Zaph|Audio

Design Mantras - my common speakerbuilding concepts and theories

Mar 20, 2005 - Initial Publication

In reviewing my projects, I noted there were a few things I do every single time, or a few concepts that I believe in. I thought that it might be time to stop repeating myself and document them here. I will be adding to this document as more design standards pop into my brain.

Crossovers

All crossover slopes and frequencies are selected based on a driver's frequency response, linear and non-linear distortion and off axis response. Many people design only based on on-axis frequency response, and while this may result in a speaker with flat response, it still may not sound very good if a driver has excessive distortion within it's operating bandwidth or a poor power response. I will typically take a look at a driver's off axis response, power handling, all harmonic distortion up to 5th order, and a few other forms of distortion testing. Sometimes I'll get IMD plots and test at multiple levels. I also get in-room plots. There are a few forms of testing that I do not use, as they are merely derivatives of other types of testing. A cumulative spectrum decay is one such useless measurement, though I still collect them in my driver tests.

The most common crossover I use is 4th order Linkwitz-Riley. This often works out to be the simplest. Most drivers only require a couple components to combine with the driver's natural rolloff and reach 4th order target slopes. 6th order or higher is typically only needed with drivers that have "behavior problems". Only drivers with smooth extended response, good off axis response and low distortion are suitable for 2nd order, and drivers like that are pretty rare.

The least common crossover I use is 1st order. And by that I mean true first order acoustic slopes, not "throw a cap and a coil on the tweeter and woofer" type of first order, which will never result in a system that sounds good or adheres to any particular crossover topolgy. True first order order designs are rare and nearly impossible to do. I've never actually seen one.

When I refer to an order of slope, say 2nd order Linkwitz-Riley for example, I refer to the acoustic slope. The electrical components requires to reach my target slopes vary according to the driver's own response and the driver's impedance. Some people might call a system with one coil and one cap on each driver a 2nd order. That's not true. It's might be 2nd order electrical, but more than likely it's 4th order acoustic, or mixed order assymetrical, or something else. The point made here is to be accurate in your descriptions of filter topologies.

My most common method of matching phase between the woofer and tweeter is with assymetrical crossover slopes. Often using this I can get away with a normal "tweeter on the top" driver arrangement, and a flat baffle. How I make them assymetrical varies from driver to driver. Sometimes the tweeter slope is made shallower or sometimes it's made steeper. Ditto for the woofer. What works best depends on a lot of factors. The components required to modify the slopes also vary but in general it's going to require 1 or 2 extra parts strategically placed within the design. Knowing what and where to put these components is something that only comes with experience. There are other methods such as delay networks or slanted baffles. I rarely use trellis-ladder type delay networks due to the complexity they add to the crossover. Likewise, I rarely build slanted baffles due to the complexity of enclosure construction.

I design crossovers to be as simple as possible. The fewer the parts, the better. This is not because I think that more components in the signal path degrade the sound. That statement is true, but several components will not affect the sound enough to be noticed. I use fewer parts because it's a better balance of cost vs performance improvement. Some designers throw so many components at a design just to get a pretty looking flat line response curve. Personally, I only use enough components to get within a certain flatness spec but no more. There is an economy to crossover design.

I pay more attention to peaks in a frequency response curve than dips. Errors of omision are much less noticable than errors of inclusion. Hell, some people can't immediately hear a 40db null when their tweeters are hooked up out of phase. It is easy enough however to hear even a 3db narrow band peak when just the right music comes along. I also pay attention to small decible variations that occur over a wide bandwidth. Even a 1db difference is drastically audible if it happens over 3 octaves.

Most of my designs use baffle step compensation unless otherwise noted. Sometimes this will be a traditional coil and resistor combo on the woofer, or other times it will be integrated into the rest of the woofer circuit with an oversize inductor doing double duty as the first component in a multi-element filter. I think the use of baffle step is really unrealized in many designs. It's not just there to flatten the anechoic response, it's there to add "warmth" and a relaxing tonal balance in a typical room. The frequencies affecting baffle step and room boundary reinforcement are different, but they do overlap and both have to be considered.

The selection of BSC is more than just the baffle size and the boundary reinforcement. It's also affected by non-linear distortion such as harmonic or intermodulation. A bass driver with a clean low end but a "dirty" high end will do better with more BSC than a driver that has high midbass distortion but a clean top end. All issues that affect the tonal balance have to be considered when setting the BSC, and I typically use both measurements and the judgement of my ears for this.

Design

Tonal balance is an important consideration. Some people say they like flat and some say a tilted down response curve works best. The fact is, what works best for a design will depend on the drivers used.

The kind of tonal balance I choose is determined by the driver's distortion in it's operating range. Even if a driver measures flat, harmonic distortion may make it sound brighter than it actually is, by creating high frequencies where there should be none. So to my old ears, flat with low distortion sounds perfect. Flat with higher harmonic distortion sounds bright or edgy, and if I am designing with drivers like that, I'll generally tilt the response down at the frequency where the distortion rises. It's a delicate balance to find the right sound. Generally speaking however, flat sounds best with higher quality low distortion drivers. Cheaper drivers often require the tilt if you cannot work their distortion out of their operating range.

Tilted or flat response is one issue, but response variation is another issue. Once again, +/- 1dB could work with low distortion drivers. But designs with less than perfect drivers can be varied far from flat in the interests of removing listening fatigue. For example, I've done a design with a driver that had a large distortion peak at 1400hz. I intentionally put a high Q, 6 db deep notch right there and it sounded pretty good afterwards. Without it, high female vocals and strings caused listening fatigue. Dips in response are generally far less noticable than peaks, and may be needed to solve a problem.

Don't forget the room. Generally, flat sounds good in a carpeted room, but hardwood or tiled floors can increase the aparent brightness.

In case it's not obvious yet, distortion management is the most important issue in design. Folks that fire up LspCAD and pop out a design without doing distortion testing have success rates that are essentially just based on luck. You've got to know your drivers well before you pair them up, select a crossover point and crossover topology. Without distortion tests, it's a shot in the dark, and design iterations may go on endlessly in the course of finding something that works. When you start your design process armed with full distortion data, the chances you're going to get the design right on the first try or soon thereafter. Without distortion data, you may never figure out what's wrong with a design.

Smart driver pairing will increase the chances of success. No offense to fans of the ProAc 2.5 clone, but it makes a great example of poor driver pairing. The Scan Speak 8513 has heavy harmonic distortion anywhere below 4kHz, while the 8535 woofer has energy storage issues above 2kHz, along with poor off axis response. A compromised crossover point at 3kHz can't save this system and the result is a harsh, edgy sound. Crossing a 7" woofer at 3kHz also results in a narrow vertical listening window and a rough power response.

Generally, all things being equal, I'll cross a tweeter over as low as I can before distortion and Xmax start to become issues for the level I listen at. As you cross over lower, the vertical listening window becomes taller. Once again, pairing drivers and selecting a crossover point is a delicate balance.

Enclosures

All drivers are countersunk, particularly the tweeter. The woofer is also countersunk unless I specify otherwise. About the only time I specify otherwise is when the woofer flange is specifically designed for surface mounting.

All baffles are removable, and fastened with socket head cap screws. Normally I'll use 1/4-20 black anodized, with a metric M6 regualr series washer. Why an M6? Because it's 6.87mm inside diameter fits a 1/4" bolt with less slop. The washer is required so I can make the hole in the baffle a little bigger to allow for slight positioning of the baffle before it's tightened down.

Removable baffles are held in place with corner gussets that hold a Parts Express Hurricane nut. These are like normal T-nuts, but are press-fit rather than having teeth that dig into the back side.

I finish most of my baffles with black Rustoleum texture finish. Before that I'll use sanding sealer on bare MDF, and I'll do about 5 coats of it on the edges with 1 or 2 coats on the face. Sand it all smooth before painting. Depending on how well I seal the edges, I'll likely need two coats of texture finish. Then finally I'll put a single clear coat over that for a for a bit of smoothness that makes wiping dust off easier.

In the case of using a Parts Express pre-fab enclosure, I'll cover the baffles with white "Contact" brand stick-on shelving paper. This will protect it from router base scuffs and scratches. If all goes well, I'll leave this protective paper on and then spray-paint the countersinks so no bare wood shows through the cracks. If all doesn't go well - you might have some chipping, or you might have scratched the baffle anyway, then I'll fill scratches with bondo, sand it a little, and use the baffle painting routine specified above.

All woofers 7" and under have the inside of the woofer hole chamfered. Depending on the flange, I may just go all the way around the inside or I may "scallop" the inside to allow some meat for screws to fasten into.

Depending on how the driver looks, I may choose to bolt it in place using socket head cap screws (normally 1" long #8 or #10) or I just use black wood screws. Using the socket head cap screws requires scallops so there is enough wood to support Hurricane nuts on the back of the baffle.

Enclosures of .375 cu ft or larger have a crossbrace from side to side. Of course if I buy a Parts Express enclosure, it has the crossbrace already in it.

All of my designs call for roundovers on the baffle. 1/2" is the minimum, but 3/4" is better if you have a large shank, variable speed router than can handle it. The roundover is required to help smooth out the diffraction ripple response.

For vented boxes, most woofers 6.5" and under use a Parts Express 1.5" flared port. This port is 4" long, approximately 1.6" average diameter, and requires a 2-9/16" hole. If the hole is sized just right, the little ribs will press fit into the hole. Occasionally, 4" is the right length in a smaller enclosure with a small woofer, but when it's not I'll use 2 of these ports, with each cut to half the size I need, then *taped* together. This will have to be done with the port mounted in the enclosure since the flare is too big to make it through the hole. This two piece arrangement gives you a good flare on the inside, which lets you get away with a smaller and therefore shorter port than normal without a substantial increase in turbulance noise. In theory, the safest way to cut these ports is with a bandsaw. I don't have one, so I use a rather unsafe method. Copy this routine at your own risk: I cut them on the table saw with the fence as a stop. I set the fence to 1/2 of the port length required and press the flange part of the port up against the fence with my hand to guide it through. Kickback is possible in this situation, so I use a lot of force against the fence and I stand to the side in case it does kick back. Note that I don't slide it all the way through the saw, just enough to cut it partially, then back it out and rotate, and run through again til I make it all the way around.

8 to 10" woofers will use a 2" flared port. 12 to 15" will use a 3" flared port. These are the pre-made, cut to size "Precision" ports. These are of course adjustable in the middle section. My own experience finds people making their ports *way* larger in diameter than they need to be, which which relates to longer ports.

For damping vented enclosures, I use Whispermat, Sonic Barrier or 1/2" carpet padding doubled up. Whispermat is the best value. Sonic Barrier is expensive but occasionally handy since it comes with adhesive and is sold in small (but expensive) quantities. Carpet padding is surprisingly effective and cheap but it varies in performance. I never use egg crate foam because it just plain does not perform well. For sealed, I feel that "Acousti-Stuff" is the best performing, but standard craft store dacron pillow fiber will do oK for about 1/3 the cost.

Tuning an enclosure

Tuning a vented box is important enough that it needs it's own section here.

For vented boxes, the box size and tuning frequency that I select is rarely what a box modeling program might consider optimum. For folks new to DIY speakerbuilding, the first software tool they often use is a freeware box modeling program like WinISD. The bad news is that what WinISD considers optimum is merely the the alignment that results in the lowest F3 in anechoic conditions. These programs don't consider room response, power handling, group delay or any of a number of other issues when they pop out a "optimum" box size and tuning frequency. So don't email me saying "Why did you do that when WinISD says this is optimum?" The answer is: Because I'm smarter than WinISD.

The vented alignment I typically use doesn't have a name. I'm sure I didn't invent it, but I haven't seen much out there to describe it. Normally, I'll tune lower to get a more gradual rolloff that works better with room response and puts power handing where I need it. It's basically a type of tuning that is a combination between sealed and vented alignments. Note that this is not an Extended Bass Shelf (EBS) tuning. That kind of tuning implies an oversized enclosure volume, with a dip above the tuning frequency. You get the lowest anechoic F3 with that configuration, but in a real room you get an overbearing one-note bass at the tuning frequency, a huge midbass dip, and crappy power handling.

I will never vent a woofer that is 4" or smaller. I've seen many people that do however. The problem with that is the tuning frequency will be too high, and there will always be substantial music output below the tuning frequency, easily distorting or destroying the woofer.

No tuning should be selected without looking at a Xmax vs frequency chart at several power levels up to the one that is needed to reach your required listening level, including the effects of a subwoofer crossover if one is used. When you factor in all the other design elements, you see that selecting a box size and tuning is no simple task.

Generally, unless otherwise specified, most of my speakers are designed for a medium size room. My HT room is about 5 x 6 meters. If you are a person trying to reproduce one of my designs, but to use it in a very different sized room (tiny dorm room, great hall with high ceilings, etc) you may wish to reconsider the bass tuning, in addition the the baffle step compensation used. You're on your own for that.

Cost and value

I believe it to be true that the cheaper a speaker is, the more likely it is to vary from spec. There is no budget for quality control in a $15 woofer. As such, when I make a design available to the public, the biggest thing I worry about is driver consistency. If I get word that a particular driver used in one of my posted designs has a consistency problem, I pull the design. This has happened a few times.

Seas makes the most consistent drivers. Have no fear when you are trying to replicate a design using Seas drivers. While not the most expensive, Seas drivers are worth the extra money. Vifa drivers have a wide range of consistency. Some of their older products have specs that are all over the place. Some of their newer products are a little more reliable, probably because they have not had the time to mess up the manufacturing setups.

I rarely buy drivers in the $100+ price range. There was a time that I did, but slowly I learned that there was a point of diminishing returns in the range of big dollar speakers. Dynaudio and Focal - poor performers and not worth even half of what they cost. We're not going to miss them as they withdraw from the DIY market. Scan-Speak does indeed make the best drivers and they were worth the cost 10 years ago, but the Chinese are slowly making the cloned drivers work and Scan-Speak value has gone down. Still, the Revelators are the current best in class and their purchase can probably be justified. Peerless woofers remain a high value and high performace option, though they really only gotten tweeter performance right with their new HDS model.

The Chinese still have a lot to learn about quality control. Take this as a bit of advice from an engineer at a company that has moved some manufacturing to China. (me) How many times over the years has the Dayton 1" silk dome had performance variations? I can't even count anymore. The Silver Flute woofer has has a wide range of efficiencies due to variance in magnet charge, forcing me to add a section on adjusting the tweeter L-pad to match. For an Asian driver manufacturer to produce consistent drivers, quality control has to be brought in from Europe or North America. Most Asian companies have never heard of programs like Six Sigma. I'm sure this will slowly change. Likewise, the cost of Asian drivers will go up. Nothing is free, including quality control.

I use "high value" crossover components, meaning "the most for my money". Most of the time this means Eagle Metal Oxide Film resistors, Bennic Poly caps, steel laminate core inductors for large values, and small gauge aircore coils where DCR is not important. This subject is highly debateable and often starts flame wars on the forums. I don't participate in those discussions. Don't expect me to believe someone has "Golden Ears" if they say they prefer big dollar crossover components. Most of the time I'll just believe they are idiots with too much money to spend, or just too defensive of something they once simply chose to believe in. I just offended someone. Gosh, I've never done that before. :) It's probably best to just keep a value balance - If you're going to buy $100+ drivers, you might as well buy boutique crossover components also.