ONLINE - ISSUE 15
Why even the best component designs benefit from mods
Steve Nugent is President of Empirical Audio (www.empiricalaudio.com), and an engineer with numerous patents to his credit (for more information about Steve and his company, see my interview with him in PFO Issue 7, at www.positive-feedback.com/Issue7/empiricalint.htm). New readers to PFO should note that we are an open forum for discourse about the audio arts, and for the advance of same. We welcome any constructive contributions from editors, readers, or members of the fine audio industry to the pages of this magazine. Provided that mere self-promotion is avoided, audio professionals are invited to participate without worrying about "conflicts of interest." It is in that spirit that we publish Steve's commentary on the modification of audio equipment; caveat lector.
Interested readers/industry professionals may submit letters/articles of response to the Editor: firstname.lastname@example.org. Superior work will be published.
Those of you that have taken the plunge and had some of your components modded understand how much better they can sound than stock. I often am asked: "why don't competent circuit designers just do designs that do not need mods?" Well, there are a host of reasons for this, including price point, assembly technology, power delivery, and current paths/grounding. First, let's examine the "Price-Point" criteria.
The majority of consumer electronics is designed to specific price points. As a result, there are compromises made in the quality of parts, such as op-amps, resistors, inductors and capacitors, the number of layers in the circuit boards, the input/output connectors as well as the wiring harnesses and connectors that connect boards together. Passive parts, such as resistors, capacitors and inductors can be purchased in a wide range of quality levels. Low quality, low cost parts tend to behave less like the ideal, and more expensive parts tend to behave more like the ideal.
Capacitors, for instance, have internal resistance and inductance as well as capacitance. All capacitors have the equivalent of a small resistor in series. This makes them non-ideal. The lower the internal resistance or Equivalent Series Resistance (ESR), the more ideal the capacitor. More ideal parts tend to perform better in almost all circuits, with the exception of some filter circuits where the losses in the part are actually leveraged. These cases are rare however, and usually a high-performance capacitor in series with a resistor serves the same purpose and performs better. Electrolytic capacitors such as Black Gates and Jensen 4-poles have very low ESR and large diameter copper leads, which make them perform better than most electrolytics. High frequency and coupling capacitors have similar requirements, including low ESR and low dielectric absorption. Dielectric absorption is a function of the dielectric in the capacitor. It affects the ability of the capacitor to ideally discharge or charge in a specific time duration. More ideal coupling and high-frequency capacitors typically use Teflon, polystyrene or polypropylene dielectrics (in order of ascending dielectric constant). An ideal capacitor might use air or vacuum as a dielectric. I often find inexpensive less ideal capacitors such as ceramic and polyester film in expensive equipment, where more ideal capacitors would make a significant improvement.
Low quality resistors tend to be noisy and have excessive inductance. Resistors such as the thin or thick-film Caddock and Vishay have very low inductance and create very little thermal noise, so they more approach the ideal.
Low quality inductors use high-dielectric constant (Er), lossy insulating materials on the conductors or exhibit high resistance or high skin-effect. Like capacitors, inductors are not ideal either. They have finite resistance as well as inductance. Inductors that use litz wire or flat copper or silver ribbon and are insulated with oil-impregnated paper or Teflon are superior inductors because they minimize resistance, skin-effect, eddy currents (causes heating losses) and dielectric absorption.
Circuit boards are generally poor for analog signal propagation, except for short paths, due to their high dielectric constant. They are acceptable for most digital applications, but not optimal for this either. They are convenient for attaching integrated circuits and transistors, so called "active" elements, and interconnecting these over short paths using "traces" or "nets". Ground planes or large copper fill areas are typically used for completing the circuits between active elements. Point-to-point wiring harnesses are generally superior. I will speak more on the technical reasons for this later in this paper.
Op-amps are active devices for amplifying and buffering signals. They are general-purpose gain blocks. Unlike most special-purpose integrated circuits, they come in a wide variety of types and qualities. The types range from low frequency to microwave, instrumentation to current-feedback. Certain characteristics are tuned in each op-amp design so they perform more optimally in those applications. For audio, we tend to use low-noise versions that have good stability, in other words, they do not oscillate easily. The quality of these is usually a function of noise specs, speed (slew-rate) and bandwidth. High-performance op-amps tend to be quite expensive, so they are almost never found in stock components. Even though the specs on one op-amp may be almost identical to another, they can sound different. This is usually due to the on-die power delivery scheme. Knowledge of what kind of power delivery makes a particular op-amp happy is critical to getting the most performance from it. The "Power Delivery" section below addresses this topic in greater detail.
So, why don't designers take advantage of these more ideal parts, given that many of them are not that much more expensive? I believe part of the reason is that they just do not have the time to experiment and keep up to date on the latest component technologies. Understandably, their main concern is the circuit design and the active parts. Experimenting and seeking-out better components are the bread-and-butter of the modder.
Surface-mount technologies (SMT) are often used to enable components to be built in high-volume, with low manufacturing costs allowing smaller physical packaging. The problem with SMT is that the passive parts, namely the resistors, inductors and capacitors are generally lower quality than their "through-hole" counterparts. For instance, there is simply no way to reproduce the performance of a good leaded coupling capacitor using current SMT technology. They are just too large. In fact, most high-quality, more-ideal passive parts are quite large and tend to be larger than the same lower quality part, even in through-hole versions. At least, with capacitors and inductors this is definitely the case. Therefore, many of the modifications amount to replacing these SMT parts with larger leaded parts.
Printed circuit board technology may be convenient for attaching SMT and through-hole parts, but its two-dimensional layout typically prevents optimal wiring of all of the circuits. As a result, many of the circuits have very long serpentine traces. Since the fiberglass substrate used in the circuit board has a fairly high dielectric constant (typically ~4.7), the performance of these long traces is fairly poor. There can also be other compromises in the layout, such as split ground-planes and ground-plane voids, which can cause noise, and even hum, in the final design. This will be explained in the next section on Power Delivery.
This is probably the one design aspect that accounts for the lion's share of improvements that can be made to almost all stock components, including CD players, DACs, preamps and amps. Power delivery does not mean "power supply," but rather involves all of the components and circuits that deliver power to active devices. It's not that the circuit designers don't try to design in power delivery. They do make an attempt, at least most of them. However, most seem to have no formal training in this area and often resort to "cookbook" methods, relying on the tips that are included in the chip data sheets, or things they have learned from other designs where those designers sometimes don't have a clue either. It is one of those areas that is a bit of a "black art". Even large computer companies with over 45000 employees have only a handful of power delivery experts.
Good power delivery to active components results from doing several things right, including:
Optimum capacitor selection amounts to choosing the right capacitor for each power delivery job. If the job is filtering the rectified input voltage, then the capacitance must be large and able to absorb high voltage ripple without overheating. If the job is high frequency "decoupling," then the capacitance must be a small with low-inductance and ESR. For speaker crossovers, the capacitor must be capable of carrying large continuous currents.
Capacitors used in power delivery situations are analogous to small batteries that store energy. They "decouple" the power supply from the load (not to be confused with signal "coupling caps"). The large ones store energy for low-frequency demands of the active components and the smaller ones store energy for the high-frequency demands. The larger ones are also referred to as "bulk decoupling caps", which means that their primary function is to recharge the smaller caps when the smaller caps become discharged from high-frequency current demands as well as providing low-frequency power. I often see capacitors that are optimized for signal coupling in locations where power decoupling is needed. In these situations, the capacitor is essentially ineffective.
Low-inductance paths are primarily needed for high-frequency power delivery, but the designer should make an attempt to minimize inductance in all paths. When a designer adds small decoupling caps to provide high-frequency energy to active circuits, but places them far away from the active circuits they are servicing, they are inserting inductance between the capacitor and the active circuit. This inductance prevents the capacitor from delivering the power to the active circuit when it demands it. The result is that the on-die voltage will sag and the active circuit will sacrifice dynamics at the high-frequencies, even though it has good dynamics at the low and mid frequencies. This non-uniform or deficient dynamic response is typical of most stock components. When a designer installs a small value capacitor, but leaves long leads on it, he is also inserting inductance between the capacitor and the active circuit. In order to optimize the inductive paths, the capacitors must be arranged in a hierarchy starting with the large ones furthest from the active circuit and ending with the smallest ones closest. Inductance of the traces on the circuit boards and power wiring harnesses must also be minimized.
Proper sizing of capacitors is also important. If the high-frequency decoupling capacitors are too small, the dynamic demands of the highest audible audio frequencies will not be satisfied, only ultrasonic frequencies. If they are too large, the upper mids will be satisfied, but not the highest audible frequencies. This sizing is also a function of the active circuits that they are powering. Some op-amps, for instance, have superb on-die power delivery, so they need very little high-frequency decoupling capacitance. Others need all the help they can get because of their on-die inductance. There is also usually a magical combination of bulk decoupling size and high-frequency size that work together well. The bulk cap must recharge the small caps quickly enough so they recover before the next high-frequency demand on the small caps. They also must be large enough to supply the mid and low frequency demands. These time-constants are very difficult to predict or simulate and usually only come with experience. If this is designed wrong, resonance can occur between the larger and smaller caps, actually injecting noise into the audio. This is a recipe that most designers have never learned.
Current Flow and Return Path Analysis
From grade school "batteries and bulbs," most of us learned that it requires two wires from the battery to the bulb in order to get current flow and light the light bulb. For some reason, many designers have lost sight of this simple concept, or simply don't give it any consideration when they lay out a circuit board. They may be very careful with how the signal traces or wires are routed ("source currents"), but do not pay any attention to how the current return paths are routed. When there are ground-planes in circuit boards, high-frequency signals, such as digital ones, tend to flow directly underneath the source-current traces. Therefore, if the ground-plane is not continuous under these high-speed traces; the return currents must find other, more roundabout paths to get back to the source. When the return currents take these other paths, they are usually sharing return paths with many other digital and analog circuits that are also trying to get back to their sources. This sharing causes cross-talk between the signals. Cross-talk can cause jitter in digital signals, such as S/PDIF, and inject noise into analog signals, which can impact sound quality.
Another current path consideration is that of signals versus power delivery. Currents must flow for both signals and for power delivery in order for components to function. Again, if the power delivery paths are shared with signal return paths, this can inject noise into both digital and analog signals. However, this design aspect is a bit tricky, because if there are too many current return paths and the "loop" sizes are large, this can cause ground loops and resultant hum. Therefore, an in-depth analysis of the signal and power currents is necessary in order to optimize both without introducing hum or cross-talk.
Chassis grounding also factors into the return-path analysis. Many designers simply play trial-and-error with the chassis ground location until the hum is minimized, without really understanding what is causing it. Others tie all of the input and output connectors together with the chassis, and then cannot understand why they cannot get rid of the hum. Chassis grounding should not create new current paths or loops, and ideally should only be a single point. The only time that multiple chassis grounding makes sense is when electromagnetic radiation is a problem and must be minimized, to pass FCC regulations for example. About 50% of the components that I modify have hum from internal ground loops, which then have to be eliminated.
By addressing the described weaknesses with effective mods, I have found that virtually all stock components and speakers benefit significantly, no matter what the original price was. I hope that this paper increases your understanding of why even the best-designed components benefit from modifications, and how modders are different than designers. Effective modding, like good circuit design, is non-trivial, requiring extensive experience and formal training, as well as expert re-work skills.