This article will describe the fruits of my efforts to build a moderate sized speaker which might be moderately above average in performance for a basic two-way system. I desired a speaker like this both for my own use in secondary systems, and to fill a need for some friends, who desired fairly full range response (i.e., “it’s gotta have good bass”). Also, though I have access to a good wood shop, many potential DIY enthusiasts may not, so I hoped to use a modified off the shelf enclosure, rather than a build from scratch design. This makes construction simpler, especially in regards to having a nice looking finished speaker, even if you don’t have previous wood finishing experience.
Why an 8” two-way? It’s a fairly unconventional approach. Most two-way systems are typically based on a 5-1/4”, 6.5” or 7” mid-woofer, often in an MTM configuration. They are usually configured with a crossover at 2.5 to 3 kHz to one of the popular European soft dome tweeters. But, if that’s what you’re looking for, you won’t find it here! A good 8” mid-woofer will have some marked advantages in swept area compared with the smaller drivers, and usually a higher Xmax. These qualities extend the dynamic range and low frequency reach compared with smaller drivers. An 8” two-way does pose some special challenges; particularly if your goal is to maintain relatively uniform directivity over the critical midrange to high frequency octaves. This in fact required new thinking in the crossover design, and resulted in an approach which adapts ideas I used years ago in the output filter of a Class D amplifier [1].
This article will describe the design and construction of the fourth version of this speaker. Why did this project go through so many iterations? The MkI version was fairly successful at meeting many of my initial goals, but the drivers (Focal 8V4211 and MB Quart MBTTR1 were discontinued for DIY availability) as I finished the speaker. In the MkII version I decided to toss my initial cost constraints out the window in the selection of drivers, going with the Eton 8-800 and Accuton C23-6. Unforeseen shortcomings in the performance of these drivers left me dissatisfied with the cost/performance tradeoffs. The Eton 8-800 has a woofer cone mode at about 1400 Hz visible in the (not published) impedance curve, but not directly obvious in the 1 meter axial response (its big peak is at 3.2 kHz). The Accuton C23-6 did not measure as smooth and extended in response as published curves would have me believe.
The MkIII and MkIV are similar. Both use the HiVi Research M8a woofer, with rear firing ports. They differ in the front panel layout and choice of the tweeter. The MkIII used the Focal T c120dx2 (another discontinued part), while the MkIV features the Vifa XT25 ring radiator. In this article I will focus on the design and construction of the MkIV of this project.
| Fig. 1: The first version, using Focal woofer and MB titanium dome | Fig. 2: 2nd version with Eton woofer and Accuton C23-6 tweeter |
I included some overview about each speaker’s evolution in design, as well as the details of constructing the final design. It’s a little bit like that famous cartoon character Bullwinkle J. Moose “This time for sure!” As such, it’s been a voyage of discovery about making choices and trying things out, as well trying to blend a combination of experience and theory. So I’ll describe what I think worked well, what could be improved, and maybe may inspire you to build further on this design.
THE BASIC CONCEPT
two-way loudspeaker systems are typically a series of compromises, judiciously made to reflect the needs of the application and values of the designer. Constraints are imposed by limitations of size, complexity , and expense. Let me start off by explaining the needs and values I brought to this project.
First, for a moderate size enclosure, no more than 40 L or so, I wanted reasonable bass extension and power handling, and with good tonality and pitch definition. Also, since the majority of the music is in the midrange, particularly for classical, jazz, and vocal music, excel- lent performance in this area is very important to me. This requires both smooth axial response and fairly flat total room power response, where two-way systems with larger drivers often fall short. This implies trying to keep as constant directivity as possible through the operating range of the mid-woofer and tweeter, not an easy task in most cases.
A characteristic that is sometimes hard to define in simple terms is transparency, resolution of inner detail. In this regard, I wanted to minimize “additive” errors due to driver or enclosure resonance. This requirement has caused me to seek out drivers developed for pistonic behavior over a wide frequency range. In this regard I lean towards the composite or metal cone constructions over the various plastics that are often employed in loud- speaker drivers. The latter are in- tended to softly decouple portions of the cone with increasing frequency. Some, like the Seas P17RCY do this fairly admirably, with nary a glitch in the impedance curve, yet they still seem to have a characteristic cone coloration regardless of how the measured frequency response is tailored.
There are other questions about the ultimate “voicing” of the speaker. There are several issues or conventions to consider.
- Low frequency voicing positioning from adjacent boundaries, and degree of baffle step compensation to use
- Directivity – axial response versus forward power response
- Presence region voicing – to BBC dip or not to BBC dip?
Modern loudspeaker drivers, with their extended response and wide dispersion tend to provide flatter power response than many components available 10 to 20 years ago. And modern recordings tend to benefit from extended wide range response, assuming a very “clean” playback chain. However, many drivers and speakers have probably gotten a bum rap from time to time by being too revealing of flaws else- where in the system, including source components, electronics, and even the room. That doesn’t mean that a revealing speaker is flawed. It does mean that the overall system configuration must be considered, to get the most musical results.
To realize the optimal performance, the expected room placement must be considered while developing the design. This means considering where the speaker will be placed relative to adjacent boundaries such as the rear or side walls, and of course, the floor. This is not a given. While audiophile conventions acknowledge the desirability of locating the speaker so that comb filtering from early reflections and it’s destructive effects on frequency response and imaging are avoided, [2] this doesn’t reflect where most bookshelf speakers are located in the home.
A more subtle issue is that of the “audiophile dip,” (also known as the “BBC dip”) which provides a slight reduction in the response in the presence region. In theory it renders classical and other recordings with close microphone distances with a musical perspective more resembling “mid hall.” This enhances the apparent sound stage depth, and reduces the hardness which may otherwise result from the interaction of common recording techniques and a lively playback environment. This is often perceived as more pleasant, but may also be less transparent with recordings considered to be audiophile grade. But how much of your music collection falls into the audiophile category, versus the portion which is simply what you love to listen to?
This sounds like I’ve pretty much painted myself into a corner with a fairly hopeless combination of is- sues and requirements. Furthermore, some of my starting assumptions may seem out of kilter with conventional wisdom.
That these goals might be met, at least to a significant degree, was re- affirmed for me when listening to a pair of Avalon Eclipse’s at some friends last summer. Not coincidentally, I was inspired to mimic aspects of the basic configuration of the Eclipse, which also is an 8” two-way speaker. Conversations with the Eclipse’s original designer revealed two key points to making this work. First, a low crossover frequency is required for the mid-woofer to stay out of it’s cone modes. Second, the tweeter choice and crossover are critical to integrating the two acoustically while maintaining adequate power handling.
PICKING THE WOOFER
Why a single eight instead of the more typical dual 6-1/2” or 7” drivers? There are many factors in this choice. A good 6-1/2” or 7” driver will typically cost almost as much as an 8”, while having only about 1⁄2 the Sd, or working piston area. Also, if low distortion and higher output is desired in the low frequencies, Xmax and efficiency trade-offs are an issue. A few 8” drivers are in the 6 mm Xmax range; So, eliminating the MTM configuration from my choices provided either a very significant cost reduction, or conversely the opportunity to use better quality drivers, but still led to a difficult search for a suitable mid-woofer.
In selecting the woofer, my goal was to find a driver with an exception- ally clean midrange up to 2 kHz, while having a Qts in the range suit- able for either a ported enclosure or critically damped sealed enclosure (Q of 0.5). I also desired a reasonable Xmax (5 - 6 mm or better), and a tolerable price, preferably under $100. I looked at a lot of drivers, comparing their published response and impedance curves (watch for wrinkles in the impedance curve indicative of cone break up modes), and evaluating the possible LF alignments using Unibox [3], a very nice freeware LF modeling spreadsheet for Excel developed by Kristian Ougaard. I initially selected the Focal 8V4412, based on these parameters, plus a cost under $100. I had prior experience with some other 8” woofers, including Seas and Scanspeak, but I thought the Focal was a well balanced choice. I tried two more drivers before “finalizing” the mid woofer choice used in the final versions.
| Fig. 3: MLS Plot of HiVi M8a response (gating window rolls off below 200) | Fig. 4: MLS Plot of Vifa Xt25 15 degrees off axis |
| Test System and Design Software MEASUREMENT SYSTEM Measurements were made using the PC based CLIO system from Audiomatica, with a B&K 4133 external microphone and preamplifier. Distortion measurements are made with an HP8903A, and component measurements with an HP4192. DESIGN SOFTWARE LF box and driver modeling was done using Unibox 2.33 and 3.0, while crossover modeling and optimization and acoustic simulations were done using LSPCAD. |
THE MAGIC OF METAL?
Among 7” and 8” drivers, SEAS Excel magnesium cone drivers are very well regarded, and maintain pistonic operation to a fairly high frequency. I was somewhat dissuaded, though, by their high prices, and also by the large variation in response above 1 kHz, from -6 dB to plus 12 dB in the range from 1500 Hz to 5 kHz, for the W22. This promised a challenge with the crossover de- sign that I wasn’t sure I felt up to confronting.
Another 8” metal cone driver, the Hi-Vi M8a, with an aluminum/magnesium cone looked like a reasonable candidate, due to it’s more moderate cost (under $100), and a response characteristic that appears easier to manage. The response rise at the first breakup mode at about 2.5 kHz (Figure 3) is not quite as ferocious as some metal woofers, and I hoped it would require less extreme crossover measures to suppress.
The tweeter I first selected is one I’ve used frequently over the years, the MBTT1 from MB Quart. This is a 1” titanium dome tweeter with a very compliant surround, high efficiency due to a dual magnet design, and detailed, extended high frequency response.
Unfortunately, with change in ownership of MB, their tweeters appear to be no longer available to the D.I.Y. Market.
In the case of the MkIV version I’m discussing in this paper, a friend had kindly loaned me a pair of Vifa XT25’s to have a look at. I hadn’t seriously considered them because of some “theoretical” questions I had about their design, as well as a possible issue with the published response curve, but the actual tweeters test well and sound even better especially the off axis response, which I found to be smoother than on axis. Note that the response graph shown in Figure 4 is 30 degrees off axis! I’ve found better correlation in general with how my perception of the sonic quality of a tweeter with it’s off axis measured performance, in the range of 15 to 30 degrees, than for the on axis results. If this perception is correct, it’s probably because of the impact on the total power response in the listening area.
LF DESIGN & SYSTEM VOICING
Defining and voicing the low frequency design must be based on the intended room placement “To boundary load in proximity or not to boundary load that is the question.” Well, of course, it’s not quite that simple, but maybe it gives a memorable paraphrase for this conundrum.
Room gain due to adjacent boundaries is a fact of life. Some speakers utilize it very explicitly- good examples are the designs from Roy Allison in the 70’s [5], and professional studio monitors which are intended to be built in flush to a front wall. This type of mounting controls the radiation load of all the drivers, generally resulting in a nominal half pi to quarter pi space, which increases the output by between 3 and 6 dB in most typical cases, compared with anechoic radiation into a full “sphere”.
Most home speakers (excepting built in wall speakers) are not mounted or used this way, so things get a bit more complicated. It’s possible to calculate this affect, and the impact on response which depends on the distance from boundaries. For years I’ve used a straightforward document in MathCAD using concepts expounded by Roy Allison by which I can calculate the boundary reinforcement and the interaction which occurs with various spacing ratios. A good approximation to what you’ll get in an average rectangular room can be calculated just by including the effects of three adjacent boundaries (front wall, floor, and side wall, for example). Numerous commercial programs offer this facility in conjunction with other functions, such as SoundEasy, LspCAD, and RPG Room Optimizer.
How can this information benefit us in the design process? Well, I look at it this way, anything you don’t account and plan for is likely to come back to you in an unpleasant way. And obviously, it’s better to understand an effect and use it constructively instead of trying to ignore it.
Let’s consider what might happen if we take the HiVi M8a driver and use it in a sealed box of 32-34 liters. This will result in a Qtc of about 0.585 (just slightly over a Bessel alignment of 0.577), with an Fb of 47 Hz, and an F3 (where the response is down 3 dB) of 59 Hz. How does that fit in with the room boundary gain?
First, let’s look at what happens when we loan the hypothetical speakers we just “virtually” built to my daughter. She takes them into her bedroom, and will likely put them on the bookshelf by her bed; the woofer for the left one, for example, is 1 meter from the floor, 0.8 meters from the back wall, and 1.1 from the side wall. Let’s look at how this placing affects the radiation resistance and the frequency response into the ignoring for the moment issues such as room modal response. I’ve calculated this and shown it in Fig. 5. Not a pretty picture, is it? Why do you think I would ask her not to put the speakers there, but I’m sure she would find it a “convenient” spot. Not only a big hump in the 30-50 Hz area, (which she liked with her music), but some unexpected dips in the response, with the upper mid bass suffering a 7 dB dip. It would be easy to imagine a casual “audiophile” listener accusing this system of one note bass and poor pitch definition. Next, we move the speaker around a little bit- closer to the rear wall (actually, more typical), and while the dip at 100 Hz is not so great, instead we have more of a ripple, and a new problem in the lower mid at 300 Hz. No wonder there’s not much pitch definition on the bass guitar or standup bass- yet, this is a speaker with a theoretically well damped bass alignment.!
If we want to develop synergy between the speaker and room, we have to make the boundary conditions work for us- that means avoiding placements which introduce nulls excessive nulls and peaks, and making intelligent use of the room to extend the low end response of the speaker. The placement and LF transfer function should be matched to each other.
Actually, that’s not as complicated as it seems, because I’ve found there are two main factors to consider. First, there is a very simple ratio (the golden mean, 1:1.618) that can be used for the boundary spacing conditions. You can further investigate the application of the golden mean in the reference material on the Cardas web site [2], but I’ve been using this ratio for speaker placement long before there was a Cardas website or had heard of George Cardas! Using this ratio between the dimensions distributes the peaks and nulls and smooths the response.
Then, the other factor to consider is the absolute dimensioning so that the LF reinforcement is complementary to the speaker system roll off. In Figure 7, I’ve shown suggested distances for an LF system with damped low frequency extension similar to my design goals for this speaker. I can get a pretty nice response by repositioning the speakers so that the acoustic origin (woofer) is at the following distances: 1.96 m from the rear wall, 0.75 m from the floor, and 1.21 m from the side wall.
There are some interesting things implied by this, of course, that for optimizing a box speaker with a given LF profile, there really IS such a thing as too small a room. Unless, of course, you design for placement at the boundary in which case there’s no room gain to mess up the low end, or to reinforce it compared with the anechoic response. For a friend, we designed this same system in a sealed box with a critically damped Q. It can be placed in a similar manner, but then the bass is a bit “light”, as shown in Fig. 8. It still goes fairly deep, and if response closer to flat in the 50 to 90 Hz region is desired, reducing all the distances in proportion will move the room boundary support up in frequency. [/FONT]
I used Unibox for both evaluating drivers and selecting the box volume and alignment. Because I wanted low frequency extension below 40 Hz and adequate power handling for playback peaks around 100 dB, a ported enclosure seemed a good choice. I’ve noticed some similar sized sealed systems have a tendency to develop a bit of congestion at playback levels over 90 dB, and I hoped to avoid that through driver selection and enclosure design. The mid-woofer I ultimately selected, the HiVi M8a, has a reasonable combination of Xmax (6 mm) and Qts (0.43), though as a result it gives up a little in sensitivity in order to gain LF extension.
ENCLOSURE CONCEPT
I have access to a fairly complete wood working shop, but I found after design calculations that the enclosure I needed can also be constructed by using a modified Woodstyle W123REV enclosure. This is a reverse aspect ratio version of the original Woodstyle 123 box, which has an initial internal volume of 48 liters. This enclosure is avail able from several sources at reason able prices, so I proceeded using that path, since a home constructor could build these with as little wood working equipment as a saber saw and drill. Note that considerable enclosure volume of the W123 cabinet is used by bracing, the driver volume, and of course, the port.
Figure 9 shows the concept developed for this enclosure and the planned driver positioning of the MkIV, the final version. This perspective view shows the extensive bracing used to make the enclosure more rigid and better damped. Less obvious, perhaps, is the 1⁄4” sub-panel added to the front of the woofer and port area, also to thicken and strengthen the area of the cabinet with the largest holes, and the extra rear panel. While I usually prefer a port exiting towards the rear of the enclosure, the recipient of the first pair (an old friend) planned to place these in a way which made this a poor choice. For the MkIII and MkIV I implemented a rear existing port, as shown later in Fig. 9, and substantial changes to bracing.
While building the first pair I frequently heard questions like “Why so many braces in a cabinet this size?” Though 3⁄4” MDF is a good starting point for cabinet construction, it takes little in the way of a “knuckle test” to reveal it’s shortcomings you’ll find fancy instrumentation isn’t necessary! Mark Wheeler had a good series in the 1999 “Speaker Builder”,[4] which included a section about “Listening to Walls”, in which he discusses the effects of different wall and bracing constructions, including the effects of materials and choices of adhesives. His experience mirrors mine, in that dense hardwood bracing is more effective than most other types, and that two part hard setting adhesives, such as epoxy, produce the best subjective clarity . In the first two versions I used edge on braces- but in the later versions, essentially lined the enclosure interior. So, some simple guidelines I try to follow can be summarized here:
- Use 1x3 or better oak reinforcement by all large holes, such as woofer
- Use sub-panels behind the baffle to acoustically isolate the tweeter from the woofer back pressure
- Reinforce the front panel with a hardboard (HDF) 1/4” sub-panel, which lifts the driver plane closer to the flush front of the grille cloth frame.
- In the MkIV version, I used vertical 1X6 panels to reinforce the sides, and similar panels on the top and bottom
- Brace the long walls by joining with a cross brace
- Damp the remaining wall vibration with acoustical treatment
I used red oak for bracing, which is readily available from home improvement stores at moderate cost. It’s stiffness is over double that of MDF. I attach the braces using slow setting two part epoxy. Besides it’s strength and working time, another advantage for the epoxy is the ability to make fairly good joints even when full clamping isn’t possible. Using extensive bracing in this manner is time consuming, and undoubtedly impractical for a commercial product. But, for the DIY constructor, who is typically using premium quality parts, the extra investment of time and modest cost pays real dividends. There’s a huge difference in the “knuckle test” between the original Woodstyle cabinet and the modified version, especially for the front panels, which I think are critical because they are the “launching point” for the music.
CROSSOVER CONCEPT
My initial planning for the cross- over, as described in the introduction, was for a 4th order Linkwitz-Riley network. This topology has many desirable characteristics, well suited toward the goals for a wide range two-way speaker.
- Individual driver responses are -6 dB at the crossover frequency, this lessens the driver power compared with B3, B2, or L-R2.
- Complementary phase characteristic of LF and HP through crossover region maximizes constructive interference
- Complementary phase characteristic is less sensitive to driver offset, and offset can be easily compensated
All of these characteristics work well in the context of a two-way system, which generally must stretch the performance potential of two drivers to span the wide range of frequencies which must be reproduced for many kinds of music. Especially , the first two benefits lessen the workload on the drivers compared with other choices, such as the 2nd or 3rd order Butterworth crossover topologies, since the latter are only down 3 dB at the crossover frequency, compared with 6 dB nominal for the L-R topologies.
I chose a crossover frequency below 2 kHz for two key reasons:
- Avoiding upper range break up modes for the mid-woofer, even when partially suppressed by cone damping and cone taper design
- Keeping within the range which avoids “beaming” as frequency increases which is necessary if correlation of on axis response an 20-30 degree off axis response is desired
The latter is especially desirable because I wanted to avoid the problems in timbre and tonality which occur when, as in most two-way systems, you go from a portion of the frequency range where the tweeter has a wide flare in the response, to a relatively “beamy” mid-woofer. Still, I had concerns about whether my goals for uniform response on and off axis could be met in conjunction with reasonable power handling and distortion for the tweeter, even with a 4th order L-R crossover.
Doing some initial design study with measured driver data imported to LspCAD showed it was easy to hit a 4th order acoustic target in principle- and the predicted amplitude response looks fairly good. But while the acoustic target of fourth order slopes was achieved, the actual electrical high pass slope was only about 12 dB/octave which saved money on the crossover parts, but raised big questions in my mind about power handling, even for the robust MB tweeter I was first considering. While the tweeter acoustic output is down about 15 dB at 1 kHz, the electrical power roll off is about half that value, not even 10 dB at 1 kHz. I felt that a steeper slope crossover was necessary to really make this work, but I didn’t want to use a lot of components. 4th order L-R is a lot of parts, as it is.
This got me thinking back to some filter techniques I used in the late 80’s and early 90’s with Class D amplifiers. I used a modified Elliptical-Cauer filters to implement the output filter of a Class D amplifier. Elliptical-Cauer filters belong to a class of filters referred to as finite zero filters [6] In standard implementations, they have equal ripple in both the passband and the stop-band. The group delay characteristic is not very different from standard “audio” filters, and for the same number of components, they can be shown to have very narrow transition bands compared with other filter types. Furthermore, they have a null in the response near the transition frequency which can be tuned somewhat relative to the nominal design for the passband; it was my intuition that this could be helpful for certain crossover filter problems.
The key to making them work for audio is adapting the filter alignment to the requirements of the application. In the Class D amplifier, the transfer function null was tuned to the primary switching frequency of the Class D half bridge, and the filter elements were adjusted to achieve a suitable Q at the corner frequency with in a reasonable variation of loudspeaker load impedance.
These thoughts led to some experimentation with an elliptic network which was based on a fourth order L-R topology, by adding a capacitor to both the woofer network and tweeter networks to provide the finite zero element. With adjustment of the filter coefficients, I found it was possible to emulate various filter orders. I investigated emulating 6th order L-R and 8th order L-R networks in the passband and corner frequency. With the latter it was feasible to achieve 48 dB/octave slopes and minimize the peak of the “bounce back” after the zero null to below the -40 dB level. This can be seen in Figures 12 and 13, which show the response profiles for the "ideal" transfer function into a resistive load. Some modification of the actual transfer function is employed to achieve any necessary frequency contouring such as baffle step compensation.
Note that in these figures I’ve expanded the typical vertical dynamic range by covering 10 dB/div extended down to -60dB! For good fill behavior, the net acoustical response should track the target filter function at least down to -20 to -24 dB.
This filter topology appears have several interesting properties when applied to loudspeaker networks.
- The null in the transfer function notch can be tuned to driver resonance- for example, to the upper range breakup frequency of a metal or Kevlar cone driver, or the fundamental resonance of a tweeter possibly eliminating the need for an additional notch filter
- Tradeoffs in corner frequency roll off rate and null frequency can be adjusted with some independence, allowing some variation for adjusting group delay, and compensating for driver offset on the listening axis (i.e., woofer acoustic origin behind tweeter)
- The overlap frequency range between drivers is very low, which facilitates minimizing destructive interference
- Variation in frequency response above and below the primary listening axis is reduced, which results in a flatter room power response
- Out of band loading on drivers is significantly lower, even compared with 4th order L-R crossovers
- Increase in component count is minimal compared with that to implement a true 6th or 8th order L-R network
Initially I suspected their might be issues with higher group delay in the crossover region which begged the question, would it be audible, and would it be significantly deleterious in comparison to the benefits this approach seems to offer? As it turned out, the predicted group delay for the filter alignments I used didn’t differ greatly from conventional 4th order L-R filters. As for subjective qualities, there’s really only one way to find out, of course: build it and listen.
In fact, one of the tests I usually do on a new design is to listen to the low pass section by itself, without the tweeter or midrange connected. If there is edge or grunge in the upper end of the woofer passband that isn’t sufficiently attenuated by the crossover network, it may be masked (to a degree) by the upper range driver when connected, but it will reveal itself quickly when listening to the low pass alone. I found that this unconventional crossover combined with a metal cone woofer gave excellent results in this regard, sounding like a steep active filter, without audible artifacts that would only be partially masked by the tweeter. I think this contributes greatly to both the transparency and ease of the finished speaker.
Figure 11 shows the resulting low pass network schematic after optimization to an 8th order L-R transfer function in the range of 200 Hz to 2.5 kHz, for a nominal corner frequency of 1.25 kHz. C1031 is the capacitor which converts the output filter section into an elliptic filter. The resistors associated with the reactive elements only model the parasitic loss of the component.
The high pass filter for the tweeter is shown in Figure 3. C2041 in series with L2041 converts the conventional 4th order topology into an elliptic filter in the output section. The tweeter padding network is formed by R2071 and R2081, setting the gain of the L-Pad.
The series resonance of the tweeter is compensated by an LCR zobel formed from the network of R2041, L2041, and R2041. This network is critical to maintaining a basically resistive impedance in the area of tweeter resonance. At one point, I’d hoped that by tuning the null of the HP section to the tweeter resonance that I could eliminate this section, but it turned out that the component sensitivity of the elliptic, (which is high) and the impedance interaction made it impossible to get the correct corner and slope without the zobel network.
Detailed Cabinet Design and Construction
I chose the Woodstyle WS123REV cabinet because it had the sufficient volume to work with, including allowing for bracing in addition to driver and port volume. I used the REV version instead of the standard version because it reverses the width and depth rations, having a more narrow front (12”) and greater depth (14.5”), which has several benefits. First, it’s possible to fit a 3” ID port with the required length necessary for a moderately low cabinet tuning (range of 32 Hz). Next, the narrower baffle moves the baffle step frequency up to a slightly higher range, which eases the crossover baffle step compensation. Also, my experience is that the greater cabinet depth allows the greater absorption of the mid-woofer back wave in the midrange without over-damping the bass. Lastly, the proportions are more pleasing esthetically to me, particularly since it results in a smaller front panel which is more apparent than the cabinet depth from most listening positions. This makes the speaker appear smaller, rarely a shortcoming in my experience.
If you want to start from scratch in building this, you know who you are, and what you need to do to make a basic box. Use the overall dimensions in my detail drawings as a guide. I’d recommend using a locking dado construction for the front and rear panel; to match the precision of the Woodstyle enclosures, you’ll need to be able to do repeatable setups to 1/32nd of an inch. For those interested in working with the Woodstyle enclosure, I’ll detail step by step how to modify it for this speaker system.
FRONT SUB PANEL
Starting from the Woodstyle enclosure, the first step I took was to mask off the area where the supplementary front sub panel is attached. I used a scribe and marked off the area where it will be attached (see Fig. 16), then masked the outer area using Scotch safety release masking tape (that’s the heavy flat white paper masking tape, not the beige crinkly stuff more often used for quick painting tasks). Then, using a small hand held orbital sander, I removed the black paint finish in this area. This is probably the single biggest annoying task in this project, as there is a clear coat applied to the black paint on the front panel, and this tends to clog even coarse 100 grit paper. Using a wire brush on the sand paper will help keep it clean.
I made the supplementary front sub panel from 1⁄4” hardboard; not the standard Masonite I’ve seen many places, but a hardboard more nearly resembling HDF, and finished smaller sizes, such as a dual tube plunger. Some national discount stores like Walmart are about 35% less, if you want to go that way. Epoxy , though expensive, seems to produce some of the best sounding joints (minimal sound!)[4], and is relatively non critical of glue joint thickness so that’s what I go with. Feel free to experiment, but realize that if you experiment with a *lot* of things, you may wind up with a very different result than I did.
The tweeter and mid-woofer positions were evaluated using Baffle Diffraction Simulator from Paul Verdone, available as freeware from the FRD consortium website[7]. The mirror image offset for the tweeter location distributes the edge diffraction effects over a wider frequency range, and reduces the impact at any one frequency.
Next, I marked the position for the woofer and tweeter centers (see Fig. 15 for front panel dimensioning) and using the world famous Jasper jig, routed out the mounting flanges, setting up for an outside diameter on the tweeter hole of 4.25”, and 8.5” for the woofer. I used a 3/4” router bit and adjusted the location of the hole size pin to compensate for the extra bit size compared with the nominal 1/4” for the Jasper jig calibration. The mounting recesses were made 3/16” deep. After cutting out the woofer and tweeter smooth on both sides. I obtained mine from a local home improvement center (you know, the one with the orange paint theme). This material is similar to 1⁄4” MDF in color, but seems stronger and stiffer. When gluing and clamping this panel to the speaker front, be careful to line it up per the drawing dimensions (so it will clear the grille panel), and if you don’t have bar clamps, use extra wood pieces behind the clamps at the rear panel.
You can use standard wood glue like Titebond, but for all bracing and supplementary panels I used two part epoxy, the 30 minutes working time variety. You’ll probably find the best price on epoxy from places that sell supplies for fiber glass repair or lay-up. I found that local hardware stores are expensive for holes, it’s a good idea to check the driver fit at this point (Fig. 18).
The internal braces are the next task to tackle. The sequence I’m describing for installing the braces is somewhat arbitrary, but since this is a bit like putting a puzzle together in three dimensions, it may be simpler to stick to the same sequence. First I installed the top to bottom side braces. Refer to Fig. 14 for dimensions and positioning; they are mounted flush to the inside of the front panel and the bottom of the enclosure. Their length is limited by how long a piece can be inserted through the woofer hole and still clear installing it to the side. I found this easiest to do by laying the enclosures on their side, taking care to protect the finish, and first installing both side wall braces for one side. I provided clamping force while the epoxy set using some Plitron power transformers and inductors which I had on hand, but I’m sure something more pedestrian, such as bricks, would work just as well. Braces for the other side wall are installed in the same manner after the first ones have setup.
I prefer to not have tweeter mounting plates subject to back wave pressure from the low frequency drivers, so oak 1X6 (which is actually 0.75” X 5.5”) is used to create a tweeter “sub enclosure”. Identical pieces are cut 8.5” inches long, but one must be cut with the same 3.2” diameter hole as for the front baffle tweeter cutout. The easy way to get this hole is to hold the piece in position by hand, and using marker pen, trace the hole outline in the front panel. I glued and clamped first the brace with the tweeter clearance hole, then after it sets, a second solid brace.
By now you’re probably beginning to get the idea that there is a bit of tedious repetitive work involved hereand you’re correct. It’s productive to think a bit about what you want to work on, and do some tasks in parallel for example, while some parts and glued and clamped and setting up, cut out other pieces, or work on the crossover construction. Because of the many steps involved, and attendant setting time for the glue, my experience is that it’s not really practical to think in terms of building these in just one or two weekends, but rather to allocate some regular work over a period of time.
After mounting the tweeter backing panels, the next step in reinforcing the enclosure will be the 1” x 3” braces which are glued in above, below, and to the sides of the woofer mounting hole. Large holes like the one the woofer is mounted on greatly weaken the front panel, and I like to shore it up as well as possible to address this problem. Standard clamps work well for gluing these braces in place, as shown in Figure 19 and 20. After the front panel braces had setup overnight (Fig 21), I used a router with a roller bit to follow the panel cutout and cut out the braces.
Following this the bottom and top wall braces can be glued in place after fabricating them from 1 X 6” material. Note that the top wall braces can run the full width of the internal wall, so they can be cut to about 10”, while the bottom wall braces must allow for the side wall braces, and so should be the same width as the tweeter braces, about 8-1/2”.
Last I mounted the internal braces for the rear panel (above and below the pre-cut cup hole see Figure 14), and cut an additional MDF board for an external rear panel brace. Having this brace external allows the longer port and avoids reducing the box volume. For one set of the MkIII’s, this panel was made full length of the top to bottom, but for the other and the MkIV, it was made shorter as shown in Figure 24. The side and top cuts were beveled by 30 degrees from the vertical. Like the front baffle panel, I sanded the cabinet to remove the original paint before gluing the new panel.
PORT INSTALLATION
After the internal bracing is all glued in place, the next step is to mark the position for the mounting hole for the bass reflex port, and to cut this port. I used a 3” diameter Precision Port assembly, with flared inner and outer pieces. In principle, the baffle hole for the port can be cut to the inner radius of the mounting flange, and then the assembled port can be installed through the baffle hole. However, this results in a relatively large cutout hole, which weakens the enclosure wall. I decided to minimize the size of this hole, which requires a bit more care in the assembly and installation of the port.
If you have a hole saw and drill press, this is the preferred method. I have an old Craftsman setup around I use just for jobs like this. Otherwise, mark the 4” diameter circle with a compass and use a fine scroll cutting blade in a saber saw, taking care not to overheat the blade by cutting too fast. To use a mounting hole this small for the port, you’ll also need to provide some relief on the port hole to clear the flare. The simplest way to do this is with a 45 degree bevel router bit with a roller bearing follower. I’ve found Bosch makes a nice bit for this kind of job. I set the bit to cut 0.4” deep, then checked it against the flare and took a little more off until the port flare would sit flush. With the roller follower, alignment of the cut is a non-issue; it’s all controlled by the depth and your original cut-out.
Figure 24 shows the cabinet back panel after routing the port bevel, Note that the center piece should be cut so that the total port length, including flares, is about 1” longer than the calculated length from your design program or manual calculations. For this design, the center pieces must be cut 3.5” long. Using PVC welding cement, the rear flare and the cut tube can be glued together. I didn’t install the port yet, because there’s a few more steps that have to be completed first.
Now it’s time to reach for the safe release masking tape again; as shown in Figure 25, mask off the painted areas of the front panel and the sides of the cabinet, in preparation for painting the unfinished portion of the front panel. Use your favorite satin or gloss black enamel, with sufficient drying time between several light coats. Even though this is “safe release” masking tape, I still remove it fairly soon after the last coat is reasonably dry. Let the paint harden overnight at least, then mask and paint the rear walls in a similar fashion. When you’re finished painting, assembling the port and lining the enclosure with damping material will be next on the agenda.
Before attaching the main port flare, I check the position and fit, and mark drill holes with a center punch, and drill pilot holes for the “grabber” screws which have extra large flat heads. I quickly applied a healthy bead of hot glue on the bevel area and a small bead on the flat area at the edge of the bevel, then install the flare and seat it firmly to spread the glue. After installing the mounting screws, I used the welding cement to attach the body of the port to the port flare.
An important point to keep in mind is the clearance between the port and the rear of the front panel. I carefully controlled the components and total height of the tweeter zobel network board. Because I mounted the zobel board on the rear of the built up front panel behind the tweeter, it’s important to have enough clearance to slide it into place. If you’re not sure about the room you have, or the total board height, you may want to leave the final port assembly until after you have the zobel board mounted and wired.
The last thing I did in for the cabinet assembly was to install a brace side to side to further reduce the side panel flexure. This was constructed quite simply of 1”X 4” oak, cut to 9” in length, which is a snug fit between the braced cabinet walls. It was mounted between the second set of side wall braces, just below the hole for the reflex port. I installed it using a rubber mallet to tap it in place, first coating the ends with some epoxy , and locating one end firmly in the position I wanted it in the cabinet, then tapping the other end with the mallet to scoot it into position. Figure 26 shows how it looks through the woofer hole after installation. This brace isn’t absolutely necessary, but I found it does further stiffen the sides and reduce the panel resonance.
Driver measurements and crossover development
LF DRIVER MEASUREMENTS
I evaluated drivers and worked with the first crossover concepts using a simple particle board sealed box, as well as an older Woodstyle enclosure. There are some things I look for when making an initial evaluation of drivers which sometimes show up in published curves, but even then may require you to do a little “reading between the lines”. Oddly, the impedance curve may be more useful in some cases than the nominal amplitude response curve.
First, let’s take a look at an 8” driver I used previously for a three way system, the Scan-Speak W21/8554 (Fig 27). Figure 28 shows a measured impedance curve, while the following Figure 29 is a graph of the nearfield measured frequency response at three positions: at the center, midway towards the surround, and at the edge near the surround. What can be deduced from these simple measurements? Is there any correlation between these measurements and what I hear when using this driver?
I think there are meaningful clues in both the impedance curve and a comparison of the nearfield response of the driver from different measuring positions. Classically, the ripples in the impedance curve, as seen in Figure X, are due to the discontinuity in mechanical impedance from the onset of the first cone modes. These modes occur when some part of the cone is vibrating out of phase with the rest, this is a
departure from pistonic behavior. Some reviewers see evidence of enclosure panel vibration in this effect, but when the same “glitches” are present if the driver is tested free air, then it’s clearly not the enclosure at fault. For the 21W8554’s I’ve measured, the first impedance glitch at 700-800 Hz seems to correspond with a slight dip in the response, followed by the “classic” Scanspeak rising amplitude characteristic. The near-field response tracks quite closely across the cone until this mode is encountered, then diverges somewhat up until about 1600 Hz. Then, above 2 kHz there are large perturbations in the response, as additional cone modes kick in.
My experiences testing drivers and commercial speakers has led to developing a pr efer ence for operating drivers within their pistonic mode, and preferring to see cone modes attenuated by at least 20 dB or more preferably 40 dB. For this reason, the crossover frequency should stay below the first cone breakup in the mid woofer pass band. This is often not the first obvious peak in the response curve, which is why a careful examination of the impedance curve and the nearfield response can be very helpful.
This is a philosophical matter, because there are also those who select drivers that they believe they can run up to the top of their passband, sometimes even using an absolutely minimal (if any) crossover. In this case they rely on inherently lossy cone materials with highly self damped characteristics the original SEAS P17RCY would be a good example, as well as some Morel drivers. My experience, though, has been that it is very desirable to stay within the pistonic region for the mid-woofer, especially to maintain clarity and inner detail in the midrange. (I like to hear the massed voices in choral music as individual performers if I listen closely!)
The mid-woofer I eventually settled on is Hi-Vi Research M8a (Fig. 30), an 8.5” frame mid-woofer with a concave aluminum magnesium cone. The impedance curve looks clean until the first major mode at about 2.3 kHz, and the near-field response tracks well (Figure 32). The response rise at the first breakup mode at 2.5 kHz is relatively low in Q compared with some other 8” metal cone drivers I tested, and I hoped it would require less extreme crossover measures to tame. The response curves measured at three nearfield positions track well, showing close tracking up to about 3 kHz. With this type of driver, and the challenge of mating it as a two-way, the elliptic crossover design proved to be a very important factor in the success of the project, and as we’ll see enabled the response to be tailored closely to the overall requirements.
CROSSOVER DESIGN
The basic process for measuring driver response and importing data into PC based software for crossover design has been described at some length by many others, so I won’t recapitulate very much of that. There are a few points I’ve found which make the process go more easily and consistently, so perhaps a brief discussion of those issues will be helpful for others.
For ease of data import to LspCAD, I use the DOS version of the Audiomatica CLIO software when making measurements for import, as it’s output data format is supported directly by LspCAD as well as other crossover design software. I also use CLIOWIN, and find it’s flexibility very useful for driver investigations, but it’s data format for MLS output isn’t compatible with many programs expecting the output from the DOS version.
If I had measured the drivers in the final enclosure design at “far field” ( 2 meters or more), then the required compensation for baffle step is straight forward, as it’s a function of the measured response versus the desired acoustical response. Since I was measuring somewhat closer to reduce room reflections for MLS measurements with 200 Hz bandwidth, this required that I estimate the compensation for baffle step using a combination of the BDS simulator by Paul Verdone, and comparing against my nearfield driver results.
The next step is to create a file for the tweeter which is just for impedance compensation, a single driver project. For tweeters like the Vifa XT25 which do not use any LF damping with ferro fluid, the impedance rise at resonance is substantial, and will interact with the crossover filter unless it’s corner frequency is several times higher.
My preference is to start the tweeter network design with a full impedance compensation zobel, optimized for as close to ruler flat impedance curve for the tweeter as I can get. This also has the benefit of assuring very consistent frequency response when an Lpad is used with a wide range of attenuation variation. While this is unlikely to happen in a single design, if you like to “recycle” your network designs, it’s a must. Since from the outset I know there was interest in a boundary loaded smaller version of this speaker, as well as an MTM, this flexibility was mandatory . While it’s not uncommon for “production” speakers to use a single resistor for tweeter attenuation and optimize all the component values so that the desired acoustic transfer function is achieved with a single resistor, any need to adjust the tweeter level up or down may result in the shape of the filter transfer function being altered due to the change in impedance seen by the network. My experience has been that minimizing these kinds of interaction in the early design phase pays benefits in getting a consistent performing design finished quickly optimizing for cost reduction can be done later, and for most DIY constructors, the price of a few resistors is a non-issue anyway .
Figure 34 shows the predicted impedance curve after selecting the zobel values. No, this isn’t a typo or mistake the resistance and impedance of the XT25 are rather low, and I would probably even call it a 3 ohm tweeter myself, not a four ohm as it is specified. It is probably not the tweeter to use for a SET friendly design!
Even when I know well the approximate level I’ll be using and required padding, I start with no pad and setup the network topology and optimize for the acoustic transfer function. Then, I freeze component values and add the Lpad, and reset the level and let the optimizer dial in the attenuation. This results in Lpad values which are “classic”, and avoids impedance interaction in the network. I’ve found that if I optimize the circuit at once, it may use the impedance interaction of a specific set of attenuation resistors to control part of the slope, and result in a network that only works properly at one attenuation level. Since the optimizer is “dumb” about these issues, you have to be the smart one.
Other considerations for the woofer network and optimization, beyond the nominal acoustic transfer function target for LP network, include baffle step compensation and compensation for driver acoustical distance behind tweeter, plus attenuation of the cone resonant peak. In general, the network topology is that of a conventional fourth order LP network, with the addition of a zero in the transfer function by means of C1031. L1031 and C1031 are selected to optimize the slope of the LP cutoff to the acoustic target (8th order L-r), and also to tune the notch in the response to the driver peak at roughly 2.5-2.8 kHz. Baffle step compensation is achieved b y the interaction between the primary inductor, L1011, and the RC network formed by C1061 and R1061 in parallel with the driver impedance. Within a limited range, the baffle step compensation can be adjusted by changing the values of this RC network. The change in the effective load impedance presented to the filter network as a result of the use of the baffle step compensation requires adjustment of other component values, which is how the optimizer “pays it’s keep”.
As a final step, I try adjusting inductor values to the nearest conventional available value, and re-run the optimizer once again to finalize the capacitance values. Last, I double check that the optimizer and I haven’t created a monster network with unusual impedance dips that only an Aragon, Krell, or Bryston amplifier will drive comfortably.
Figure 38 shows the predicted composite response; note that the tweeter measurements were made in a test box which didn’t use any diffraction suppression techniques, and I believe the roughness in the 6 kHz and 12 kHz areas are due to this. I don’t think this is something to address in the crossover design, but rather by the baffle layout and diffraction suppression techniques, which I’ll describe in the final assembly . Note the expanded vertical scale compared with typical crossover plots, covering a range of 60 dB. A more conventional plot range of 30 dB would not show the notch and “bounce-back” of the filter network.
This predicted response includes the effect of adding a 60 mm acoustic center offset in LspCAD for the mid-woofer, to model explicitly the difference in acoustic origin relative to the front panel, and modifying the tweeter slope and woofer slope slightly to achieve optimum summing on the design axis, which is midway between the woofer and tweeter.
CROSSOVER CONSTRUCTION
In order to fit the electrical crossover conveniently into the cabinets, I split the networks up into three boards, using 6” by 8” sized quarter inch hardboard as the construction base. I’ve gotten in the habit in the last several projects of using Euro power connectors for interconnecting the crossover boards with the drivers and input connectors; this makes experimental work much easier than soldering and unsoldering connections, and they’re available at your friendly local Radio Shack, as well as conventional distributors.
Construction is straightforward (Fig. 39, 40, and 41); I use hot glue to mount the components to the board, and wire up the networks with AWG 14 wire.
For these speakers I obtained most of the inductors from distributors for Solen. If you wind your own coils, as I sometimes do, I suggest using AWG 14 wire to get a similar DC resistance. I used a combination of GE polypropylene and Solen polypropylene capacitors. Just for fun I measured the ESR (equivalent series resistance) of the GE and Solen caps, and they’re both quite low under 10 milli-ohms. In combination with the low dielectric absorption coefficient for polypropylene, the electrical behavior is quite good. There are more expensive film capacitors available, which do have their fans and advocates, but unless you’re using very high grade electronics and cables, you may not hear a difference in your system. I leave the upgrading of crossover components to the judgment of the individual constructor.
Note that the tweeter crossover is shown without the tweeter level pad installed and wired. In order to handle the power required by 8-9 dB attenuation, and to have some flexibility in adjusting the attenuation, I used an unusual construction for the Lpad with multiple parallel resistors (Fig. 42). This was built with paralleled 15 ohm and 12.5 ohm Mills resistors, five in each leg, with an additional 10 ohms in parallel on the shunt leg to fine tune the attenuation. If this is a little much to deal with, there are high power four ohm Lpads available from a few sources, such as Phoenix gold, but while I’ve sometimes used them to experiment and determine the range of levels, I don’t usually leave these in the finished speaker. Also, technically, a 4 ohm Lpad is about a 25% impedance mismatch to the Vifa tweeter. In practice, this doesn’t make a very significant variation in the response,and if one wanted the flexibility of adjusting levels for adverse acoustics (overly reverberant in the presence region and highs), this might be the ticket to consider.
I used a popular “DB-Cup” assembly for the input connection for which the Woodstyle enclosure is pre-cut. After testing it I made some modifications. Testing it, you may ask? Something I try to avoid in connectors and wiring is the use of any ferrous materials. The tabs for connecting the wiring to the binding posts as supplied can be picked up by a speaker magnet, so they’ve got to go. This got me looking a little closer at the binding posts themselves, which though nicely plated, may not be the best quality base metal. In the end, I replaced the binding posts with some Vampire posts, and directly soldering the Cardas hookup wire to these posts. In testing wire for home brew power amps, I’ve found in the past I prefer the subjective qualities of the Cardas cable hookup wire, and so I use it in speaker construction also. Naturally, you can substitute your favorite hookup wire (I used 13 AWG for crossover inputs and woofer wiring), and the 15 AWG for the tweeter boards and connections.
ASSEMBLY & MEASUREMENT
At this point I knew I was in the home stretch. For the experienced builders, there’s little tricky about this phase, unless you forget to wire the tweeters in the correct phase, or don’t adequately seal the drivers when mounting them.
Depending on your confidence level, you may do as I do, and mount the drivers first with long leads coming out of the reflex port, and check out everything with the crossover boards wired up externally. This is highly recommended as a “sanity” check before putting everything together; it’s much easier to work on the crossover boards before putting them in the cabinet, if you find a wiring error.
As mentioned during the cabinet construction, the first of the crossover boards to mount and connect is the tweeter zobel board. This contains the impedance equalizing networks formed by the LCR combination of L2101, C2101, and R2101, which controls the impedance rise at Fs, and the RC network formed by R2211 and C2211, which flattens the impedance rise due to the voice coil inductance. Cut and connect output and input leads before mounting it; I arranged the terminal connections so that the “positive” input and output were adjacent on one side, and the nominal “negative” on the other. Actual polarity at the board doesn’t matter; this is a network which is connected in parallel with the tweeter, and the connection to the tweeter is just a "pass through". However, it’s necessary to be careful about the absolute polarity; note from the crossover diagram that the tweeter is connected in reverse phase to the woofer for an 8th order all-pass network, just as it would be for a second order. Since the tweeters are mounted in what is effectively a sub enclosure, it will be necessary to drill holes sufficient to pass the tweeter wiring; I angled these to one side so that a large flat area was still available for mounting the zobel board.
The crossover boards can be attached with “Industrial Velcro”, which is available at most hardware stores as well as Radio Shack. Or, if you’ve tested everything and you’re feeling a little brave, just hot glue it in place (guilty as charged ;-). The input and output connections must be made before mounting the board, as there’s very little space for working by hand in this area.
The woofer crossover boards were mounted in the bottom wall of the enclosure, towards the back and away from the front panel (Figure 45). The tweeter crossover board was mounted on the back panel, on the oak brace just below the input cup hole. Connect the leads from the zobel network, being careful to observe correct phase.
Prior to connecting and mounting the woofer, internal damping materials must be installed. I hot glued heavy felt on the side walls behind the woofer, attaching it by using a series of closely spaced thin beads (speed is of the essence, even when working with slow setting hot glue). In the area behind the tweeter, and also directly behind the woofer, I installed small folded blocks of polyester quilt batting. This helps with absorbing the midrange back-wave, and damps the LF alignment to a degree. Keep this material away from the port opening in the cabinet, or the reflex action may be impeded and low frequency output will suffer.
For sealing the drivers, I use self adhesive weather stripping foam. I drill pilot holes which are just slightly under the diameter of the body (excluding threads) of the mounting screws. I’ve gotten in the habit of using 1" drywall screws for driver mounting in MDF, though you may prefer black finished round head screws which are included in the mounting kits supplied by Parts Express with the woofers they sell.
INITIAL TEST AND DIFFRACTION CONTROL
After completing the assembly of the first MkIV cabinet (Figure 46), I made an initial check on the frequency response (Figure 47). Seeing this, the first inclination might be to reduce the tweeter level or fine tune it’s voicing, but seeing where the elevated regions lay, in a series of ripples in the 3 kHz to 14 kHz range, I suspected diffraction effects at the edge were inducing this problem. As hoped for after using the Baffle Diffraction Simulator, they are distributed over a wide frequency range, and at relatively low amplitude.
Figure 48 shows my experimentation with using soft felt to reduce edge affects. I cut up scrap pieces of various sizes, and spent some time experimenting with the felt cutout in the area of the tweeter. A circular cutout would be the worst of course, since it would result in a diffraction effect at a uniform distance and peaking at one frequency. Though I also tried a square cutout as Wilson Audio uses and I have previously employed with the Focal Tc120dx2, I found that the shallow diamond shape shown here seemed to work best in minimizing the diffraction ripples, as shown in Figure 49. This is reflected in the finalize configuration implemented for the felt on the front baffle, as shown in Figure 46. Double sided foam tape could be used to attach the felt, but very satisfactory (if hard to reverse) results will be obtained with that popular standby, hot glue.
Figure 50 shows the final configuration; note that the raised surface provided by the additional 1⁄4” sub-panel and the felt makes a fairly good blend to the standard Woodstyle grille frame, which has a slight bevel under cut on the inside/top, reducing the height of the grille frame at the inner edge.
ADDITIONAL MEASUREMENTS
Some additional measurements were made to check and verify performance against the design goals. Figure 51 shows the measured impedance curve. As expected from the tweeter zobels used, the upper range is quite flat, though one friend quipped, on seeing the low frequency variation and minimum impedance, that "this isn’t very SET friendly...". The LF impedance curve shows some modification from initial test box measurements, cutting the impedance peak from about 55 ohms to about half that value, and reducing the Q of the curves. I believe this is due to the batting used behind the woofer and behind the tweeter area for additional damping. The effects of this damping can be seen in the plot of near-field woofer and port output (Fig. 52). This somewhat "over-damped" alignment has good bass extension and weight, but without the heaviness or lack of articulation often attributed to ported designs.
I also made plots of the response with the tweeter in "normal" phase (actually reverse polarity) and for the "out of phase" condition as the classic check on crossover alignment. Figure 53 shows a deep and narrow null as would be expected from this high slope design. For the MkIV version I centered the design axis between the woofer and tweeter, in my belief that it was more useful to have good off axis behavior above the primary listening axis, such as when standing up, compared with the performance well below the listening axis, which might only be encountered if lying on the floor. Figure 54 shows the resulting vertical response window, with the dip in response in the crossover region occurring at 15 degrees below the woofer axis. Performance in the crossover region is smooth both for seating position and standing at reasonable distances.
The following Figure 55 shows a series of sweeps at 15 degree increments on the horizontal axis. The dispersion limitations of a 1" nominal tweeter above 10 kHz become readily apparent from this plot. However, this also shows the wisdom in general of not exceeding roughly 1200 Hz for the crossover of an 8" driver, as the behavior in the upper midrange and presence region (1 – 3 kHz) remains remarkably consistent from on axis to 45 degrees off axis. Note that a different test microphone was used for this measurement, a Behringer ECM8000. I believe that this consistency on and off axis goes a large way towards explaining the large sweet spot these speakers exhibit, as well as their overall sense of transparency compared with many two-way systems using smaller mid-woofers but higher crossover points.
MEASUREMENT NOTES:
Measurements were made with CLIO system from Audiomatica, using both DOS 4.5 software and CLIOWIN. An HP microphone preamplifier and B&K 4133 capsule were used except figures 52 and 55, made with a Behringer ECM8000 microphone and MAudio DMP3 microphone preamplifier. In room MLS measurements used relatively short windows to reduce wall and floor boundary interactions, limiting the measurement accuracy below 400 Hz. zz0.1rxazpw73srzz
SETUP & EQUIPMENT
All of the versions including the MkIV described here underwent listening checks during the crossover evaluation process. By the time I was finishing the MkIV, my ability to correlate what I heard with specific measurements had improved considerably. Much of the listening evaluation for the MkIV crossover occurred in the test box phase, with only a very minor adjustment (lowering) to the shunt impedance of the Lpad made in the final crossover.
My program sources are all digital, including a Sony SCD777ES SACD player, an APN Audio MP-DAC II, and an experimental DAC using a CS8420 for 2X sample rate conversion (44.1 to 88.2 kHz) and re-clocking, with a CS4397 24/192 converter and transformer coupled connection to a non loop feedback discrete class A balanced output stage.
The preamplifier used for evaluations was a Marchand PR41 passive unit, using Shallco switch based attenuators, which has little sound of it’s own, as long as short interconnects are used. Interconnects from the digital sources to preamplifier were Jon Risch recipe cables (the second version with mixed core materials) constructed with WBT connectors. Cardas interconnects were used from the preamplifier to power amplifier. Speaker cables are based on Kimber 8TC with WBT connectors.
I tried both a conventional power amplifier with high damping factor (an Aragon 8008X3B) and a non-loop-feedback design, the A yre V-5 in my listening tests. In no sense did the bottom end control seem to suffer with the Ayre amplifier, though it’s measurable damping factor is no where near as high as the Aragon. Because of it’s pristine midrange and high frequency behavior (very "un-solid state" in character), I used the Ayre V-5 for the majority of listening evaluations. The V-5’s frequency response extends to 200 kHz, so I don’t think the smoothness in the upper range can be attributed to rolled off frequency response, as some claim for vacuum tube amplifiers.
As discussed earlier in the design phase, setup with regards to boundaries plays a crucial role for the in room performance of any speaker. Fortunately, if you’re not fond of throwing together your own MathCAD doc or Excel worksheet to analyze your setup, there are some excellent programs available at reasonable cost to enable one to take as cursory or detailed a look at this issue as you like.
One I’ve found very helpful is Room Optimizer from RPG Acoustics. As it’s name implies, Room Optimizer can actually help one optimize the setup of the room, and not just by telling you how mediocre your first thoughts for speaker placement were! Using constraints you choose, it will perform a search for optimization of both the speaker and listener position in the room.
It does this by analyzing both the boundary and modal response issues in the room, and within the constraints of the configuration setup, explores a variety of combinations of locations to seek out the overall smoothest response in the optimization range for the low frequencies between 20 Hz and 300 Hz . For the “optimized” layout, it will also suggest locations for room treatment and damping to minimize response irregularities stemming from multiple path comb filtering.
What’s worked best for me with Room Optimizer is to define a starting point based on my MathCAD calculations for locations next to the major boundaries. Then, enter a range of location for the speakers, and make one mirror dependent on the other for a symmetrical stereo layout. Define an area for the preferred listener locations, then let the optimization process have a go at it and see what turns up. Figure 56 shows the results of a positioning analysis I ran for this system in one room, including suggested positioning for wall treatments to reduce early reflections. Figure 57 shows the comparison in the modeled frequency response during optimization, including the best and worst case results.
Keep in mind that the optimizer is an idiot savant, it’s just a “dumb as dirt” number cruncher. If you give it some good clues and starting points to work from, it can weed through a lot of possibilities and allow you to explore some “what if?” scenarios pretty quickly (literally, just a couple of minutes on a PIII or Athlon system). Such as, “what if I rotated the listening axis in my room 90 degrees what benefits or problems would result?” A comprehensive look at ways to use Room Optimizer could be a complete article on it’s own, and is well beyond the scope of this article. The white paper available on RPG Acoustics web site [8] gives a thorough overview of the capabilities of the program and the basic techniques used at it’s core.
In my own setup, I have a few additional “problems”, such as a front projection screen behind and between the speakers when not viewing video, I drape the fixed screen with a white down comforter.
LISTENING IMPRESSIONS
Vocal music, both popular and classical, constitutes a large part of my listening preferences. Because of our innate familiarity with human voice, it tends to be fairly revealing of coloration in a wide part of the musical spectrum, from upper mid-bass through the presence region and top end. Solo voice works well, but probably combined voices work even better, as they’re more revealing of possible intermodulation issues and quickly separate the wheat from the chaff in the ability to resolve individual vocals with natural timbre.
Modern high quality recordings of pop such as the Keb Mo “Slow Down” and Alison Krauss’s “Forget About It” sound very good, with the expected clarity, body, and dimensionality . I’ve found listeners with little previous exposure or interest in Bluegrass entranced by tracks like “It don’t matter now”, and the title cut “Forget about it”. The accapello vocals of Jonatha Brooke and Jennifer Kimball of the Story on several cuts from “Grace in Gravity” float in space while at the same time make a strong personal connection for me.
I’ve been very pleased in the last few years how some older recordings fare that have been carefully re-mastered, such as Maddy Prior’s “Woman in the Wings”. Woman in the Wings is a collection of original songs from the vocalist who is probably best known for her long time career with Steeleye Span, a British band doing rock arrangements of traditional British folk songs and original compositions. In this recording, her original songs showcase her voice in a variety of styles including traditional jazz influences; the recording is so natural and vivid sounding that it’s a startling contrast for those familiar with the processing and reedy tonality on her work with Steeleye Span and in other projects. In this regards it sounds like a much more recent recording than it’s original 1979 release date would suggest. Perhaps the production values Ian Anderson brought to the studio as producer made the difference, complete with very competent backing performances by a host of musicians from late 70’s incarnations of Jethro Tull.
“The Hunter” by Jennifer Warnes is a modern well produced album in her ultra clean style that gives a good demonstration of natural, relaxed midrange with detailed top end reproduction, while including a good workout in the bottom end from the low 30’s and up. Cuts like “Big Noise New York” demonstrate Warnes command and assurance as both a writer and performer.
Recent releases from Telarc have raised the bar for recording quality and have also featured some surprisingly good performances compared with many “audiophile” recordings. I especially appreciate their new DSD recordings released on hybrid CD/SACD disks, because they give the listener a choice in playback equipment. Though I’m no music critic, I think the new release of Carl Orff’s Carmina Burana could well come to be the performance and recording by which others are judged, and likely found wanting. This opinion comes from auditioning it on a “reference” system which has been assembled at a friends over the years.
Obviously, it would be pretty ludicrous to compare this recording on an 8” two-way to a four way system with active crossovers using an IB sub with a dozen long throw 12’s, two pairs of Acoustat 1+1 electrostatic panels from 100 Hz to 600 Hz, a pair of Bohlender-Graebner RD75 ribbons from 600 Hz to 7 kHz, and a Technics leaf ribbon line array covering the top end. Yet, on this “modest” two-way there’s a surprising clarity and transparency in vocal reproduction that clearly reveals the recordings merits, and surprises even more at the impact achieved with the remarkable percussion on this recording. This is assuming relatively reasonable playback levels, considering the limitations of a speaker of this size. Otherwise, increasing levels of even order distortion from the Vifa tweeter as well as dynamic compression of the woofer on high level bass transients will result, but at playback levels over 100 dB.
All in all, the performance of these moderate size speakers has reached, and perhaps exceeded the level of transparency and musicality I was hoping for. With four iterations of the design completed, it’s clearly a triumph of perspiration over inspiration; there’s no substitute for development and refinement. In the form presented here, they’re not for every room and situation, because they are designed and benefit from spacing well out from the walls. My friend with the “reference system” was pleased enough with the results that we’ve designed and built a version specific to boundary loading on a wall for use in his home, using a smaller cabinet and removal of baffle step compensation in the crossover filters. Further development underway includes a floor standing MTM version, being constructed as I complete this article, and a dipole system using these drivers in the midrange and high frequency area.
I’d like to take this opportunity to thank Charles Hansen (formerly of Avalon, currently of Ayre Acoustics, not the Charles Hansen frequently published in AudioXpress) for the time he gave discussing the design of the Avalon Eclipse, as well as some suggestions with regards to measurement techniques. I’d also like to extend my thanks to Thomas Waale, who suggested some of the bracing techniques used in the MkIII and MKIV versions, as well as providing the use of his main listening room for evaluating the MKIII version in a larger environment than my own home would permit. I’d also like to “complain” to Edward Dell about the huge stack of Speaker Builder magazines I have at home and can’t bear to give away or dispose of. For these guys, it’s all about the love of music and it’s reproduction, and sharing that with their colleagues, friends, and readers.
RECORDING REFERENCES
- Carl Orff, “Carmina Burana”, Atlanta Symphony Orchestra, Donald Runnicles conducting, Part 1, Telarc Hybrid SACD-60575
- Debussy, “Iberia” from Images, Cincinnati Symphony Orchestra, conducted by Jesus Lopez-Cobos, Telarc Hybrid SACD-60574
- Gabriel Faure, “Piano Trio in D minor”, the Florestan Trio, Hyperion Records SACDA67114
- Harry James, “Best of Harry James”, Track XX, Sheffield Labs
- Patricia Barber, “Companion”, Track 7, First Impressions Music, FIM XRCD 027
- Arvo Part, “Frateres”, The Orchestra of Flanders, conducted by Rudolph Werthen, Telarc CD-80387
- Oscar Peterson, Ray Brown, Milt Jackson, “The Very Tall Band”, Telarc CD-83443-SA
- The Ravi Shankar Project, “Tana Mana”, Tracks 4 & 8, RCA 2016-2-P
- George Fredrick Handel, “Messiah”, Swedish Radio Symphony Orchestra, conducted by Anders Ohrwall, FIM Music FM SACD 039
- Jennifer Warnes, “The Hunter”, Track 2, 3, & 9, HDCD XRCD2 (Japanese Import) CDA1065
- The Story (Jonatha Brooke & Jennifer Kimball), “Grace in Gravity”, Track 1, 10 & 11, Elektra 961321-2
- Allison Krause, “Forget About it”, Track 2 & 8, Hybrid CD/SACD, Rounder SACD 11661-0465-6
- Keb Mo, “Slow Down”, Track 3, 6, 9, BK069376
- Jonatha Brooke, “Steady Pull”, Tracks 1, 2; CD BDR60801-2, DVD-A BDR-DV-61001
- Maddy Prior, “Woman in the Wings”, Tracks 2, 8, 11, BGOCD215
- Oregon, “Beyond Words “, Tracks 2, 4, 6; Chesky Records JD130
PARTS AND VENDORS
| Woodstyle W123REV Enclosures | Madisound Speakers |
| GE Polypropylene caps | Madisound Speakers |
| Hi Vi Research M8a woofer | Parts Express |
| Vifa XT25 XXX Tweeter | Madisound, Parts Express |
| Solen Capacitors | Madisound, Parts Express |
| Solen Perfect Lay Inductors | Madisound, Parts Express |
| DB Binding Post cups | Madisound, Parts Express |
| Vampire or Cardas Binding Posts | Welborne Labs |
| Cardas Hookup wire | Welborne Labs |
| Mills Resistors | Parts Express, Welborne Labs |
| 1X3 oak board ~ 4 feet | Home improvement stores |
| 1X4 oak board ~ 10 feet | Home improvement stores |
| 1X6 oak board ~ 17 feet | Home improvement stores |
| 1/4” smooth sided hard board | Home improvement stores |
| Two part slow setting epoxy | Home improvement stores |
| Slow setting hot glue and gun | Home improvement stores |
| Polyester batting, 1 pkg, as for quilting | Fabric stores |
| 3/8” - 1/2” Hard Felt | Gasket Supply Firms |
| 3/8” Soft Felt | Local Hardware stores |
References
[1] J. Hancock, “A Class D Amplifier Using MOSFETs with Reduced Minority Carrier Lifetime”, Journal of the Audio Engineering Society, Sept., 1991.
[2] George Cardas, Loudspeaker placement guidelines under “Insights”, http://www.cardas.com
[3] Christian Ougaard, Unibox 3,4; web site: http://www .danbbs.dk/~ko/ubmodel.htm
[4] Mark Wheeler, “Listening to Walls”, 1999 “Speaker Builder” magazine (reprint available through AudioXpress).
[5] Roy F. Allison; “The Ifluence of Room Boundaries on Aloudspeaker Power Output, “Journal of the Audio Engineering Society, vol. 22, pp. 314-320 (June 1974).
[6] Richard C. Dorf, “The Engineering Handbook”, IEEE Press, pp. 1204-1212.
[7] Paul Verdone, FRD Consortium, Baffle Diffraction Simulator, http://www.pvconsultants.com/audio/frdgroup.htm
[8] RPG Acoustics, http://www.rpginc.com , White papers for Room Optimizer, Room Sizer
