Return to biline.ca Audio/Video Section
Read some articles from The Audio Critic Magazine.
A New Way to Examine Them
By David A. Rich, Ph.D., Contributing Technical Editor
and Peter Aczel, Editor and Publisher
We look at high-end audio circuitry not from the testing point of view but as an engineering discipline—and find no consistency.
Editor’s Note: This article is based on two engineering papers delivered by David Rich in 1995 before two different professional societies. The first was an advanced tutorial at the DSPx ‘95 conference in San Jose, CA, addressing the question: 'Are there any design considerations in audio that go beyond standard measurements?' The second, coauthored by your Editor, was presented at the 99th Convention of the Audio Engineering Society (October 1995, New York) under the title 'Topological Analysis of Consumer Audio Electronics: Another Approach to Show that Modern Audio Electronics Are Acoustically Transparent' (Preprint 4053). What follows here is a digest of these two professional papers, revised and edited for the audiophile consumer. Technical details left out here can be found in the AES paper. We ask our readers’ indulgence for the few audiophile commonplaces that have been left in; these are not always obvious to professional engineers who don’t read the consumer magazines. The professional perspective, on the other hand, is one of the benefits the audiophile can derive from this material.
Degreed electrical engineers tend to regard the design of audio electronics to be trivial compared to other design challenges of consumer electronics, yet it is more complex than appearances would suggest. Some of the design challenges may not relate to measured performance but instead to consumer expectations and beliefs.
For example, it is not possible to sell an audio product with a sharp-cutoff analog reconstruction filter. Consumers have been told by dealers, sales literature, and audiophile magazines to believe that the asymmetric ringing on square waves associated with such a filter has a detrimental effect on the audible quality. As a result, interpolating digital FIR filters placed before the DAC are used for the reconstruction function. This digital approach results in symmetrical ringing of the square waves, which consumers have been informed is insignificant. In this example, we see the first occurrence here of those troubling words that keep coming up in audio design: believe and audible.
Controlled vs. open-loop listening tests.
The problem with audio design is that the performance is assessed by listening to the unit. The subjective nature of the assessment of audio components makes it possible to make outlandish claims for the audible characteristics of a piece of audio equipment. Outlandish claims are more easily made for comparisons involving hearing than those involving vision. When people claim that they see something different, they can physically point to the difference to substantiate that the difference exists. This is not possible when someone claims to hear something that another listener cannot [Eargle 1995]. Audio dealers constantly invoke the phrase “Can’t you hear that?” as they attempt to close a sale. Fortunately, controlled listening-test methodology can be used to eliminate much of the uncertainty of making a subjective evaluation.
Unfortunately, comparisons of audio equipment at the consumer level are never done with double-blind tests. The 'buff books' that consumers of audio equipment read are also not using controlled subjective listening testing (except The Audio Critic). They instead use unreliable open-loop testing. The brands of the equipment are known to the reviewer, and levels are not carefully matched. As a result, elaborate descriptions of the sound quality of the equipment appear in print. As would be expected, given that the testing methodology is so sloppy, little correlation exists in the descriptions of the sound of a component from reviewer to reviewer. Indeed, even those who support the open-loop listening method can only cite loudspeaker reviews as consistent in describing the sound of a given product [ Harley 1991 ]. The only thing that does appear to correlate in the reviews is that more expensive equipment is felt to sound better than less expensive equipment.
Why should the magazines want to perpetuate such sloppy testing? The answer has two parts: (1) The highend audio industry depends on convincing consumers that large differences in components exist and that expensive components are clearly better than cheaper ones. Buff books expounding this philosophy have increased the number of their advertising pages from both dealers and manufacturers. (2) Controlled listening tests may be difficult to set up, and doing multiple listening trials represents real work. Most of the writers in consumer magazines are having too much fun doing open-loop listening tests to want to change to a methodology that requires real work.
Controlled listening tests have consistently shown that electrical components will be audibly indistinguishable if the have: (1) flat frequency response, (2) noise and distortion levels below audible thresholds, (3) high input impedance and low output impedance [D. Clark 1982]. So how can the audio magazines justify the differences they report, when the measurements they make on the equipment show that well-designed audio electronics have virtually flat frequency response and noise plus distortion 80 dB or more below the fundamental test signal? The answer is that they use pseudoscience. Consider the following quote from Stereophile magazine [Harley 1990], discussing modifications to CDs that make no technical sense.
'I see CD tweaks as a Rosetta Stone to an audio engineering establishment that dismisses the possibility that freezing a CD, or painting it black, or putting green paint around the edge, or making it from a different material, could affect its sound. Because these treatments are considered the epitome of audiophile lunacy and because they are readily audible, some measurement-oriented scientists may, if they listen for themselves, realize that audiophiles are not always the demented mystics they are often accused of being.'
This idea that undiscovered phenomena are responsible for the discrepancy between measurement and audible observations is commonly expressed in the audiophile press. It should be noted that none of the items discussed by Harley have ever been shown to be audible in controlled listening tests.
Often the semitechnical (to use a kind word) audio writer may try to claim that a measurement that can be made only inside the equipment is responsible for the sonic performance. For example, it has been reported— once again in Stereophile [Harley 1994], but there are also other examples—that CDs which have more eye pattern closure at the photo detector output of the CD player have poorer sound. The fact that the designers of the CD playback system designed the system to be insensitive to eye pattern closure, as a result of reclocking of the data by a stable crystal reference, is never discussed. Because the signal is reclocked, it is impossible to see any effect of the closure of the eye pattern at the CD player’s output jacks. Other examples of this kind of untutored reporting are discussed in [Aczel 15, 17].
Disinformation for the innocent consumer.
It is important to keep in mind that 'technobabble' concerning audio equipment is not confined to the buff books but has spread to such highly respected business publications as Forbes and Business Week [Aczel 16, 22], which have both published articles claiming that tube amplifiers produce sound quality superior to that of solidstate units. As a result, consumers who are not directly involved in the hobby may still be led to believe many of the strange notions of the audiophiles. Just as in the audiophile press, these articles quote no electrical engineers to give a plausible explanation of why a tube amplifier might sound different, such as the high distortion and high output impedance measured in typical tube amplifiers.
Improperly conducted subjective testing methodologies have given rise to an entire industry that produces what can only be described as consumer rip-offs. These products cannot possibly affect the sound quality of a system. They exist because it is possible to convince the purchaser through techniques of salesmanship that a sonic change has occurred when no real change has occurred. Examples include: (1) a $400 LED clock claimed to improve the sound of a system when it is plugged into the same power line as the stereo, (2) brick-shaped objects of wood filled with ferrous metal inside that are claimed to improve the sound quality when placed on a stereo component, (3) a device which generates signals to be sent through an interconnect or speaker cable for the purpose of “burning the cable in” to obtain better sound, and (4) the CD “tweaks” mentioned by Robert Harley in the quote above.
Although it is possible to sell consumers the above products despite the fact that they do nothing, it becomes even easier to sell products if they do change the sound, and it turns out that some audiophile products can yield a definite sonic change which is observable in controlled listening tests. Audiophiles have a tendency to assume that if a product sounds different it must be better, but in most cases the differences can be attributed to the introduction of frequency response errors or the addition of distortion. The single-ended low-power class A triode amplifier craze is an example of this.
One easy way to introduce frequency response errors is to increase the output impedance at the amplifier loudspeaker interface by inserting a series inductance or resistance. This is exactly what is done by very expensive speaker cable. These cables, which can cost over $1000, do nothing but add a series inductance and series resistance. Expensive power amplifiers often have very high output impedance, which will affect the system transfer response in a similar manner. Cable manufacturers often take advantage of the high output impedance of such power amplifiers to further modify the transfer function. This is done by designing the speaker cable to have significant parallel capacitance. Frequency response plots showing these effects can be seen in Issue No. 16 [Aczel 16]. Significant errors in the RIAA equalization curve of a phono preamp are often present in designs preferred by audiophiles. Similar things can be done in the digital domain. For example, coefficients in a digital filter have been derived in the time domain, using spline functions, by manufacturers such as Wadia [Moses 1987]. It is claimed that better sound results from this. While such a claim is highly debatable [Lipshitz 1991], it is not debatable that this technique results in the frequency response being down by 3 dB at 20 kHz, and that response change is audible.
Nonlinear distortion mechanisms are often present in designs preferred by audiophiles. One recent fashion that has proved popular among audiophiles is a preference for single-ended amplifiers. Such amplifiers have significantly higher levels of distortion than push-pull designs. Distortion well over 1% is observable in these amplifiers. Audiophiles have often shown a preference for amplifiers with little or no feedback because they believe that feedback somehow makes the sound quality worse. In the real world, amplifiers using little or no feedback will exhibit increased distortion, especially at lower frequencies. Again distortion of 1% or more has been observed in low-feedback designs, although careful use of local feedback and other innovative techniques can bring the distortion below the level of audibility.
The designer's dilemma.
Given the confusing trends and frequent irrationalities of the consumer audio market, how is a designer to respond when assigned the task of designing consumer audio components? Three approaches can be taken by the professional engineer: (1) Modify the frequency response of a product, or add distortion to the product, so that it will have a different sound characteristic and thus achieve a competitive advantage; (2) ignore any considerations of sonic issues and design to achieve a design with flat response, low distortion, low noise, high input impedance, and low output impedance; (3) design the circuit so that it is judged to be transparent under controlled listening tests, and let the sales and marketing department justify that the design is superior.
A key question for the designer to answer is whether or not approaches (2) and (3) will lead to the same results. If they do not, then what are the design parameters required to achieve sonic transparency? This is the question that we seek to answer here. Our own experiences have shown that when a piece of stereo equipment passes the traditional set of measurements (distortion, noise, frequency response, input impedance, output impedance), then it will not show any acoustic signature in a controlled listening test. But the marketing department must be aware that audio components that achieve excellent specifications at relatively low price points are often not the market leaders. Indeed, one of the most remarkable products we have tested, the Boulder 500AE stereo power amplifier, based on a topology discussed in [Jensen 1980], achieved distortion levels as low as a perfect 16-bit converter at full power across the audio band, but it has not been a major success. We can see the problem in the reaction to the excellent measured performance of this amplifier by one reviewer, who wrote the following [Harley 1992]:
'...the 500AE was designed on paper, rather than in an iterative listen/design/listen process. ...I believe the Boulder 500AE to be good, but not extraordinary...'
Contrast the above reaction, to an amplifier that measures really well, with the following quote, from a review of a Jadis tube amplifier [King 1986]:
'So why do these amplifiers sound so good? Now my job gets tough. The bottom line: I don’t really know. The only really good measurement is of its harmonic structure. Output impedance and amount of harmonic and IM distortion are OK, but, the gross slewing or reduced high frequency power output is bad. The truth is that the measurements most of us make are not very relevant to the sound of circuits and I’ve spent a good part of my career looking for ones that do with little success so far.'
In no other field would this statement be taken seriously. It appears very likely that this amplifier was perceived as sounding good because it falls into category (1).
It should be noted that one possible approach for the designer is to work in category (1) and then disclose what he is doing. This is the approach that has. been used by Bob Carver. He modified the transfer characteristics and output impedance of a low-cost solid-state power amplifier to match a very expensive tube amplifier that the audiophile community thought had the best sound quality. In double-blind listening tests he then demonstrated that the two amplifiers sounded identical. Unfortunately, the consumer audio press and the distribution channels reacted very negatively to Bob Carver’s approach. Their reasoning was obviously that if Mr. Carver’s approach were given Legitimacy, it would no longer be possible to sell the expensive tube amplifier and other very expensive audio components. Bad publicity from the audiophile press reduced the sales of the Carver amplifier dramatically. Understanding how a given engineering approach will be accepted by the press and the public is an important issue for the marketing department, but should not be the concern of the designer.
If the designer wants to work in category (2), he needs to know what are the limits on the electrical specifications that must be achieved for a unit to be audibly transparent [Olive 1995]. There is a wide range of answers, depending on which investigator’s result you are reading. Signal-to-noise ratio clearly needs to be better than 98 dB if one wants to pass a 16-bit digitally encoded signal without degradation. It has been shown that in sonic cases a dynamic range greater than 98 dB may be required to capture the full dynamic range of music [Fielder 1989]. Because the ear is more sensitive in some frequency bands, noise shaping can be used to increase the subjective dynamic range of a 16-bit system. If noise shaping is used, the signal-to-noise ratios of the D/A and analog sections may need to be in the range of 110 dB [Benjamin 1993] if the noise in these sections is flat over the audio band. With respect to distortion, most studies have shown that 0.1% distortion is inaudible even when using test tones. Other studies have shown that distortion on the order of 2% is not audible on music signals [D. Clark 1982].
The ear appears to be very sensitive to level differences, and overall level mismatches larger than 0.1 dB may be perceived. Localized deviations of frequency response may require larger deviations to be audible. The upper frequency response limit is typically determined by the sampling rate of the signal if it has been, or will be, digitized. At the low end, the loudspeaker usually rolls off well before electronic components. One consideration that causes some designers to place the low-frequency rolloff at a very low frequency is phase errors at 20 Hz. It has been shown that the ear is slightly sensitive to phase errors in the bass spectrum [Fincham 1985].
Input and output impedance specifications are principally determined by frequency-response considerations. In most cases these errors are most likely to occur at the amplifier-speaker interface, but it is possible to observe high-frequency rolloff at low-level signal interfaces [Aczel 17]. This is almost always due to a component with a very high output impedance (3 to 50 kΩ) driving a cable with very high capacitance (1 nF).
A new way to evaluate design differences.
Is it possible that a design accomplished with the approach in category (2) will not lead to a transparent audio component? One way to answer this question is to look at already existing designs that are claimed to have excellent sonic qualities on the basis of open-loop listening tests. By examining the circuit topologies, we should be able to determine if some new design approach is being used that would not have been used if the design had been developed simply to achieve good results in traditional bench measurements. Such examinations [Rich 15, 18, 20] have shown that these designs do not show any particularly unique circuit-design topologies. Furthermore, little commonality can be seen among the designs’ topologies.
It is a reasonable assumption that a superior audiocircuit concept would slowly but surely recruit a much larger following than an inferior one, so that eventually there would be some kind of consensus among practitioners and a discernible convergence toward the superior topology as new designs emerge. Total randomness in the choice of topology would indicate that no single approach is clearly superior to any other, in much the same way as total randomness in the results of a double-blind ABX test indicates that there is no clearly audible difference between A and B. It’s basically the same statistical criterion. Thus, if a designer had discovered a unique topology that sounded better but did not measure better, we would expect that topology to spread to other companies through “reverse engineering” of the product with the superior sound. There is no evidence whatsoever that this is actually happening.
Although nothing especially unique can be found in the design of audio components as distinct from other electronic products, a number of circuit-design techniques can be seen in high-end audio equipment that are distinct from common design practice. As stated previously, we have found that none of them yielded a product with superior sound quality, but it is still interesting to look at them to see if they can reveal any clues regarding the thought processes of the designers of the equipment. Note that the summary below has been generated on the basis of many different pieces of audio equipment, with no single one of them using all or even most of the techniques outlined.
1. Discrete circuit design
High-end audio designers tend to shun integrated op-amps. This gives them the added flexibility to design circuits such as are discussed below, which they would not have designed with a single monolithic device. The principal advantages of discrete design from a measurement point of view are reduced noise levels and increased output drive capability. The downside is that the circuit will be slowed down because of larger parasitics, and it may cost more. The former problem is not a significant concern at audio frequencies.
Of course, many high-end products still have integrated operational amplifiers in the signal path. The opamps may often precede or follow the more exotic discrete circuitry favored by the high-end designers. Since the op-amps have none of the design features that these designers believe are required to prevent sonic degradation, such practice appears very strange. For example, high-end designers who hold that an amplifier must use very low levels of feedback to prevent sonic degradation may use op-amps that have high feedback levels at low frequencies in the signal path [Rich 21]. Despite their deviation from the “politically correct” techniques common to high-end design, electronics with integrated operational amplifiers in the signal chain often receive very favorable reviews. Go figure.
2. Extensive use of FETs
Designers' typical explanation for this is that FETs perform more like tubes. A more scientific explanation involves the fact that these devices increase the input stage’s dynamic range (see 6. below) and that they are more robust into short circuits (see 3. below). Some IC manufacturers often encourage the use of BiFET op-amps for better sound quality using the explanation that they have a distortion characteristic more like tubes [Burr-Brown 1992]. This begs the question that, if the distortion numbers are very low, why does it matter what the characteristics of the distortion are.
3. Class A output stages
Audiophile folklore has always stated that class A amplifiers sound better. This can be carried as far as biasing the output stage of a power amplifier into class A. Clearly, crossover distortion is possible in class AB stages, but this can be minimized [Sandstrøm 1983] in good designs. If the designer is using an op-amp, he may put a load resistor from the output to the negative supply rail to cause a large dc current to flow. This dc current forces the npn output transistor on for the full swing of the output, yielding class A operation [Jung 1986]. One design we have examined got this backwards by placing the resistor to the positive supply rail, forcing the slow pnp transistor on instead. Despite this the amplifier has received good reviews. (The old football-team adage “what the ref don’t see don’t bother the ref” appears to apply to high-end audio reviewers as well.)
Some power amplifiers are stated to be class A by manufacturers so that audiophiles will think they “sound good,” even when they are in reality class A/B amplifiers with high quiescent currents levels. Other amplifiers use dynamic biasing circuits which keep the output device that is not driving the load biased to a small constant quiescent current. This technically fits the definition of a class A amplifier, but since the output devices still experience wide variations in current flow, the problems identified by Sandstrøm still apply. This dynamic biasing approach has been adopted in nonaudio applications that must run at low power-supply voltages [Sakurai and Ismail 1995]. The designers of these dynamically biased output stages still refer to them as class A/B and never class A. Audiophiles are not aware of such distinctions and in open-loop listening tests they find amplifiers labeled class A to have excellent sonic qualities in comparison with class A/B amplifiers—even if the so-called class A amplifier passes into class A/B above a certain signal level or if the amplifier labeled class A uses dynamic biasing.
Other design trends in output stages can also be seen. In low-level signal stages, current limiting circuits are not often used. Resistors in series with the output provide the current limiting. In power amplifiers, current limiters may still not be used, with designers relying on rail fuses to protect the amplifier. If current limiting is used, it will not be the simple one-transistor foldback design. It can be shown, using some novel test procedures, that an improperly designed current protection circuit can activate prematurely into real loudspeaker loads [Otala 1987], [Baxandall 1988]; thus, relatively sophisticated protection circuits are required. The effect of carefully designed current-protection circuits can be measured by checking to see if the voltage output is reduced when driving reactive loads. Often these tests must be done on a dynamic basis because the amplifier may have inadequate heat sinks for steady-state testing [Rich 20]. It should be noted that many amplifiers said to “sound good” in openloop listening tests have protection circuitry that operates poorly.
4. Low levels of global feedback
One of the parameters often identified as important is the amount of global feedback that may be applied in the circuit. This concept was first identified in [Otala 1970], although Otala’s analysis has not been accepted by some peer reviewers [Cherry 1982], [Cordell 1980], [Cordell 1983]. In its simplest and most common form, the open-loop gain is made constant up to the 10 kHz to 50 kHz range. This is accomplished by resistively loading the voltage-gain stages in the amplifier. This approach increases distortion at low frequencies, since the return-loop gain is held constant across the frequency band. Gain stages often have large amounts of emitter or source degeneration in an attempt to linearize the gain stages with local feedback, so that global feedback rates can be reduced. It is well known from feedback theory that multiple small feedback loops will not be as effective as one global loop, so the local feedback approach cannot be justified on pure engineering grounds [DiStefano 1990]. Some designs may have no global feedback or very small amounts of feedback (6 dB or less). In these cases designers must move beyond local feedback and use more exotic methods of error cancellation [Cordell 1984].
5. Fully complementary circuits
These circuits are designed to be fully complementary from the input stage onward. This technique may be useful in reducing distortion in amplifiers that are run at low feedback levels because even-order harmonics are canceled. One reason this is helpful in low-feedback designs is that, when large amounts of local feedback are used in the second gain stage, the voltage swing at the input of the stage must be larger, since the stage’s gain is reduced. As a result, the first gain stage’s output swings are higher, and this stage can now contribute significant distortion. The downside of a fully complementary amplifier is increased noise, dc offset, and decreased CMRR.. Another clear disadvantage is increased parts count.
While a fully complementary amplifier may be overly complex and may not always yield the best performance, one aspect of its design can lower distortion over the standard topology. This aspect of the design is the push-pull second stage. This is especially true in power amplifiers, where a large voltage swing occurs at the output of the second gain stage. Achieving a push-pull second gain stage while retaining the high low-frequency gain in the first stage is not a trivial design problem. There exists a published circuit which does this [Cordell 1984]. This circuit, combined with cascodes in the second stage and a buffer stage between the first and second stages, results in an amplifier with remarkably low distortion levels. Despite its many advantages, the amplifier has never found a commercial realization—perhaps because it measures too well to “sound good” to the indoctrinated open-loop listener.
6. Input stages with very wide open-loop linear range [Jung 1987]
A designer who searches the literature will see the term transient interinodulation distortion. In brief, this effect occurs when an amplifier slew limits. Early work suggested that transient intermodulation distortion could be eliminated only if small levels of global feedback were used [Otala 19701. Later work showed that the effect could be eliminated if the input stage linearity was made large enough so that under worst-case conditions the summing junction is never moved outside the linear range of the input stage [Leach 1981]. Designers will use FET devices and/or degenerate the gain device in the front-end differential pair to achieve the wide linear range.
Some researchers have suggested that this requirement results in significant overdesign and that the transient intermodulation effect cannot occur with bandlimited music signals [Cherry 1986]. This explains why bipolar op-amps with no degeneration of the differential stages are acoustically transparent in controlled listening tests. Also note that sophisticated tests are not required to test for transient intermodulation distortion. If an amplifier has low levels of THD at 20 kHz on full voltage swings, it is free of transient intermodulation distortion. If an inband test is required, then some unusual three-tone intermodulation tests can be used [Borbely 1989]. Examining the input stages of amplifiers said to “sound good” will show that the amount of degeneration varies significantly from no degeneration on a bipolar stage to orders of magnitude beyond the emitter (or source) resistance of the active device.
7. Radically overdesigned power supplies.
It is not uncommon to see power transformers and rectifiers much larger than required to drive the power supplies. Multiple stages of regulation are often used [Jung 1995]. Sometimes this is carried to the point where each op-amp has its own local regulator. Discrete regulators are often used in places where a cheaper monolithic device would have been sufficient. Designers will explain that the regulator has to be “fast” to improve PSRR. Exactly why simple bypass caps are not as good is never made clear.
Again, no consistent design practice is observable, since inexpensive monolithic regulators often drive complex, discrete, active electronics in some designs. On the other hand, very complex discrete regulators often are used to drive low-cost op-amps. This approach is common in high-end Japanese designs. One interesting feature in the high-end Japanese designs is the use of a complete push-pull output stage in the regulator. It is unclear why a positive regulator should ever be required to sink current. One assumes that it has a role in reducing transient noise signals on the supply line.
Sometimes high-end discrete regulator stages use no global feedback. This approach, perhaps an attempt to mimic the low-feedback design of the active stages, results in a less capable regulator with much higher output impedance. Some high-end designs will be dual-mono right to the power cord. Other designers will use supply rails for both channels derived from a single voltage regulator.
Perhaps the greatest deviation in power-supply design occurs in power amplifiers. Some amplifiers will have complete regulation of all active elements, including the output stage. Holding the output-stage voltage rails constant is counterproductive if a purely engineering analysis is applied. A much more logical approach would be to dynamically vary the power supply voltage to the output stage so that the VBC (or VDS) of the device could be held constant. This would linearize the output stage and allow for the use of faster devices with lower VBCO (or VDS(max)). Despite the technical quicksand that power amplifiers with regulated output stages stand on, the openloop sonic descriptions have often been very favorable.
Most power amplifiers do not have regulated output- stage supply rails, but some have regulated rails for the voltage-gain stages. This approach improves distortion performance, reduces crosstalk if the power-amp channels share the same power supply, and increases immunity to power-line noise. The downside of regulating the voltagegain stages is a significant increase in complexity because the regulated voltage must be higher than the unregulated voltage applied to the output stage if the available outputvoltage swing is not to be limited by the regulated supply. This requires additional transformer windings for the regulated power supply. Again, so-called “goodsounding” power amplifiers use no consistent powersupply design.
8. Exotic materials.
Parts like 10 µF polypropylene capacitors and Teflon boards are just a few of the strange things that may be found in a high-end audio design. Other weird stuff may include silver wire and very expensive bulk metal resistors. These parts are justified on the claim that they sound better, although some measured differences have been reported [Curl and Jung 1985]. Many designers impose the requirement on themselves that the circuit must be flat to subsonic signals. Often these designers will not place electrolytic capacitors in the signal path [Jung and Marsh 1980]. As a result, parts cost for capacitors can become very high. Often a mix of electrolytic and film capacitors is used to reduce cost. The electrolytics are typically in those places in the circuit that would be most sensitive to capacitor nonlinearities, such as feedback loops. It appears illogical to assume that film capacitors in the less sensitive locations will improve the sound of the circuit if the electrolytics do indeed affect sound quality. One way around this is to eliminate coupling capacitors altogether. The dc offset is reduced with trim pots or active circuitry in the feedback path [D. Clark 1982]. Sometimes circuits are claimed to be directcoupled because capacitors have been removed in the noncritical direct paths, but electrolytics are still found in the sensitive feedback loops. Such designs have received rave reviews from audio journalists whose ears are heavily influenced by simplistic technical claims.
Is there a conclusion to be arrived at here?
The examination of the topologies of audio equipment said to 'sound good' has shown little commonality in the designs. Some designs use no feedback, others a small amount, and yet others a large amount at low frequencies. Some designers include the output stage in the feedback loop; others do not [Dalzell 1995]. Some designers use no capacitors in the signal path; others use only expensive film caps, while still others use less expensive electrolytics. Some designers use complex powersupply regulators, while others use no regulation at all in power amplifiers. Some designers will work mostly with FETs; others use only bipolar devices. Some designers use fully complementary circuits, while others use only single-ended circuits that are the latest vogue. Some designers use ICs in the signal path or for voltage regulation; others only use discrete designs. Some designers may even mix design styles within a given unit.
The random nature of the designs strongly suggests that no 'X factor' parameter is being optimized. Instead, we can assume that the designer, using open-loop listening tests, has convinced himself that the changes he is making to the circuitry are affecting the sound. In this process— design, listen open-loop, design—the circuit designer has no checks and balances to guide him in his work. Openloop listening is to a very great extent subject to the biases of the listener, and a designer wanting to prove that his new idea is better-sounding is clearly biased. Controlled blind listening tests would show if a sonic change were truly happening when a circuit change was implemented, but designers are unhappy when a new circuit idea is shown by such tests to be of no consequence. They thus try to dismiss the controlled test results, instead of facing the reality that electronics exhibiting proper measurements are sonically transparent.
This can often work in reverse. A designer might not use in his circuit an expensive component that would result in a measurable change in the device’s performance because he has convinced himself through open-loop listening tests that the better component produces no sonic change. That is probably the reason why a lot of very expensive high-end equipment uses inexpensive D/A converters or digital interpolation filters. Considerably less expensive mainstream components, often said to sound less good in open-loop listening tests, use much better-performing parts.
The results of the study of the circuits discussed above confirm the results of controlled double-blind tests, which have shown that no sonic differences exist in audio electronics that measure well. When double-blind listening test are performed, random answers occur to the question “Which component sounds better?”. When circuits that are claimed to “sound better” are analyzed, random design techniques are noted. Both analysis techniques, approaching the subject from opposite ends, converge to the same conclusion: audio electronics that measure properly will sound acoustically transparent. No “X facto?’ exists. Designers are wasting their time developing audio equipment using the “design, listen, design” approach because they are not using controlled techniques in the “listen” part of the process. If controlled techniques were used, the designers would discover that audio design is no different form other electronic design. It is done with a set of specifications, with paper and pencil, with computer analysis programs, and with laboratory measurements.
It should also be noted that audio--high-end audio in particular--appears to be the only technological discipline suffering from the peculiar attitudinal syndrome analyzed above. You will not find automotive engineers, for example, claiming that one brand of spark plug (or ignition wiring or distributor) with exactly the same specifications and measured performance as another “feels better” when driving, and certainly not that it makes the car go faster!
We are faced with so many real problems in audio design. It is time for the designers of audio electronics, when they are not accomplishing anything, to recognize it and move on to the solution of those real problems.
Goals for the audio designer.
In the final analysis, the designer of audio equipment should attempt to design the most transparent circuit consistent with cost goals and reliability requirements. Careful analysis of the literature and competitive designs will help him in this goal, as will controlled listening tests.
We believe that the design process can be relatively straightforward. Our recommendation is to design an audio product in such a way that its measured performance exceeds the target specifications. In this case the target specifications come from psychoacoustic studies of human hearing. Value added can be achieved by bringing out a significantly overdesigned product that well exceeds specifications, offering enhanced build quality and improved reliability. It is not the designer’s job to worry about the delirious condition of the consumer audio industry. That should fall to the marketing department, which must come up with a plan to sell a well-designed product in a difficult market. In consumer audio the marketing department may have to work harder than the designer. That might be something of a first.
