Design of the Sound System

This section is a work in progress, and probably will be revised from time to time. Necessarily, a lot of stuff here represents my personal opinion. Topics include:

Additional details are included in separate, more technical sections on crossover design, and Thiele-Small enclosure analysis and design.

Goals

My objective is to enjoy (usually by myself) classical, jazz, rock, etc. at live concert volume levels, while not disturbing my wife or neighbors. I don't care a whole lot about physical appearance, or impressing my friends (although good feedback is always nice).

As I listen to music, the pleasure I experience depends on:

I did have a hearing test; it really doesn't make much sense to spend a lot of money on a fancy sound system if you can't hear it. We can control what music we buy, but whatever the musicians and sound engineer have done to it is beyond our control.

If you are lucky enough to have a choice about room size, the acoustic design of a music room is considered in another section. However as shown in that section, a room with "optimum" dimensions is really not all that much better than a room with horrible dimensions. Background noise is important. Hearing a neighbor's chain saw really messes up my enjoyment of a Debussy piano piece (of course candy wrappers at a live performance are even worse!). To minimize background noise, and to disturb my wife as little as possible, maximum soundproofing is a major goal. Construction techniques for this purpose are discussed in another section.

The remainder of this section is devoted to the last factor, sound system quality.

Back to the top

Sound System Quality

What does "sound system quality" mean? Audio newsgroup flame wars rage between objectivists, who insist that measurements are primary criteria, and subjectivists, who insist that listening is the only criterion (but who generally dislike ABX tests). I'm a little of each; subjective since listening pleasure is my ultimate goal; objective since I believe the best way to reach that goal is by careful measurements. Doug Self's site has interesting thoughts on this topic.

My approach to sound system quality is to learn as much as I can about human hearing (see Music and the Human Ear). I am also a musician, and I believe we do not know how to measure everything that affects our enjoyment of music. I am fascinated by the ability to numerically process .wav files to simulate various types of distortion, diffraction, etc. I intend to do a lot more with this in the future, as my own research project of what affects sound quality. Minimizing distortion is an obvious goal. I also believe that early room reflections and edge diffraction affect the sound image, and should be minimized. Methods to accomplish this are discussed in another section. I still have an open mind regarding the value of true time-alignment, but decided to go for it. This can only be achieved for a relatively small "sweet spot," but that is fine with me.

Measurements of the frequency spectra and transient levels of music I listen to convinced me that lots of amplifier power was desirable. I believe clipping is the most serious distortion problem.

A chart of the frequency range of various instruments [10 kb], and playing my own piano led me to the conclusion that the 300 to 3kHz range is critical. I want the whole audio spectrum, but I want this region to be as flawless as possible. It is probably not a coincidence that this overlaps a good part of the range of the human voice, and also happens to nearly overlap the 300 to 6kHz frequency range where the ear is most sensitive.

For another viewpoint on sound system design goals see Siegfried Linkwitz's site

Back to the top

What kind of Loudspeaker

This is probably the most difficult and the most important part of the whole design process. There is a wide divergence of opinions on just about every aspect of the possible choices.

Things that I found very helpful for addressing this question are:

There are very good sounding loudspeakers of almost any type you can name. The choice is at least partly a matter of taste. For a DIY project, I believe it is also wise to "keep it simple, stupid."

Driver type?

There is an embarrassment of choice. I love the transparency of electrostatics, but they cannot, in my experience, handle the volume levels I want. I selected conventional drivers mainly for this reason. I was very attracted by kevlar and other exotic materials until I learned more about nasty resonance problems (see the snippets site). I stuck with tried-and-true cone materials (which actually can be pretty hi-tech).

Two-way, Three-way, N-way?

Some two-way systems are surprisingly good, but still sound strained to me. It is extremely difficult, if not impossible, to adequately cover the audio band with only two drivers.

Four or more drivers cover the band nicely, but create lots of possibilities for screwing things up. Also, for either a two- or four-way, a crossover frequency is typically within my critical listening band of 300 to 3kHz. The crossover region between one driver and the next is particularly susceptible to problems.

I selected a three-way system, and a 5-inch Vifa P13WH-00-08 midrange to cover the entire 300 to 3kHz band. This driver is used by Lynn Olson in his Ariel speaker. I think it is a terrific midrange. I originally selected the Dynaudio D-260 1" tweeter, and Peerless 315SWR39134 12" woofer to fill out the system. The primary criterion for the size choices were that the data specs indicated that they would cover the frequency ranges outside of the Vifa's range, as shown in the chart of instrument frequencies referred to above [10 kb]. I originally wanted a macho 15-inch woofer, but I couldn't find one with an adequate high end. The 12-inch turns out to have plenty of oomph. The particular brands chosen were based on various recommendations, guesswork, and hope. Despite a fuse directly in front of the Peerless woofer, which prevents exceeding its power rating, I blew out three of them. Solen replaced two, but after the third I switched to a McCauley 12-inch.

Although the D'Appolito configuration has some merit with regard to broadening the vertical size of the sweet spot, this doesn't mean much if you are designing for a single listening position. It is harder to reduce diffraction with this design. So I opted for a very conventional three-way configuration.

Ported, sealed enclosure, or?

I did not consider designs such as horns, or transmission line speakers.

I have lovingly designed and painstakingly tuned ported enclosures. The bass always sounds muddy. My sealed enclosures produce a wonderfully tight, clean bass. Some possible reasons for this are the poorer transient response, and vent distortion, of ported enclosures, as discussed in the section on Thiele-Small analysis. I chose a sealed enclosure.

Back to the top

Enclosure Analysis and Design

Enclosures have two major influences on sound system quality: (1) they have an important effect on the driver response, and (2) they radiate sound along with the driver. The latter effect is quite important. My guess is that many people have little idea of the incredible amount of sound, across a broad frequency range, that can come right through the walls of a speaker cabinet. This fact, and construction techniques for coping with this (very challenging!) situation are discussed in the section on speaker construction.

Enclosures are also a major source of edge diffraction, along with shelves, furniture, etc. Minimizing this problem is also discussed in another part of the speaker construction section.

For driver response, enclosures mainly influence the bass. In the early 1970's a series of technical papers published by Neville Thiele, and Richard Small made the design of speaker enclosures much more scientific. Today, high-quality speaker manufacturers and retailers include "Thiele parameters" in their sales literature, which provides the information needed to select an appropriate speaker and to tailor-design an enclosure for the speaker. The required design information is widely available in books such as Dickason's Loudspeaker Design Cookbook, and in the form of computer programs. For the technically inclined, see my version of Thiele-Small enclosure analysis. A method for measuring the parameters is given on the Subwoofer DIY page.

Small "acoustic suspension" sealed enclosures are designed to act as an integral part of the cone suspension. Although the frequency response can be equal to a larger enclosure, larger enclosures have the advantage of less edge diffraction (simply because the edges are further away from the drivers), and more absorbing material can be put inside to sop up all that energy pouring into it. Panel resonances tend to be lower (usually not good), but bracing can fix this, and internal pressure is lower. Finally, Small showed that for a given low frequency performance, there is an upper limit on efficiency that increases with enclosure size. Increased efficiency is the same as more amplifier power. Note however, that designs don't necessarily achieve Small's upper limit. The acoustic suspension and infinite baffle designs compared in the section on Thiele-Small analysis have exactly the same efficiency. I selected a large enclosure (about 48 cubic feet volume). In any case it is important to choose a driver designed to complement the selected enclosure type.

Back to the top

Crossover Design

Crossovers are a very important part of a sound system. This has received increased recognition in recent years. Yet folks can be passionate about the gauge of their speaker interconnect wire, while being clueless about the gauge of the many yards of wire in their crossover inductors, which carry exactly the same current as the interconnect wire (tweeter excepted). Inductor wire gauge is rarely, if ever, mentioned in a loudspeaker manufacturer's specs. Having huge interconnect wire hooked up to a wimpy crossover inductor is like connecting a fire hose to a straw. Even if you accept the argument that the gauge of the crossover wire is "part of the design of the loudspeaker," adding or subtracting a tiny fraction of an Ohm when the inductor resistance is many Ohms still doesn't make a whit of difference. This fact, and other technical information is discussed in more detail in the crossover design section. Dickason's book is also a very nice source of information.

More than 2 years after originally writing this section I have become interested in digital filters. There is no question in my mind that these are the wave of the future for serious audio. See the new section devoted to these interesting devices.

Crossover frequencies go along with driver selection, which in my case means 300 Hz and 3kHz. The main design decision is: what crossover order? The order determines how rapidly frequencies outside of the desired range are cut off. It is desirable to eliminate out-of-band frequencies as rapidly as possible, to improve power-handling capability and reduce distortion. Sadly, there is a price to be paid for rapid cutoff. All analog filters introduce phase shifts and time delays. And the higher the order, the more you get. Digital filters can provide rapid cutoff without phase problems.

The effect of crossover order on power required in each channel in a 3-way system is discussed below. Three music samples were used in the evaluation. For the Talking Heads cut, the tweeter peak power is a factor of 1.4 higher for a 1st order crossover than a 4th order. But the peak cone excursion is almost 7 times larger with the 1st order. Sound pressure is proportional to cone acceleration, which is the second derivative of cone excursion. This means that to generate sound with a flat spectrum, the spectrum of the cone motion of an ideal speaker would increase 12 dB per octave as frequency decreases. The spectra of the tweeter sound for the Talking Heads sample illustrates the problem [47kb]. The spectrum of the signal going into the crossover rises enough as frequency decreases to nearly match the 6 dB per octave reduction by a 1st order crossover. Therefore the signal sent to the tweeter by a 1st order crossover has a nearly flat spectrum down to 100 Hz. If the tweeter cone excursion increased 12 dB per octave the cone motion would peak around 70 Hz for the 1st order! However, below driver resonance, the cone response changes, and, for a constant input voltage, becomes flat with respect to frequency. This reduces the severity of the effect.

For the three music samples, I calculated peak excursions for a 1-inch dome tweeter, a 5-inch midrange and a 12-inch woofer. The average sound level is set at 100 dB SPL RMS at 1 meter in front of the drivers mounted in an infinite baffle in an anechoic environment. Peak sound levels were close to 116 dB SPL. Results are shown in the table below, compared to the catalog value of xmax, the maximum excursion which stays in the linear region. (Note that 100 dB is not unreasonable for Talking Heads or Shostakovich, but is pretty loud for Diana Krall).

Maximum Cone Excursion

Talking Heads

Diana Krall

Shostakovich

Xmax

1st order

4th order

1st order

4th order

1st order

4th order

Tweeter

0.5

0.08

0.8

0.1

0.5

0.09

0.3

Midrange

9.0

1.0

6.0

0.8

1.2

1.1

4.0

Woofer

7.3

7.4

5.4

5.5

1.6

1.6

9.0

The woofer excursions are almost identical for a 1st and 4th order, and always within the linear region. The midrange and tweeter excursions are typically 6 to 9 times larger for the 1st order, and often outside the linear region. Peak excursions are inversely proportional to the cone area, so a 10-inch woofer would have an excursion about 1.7 times greater (based on catalog values of effective area).

At one point oscilloscope measurements using my CLIO system showed that at high volume levels the tweeter was creating quite a bit of distortion with the 1st order crossover - which is what I would expect from the above analysis. However when I attempted to duplicate this measurement recently, the waveforms looked fine. Direct measurement of the 2nd and 3rd harmonic distortion showed virtually identical results for the 1st and 4th order crossovers. At some point I am going to do more calculations regarding the effect of clipping, and more measurements, to attempt to clear up this distortion issue. In the meantime I am assuming that the lower the excursion the better, which strongly favors selecting a 4th order crossover.

Back to the top

Time Alignment

And then there is the question of time alignment. If a snare-drum rim is whacked with a drumstick, a sharp musical transient is produced. That is, the sound can go from silence to ear-shattering and back to silence in one-thousandth of a second or so. Play this sound through the vast majority of the loudspeakers, and what you get is a sound that is smeared out over a much longer period of time. Two effects cause this: (1) sound from the tweeter reaches your ears before the sound from the midrange, and (2) the crossover time delay. The first problem can be cured by physically offsetting the drivers, as described in detail in another section. The second problem can be avoided by using a first-order Butterworth crossover, which (ideally) perfectly preserves the time relationship.

Both of these problems can also be cured by using digital filters. For more technical detail, see the Digital filter Section.

Square-wave response is a good way of measuring transient response. Plots of computed square-wave response of various crossovers are given in the crossover design section. Except for the 1st order Butterworth, which produces a perfect response, it's really an ugly picture. Plots of the measured response of my system, square wave, frequency etc., are in the section on final system measurements.

The key question is whether or not you can hear the difference. As discussed in the crossover demo section, one of my .wav file tests convinced me that time-aligning the drivers is audible and significant. This affects both amplitude and phase. A second test indicates that I can't hear the difference between a 1st and 4th order crossover. But this was with headphones. I originally selected a 1st order Butterworth to obtain true time-alignment. I also have built a 4th order Linkwitz-Riley, and can now perform a direct A/B test with my actual loudspeaker system. Fortunately, hot-switching does not produce any nasty pops, so I can switch instantly. A blind test indicated that a friend, Alan Ross, and I could both hear an difference between the 1st and 4th order crossovers, and we preferred the 1st order. However there is a difference in the frequency response due to room reflections (see discussion), and I believe this was the cause. If true, then readjusting the system for the 4th order would probably reverse the preference. With headphones, I cannot hear any difference between music files with and without the phase shift of a 4th order crossover. With a file of a recorded finger-snap, I think I can hear a difference, and at one point I obtained nearly perfect results in an ABX blind test. But I have not been able to reliably repeat the results. Maybe my hearing is more acute sometimes? Another mystery. After futzing around with this issue for a long time, I finally settled on the 4th order Linkwitz-Riley.

Back to the top

Multi-amped System Architecture

I am a big fan of bi-amping. (So-called "bi-wiring" is a farce, pure and simple). A block diagram of the main components shows the original architecture I selected [6.2kb]. Bi-amping is a handy way of boosting total amplifier power. Also, when a huge bass note is being played, sucking one amp dry, the other amp can cruise along and play the highs virtually undisturbed.

Bi-amping also greatly eases the crossover design in several ways. With a single amp, there are two gnarly problems with a low-frequency crossover : (1) it requires enormous inductors; (2) the impedance that it must work into, the driver input impedance, is awful. A Zobel network helps, but by itself is insufficient; details in the crossover design section. Finally, as shown in the Thiele-Small analysis section, bi-amping improves bass response by reducing ringing induced by the enclosure.

After building my 4th order Linkwitz-Riley crossover I converted to a tri-amped architecture, which is better yet. I now have a Marantz AV560 Preamp, and a Rotel RB991 200 Watt/channel driving the mids. And a Rotel 980BX 120 Watts/channel driving surround speakers for a grand total of 1.4 kilowatts of glorious power.

My amps and preamps are solid state. I have an open mind about tubes, and intend to look into the subject in more detail in the future. End-to-end, my electronics produce a very clean looking square wave, which means they are pretty close to the proverbial straight wire with gain. But! That's as long as the power amps aren't clipping. The results for three music segments show examples of the power requirements. Average power is not all that important. It is peak power that will determine if an amp clips or not. There is no clear-cut advantage of a 4th order crossover over a 1st order in peak power. Perhaps the result that will be surprising to many people is how much power is needed by the midrange and tweeter in comparison with the woofer. As stated just below the table, a peak power requirement of 700 Watts is not unusual; this applies to a mono-amped system. The percentages in the table indicate what fraction would be required for a tri-amped system. Tube amps have a big disadvantage in that they (generally) have lower power output, but may have an advantage with regard to the type of distortion produced in the presence of clipping. Another future project.

Back to the top

Active vs. Passive Crossovers

With bi-amping the low frequency crossover handles very low power since it follows the pre-amp instead of the amp. A passive crossover made with inductors and capacitors is impractical due to the large inductance required. But, a very simple and cheap passive crossover can be made using resistors and capacitors. Active crossovers are the alternative. Some cost tons of money. A source for an active crossover kit, and design information for the passive crossover are given in the crossover design section.

Back to the top

Fine-Tuning

Your careful design ain't going to work like it's supposed to. I found it essential to make careful measurements at every step; crossover component values, overall crossover performance, and then finally the full system frequency response. At every stage fine-tuning was required to achieve the performance I was looking for. I invested in a good multi-meter, oscilloscope, and the CLIO system, and I consider it some of the smartest money I spent.

Back to the top

Cables

My interconnects either came out of the component box, or I dug old ones out of the bottom of my spare parts drawer. See the Thiele-Small analysis section for the effect of speaker interconnect cable. Also see skin effect.

Back to the table of contents