voice coil from going "non-linear," the voice coil has already gone non-linear and the circuit is attempting to push it back! You've already lost the war on bad sound that way and it will not easily work with midrange and treble drivers. Of course, this “old school" servo-control does work in the deep bass range, but the method is used primarily to control woofer motion, a la Infinity/Velodyne/et al. I had to first discover, then quantify, the waveform distortion. Finally, I had to design pre-EQ correction for the non-linearities by using good ol’ circuit design and application the old-fashioned way: using my brainpower and thought process. As many other scientists in this industry realize, circuit design is almost a “lost art” and has been replaced by software programs that attempt to help the “speaker designer” come up with a suitable circuit to chop the music into bass, midrange, and treble frequency bands. Then the “speaker designer” sends these chopped signals into drive units that attempt to reassemble the signal into music. This technique does not work as well as “speaker designers” believe! No software program and a $300 sound card is going to teach someone how to actually design a good speaker, it takes years of experience, years of education, and years of R&D, along with some really good ideas and novel thinking process. HOW THE SERVO CONTROL BACK EMF CIRCUIT WORKS The actual implementation is accomplished by sensing the back electromotive force of the cone motion, not by using an active motion sensor attached to the cone. My circuit design is similar to Zobel impedance leveling circuits, in the fact that they are not in the direct signal path, but are in the ground return circuit path from the speaker back to the amp on the “negative” side. Since we have a “perfect” signal going into the crossover circuit that can be compared to the signal reproduced by the transducer, it is a simple job (if you know how to do it) to design a comparator circuit that can equalize the “difference” signal created by the transducer distortion. Note that I'm not “just” EQ'ing the driver impedances, it involves the complete architecture of the “crossover” circuit that “blends” the bass, midrange, and treble frequencies together at the transducers point of wave launch on the outside of the baffle. SPEED CONTROL??? Since the transducers all have different mechanical and electrical responses that differ wildly from the woofer motion to the tweeter motion, i.e, the “transient responses,” there had to be some method of “forcing” the different transducers to behave as a single driver. Since the waveforms of the tweeter are generated at a faster speed due to lighter moving mass, it is reasonable to believe that the woofer response and tweeter response cannot be the same. The heavier woofer cone cannot accelerate and de-accelerate as quickly as the tweeter diaphragm at their respective crossover points, and you can’t “speed up” the woofer unless you do this actively with a push/pull motor system, deemed too complicated and expensive to implement passively. My method “slows down” the tweeter to match the speed of the woofer (just kidding, but barely). NEW TYPE OF CROSSOVER DESIGN THAT EQUALIZES THE TRANSIENT RESPONSES How does it work? Simple, by shaping the “Q” of the crossover filters to induce “time shifting” (alternately called “phase shifting” if it’s done incorrectly). Since the signal has to go through the crossover circuit before it reaches the drivers, you can “shape” the signal if you know enough about circuit design, how capacitors store energy, and how the relationship between the inductors and capacitors affect the combined transfer function of the overall target response. There are two types of impedances generated by the circuit and the transducer’s motor: real and “imaginary” impedance. By varying the relationship between the inductance and capacitance of a reactive circuit, i.e., phase lead and phase lag based on how capacitors and inductors “work in the real world,” you are able to “shape” the curve of the crossover slopes and generate a “control signal” that will combine the different transducer response shapes into a coherent signal. The “Q” of the circuit is based on how sharp a roll-off you need in order to reassemble the different frequency bands you’re working with into a “coherent” signal once the voltage is applied to the transducers. When the transducers attempt to reassemble the waveform, you need to take into consideration the phase lead and phase lag of BOTH crossover and transducer responses. This is where my proprietary technology comes into play, and I’ve worked on this since 1976 at Cal Tech: 32 years. I’m not willing to reveal all of my secrets just yet: patents need to be applied for (so my investors will feel safe in knowing that we’ll have an exclusive on this type of “new” sound). I believe the VR-5 Anniversary model is the first passive speaker to have accomplished this level of linearity. You can hear the lowered noise floor, increased spatial dimensional cues, and enhanced transparency by just listening to the new design versus listening to any other speaker - you have heard what I'm describing! Your comments are very similar to what all of us here keep saying: this speaker design simply sounds better than a conventional design, but it’s difficult to describe. My best guess is that the sheer lack of coloration/distortion based on a more linear transfer function enables the amazing improvement in “realism” although this still remains to be measured! CONCLUSION Yes, once again I have postulated some outrageous ideas on speaker design and will no doubt have the newsgroups buzzing about the validity of my approach. I say to anyone claiming that my method cannot work: buy a pair, listen to them, and be prepared to believe in a New World Order. Albert Von Schweikert - Copyright 2008 |
