Subscription Information and Rates
First of all, you don't absolutely need one of our printed subscription blanks. If you wish, simply write your name and address as legibly as possible on any piece of paper. Preferably print or type. Enclose with payment. That's all. Or, if you prefer, use VISA or MasterCard, either by mail or by telephone. Secondly, we have only two subscription rates. If you live in the U.S., Canada, or Mexico, you pay $24 for four consecutive issues (scheduled to be mailed at approximately quarterly intervals). If you live in any other country, you pay $38 for a four-issue subscription by airmail. All payments from abroad, including Canada, must be in U.S. funds, collectable in the U.S. without a service charge. You may start your subscription with any issue, although we feel that new subscribers should have a few back issues to gain a better understanding of what The Audio Critic is all about. We still have Issues No. 11, 13, 14, and 16 through 20 in stock. Issues earlier than No. 11 are now out of print, as are No. 12 and No. 15. Please specify which issues you want (at $24 per four). One more thing. We don't sell single issues by mail. You'll find those at somewhat higher cost at selected newsdealers, bookstores, and audio stores. Address all subscriptions to The Audio Critic, P.O. Box 978, Quakertown, PA 18951-0978. VISA/MasterCard: (215) 538-9555.
From the Editor/Publisher:
Issue No. 20 was optimistically dated Late Summer 1993 but was mailed in the second week of the fall. Unforeseen delays resulted in one omitted quarter; hence the more realistic Spring 1994 dating of this issue. The staff expansion I so fondly previewed is still in the incipient stage. With all the reviews in this issue, it was suggested to me that I split it down the middle to make it into two issues (maybe I should have), or label it a double issue in fulfillment of two quarters of a subscription (I would never do that). Yes, there will be a Summer 1994 issue, possibly even sooner than you think.

THE AUDIO CRITIC

Box 978

Letters to the Editor

In the last issue your Ed. griped in this space about uninteresting letters seeking advice on purely private purchasing plans (alliteration unintended). Now I want to gripe about reasonably interesting but unpublishable letters that ramble on for seven or eight illegibly handwritten pages, propounding the correspondent's opinions on eleven different audio subjects. What is the purpose of such a letter? What am I supposed to do with it? Get a life, guysor get a word processor. Letters printed here may or may not be excerpted at the discretion of the Editor. Ellipsis (.) indicates omission. Address all editorial correspondence to the Editor, The Audio Critic, P.O. Box 978, Quakertown, PA 18951.

shows this analysis technique for the case of a conventional D/A converter with two different input frequencies and no interpolation filter. Figure 4a shows a 1 kHz sine wave sampled at 50 kHz. The difference between jittered and nonjittered D/A outputs is seen to be a series of narrow spikes whose width is proportional to the instantaneous difference between the arrival time of the ideal clock and that of the actual jittered clock, and whose height is proportional to the change in signal amplitude from the previous to the current sample. Here we show the case for white phase jitter, which does not accumulate over time. Note the modulation of the error spikes by the signal slope, which causes the error to become very small at the signal peaks. We are now in a position to analyze the added noise and how it relates to the signal. The spectrum will be white, as there is no statistical relationship between one error pulse and the next. The rms amplitude of the noise spectrum is related to the average slope of the DAC output signal, as large step sizes between adjacent samples cause large error pulses in the error signal. This fact can be seen clearly in Figure 4b, where a higher-frequency signal (6 kHz) has been applied to the D/A converter, resulting in larger step sizes between adjacent samples and hence larger error pulses. We can summarize by saying that for white phase modulation of the clock, the D/A output will be corrupted by white noise whose rms amplitude varies with the average slope of the signal. The worst-case signal for audio would therefore be a full-scale 20 kHz sine wave at the D/A output. The situation is slightly different when an interpolation filter is used in front of the D/A converter. Analysis of that is beyond the scope of this article, but the results are simple: the sensitivity to white phase jitter is reduced in direct proportion to the oversampling ratio. This means that a D/A converter using a 16x interpolation chip will be four times less sensitive to jitter than one using a 4x oversampling filter (assuming that the absolute jitter in ps is the same
for both clocks). If the jitter is not white phase jitter but rather a relatively slow variation of the clock frequency (lowfrequency correlated jitter), then the situation is quite different. Assume that we feed the DAC with a sine wave. Spectrally speaking, a slowly wandering clock signal will cause narrowband noise "skirts" to appear around the frequency of the sine wave signal. Oversampling no longer has much effect on the output spectrum, as the errors introduced by the clock modulation are all "inband" (below 20 kHz). Many types of jitter fall in between the pure white phase jitter and slow frequency-variation type of jitter described above. In that case, oversampling may improve the jitter sensitivity to a certain degree, but not as much as in the case of truly random white phase jitter. Jitter in which the time base is sinusoidally modulated will potentially produce discrete frequency components spaced around spectral sticks in the DAC output signal. For resistive ladder converters, it is obvious that with no input signal (or dc), jitter cannot have any effect on the output. The output is not changing, so it doesn't matter exactly when it doesn't change! While this observation may seem trivial, the same statement cannot be made for other types of converters, as we shall see presently. In summary, regarding resistive ladder D/A converters, we can state the following: For white phase clock jitter, the jitter spectrum on the DAC output is white and proportional to the average of the absolute value of the signal slope. Oversampling filters decrease the jitter sensitivity in direct proportion to the oversampling ratio. With no input to the DAC, jitter has no effect and does not raise the noise floor. For "slow" variations in the frequency of the clock signal, narrow noise sidebands appear around sinusoidal components in the D/A output spectrum, again with an amplitude proportional to the frequency and amplitude of the sinusoid. Oversampling filters do not decrease the jitter sensitivity in this case.

pdf 15

Figure 4
Figure 4a: Time-domain jitter error for a 1 kHz signal.
Figure 4b: Time-domain jitter error for a 6 kHz signal.

pdf 16

4(b) One-Bit Noise-Shaping D/A Converters with No Switched-Cap Output Filter This type of converter has become very popular in recent years, both because of its inherent linearity and because it can be implemented in an all-digital CMOS process. One-bit noise-shaping converters can be further divided into two classes. The first is a single-loop modulator with 1-bit quantization, and this 1-bit signal is fed directly to a 1-bit output stage at the modulator clock rate. The second class of converters, so-called MASH converters, involve multiple loops with feedforward noise cancellation. They internally produce a multiple-level digital signal, which is converted to a 1-bit signal through digital pulse-width modulation. To achieve the desired time resolution for the digital pulse-width modulator, a clock signal with a very high frequency is typically used.1 In the previous section, we saw that the sensitivity to clock jitter was proportional to the changes in the DAC output from one sample to the next. For 1-bit converters, this "change" is always full-scale, regardless of the actual input signal! Figure 5 shows a 1-bit waveform with and without clock jitter, and the resulting error pulse, for the case of uncorrelated jitter (white phasemodulation jitter). Differing from the previous case, the effect of jitter is largely signal-independent, and in fact the input signal to the noiseshaper loop may be zero with no reduction in jitter sensitivity. This is because the modulator is still switching vigorously even when the input is zero. In some cases there may be a slight reverse sensitivity to signal amplitude, as the number of transitions per unit time in the output bitstream generally decreases as the signal approaches the maximum level in either positive or negative directions. A rough estimate of the jitter
1 The new DAC devices of NPC use single-loop modulators with multilevel quantizers. As in the MASH devices, the multilevel digital signal is converted to a 1-bit signal through digital pulse-width modulation. The new Philips-designed $400 Marantz CD-63 uses an NPC DAC. David Rich

pdf 18

Figure 7: Time-domain view of sample-rate conversion.
moved. If the sample-to-sample change is dominated by unfiltered out-of-band noise, then the jitter sensitivity will be in proportion to the rms value of the sample-to-sample changes. Note that in cases where a switched-cap filter is followed by a continuous-time analog filter to reduce further the out-of-band noise, only the switched-cap filter is useful for reducing the sensitivity to jitter. This fact implies that it is impossible to predict the jitter sensitivity of a chip-level D/A converter that has The Crystal CS4303, Burr-Brown PCM67, all MASH chips, and all NPC chips use continuous-time filters. An application note available from Crystal Semiconductor on the CS4303 shows how difficult it is to create the required low-jitter clock. The Crystal CS4328 uses a fourth-order switched-capacitor filter for removal of most of the out-of-band energy. It is followed by a second-order continuous-time reconstruction filter which removes most of the image signals that arise from the switch sampling of the switched-cap filter. The Philips 1-bit DACs are a hybrid. They use an on-chip first-order switched-cap filter (similar to that in Figure 6), which removes some of the out-of-band energy. This is followed by an off-chip continuous-time filter. David Rich 20
both switched-cap and continuoustime RC filters onboard, as it is impossible to tell how much of the filtering is done in each section.Asynchronous Sample-Rate Converters (ASRC) and Jitter Reduction In a previous paper [Adams and Kwan 1993] I described an algorithm and VLSI implementation of it which allow sample-rate conversion between arbitrary asynchronous rates. Unlike synchronous converters, the device accepts external clocks at Fsin and Fs out , and by performing various signal-processing operations on those clocks it is able to derive a highaccuracy estimate of the current sample-rate ratio, and this estimate is continuously updated so as to track real-time variations in the input or output sample rates. Figure 7 shows a time-domain view of sample-rate conversion. Conceptually, asynchronous conversion consists of interpolating the input sequence to an extremely high frequency, which causes the amplitude differences between adjacent interpolated samples to become very small.
The output resampling process then consists of picking off the nearest interpolated point. The ASRC chip described uses a polyphase filter approach, with 65,536 unique polyphase filters of length 64, each stored in compressed form in ROM. This approach to rate conversion is more efficient to implement than the interpolation/decimation model, as unneeded interpolated outputs are not computed. While polyphase filtering sounds complicated, it is actually quite a simple concept. Every FIR filter has a particular group delay, which defines how much delay the filter introduces to signals appearing on its input. A typical interpolation filter might exhibit about 600 s of group delay, for example. Most FIR filter are designed to be linear-phase, which means that the delay introduced by the filter is independent of frequency. Normal linear-phase FIR filters have a group delay that corresponds to an integral number of clock cycles. But it also is possible to design an FIR filter which is linear-phase but has a group delay of an integer plus a fractional number of clock cycles. For example, a normal FIR filter of length 100 taps might have a group delay of 50 samples (half its length). But it is possible to design a linearphase FIR filter with a group delay of 50.5 samples. Now suppose that we had a large bank of FIR filters all connected to the same input signal, and each of these filters had the same frequency response but a slightly different group delay. When we change the sample rate of a signal, we are effectively attempting to resample the signal at a point between the existing sampled points of the input signal. Using the filter bank described, we could simply pick a filter output whose delay matched most closely the desired resampling point for that particular output sample. For example, if the output clock signal (which differs in frequency from the input clock signal) were to fall halfway between two edges of the input clock signal, we would select the filter that has a group delay of 50.5 input sample periods. To ensure that we have enough possible group delays to select from, the new AD 1890 chip uses a bank of

pdf 23

speaker system, $4000.00 the pair. Tested samples on loan from considerably, of course. The fundamental resonance of the sealed box appears to be 29 Hz from the impedance manufacturer. curves, suggesting a very slightly overdamped condition, which is desirable in this type of design. If you have read the interview with NHT's Ken The other three drivers are configured in accorKantor in our last issue, or have been exposed elsewhere dance with NHT's so-called Focused Image Geometry. to his ideas, you already know that he is a formidable and That means the front baffle is angled inboard by 21, highly original thinker on the subject of sound reproducwhich in the Ken Kantor canon is the optimum angle for tion. This new loudspeaker system, his most ambitious minimizing interaural cross-correlationas long as the design to date, confirms and documents that perception. listener is equidistant from the two speakersand thereIn my experience as a reviewer, the NHT Model 3.3 is by optimizing stereo separation and ambience retrieval. the most unconventional implementation of basically (More about that below; see also the aforementioned inconventional transducer/enclosure technology and probterview in Issue No. 20.) A strip of absorptive foam on ably the most nearly perfect. That doesn't necessarily the outboard side of the midrange and tweeter units helps make it irresistible to me as a music lover, but it is certo direct the sound radiation toward the listening area and tainly a world-class design by any criterion. aways from adjacent walls. The whole thrust of NHT's The unconventionality of the speaker is immediateapproach is to bring the stereo information to a focus in ly apparent from its shape and driver deployment, both of the listening area and minimize the effects of room which are ingeniously conceived to compel correct room acoustics on the quality of the reproduced signal. placementthe user actually has no other practical The drivers in the Model 3.3 are crossed over as choice. The enclosure is a big, flat slab, 42" high by 31" follows: 12" to 6" at 100 Hz, 2nd-order slopes; 6" to deep by only 7" wide, which must be placed with the nar4" at 320 Hz, 2nd-order slopes; 4" to 1" at 3.5 kHz, 3rdrow front side facing the listening area and the opposite order slopes. A positive-going pulse at the input of the narrow side against the back wall. (If you don't have an system causes a forward excursion of the 12" and 6 available back wall, the speaker is not for you, at least in drivers and a rearward excursion of the 4" and 1" drivers. my opinion.) The left and right units are mirror-imaged, This is as it should be with the crossover orders used, exeach with a 12" woofer mounted inboard in the "corner" cept for the 100 Hz crossover, where out-of-phase wiring formed by the side of the slab and the wall. That provides would be expected with the 2nd-order filters; the large four times the radiation resistance of conventional floor distance between the drivers probably accounts for the placement out in the room, with vastly increased drive decision to do otherwise. capability at the bottom frequencies. The narrow front face of the slab is angled inboard (i.e., with built-in toeThe frequency response of the speaker is amazingin) and has three more drivers mounted on it, each in its ly flat; indeed, the on-axis response of the tweeter (a 25own subenclosure: a 6 upper-bass/lower-midrange mm aluminum-dome unit with magnetic fluid damping, unit, a 4" upper-midrange unit, and a 1" dome tweeter. made by SEAS of Norway) is almost amplifier-flatand (See also the front cover of this issue.) Thus the woofer I mean 0.5 dBfrom about 8.5 kHz to 22 kHz. Way up is where it should be, in a corner, and the midrange/treble in the ultrasonic 25 kHz region the tweeter has a resonant part of the speaker is also where it should be, out in the peak; below 8.5 kHz there is a tiny downward step; but room like a typical audiophile minimonitor. Clever, these even so the 1-meter quasi-anechoic (MLS) response of Americans. the total system on axis is flat within 1.5 dB from 22 kHz on down to the woofer rangewhich is really someI developed a particular admiration for the woofer thing and shows that the 4" SEAS upper-midrange unit is design, which uses only 70 liters (2 cubic feet) of no slouch, either, over its 3-octave band. Off axis in sealed space behind the 12" sideways-firing polymerthe horizontal plane one has to distinguish inboard vs. cone driver. The skinny cabinet allows no larger volume; outboard response, the latter being deliberately barricadnevertheless, the bass is the finest I have ever heard out ed by the foam strip. Inboard there is very little off-axis of an unequalized and passively crossed-over sealed box rolloff even at the highest frequencies, but remember that of less than giant size. It remains tightly controlled and "on axis" in this speaker means 21 inboard to begin natural-sounding on the biggest bass-drum whacks and with, so that off-axis measurements beyond an additional most powerful organ pedal blasts. Ken Kantor credits 20 are hardly meaningful. In the vertical plane the offNHT's Bill Bush (an amateur bassist!) for the driver deaxis measurements show mainly a tendency toward a sign; the actual OEM for the 12" unit (as well as the 6" suckout around the 3.5 kHz crossover pointnot a major one) is Tonegen, a Japanese firm. My nearfield measurefault in my opinion. The phase response of the system is ment of the woofer showed the characteristic 12-dB-perunproblematic and, given this particular configuration, octave bottom-end rolloff of a sealed box; the -3 dB basically what I would expect. Overall, I haven't seen a inflection was at 35 Hz; there is a tiny notch at 23 Hz, better passive 4-way solution by any other designer, or which is probably the limit of useful response. The virtual even one as good. corner boosts the in-room response at the bottom end

ISSUE NO. 21 SPRING 1994 27

pdf 24

The Model 3.3 also excels in THD + N. It is very difficult to make it distort more than 1% at any frequency above 50 Hz without pushing the output to nearly unbearable levels, and that figure drops sharply as the drive is reduced. At the lowest frequencies (20 to 40 Hz) all bets are off at high SPLs, as is usually the case, but the higher-order harmonics are still very low. I must add that none of my measurements differed significantly from the manufacturer's specs and curves, which appear to be remarkably honest. I wish I could say the same of all loudspeaker brands. One thing that NHT does not talk about but I found to be outstanding is the complete freedom from ringing on tone bursts, regardless of frequency. The reproduction of gated square pulses is far from coherent; the waveforms emerge somewhat disorganized, but less so than with other 4-way speaker systems known to me. Let us not forget about the impedance curve: above the woofer range it varies only from 4.5 ohms to 7.5 ohms in magnitude (6 ohms nominal); at the 29 Hz box resonance it rises to 10.5 ohms (only!); the phase is well within 30 at all frequencies. A piece of cake for any amplifieror two amplifiers, as there are terminals provided for biamplification via the passive crossover. (Or biwiring if that harmless delusion makes you feel better.) The only thing I found objectionable about the Model 3.3 had to do with the stabilizer bars which must be screwed to the bottom to keep the heavy (123 lb.) but wafer-thin cabinet from toppling over sideways. For $4000 one has the right to expect high-quality machine screws and threaded inserts for attaching the stabilizers; instead I had to screw huge self-tapping wood screws into hardwood and was rewarded with blistered palms for my unskilled effort. I never affixed the pointy cones the stabilizer bars are supposed to rest on; they are not for the reviewer who constantly moves things around on the floor. Well, you probably thought I'd never get around to the subjective listening quality, but here goes: Very accurate, transparent, and neutrallike the best of Snell, but even more so. Not very forgiving when the recording is overbright. Perfectly balanced when the recording is. A little more open and airy than the Waveform Mach 7 but maybe not quite as dynamic. Medium-sized wave launch into a large listening roomI wish it presented a larger apparent source, but it certainly doesn't sound small. Beautifully delineated, strong bass, as discussed above, though not quite on a par with the most sophisticated sub woofers. As for the Focused Image Geometry, I am the wrong person to ask. Amazing snap-into-focus 3-D imaging is not my Holy Grail because I do not hear it in the concert hall; it seems to be achievable only with microphones. I am basically a tonality/balance/transparency man. Ken Kantor wants the new owner of a Model 3.3 to set it up with half-inch-by-half-inch trial-and-error

changes in position, separation, parallelism, etc., until the focus snaps in like a seat-belt buckle. I have no patience for that. After attaching the stabilizer bars I placed the speakers perpendicularly against the back wall a little over 8 feet apart, and they imaged/soundstaged just fine without further ado. There are other speakers that will not image one tenth as well, no matter what you do. Bottom line: the NHT Model 3.3 is a marvelous speaker and perhaps not even overpriced at $4000, but for me it doesn't have that ultimate I-can't-live-without-it quality because, ideally, I want a bigger apparent sound source plus dead-flat low-frequency response down to the subsonic region. Ken Kantor claims that this is best loudspeaker system he knows how to make, regardless of price, but I don't for a moment believe that. It is probably the best 3-foot-high monolithic 9-octave-range speaker he knows how to make. Or anyone else knows how to make, for that matter.

Sequerra Model NFM-PRO

R. Sequerra Associates, Ltd., 792 Pacific Street, Stamford, CT 06904. Voice: (203) 325-1791. Fax: (203) 325-0263. Model NFM-PRO 2-way nearfield-monitor loudspeaker system, $2000.00 the pair. Tested samples on loan from manufacturer.
Dick Sequerra clearly has nine lives as an audio entrepreneur; this particular engineering/marketing effort represents the sixth one I am aware of but I may have overlooked one or two. Right now the NFM-PRO constitutes the sum total of his active product line, or close to it, as I understand from my brief conversations with him. This is a teeny-weeny speaker, 11" high and deep, 6" wide, with 10 of its 12 edges rounded, no grille, black Nextel finish all around. The driver complement consists of a 6" plastic-cone woofer and, believe it or not, a 2" paper-cone tweeter with an aftermarket phasing plug. Dick Sequerra hates dome tweeters and is the only one in the hi-fi loudspeaker industry (to my knowledge) to use this Japanese-made cone tweeter. His main reason, I believe, is power handling. The crossover appears to be first-order (Dick Sequerra also hates 4th-order LinkwitzRiley, etc.), and the woofer protrudes l" from the box by means of a short tube extension in pursuit of better time alignment. The tweeter level can be adjusted up and down over a narrow range with a 6-position rotary switch in the back. How can such a simple little speaker cost $1000 per side? For one thing, it is very solidly built and, in its austere way, carefully finished. For another, it is made primarily for the professional market, to sit on top of a console for nearfield monitoring, and that market isn't necessarily driven by apparent-hardware-value-per-dollar. Furthermore, Dick's prices have always been somewhat arbitrary, based onshall we say?audiophile-awe-perdollar. The NFM-PRO is, in effect, a slicker and more

pdf 27

an altogether happy picture. In the time domain the picture is no happier. A tone burst consisting of 10 cycles of 500 Hz makes the lowfrequency panel produce 10 more cycles only about 14 dB down from the input. The same panel, when excited with a single cycle of 20 Hz, produces another cycle at the original level and then a decaying series of six more cycles starting at only 9 dB down. That panel is alive! The high-frequency strip also rings badly: a 4 kHz tone burst of 10 cycles makes it produce 16 more decaying cycles, starting at 6 dB down and persisting at 12 dB down most of the way. As for coherence, I found a 1 kHz square wave to be pretty badly mangled by the crossover, but a 2 kHz square wave came off the high-frequency strip looking quite nice. How does all that affect the speaker in terms of subjectively perceived sound? More or less as you would expect: not enough bass, not enough highs (at least from some listening angles), not enough focus, too much coloration, an overall not-quite-neutral quality. Lest that should appear like a totally negative reaction to the speaker, I must add that I still liked the planar wave launchthe height and width of the apparent sound source were just right, and that made for reasonably happy listening despite all of the above. Even a faulty planar speaker has a certain dimensional Tightness about its sound. Since I had not looked at a Magnepan product for years, I felt a little uneasy about the possibility that this single unenthusiastic review would define my view of all Magneplanars in the eyes of our readers. I therefore borrowed a pair of Magneplanar Tympani IVa's from a friend and put it through its paces. The Tympany IVa, with three screenlike hinged panels per side and a bona fide ribbon tweeter, was until 1992 the top of the Magneplanar line at $3750.00 the pair. It is no longer made, having been replaced by the single-panel MG-3.3/R ($3000.00 the pair) and single-panel MG-20/R ($8500.00 the pair). All I wanted, however, was a generic high-end Magneplanar for reference and comparison, and the IVa served that purpose very well. It is an incomparably better speaker than the MG-1.5/QR (at almost three times the price it had better be!) but it still has some of the same generic shortcomings. It rings all over the place in much the same manner (not the ribbon, though), and the bass is still far from superb, with a big bump at 35 Hz and poor damping. A single Velodyne ULD-15 Series II made the whole system sound significantly more authoritative below 85 Hz. The frequency response of the IVa, however, is much flatter and less ragged than that of the MG-1.5/QR, and in the 2 octaves from 100 to 600 Hz (upper bass and lower midrange) the IVa is extremely impressive, with tremendous impact and dynamics resulting from the huge drive area. No conventional speaker I know of is its equal in that respect. On the other hand, imaging and directional cues are quite vague on the IVa,

explanation in the next issue of the circuit details of the unorthodox ELF solution, which is incorporated in the $2460 ELF-1 electronic unit and which we haven't tested yet as a separate component. The heart of the circuit is described as a dual integrator (with the emphasis on "dual"). This is rather different from the conventional lowpass-filter type of bass equalizer and is claimed to have many advantages. Very briefly, the main advantage, in addition to inherently excellent linearity down to virtually dc, is that the very low permissible low-frequency cutoff (almost dc-ish at 8 Hz) results in significantly improved signal-delay characteristics in comparison with conventional bass-response profiles. Time offset, which in the case of low-frequency transducers is of more than just tweako concern, thus becomes a relatively innocuous issue with the ELF system. The basic ELF electronics could be implemented much more simply and cheaply than in the ELF-1, which is an incredibly elaborate and versatile professional unit with balanced inputs/outputs (only!) and a zillion possible settings and adjustments. Its so-called concealment circuit is a particularly sophisticated adjustable control feature which is actually essential with the 8 Hz cutoff to protect the typical amplifier from overload. The ELF-1 also provides electronic crossover and level setting facilities for the main amplifier/loudspeaker system. Details about all that will be part of the more technical followup review in the next issue; meanwhile I should note that Modular Sound Systems is definitely interested in the home audio market and new, more audiophile-oriented versions of the Bag End ELF products can be expected. The earliest sign of that is a substitute front panel for the ELF-1 with a much more High End kind of look than the very utilitarian decal-labeled original. As for the woofer enclosures, I absolutely love the indestructible carpetcovered road-show versions they sent me, complete with corner protectors and handles, but for the home market they will have to make the entire line available in various veneers, not just the one 18" model in oak veneer as at present. A single ELF-1 control unit can drive any number of amplifier/subwoofer hookups, as long as it sees at least 600 ohms. One of the beauties of the system is that the different subwoofer models, regardless of size, all have essentially the same response characteristics, differing only in efficiency and maximum SPL capability. Thus multiple arrays can be used to increase subbass output and reduce distortion ad libitum. I am not quite finished yet with my measurements of the distortion characteristics of a single S10E-C and single S18E-C; the complexity of the electronic signal path makes the process a bit more than routine; I can confidently predict, however, that the final results will be somewhere between very good and excellent. To get back to the subjectively perceived sound of the Bag End ELF system, I find it more solid, clearer,

pdf 29

more detailed, less nimbly, more revealing than that of any other subwoofer known to me. It would seem that getting rid of a very low fundamental resonance as the crutch for low-frequency extension also gets rid of a certain vagueness of definition in that range. That of course is speculation without proof. If you want objectively verifiable and repeatable observations, I have only gone as far as to measure the nearfield frequency response. That goes down flat well below the lowest audible tones, as I said, with worst-case deviations of less than 1 dB. And that goes for both models. So much for nowto be continued. (Just before press time, a pair of Velodyne Servo F-1500R's arrived. That will be some shootout!)
.lastly, an acoustical accessory:

MSB Acoustic Screens

MSB Technology Corporation, P.O. Box 141, Moss Beach, CA 94038. Voice: (415) 728-5265. Fax: (415) 747-0405. MSB Screen, 6 ft. high with three 2 ft. wide folding sections, $400.00 ($800.00 the pair). Tested samples on loan from manufacturer.
I got into trouble with MSB Technology by being disrespectful to their "EMA Isolation Plate" in an editorial reply to a subscriber's letter in Issue No. 18. They wrote a long letter, published and answered (with "caseclosed" finality, in my opinion) in Issue No. 19; they complained to us verbally as well; however, as of the
time this is being written, I still have no test data from them proving that they are right and I am wrong, only promises that such proof is coming. I shall now redeem myself (in their eyes, I hopein mine I need no secular redemption) by warmly recommending their acoustic screen product. What we have here is a handsome and useful piece of furniture, a three-section folding screen covered with the same acoustical fabric on both sides. Two short, rounded maple legs support the frame of each section and keep it off the floor. The hinges work in either direction, and the screen is self-supporting in just about any folded or unfolded position except a straight line. Under the acoustical cloth are three separate layers of special damping material. I do not have an established test protocol for quantifying sound absorptionone of these days, maybebut I can tell you that the device works. Two of these screens can make a serious difference in stereo loudspeaker deployment. Putting a screen behind each speaker system in a shallow U shape will block just about all backwall and nearby sidewall reflections, yielding a strictly forward-firing launch. That can eliminate a lot of soundstaging confusion in certain setups. You can also cover reflective surfaces anywhere else in the room for improved acoustics. The screens can be moved easily by one person and folded when not in use. It would be an exaggeration to report that the MSB Acoustic Screens changed my audio life, but I will say that after you have used them for a few weeks you don't want to be without them. They are a bit pricey, though, for something that couldn't cost all that much to make.

pdf 36

regulated rail on its negative input. The reference voltage is connected to the plus input of the op-amp. The regulator is configured as a tracking regulator, so that the positive power supply acts as the reference for the negative supply. The TO-3 packaged pass transistor is on a big heat sink. The supply for the microprocessor and other circuitry used for the remote control is on a separate transformer secondary and has 7812 and 7805 integrated regulators. These regulators are also on good-sized heat sinks, as are all the output devices. The remote-control features of this preamp are no gimmick. Once you have experienced the ability to adjust signal levels at the listening position, you will never want to go back to a preamp without a remote. The microprocessor-based circuitry for the remote control is essentially similar to that used in other remote-control products. In less expensive products much of the signal switching would be done with semiconductor devices, not the 18 relays used in this unit. A shaft encoder is used to simulate the action of a rotating volume control. In a highvolume product the selector pushbuttons and display devices would be a custom implementation that would cost significantly less per unit, but at very low volume the tooling cost would make production of such units prohibitive. In the KRC-2 these components are all on a PC board mounted to the front panel. Even the handheld pushbutton control unit for the KRC-2 is manufactured by Krell in the USA. Unfortunately, without custom tooling the remote control is somewhat large and cumbersome. The code for the microprocessor still needs some work. The balance control requires that the L or R pushbutton be pressed each time to advance it, but the volume control advances only if the Level button is pressed continuously. A pop could be heard through the speakers when the Gain button was pressed to switch between low and high gain. This indicated the intrusion of dc offset when switching, a bug that should have been cleaned up before the preamp was put into production. The gain setting of the preamp would occasionally change arbitrarily when the unit took a static discharge hit. (Sonys do not do this.) The feel of the shaft-encoder-based volume control, however, is excellent. The same cannot be said of the power switch because this preamp hasn't got one!! This is unacceptable, given the power dissipation of this unit. [Come on, David, you know the rules of the highend game. Preamps must be powered up 24 hours a day, 365 days a year, otherwise they won't sound good. They begin to sound really good only after several weeks, or is it months? Ed.] The Krell KRC-2 showed no dynamic distortion in our tests, but overall distortion was quite high for a preamp, even if not nearly high enough to be audible. With an unbalanced input and unbalanced output in the low-gain mode, THD + N reached a minimum of -81 dB at approximately 1 V rms output, then rose to -60 dB at 7 V rms, just before clipping. The Harman Kardon

pdf 39

culated with R1/R2 = R3/R4 for simplicity.
The last equation is just what we want: the difference of the input signals is amplified and converted to a single-ended output. Now we can build this circuit using an op-amp and place it ahead of the single-ended power amp. This is what was done in the original Parasound HCA-2200. Indeed, Parasound used an even better circuit, based on a two-op-amp balanced-to-single-ended converter, which had improved distortion performance. An alternative is to use the power amplifier itself as the op-amp element. A problem with this second approach is that the input impedance of the negative terminal can be very low. From the circuit above it is easy to see that the input impedance of the positive terminal is R1+ R2, which will be a high value, but the impedance of the negative input terminal is much less. This impedance is calculated as
For a high-gain amplifier, this can be approximated as R3/2. Now, if we want the input impedance of the amplifier to be 50 k and the amplifier's gain to be 10, then R4 is going to have to be 1 M , which is not a practical value. The solution is to use a smaller value of R3 to bring R4 down to a practical value. The low input impedance is dealt with by adding a voltage buffer in front of the negative input of the amplifier. * * * As I said, the two-transistor open-loop JFET buffer regrettably used by Parasound in the II revision is the simplest possible discrete circuit for this application. In its simplest form a diode-connected JFET forms the current source that biases the single-transistor follower stage. For matched devices the voltage drop across the biasing device matches the voltage drop across the source-follower device. Since the gate-to-source voltage of the current-source transistor is zero, the drop across the source-follower device is also zero, and the input and output voltages of the buffer are matched. In the Parasound implementation of the two-transistor buffer, a resistor is placed in series with each source of each JFET to reduce static current flow and improve matching. This simple circuit cannot follow the input signal accurately enough to transfer it without distortion, as we will see below, because an open-loop JFET source follower in not linear enough. The bridged mode of the amplifier is also poorly implemented, in both the original version and the II revision. For the bridged mode to work, the input of one channel of the amplifier must be inverted with respect to the input of the other channel. Again, this is usually accomplished by placing an inverting amplifier, formed

pdf 45

the algorithm can be updated by changing the firmware PROMs that contain the computer code which controls the DSPs. I was given no information on the algorithm used by Krell in the filter, so let's move on to the DAC. The DAC is a Burr-Brown PCM63P (K?the grade is not given on the schematic), placed inside a shielded box. Trim pots are included for adjusting the MSBs of the colinear DACs. Adjusting these pots is very difficult because they affect signals at half of full scale, not around ground. I do not know what method Krell uses to set them. The pedestrian 7805 and 7905 3terminal IC regulators (what were you expecting, some high-tech discrete regulators?) subregulate the 15 V analog supply to 5 V for the DACs. The current-to-voltage converter is the SSM-2131 op-amp, which appears to be identical to the PMI OP42G (the only specification differences are a slightly higher offset spec for the SSM2131 and the addition of a typical THD spec). But wait! Stop!! Hold the phone!!! What's a low-cost, highfeedback op-amp doing in a Krell? What happened to discrete circuits, low feedback, and all that good highend stuff? Now, don't get me wrong here; the OP42 is an good op-amp for this stage, but it is by no means the lowest-noise, widest-bandwidth, lowest-distortion (see Ben Duncan's article in the December 1993 issue of Audio Amateur), and fastest-settling one around. It is not processed with a complementary bipolar process, so its principal attribute is relatively high speed at a low price. (See Issue No. 15, Table 4 for a complete set of specs and prices for this and other ICs used in CD players.) A good price-performance ratio, however, is not exactly what I had in mind as the principal attribute of a component in a $3900 Krell. I wonder how many of Krell's customers realize that the designer could have used, for example, the Burr-Brown OPA627, which outperforms the SSM-2131 in all respects by more than 2 to 1. The only disadvantage of the OPA627 is that it is three times the cost of the SSM-2131. Note that the $14,000 Krell Reference 64 does use expensive complementary bipolar op-amps (AD841 and AD846) for its I/V converter and deglitch circuit. The important point here is not the price of the opamp but the fact that using any op-amp breaks all the supposed rules of "high-end" design. If it is okay to use one op-amp in the signal path, why not make the whole signal path out of op-amps? One would think that once a signal has been "corrupted" by a high-feedback amplifier, no amount of discrete amplifiers downstream could restore its purity. Once the signal has been negatively affected by passing through an on-chip diffused resistor, why should it matter that other resistors in the signal path are of the audiophile-approved variety or that audiophilegrade interconnect cable is used? Once the signal has passed through the class AB (quasi-complementary in the case of the OP42) IC output stage, why should it be important to run the circuits downstream from the chip in