Foreword by the Tech. Ed.: "The Jitter Game"
I have used the same title for this foreword to an important article on jitter as Stereophile's Robert Harley uses for his articles on jitter, but with a different meaning. Harley is playing a game of pretend engineering when he attempts to analyze the jitter of a CD player and correlate the resultant measurement to the sound quality he perceives. Why does Harley spend so much time on jitter? Because he thinks that it strongly correlates with the sound quality of the equipment. Since open-loop (i.e., nonblind) listening tests are subject to externally originating listener bias, it is easy to see how he can delude himself to arrive at such a conclusion. Stereophile is unfortunately quite influential, and jitter has thus become in the early '90s what TIM (transient intermodulation distortion) was in the late '70s and early '80s. But Harley has a huge problem because clock jitter cannot be measured directly at the output of a black box. The effects of jitter can be assessed indirectly from black-box measurements, but in a correctly designed CD playback system these effects are commingled with, and usually swamped by, noise and distortion products. Indeed, in exotic designs, the loony-tune analog stages are so riddled with noise and distor10
tion that even large amounts of jitter would have little effect on the measurements. To overcome this problem, Harley plays his little game. He takes off the cover, gets inside the unit, and attempts to measure the jitter on the internal clock line. Now, two problems exist when he does that: (I) since jitter is an internal parameter, its effect on the external performance of the system is dependent on other aspects of the system's design, so it is not possible to compare the measured results directly between two models under test; and (2) measuring clock jitter is a nontrivial task, subject to many errors even when conducted by one skilled in the art. Note that Harley could continue to play his game and make other measurements while he has the cover open, such as I/V settling time, power-supply rejection ratio, the amount of closed-loop feedback, powersupply output impedance, etc. All these parameters could affect the sound quality, and under Harley's rationalethat being unable to observe the effect of such parameters at the output of the system does not mean they do not affect the sound qualityone could ask why he doesn't make these additional measurements. My contention is that Harley would indeed make these measurements, and then delude him-
self into thinking they were remarkably revealing of sound quality, if some manufacturer delivered to him a test system for a given parameter and showed him step by step how to use it. Jitter and its effect on the performance an electrical system is a difficult subject, truly understood by only a few experts involved in the design of systems sensitive to jitter. As a result, much misinformation on jitter has been circulated in the press, originating from manufacturers' press releases reproduced without any competent review. In an attempt to clear the air on the subject, we commissioned an article by a genuine expert in the field of digital audio, Robert W. Adams, of Analog Devices, Inc. This article is based on a paper Adams presented at the 95th Convention of the Audio Engineering Society in New York last October. (The preprint number was 3712.) Bob Adams is perhaps the youngest Fellow of the AES (the highest honor awarded in the field of audio engineering), and his many pioneering achievements in digital audio at AD, and before that at dbx, are too numerous to be summarized here. His investigations in the field of jitter reduction have resulted in a new method to attenuate jitter, a practical asynchronous sample-rate converter chip, which is explained in his article. Before this

tals before making any judgments based on jitter about the quality of a particular piece of equipment. Another complicating factor will soon be introduced commercially: a new chip from Analog Devices called an "asynchronous sample-rate converter," rapidly making its way into outboard D/A processors. This chip acts as a universal digital buffer between an input at one sample rate and an output at any other sample rate. As a byproduct of the algorithm employed in the chip, jitter on either the input or output sample clocks is largely eliminated. While most engineers understand how a conventional analog PLL may be used to remove clock jitter, it is not obvious how an all-digital sample-rate converter can accomplish the same task. Later in this article we will discuss how use of this chip affects jitter in D/A converters. 1 Review of Clock Jitter Clock jitter may be defined as the time displacement of a clock signal relative to an ideal clock signal with no jitter. Note that all the information about jitter is contained in the edges of the clock signal, and it is common to specify jitter in the time domain as either the p-p or rms deviation of any edge from its ideal position over many thousands of clock cycles. Most digital systems will change state only on one edge of the clock signal (the rising or falling edge), in which case the jitter is measured on the clock edge to which the system responds. In practical systems, the common types of clock jitter are: (a) Random variations in the arrival of clock edges relative to their ideal positions. For advanced readers, this type of jitter is referred to as white phase jitter, as it may be produced by feeding a random-noise (i.e., white-noise) signal into a phasecontrolled oscillator. (b) Random variations of the width of a clock pulse. This type of jitter differs from (a) in that each edge is referenced to the previous edge rather than to a hypothetical ideal clock signal. Again, for E.E. types, this jitter is referred to as white FM jitter, as it can be produced by feeding a white-noise signal into a

pdf 12

frequency-controlled clock generator. (c) Correlated variations in the clock edge events relative to an ideal clock. By correlated we mean that the instantaneous time displacement measured on each clock edge is not an independent event but is in some way related to previous clock edge times. For the technical reader: this causes a jitter "spectrum" which is nonwhite and may have spectral peaks at particular frequencies. We will refer to this type of jitter as correlated jitter. If the variation in clock frequency is "slow" compared with audio frequencies, we will call this low-frequency correlated jitter; if these variations are fast compared with the audio spectrum, we will call this high-frequency correlated jitter. 2 The Pitfalls of Time-Domain Measurements It is common to estimate clock jitter by using an oscilloscope with a very accurate time base. This practice is dangerous, as the results obtained depend on the type of jitter as well as on the measurement technique. It also is often the case that the oscilloscope used will have more jitter in its time base than is present in the clock itself. Advanced instruments are available to make accurate measurements of jitter but are not used enough. Figure 1 shows one measurement technique where an oscilloscope is set to trigger on a clock edge and the time base is set so that only the next edge is visible on the scope. The variations in the arrival time of the later edge can be used as a measure of p-p jitter. More sophisticated oscilloscopes can plot a histogram of zero-crossings, allowing a more accurate estimate of the rms jitter without resorting to "eyeball" measurements. Since we are triggering on one edge and measuring the arrival time of the next, we are assuming that the first edge (the one we are triggering on) is in its "ideal" time position. This technique is fine if we are measuring white phase jitter as defined above, where the errors in the clock edge positions do not accumulate over time relative to an ideal clock signal. But suppose that the frequency of the clock signal is slowly wandering by a small amount (low12

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 19

65,536 possible filters. Depending on the internally calculated, desired output-resampling point, one of these 65,536 possible sets of coefficients is selected to weight the surrounding input data values to produce an output point. By using such a large number of possible group delays, the error introduced by the time quantization in the resampling process is about at the 16-bit level for the "worst-case" signal, which is a full-scale 20 kHz input signal. For lower amplitudes and/or frequencies, the error is significantly less and is ultimately limited by the stopband rejection of the polyphase filters. The algorithm used in this chip is quite complex, and interested readers are referred to AES Preprint 3712 for details. At the heart of the chip lies a circuit that computes an internal estimate of the ratio of input to output sample frequencies. This computation is done continuously, and in fact will track real-time changes in sample rates that are quite rapid. The interesting part of this chip from the perspective of rejecting jitter lies in the fact that this internal estimate of the frequency ratio is computed using thousands of past input- and outputsample clock events, and is therefore immune to small perturbations in the arrival time of any individual clock edge. This jitter-rejection capability may be thought of as a filter where the cutoff frequency is as low as 3 Hz. Any jitter components above 3 Hz are filtered with a lowpass characteristic of -6 dB/octave. For example, a 100 Hz jitter component is 5 octaves above 3 Hz, and will be attenuated by 30 dB. Higher-frequency jitter, naturally, will be attenuated even more. This ability to reject jitter suggests an interesting system architecture for outboard D/A converters. Normally, they must recover a lowjitter clock from the incoming serial bitstream. Typical integrated S/PDIF receivers have PLLs with relatively poor jitter-rejection capability, and even if the original S/PDIF signal is of good quality, by the time it has traveled through several feet of cable it may have high jitter due to the intersymbol interference caused by the finite bandwidth of the cable. That

Stable clock

Figure 8: Outboard D/A processor with sample-rate converter. makes it difficult for the circuit designer to provide a stable low-jitter clock to the D/A converter chip. Figure 8 shows a system where a sample-rate converter is used between the clock-recovery and S/PDIF receiver chip and the D/A converter. Note that since the input and output sample frequencies are decoupled, the designer can use a crystal oscillator to provide a low-jitter clock to both the D/A converter and the output side of the sample-rate converter. The sample-rate converter will prevent the input jitter from affecting the output data, and the D/A converter is allowed to operate with a crystal clock signal. Note also that the sample-rate converter will reject jitter on its output clock as well, in that the output data will not be affected by the jitter. But if this jittered clock is used to clock the D/A chip, then errors will still arise in the D/A converter itself. Evidently such ideas are confusing, as conceptual errors have already appeared in print. Suppose that one measures the jitter at the D/A converter clock pin both with and without the sample-rate converter being used. Suppose that without the sample-rate converter, the jitter is measured to be 1 ns rms. In this case we are measuring the jitter of the PLL used in the clock-recovery circuit of the S/PDIF receiver. Now we measure the same D/A clock pin using the sample-rate converter, and we find that the jitter is measured to be 100 ps. One might be inclined to state that the sample-rate converter chip has reduced the jitter from 1 ns to 100 ps. But this misses an important point: the sample-rate converter does not produce an output clock; rather, it accepts whatever clock the user feeds to it. The reduction in jitter follows because the designer is free to supply a crystal-generated clock signal to the D/A converter, and this clock can be as clean as he or she can make it. In fact, even with the world's worst sample-rate converter, as long as it was asynchronous the clock jitter measured at the D/A converter would be that of the crystal oscillator itself. How, then, can we judge the quality of the sample-rate conversion? There is no clock signal produced by the sample-rate converter chip that can be measured (it does not even exist internally!). The only way to judge the signal quality is to look at the signal itself. Since the jitter is filtered to about 3 Hz with the sample-rate converter chip, it should be possible to "zoom in" on a sinewave test tone using an extremely long FFT (or a slow-sweeping analog spectrum analyzer connected to the D/A output) and see very narrow noise "skirts" around it. These are in fact visible if you have test equipment good enough to resolve such narrow sidebands. Note that these sidebands are much narrower and lower in energy than the narrowband noise-modulation products in such bit-rate reduction schemes as DCC and MiniDisc, so they will not be audible. To demonstrate the jitter-rejection capability of this chip, we have constructed an artificially severe test where both the input and output clocks are jittered with 100 ns p-p binomial-distributed jitter (produced by feeding a random binomial bitstream into an RC oscillator). Notice that the degradation consists entirely

when it falls within a critical band. Comments Traditional THD + N versus frequency tests and FFT spectrum plots for input signals of various frequencies are adequate to cover the effects caused by jitter. There is no reason to single out distortion components caused by jitter as distinct from those caused by such other effects as D/A nonlinearity, op-amp distortion, etc., the only possible exception being the broadband noise-modulation test mentioned above. A full-scale 15 kHz or 20 kHz tone will have no masking for frequencies in the 1-5 kHz range, and for those very rare audio signals with most of their energy in the last half-octave, it is possible that, if excessive jitter is present, noise modulation might be audible at very high playback levels. A high-frequency THD + N measurement may not adequately show this noise modulation if the residual distortion signal is dominated by discrete harmonic distortion. On the other hand, if the THD + N at 20 kHz shows little or no rise compared with the same measurement at 1 kHz, jitter is not a problem. 7 Conclusion Any signal degradation caused by using jittered clocks to drive D/A converters is a complex combination (continued on page 33)
Figure 9: FFT with 100 ns p-p binomial jitter on the L/R clock.

pdf 21

Loudspeaker Systems Using Forward-Firing Cones and Domes: How Good Can They Get?
By Peter Aczel Editor and Publisher
As drivers, designers, and system design software become more and more sophisticated, the sound improvesbut how much? We tested a number of interesting new speakers to explore the answer.
It has become a truismtweako cultists to the contrary notwithstandingthat switching to a significantly better loudspeaker will improve your sound more than any change in electronic components, regardless of cost. This publication is one of the most fervent advocates of that tenet. Even so, if you already own a fine forward-firing dynamic speaker systemlet us say one of the higherpriced models by Snell, B&W, Fried, or anything else of that orderdo not imagine that your audio life will dramatically change if you switch to the latest and most highly refined system of the same basic architecture. There are no miracles in electroacousticsif the music comes out of 1-inch and 6-inch and 12-inch holes, it will invariably sound more constricted than it would in Carnegie Hall, even if those holes are exceedingly transparent. If that sounds like an indirect endorsement of large planar and line-source speakers, so be itbut that approach has its own goblins and demons. If I were a record producer and wanted to monitor a new recording in progress, I think I would still opt for the most accurate forward-firing dynamic speaker I could lay my hands on. The truthwhat do I really have on my tape? is more likely to be told by such a speaker. But for home listening, strictly for pleasure? I'm still not sure. My vacillation is rather strikingly exemplified by the DCM Time Window Seven, which I reviewed in the last issue and which quite successfully splits the difference between a foward-firing monitor and a multiplewave-launch bi- or omnidirectional type of design. It sounds bigger and more room-filling, though perhaps a bit less precise, than the former but more focused and sharply delineated than the latter, yet almost as expansive and 3-D. I have become quite addicted to it and now recommend it even more highly than I did before. Anyway, perfecting the standard, conventional formats is always a most important part of technological progress, and the speaker systems reviewed below are

for example, Issue No. 11). Magnepan's literature clearly explains the design principle, but naturally only the pros without the cons. Suffice it that the MG-1.5/QR is a 64" high, 19" wide panel divided into a large low-frequency section and a tall, narrow (i.e., ribbon-like) highfrequency section. The left and right units are mirrorimaged. The electrical crossover frequency appears to be 700 Hz, although the specs say the acoustical crossover is at 1 kHz. The lowpass filter is 2nd-order; the highpass filter is lst-order (series capacitor). The measured impedance curve indicates that the low-frequency and high-frequency transducers are both essentially resistive (5.2 and 3.7 ohms respectively) but that the crossover network synthesizes a fairly complex load: magnitude, 3.7 to 29 ohms (peak at the crossover); phase, -55 to +40. The nominal impedance is 4 ohms. Terminals are provided for biwiring (pure nonsense) or biamping via the passive crossover (marginally advantageous). The low-frequency and high-frequency sections are wired out of phase (probably because the overall acoustical profile of the crossover is closer to 2nd-order than anything else). When the sound source is a tall structure like the MG-1.5/QR, its frequency response is not as easily measurable as that of a minimonitor or even a fair-sized box speaker. I opted for 2-meter MLS (i.e., quasi-anechoic) measurements with the microphone aimed at the highfrequency strip from various heights and at various horizontal angles. I also took nearfield measurements of the low-frequency panel. I would have preferred outdoor measurements from a distance of 3 or 4 meters, but that was not an available option. I am reasonably certain, however, that I have a good handle on the speaker's overall response characteristic, even if the accuracy isn't the highest possible. The bass response of a speaker like the MG-1.5/QR is hard to quantify with my usual nearfield technique because the real-world bass output of a dipole exists only at some distance from the launch point on account of the back-to-front cancellation. I was able to determine that the fundamental resonance of the "drumhead" (because that's what the low-frequency panel really is) was at 44 Hz, with steeply falling output below that frequency, and that there was another big bump at 75 Hz, the really smooth response starting only at 125 Hz and continuing (now on the basis of MLS) up to 1.3 kHz or so. Those three-plus octaves look very nice and are probably the long suit of the speaker. From there on up things get rough again; the response up to 20 kHz is a series of peaks and dips. The biggest dips are -7 dB on axis and -10 dB off axis (30 inboard); the biggest peaks are smaller, maybe +5 dB; the whole response profile is much more ragged than anything we have seen in quality speakers of conventional design. The high-frequency response above 11 kHz declines 6 dB per octave off axis. Even allowing for some measurement errors, this is not

pdf 30

Analog Electronics: More Power Amplifiers, Preamp/Control Units, and Mild Surprises
By David A. Rich, Ph.D. Contributing Technical Editor
This is the continuation of our ongoing preamplifier and power amplifier surveys. The plot thickens but does not take altogether new and different directions.
The reader is referred to Issues No. 18 and 20 for introductory material, basic definitions, general engineering concepts, circuit illustrations, and extensive reference documentation on the subject of preamplifiers and power amplifiers. Our first-time explanation of the PowerCube, with five typical examples of measurement printouts, is also in Issue No. 20. We cannot go over this groundwork in each issue for new readers, but both No. 18 and No. 20 are still available as back issues for subscribers. formed by source-degenerated JFETs that are cascoded by bipolar devices. The common-mode points of the two differential pairs are connected together through a 2.2 F capacitor. This capacitor allows the amount of current flowing through the differential pair to increase if the amplifier approaches slew-rate limiting. The bases of the cascodes are connected to a level-shifted version of the common-mode voltage of the differential pair. This is called a dynamic cascode, since the emitter of the cascode device moves with the common-mode voltage of the differential pair. The dynamic cascode keeps the drain-to-source voltage of the differential pair constant in the presence of a common-mode voltage. This prevents the common-mode input signal from coupling to the output of the differential pair through the gain devices. Only one other maker of commercial amplifiers I have seen uses a dynamic-cascode input stage, and that is PS Audio. Bob Odell worked for Harman Kardon before he designed these PS Audio amplifiers, so it is likely that the technique originated at Harman Kardon. The PA2400 is configured as a noninverting amplifier, as are nearly all other amps. Such an amplifier has a common-mode swing which is equal to the input signal. Recall that the differential input signal to the amplifier is an error signal which the amplifier tries to keep as small as possible. If the amplifier could keep the error signal at zero, the output would contain no distortion products, since the output/input transfer function would be controlled by feedback resistors alone. In an ideal differential amplifier stage, the common-mode signal is completely rejected. Now, if the differential pair transmits some of the common-mode signal to the second gain stage, that stage will be unable to distinguish between a signal resulting from the differential gain of the differential pair and a signal resulting from the common-mode gain of the differential amplifier. Consequently the output will be distorted because the amplifier no longer attempts to keep the differential input signal to a minimum.

pdf 35

by a dc servo. Identical circuits are used for the plus and minus signals. No circuit is included to convert a singleended signal to balanced. The volume and balance controls are formed by a resistor ladder and CMOS switches. The schematic of this stage was not submitted to us, but with proper design the CMOS switches can be arranged so that no static current flows through them when terminated into a buffer stage. But even under these conditions some distortion can occur at high frequencies. This distortion is the result of a voltage divider formed by the nonlinear channel resistance of the MOSFET and the nonlinear reversebiased diode that is the drain of the MOSFET. The catch is that if the width of the MOSFET is increased to lower the channel resistance, the parasitic capacitance grows proportionately. Using a device with a lower channel length will reduce the on resistance without increasing the parasitic capacitance, but the magnitude of the voltage that can be passed through the device decreases because small channel devices have lower breakdown voltages. The on resistance of the channel of the MOSFET varies with the gate-to-source voltage and hence the signal swing. For the case of an NMOS device, the channel resistance increases with increasing signal voltage. Fortunately a PMOS device does just the opposite. By using both types of CMOS devices in parallel, the variation in channel resistance is minimized, especially when the input signal is kept well below the gate voltages on the cmos switches. The disadvantage of CMOS is that if the signal ever goes above the chip's supply voltage, the chip can go into destructive latchup. It is therefore important to clamp the input signal to the CMOS gates by means of Schottky diodes. Without a schematic I cannot tell what Krell does to prevent latchup. The output of the volume-control circuit goes into a unity-gain follower with no feedback. The follower is required because the CMOS switches cannot be resistively loaded, as explained above. You would logically expect this buffer to be similar in topology to the buffer on the input, since they perform similar functions, but you would be wrong. This buffer's first stage is a JFET source follower biased by a single-transistor bipolar current source. Stacked diodes are in series between the follower and the current source; the stage then drives a complementary emitter follower. A separate follower is used for each of the balanced signal lines. The output of this buffer stage has a significant dc offset, which is rejected at the input of the next stage, since dc offset is a common-mode signal to this stage. The output of the buffer is routed into a balanced amplifier using a classic allbipolar complementary op-amp. The differential pairs are biased by current sources, and cascaded emitter followers are used for the output stage. A dc servo removes the dc offset at the output. One op-amp is used as a differentialto-single-ended converter (see the Parasound poweramplifier review for more on this circuit) for the plus outISSUE NO. 21 SPRING 1994

put, and a second op-amp is used in an identical circuit except for reversed inputs for the minus output. Why a true balanced op-amp (two inputs and two outputs) was not used here instead of the two single-ended op-amps is unclear to me, but that is not the strangest thing about this output stage. The strangest thing is the way its gain is set. Consider the feedback equation G = A/(l + AB), where G is the closed-loop gain, A is the open-loop gain, and B is the fraction of the output that is subtracted from the input. The Krell's output amplifier has a low-gain mode and a high-gain mode, switched by a relay. The high-gain mode doubles the G. In the low-gain mode the open-loop gain A is approximately 6 (the gain stages are highly degenerated and heavily loaded), but the external feedback resistors that determine B are set for a closedloop gain G of 10. Clearly you are not going to get any more gain out of the amplifier than its open-loop gain A, although the presence of the feedback loop does have a slight effect on the closed-loop gain G. Now, you are not going to believe what that gain-switching relay changes. No, it doesn't change B; it changes the open-loop gain A, which no other designer would touch in this application. It does this by reducing the emitter degeneration of the first stage, so that the open-loop gain is now increased to approximately 25. So what we have here, friends, is an output stage that has just a little feedback in the highgain mode and virtually no feedback in the low-gain mode! (I told you the design came from another planet.) Finally, the output stage is connected through a muting relay to the output jack. This prevents turn-on and turnoff pulses from getting to the output. This important feature often does not show up in megabuck preamps. The muting function is automatically engaged on power-up; you have to disengage it before you can select an input. The thick, high-quality sheet metal of the KRC-2's chassis is what is expected at this price point; it is perforated with many closely spaced slits on top and on the sides for both lightness and ventilation. Inside is a wallto-wall 4-layer PC board stuffed with components. RCA jacks are mounted directly to the rear chassis for added structural integrity. Right-channel jacks are connected directly to the main board. Left-channel jacks are connected to an auxiliary PC board mounted above the main board at the rear of the unit. This board contains the leftchannel relay network. The ac power line comes right onto the main board and into the big (for a preamp) PC-board-mounted transformer. All the circuitry shares the same power-supply regulators. This is not a dual mono design!! Each half of the power supply has its own bridge rectifier and 4700 F filter cap. The voltage reference is a zener diode biased from the unregulated rail by a resistor. This is then filtered by an RC network. The regulator itself is formed with an OP-27 op-amp which drives the base of a bipolar pass transistor and receives an attenuated version of the

the feature most heavily used by those who bypass their preamp). Various buttons permit analog volume steps of 6 dB, controlled by precision resistors, and digital volume steps of 0.2 dB between those large steps. Thus the maximum possible degradation of digital resolution is 1 bit (corresponding to 6 dB), eliminating the single disadvantage of digital volume controls. Pretty neat trick. For absolute-phase sticklers there is also a Phase button on the remote. Now, these buttons here are for regular and decaf. (just kidding, you deadly serious weenies). Bottom line: should you buy the EAD DSP-9000 Pro? If $5500 means little or nothing to you and pride of ownership means a lot, and if you need to switch between a large number of digital sources, then by all means go ahead, with my blessing. It's a superb piece of equipment, able to stand up under the most exacting technical scrutiny, unlike most of its high-end competition. Does it sound different from the $999 DSP-1000 Series II in an ABX comparison? Absolutely not, and I didn't expect it to. Does an audiophile live by sound alone? I don't have to answer that.

Compact Disc Player

Harman Kardon HD7725

(Reviewed by David Rich)

Harman Kardon Incorporated, a Harman International Company, 8380 Balboa Boulevard, Northridge, CA 91325. Voice: (800) 343-9381. Fax: (818) 893-0626. HD7725 compact disc player with remote control, $849.00. Tested sample on loan from manufacturer. One thing you can say about Harman Kardon is that they do not make me-too products. The innovation in the HD7725 CD player is their proprietary RLS ("Realtime Linear Smoothing") technology. The RLS circuit performs linear interpolation between adjacent sample points at the output of the DAC. Linear interpolation provides more filtering than is obtainable by simply holding the sampled signal's value for a sampling interval. The calculation of the amount of filtering provided by linear interpolation is relatively simple but requires a little math. Since your Editor believes this would be enthusiastically received by about 17 readers of The Audio Critic [because it isn't just "a little" mathEd.], I refer interested readers to A. Papoulis's textbook Signal Analysis, page 141. The easily understood advantage of linear interpolation over sample-and-hold is that simpler, lower-order analog reconstruction filters can be used. The hard part of taking that route is the design of an analog circuit which will perform the functions of linear interpolation and filtering. In the HD7725, two DACs are used per channel. The data into the second DAC is delayed in a digital memory by one oversampling period relative to the data

pdf 43

into the first DAC. The output of the undelayed DAC is sent to the CD player's output. Added to this signal is the interpolation signal, which is formed by the following interesting circuit. A signal is generated which represents the voltage difference between the delayed and the undelayed DAC just after a sampling interval. This signal then drives a first-order RC network. When the difference signal steps to a new value, the RC network starts charging. The RC values are chosen so that the exponentially rising signal at the output of the network approximates a linear ramp during the time interval between DAC output samples. Clearly, if the RC network's time constant is not exactly right, the generated interpolation signal can overshoot or undershoot the ideal interpolation signal. The complexity of the whole circuit and its potential to misbehave would seem to argue against its general usage and point to other solutions, such as filtering the signal in the analog domain or putting the entire interpolation circuit in the digital domain. The CD mechanism of the HD7725 uses a number of plastic gears and a small dc motor to articulate the laser pickup assembly. At this price point it is more typical to use direct-drive linear motors, which should be in theory more reliable. (I must hasten to point out, however, that I once had a problem with sticking rails on the linear motor of my Pioneer Elite PD-71, resulting in random skipping. This was easily fixed by cleaning and lubricating the rails, so if you have such a problem don't let a crooked repairman sell you a new laser!) The electronics of the Harman Kardon CD transport mechanism consist of three large-scale integrated circuits, two smaller-scale circuits used to drive the motors, and about a hundred discrete components. All the electronics associated with remote sensing, keyboard decoder, fluorescent display, and microprocessor management are in just one additional VLSI chip. The power transformer of the HD7725 has four secondaries. The analog supplies are filtered by 2200 F capacitors on the unregulated supply rails; the digital supplies have larger 4700 F capacitors. Five voltage regulators are used in this CD player; all are discrete open-loop devices. Zener-diode voltage references are biased from the unregulated supply rails. The pass transistors are a 2-transistor compound device. An additional transistor forms a current limiter for the regulator. Regulated analog supply rails are a relatively low 12 V, probably because only 30 V rms is available on the transformer secondary that powers the analog circuits. At this price point ($849), I have come to expect a more robust power supply. As for the PC board, it has a top-side ground plane but does not have plated-through holes. The relatively thin sheet metal of the housing is reinforced at the top and bottom of the unit with additional steel plates. The digital filter chip is the NPC SM5840. No, that isn't an advanced version of the SM5813. It is a a cost-reduced version, with fewer than half the taps. As a

D/A processor of choice, even for the audiophile who can afford it, because it does not deliver the performance that would be expected for the price. Yes, there is highlevel engineering in the Studio, including the industry's first properly performing DSP-based 16x digital interpolation circuit. Yes, the Studio is very well constructed (except for that power-supply board!), with premium parts. But the clock-recovery circuitry is identical to that in units costing thousands less, and the analog stage is actually less advanced than in units costing a small fraction of the price. Indeed, the Sentec DiAna, which uses the Phototronics PA630 analog IC, has all its analog stages operating with little or no feedback and costs only $1150 for all its political correctness. Measured distortion in the Krell Studio was disappointing, especially since it is a high-feedback design. In view of all this, a high-ender might be surprised that it did not sound a little worse than some of the lower-priced competition, but a series of ABX listening tests showed it sounded just like the rest of the group here. So I am keeping my EAD DSP-1000 Series II and feel no loss now that the Krell is gone (except that I have to pull the EAD's power plug to avoid the massive radiated RFI whenever I want to use the radio or TV).
Integrated D/A Converter and Line Amplifier

Monarchy Audio Model 33

Monarchy Audio, 380 Swift Avenue, Unit 21, South San Francisco, CA 94080. Voice: (415) 873-3055. Fax: (415) 588-0335. Model 33 Dual 20-Bit D/A Converter with Class A Line Amplifier, $1199.00. Tested sample on loan from manufacturer.
The reader is referred to David Rich's reviews, in Issue No. 19, of the Monarchy Audio Model 22A D/A processor and Model 10 line-level preamp. The Model 33 combines the two at a saving of $779.00. So far, so good. We have received no schematics, so I cannot report to what extent, if any, the circuits have been modified. The D/A performance is definitely improved; full-scale THD + N is now -89 dB up to 2 kHz, rising to a maximum of -83 dB at high frequencies, but this includes gain-related analog distortion. The irreducible D/A distortion due to other causes, measured with a -20 dB digital input and normalized to full scale, is -94 dB, still not doing full justice to the top-of-the-line Burr-Brown DAC. The power-supply-related bumps are gone from the noise floor with digital zero input, but the highfrequency noise level is too high (-104 dB at 20 kHz, -79 dB at 200 kHz). Strangest of all, the D/A frequency response is up one full dB at 20 kHz. All of the foregoing was measured at D/A out, bypassing the line stage. The line stage still measures the same as that of the Model 10 (i.e., just fine), and the Model 10's loony-tune source switching system is gone by default, since the

cently made on digital technology vs. digital politics in the wake of the R-DAT; see the DCC article in Issue No. 20 for that (p. 44). I tried out a political zinger on Dr. Lagadec at his press briefing, and he handled it candidly and illuminatively. I asked him if Sony would ever have come out with the MD if the R-DAT hadn't run into heavy political opposition. He replied that they probably wouldn't have (!), not in the MD's present form, but that the idea of a wafer-thin, 2 square, hard-shell, highly efficient, recordable/erasable data-storage medium was too attractive not to be pursued by the Sony technical teams, and therefore some sort of product would have come of it. I was satisfied with that answer. The fact is that the MD, when you hold it in your hand, is smaller, cuter, cuddlier, more lovable than the DAT cassette, let alone the relatively massive DCC. You want it to be the winner, even though in purely sonic terms it isn't. The Sony party line, in all their advertising and PR, is that the MD is not intended to compete against the CD or the DATheaven forbid!but that it is the modern replacement of the dinosaurian analog cassette. Time will tell whether such positioning is viable; right now a recordable MD costs six to eight times as much as a high-quality blank analog cassettenot very tempting to defectors. (I said recordable MD because prerecorded MDs are actually a different medium in the same format, just a tiny CD inside the plastic cartridge.) The principal audiophile reservation about MD has to do with the 5-to-l data reduction via the ATRAC (Adaptive TRansform Acoustic Coding) compression algorithm. As our readers know, I and other accountable reviewers (as distinct from out-of-control tweaks) have found the 4-to-l PASC compression in DCC to be transparent to music, at least the music tried so far. The additional compression with ATRAC poses a problem. The second-generation improvements have made a difference, but on the basis of my very limited double-blind CD vs. MD listening comparisons I am not ready to declare MD transparent. In all fairness to Sony, they do not claim such transparency. I plan to do more testing because the sonic degradation, if any, caused by the current version of ATRAC is quite subtle, and the prerecorded MDs sound pretty much like acceptable, but not great, CDs. For example, if I walked in while the Dvorak cello concerto with Yo-Yo Ma, Lorin Maazel and the Berlin Philharmonic (Sony Classical SM 42 206) was being played, it would never occur to me to say, "Hey, what's that? It's not a CD." It sounds like many middling CDs in my collection (fine performance, though). It should be noted that German researchers using the NMR (Noise/Mask Ratio) technique have found that ATRAC does not completely satisfy the mathematical/psychoacoustical model of a presumably transparent perceptual coder. That research, however, comes from the first-generation era of the system.

pdf 51

off on high or loud notes (perhaps to prevent overload somewhere in the chain). It's interesting to go back to the finished recording to see how these actions have affected the sound. At the very least, the backingoff motion increases the vocalist's "leakage" into the orchestra mikes. The rare person who likes to make measurements of recordings will be fascinated to find that those musical segments of the finished recording that in the movie obviously originated from video still carry the near-ultrasonic horizontal scan frequency that leaks out of video cameras and monitors and is easily picked up by microphones. This can be easily seen with an FFT-based spectrum analysis, which shows a distinct spectral component between 15.75 kHz and 15.78 kHz about 80 to 90 dB below full level during those segments. Even more fascinating is that this spectral (and spect[e]ral) component is slightly higher than the standard European horizontal scan frequency (which has long been 15.625 kHz). This possibly indicates that the CD's analog master tapes were running slightly faster (by about 17/100 of a musical semitone) during the mastering process than they did during the sessions. Rosetta stone indeed! Other technical notes: The sound is very good mono, the picture blackand-white with variable quality depending on whether the shots are from film or video. Although the narration is in English, as are most of the on-camera comments, it is a boon to know German, the language Solti usually uses with the orchestra. The camerawork is goodmeaning conservativewith few zooms and no dental-exam ultraclose-ups of spitspewing and sweating vocalists. There are a few misdirected shots in some of the extended musical portions (such as Siegfried's funeral procession), but they serve only to lend a video-vrit feel to the proceedings. Again, a must-see. Wolfgang Amadeus Mozart: Don Giovanni. The London Classical Players, Roger Norrington, conductor. EMI CDCC754859 2. Don Giovanni, the greatest of

pdf 57

Mozart's operas, receives here its most "authentic" recording to date, in musicological rectitude a veritable doctoral dissertation compared with the term-paper quality of the only other original-instrument recording (stman's on L'Oiseau Lyre 425 943-2) and with the high-school essay that was Richard Bonynge's earnest early attempt at correct performance practice (London/Decca, not yet on CD). The authenticity in this case extends to the ambience (a rather drysounding Abbey Road Studio 1), the employment of unwritten, but stylistically mandatory, appoggiaturas by the singers, and a true 18th-century operatic seating arrangement for the orchestra (and not just a version of the 19th-century/divided-violins layout). A seating diagram is helpfully given in the accompanying booklet, which also describes the simple CDplayer programming to use to make the discs play either the original (Prague) version or Mozart's own revised (Vienna) version. Music is provided for both. Most surprising to most listeners will again be Norrington's tempos, which are usuallybut not alwaysfaster than normal. The most notable exception is the so-called Champagne Aria ("Finch'han dal vino," which mentions only wine). [Champagne is a wine but it did not exist in Don Juan's early-17thcentury Seville. Amontillado Aria would probably be more appropriate.Ed.] For once it is taken at a tempo that doesn't force the singer to bark out his notes and gains thereby an extra dimension of snarling lasciviousness. Otherwise, the drama sweeps along at a theatricalnot "operatic"pace, with possibly too few moments of repose. There certainly is little very soft singing, even when it might be appropriate dramatically and sanctioned by the score. While vocally there is no weak cast member, characterization during the arias and ensembles is shortchanged by all the singers. A little more personality from them would have helped integrate these pieces more fully into the rapidly moving recitatives. Whatever one may think of these points, there can be little disagree64

46 Productions
42W557 Hawk Circle St. Charles, IL 60175 [630] 46pro@46p.com
Surplus Equipment Inventory
From time to time, 46 Productions retires equipment that has been replaced or is no longer needed or used to provide our services and products. The following list details our current surplus inventory. All equipment is studio use only and has not been used outside of the studio for any reason or purpose. Manuals and accessories are usually available. In some rare cases, we may even have the original boxes. Contact us by phone or email to discuss the equipment you are interested in to arrange for a demonstration or purchase.
TERMS: Prepaid. Cash, Visa, MasterCard. PRICES: Firm. However, I will consider all reasonable offers. SHIPPING & INSURANCE: Paid by purchaser. CONDITION: All equipment is in working order unless otherwise stated. Repairs, if any, performed by authorized service centers. Operational status will be verified before shipment or delivery. Equipment can be inspected at our location. RETURNS: None. Make sure you want it before you buy it. Equipment that is damaged in shipment should be reported to the shipping carrier.

SOFTWARE:

Sonic Foundry Sound Forge 5.0 - $65.00. With printed manual. Original discs and serial number. McAfee Virus Scan Professional 6.0 - $15.00. Original discs and serial number. McAfee Virus Scan 6.0 - $10.00. Original discs and serial number. McAfee QuickClean 2.0 - $10.00. Original discs and serial number.
Kurzweil K2000R- $995.00. With AES/EBU and optical digital I/O. Manuals and sample diskettes. Very good condition. Studio use only. Yamaha QY-20 Music Sequencer - $195.00. 8-track sequencer and auto accompaniment. 100 voices, 8 drum kits, 100 accompaniment patterns. Manual. AC adapter. Very good condition. Studio use only. Roland GM-70 Guitar MIDI Converter- $225.00. With GK-1 Pickup. NOTE: This unit requires an external MIDI sound module. Manual. Very good condition. Studio use only.
STUDIO EQUIPMENT: Sony PCM-2600 DAT - $495.00 Sony SBM and high-precision A/D/A converter. With manual and i/r remote. Excellent condition. Studio use only. Approximately 635 hours.
Otari DTR8 DAT - $395.00 2-Head Pro deck. Excellent condition with very little use as a backup deck only. With manual and i/r remote. Approximately 175 hrs. Studio use only. Excellent condition. Mar 28 2005
MARANTZ CDR-620 CD Recorder - $395.00 Balanced analog I/O. Digital I/O - AES/EBU and S/PDIF. Includes input sample rate conversion and translates track numbers from DAT, MD, DCC to CD track IDs. Full feature wired remote w/display that duplicates front panel. Has SCSI port to connect to computer for use as 2X drive. Many other features. Excellent condition. Studio use only. Very low hours with use as a backup deck only. Sony MDS-501 MiniDisc Recorder - $195.00 Consumer type deck with pro features. Includes optical digital I/O (s/pdif). Analog I/O is unbalanced. Editing functions include dividing/combining tracks, moving tracks and labeling recordings. Full featured wireless remote and manual. Very low hours. Studio use only. Excellent condition. HHB Bit Box - $145.00 Translates sub-code data from DAT start id to CDR track increment information. AES/coaxial and optical I/O. Input select controls. All outputs active. Has delay feature to align track id information with audio. Studio use only. Excellent condition. ROLAND DM80-8 Digital Audio Workstation - $395.00 Quality digital audio workstation for post-production, music recording, project studio and radio or television. Internal SCSI disk drives. Includes DM-80-R remote controller and DM-80-F remote fader. Supports MIDI & MMC. SCSI ports for adding external disk drives. Software Versions: Rack Unit 2.23 / Remote Fader 1.00 / Remote Controller 2.23. Studio use only. Excellent condition. Ask about external SCSI disk drives available. Unit is heavy. If shipped, expect to pay for large, heavy package and insurance. Yamaha ProMix01 Digital Mixer - $695.00. With Yamaha rollaround stand and Rapco custom 16-channel snake. Studio use only. Excellent condition. Prefer not to ship. Digital Audio Labs Digital Only Card - $10.00 Standalone PC card for digital I/O in S/PDIF formats (coax). 16-bit ISA card requires AT Bus Slot. Supports 32, 44.1, 48kHz sample rates. Card is compatible with most Windows software. Features duplex operation (play while record) for overdubs and punch in/out with supporting software. If you need high quality two-channel 16-bit digital I/O for your computer system you cant go wrong with this card. I have latest Windows drivers that work with both Windows 95 & 98. This card is NOT plug & play. No manual but I have how-to and FAQs from Digital Audio Labs. I believe they will still provide tech support for this card but no software updates. JL Cooper PPS-1 MIDI-SMPTE Synchronizer - $10.00 Uses fixed rate SMPTE time code to provide audio/MIDI synchronization. Allows MIDI devices to sync to tape with song position pointer ie, allows them to chase tape. Includes feature to convert SMPTE to MIDI Time Code (MTC). With manual and functioning power supply.

MISCELLANEOUS:

Rorke Data DM2000 SCSI HDD Cabinet - $395.00. 2-Seagate Hawk ST31200N HDD in 19 rack. Excellent condition. Studio use only. Pacific Coast Technology PCT1200S SCSI HDD Cabinet - $195.00. 1-Seagate ST11200N HDD in 19 rack. Excellent condition. Studio use only. Samsung WN3216U SCSI HDD - $75.00. Internal HDD. Little use. Excellent condition. Studio use only. Pioneer PL-520 Direct Drive Stereo Turntable- $75.00. With Shure RXP3 cartridge. Good condition. Cover is scratched and faded.

Mar 28 2005

Ultimate Support Keyboard Stand- $65.00. A-Frame style with 3-tiers. Silver. Very good condition. Studio use only. Ultimate Support Keyboard Stand- $65.00. A-Frame style with 3-tiers. Black. Very good condition. Studio use only.

OFFICE EQUIPMENT:

Siemens Gigaset 2420 Desk Phone- $125.00. With one handset. Very good condition. HP 2000C Inkjet Printer - $95.00. With extra ink cartridges. Very good condition. Ricoh 3100L Laser Fax / Copier- $225.00. Heavy duty fax / copier with laser printing. Uses plain paper. Extra toner cartridge. Existing cartridge is about half used. Very good condition.