AOS Studio 30

The AOS Studio 30 (S30) is a closed box MTM loudspeaker (Photo 1). The S30 is in many ways a bigger brother of the S24. I built this Studio 30 for use in a large room, so I changed the «bookshelf» format of the original AOS design to a full-size floor-standing loudspeaker. The S30 uses high-quality driver units made by ScanSpeak (Photo 2).

AOS_S30

Photo 1: The AOS Studio 30

AOS_S30_close

Photo 2: Close up of the tweeter and the midwoofers

Peerless Subwoofer (active)

I designed this subwoofer (Photo 1) for flexible use in high-quality audio systems. In particular, I use it in my own system in combination with the S24. The subwoofer uses two Peerless XXLS-P830845 12-inch long-stroke woofers. Each woofer is mounted in a separate closed box. The woofers are mounted opposite to each other to cancel the vibrational forces acting on the enclosure.

peerless_subwoofer

Photo 1: The Peerless subwoofer

Each woofer is driven by a dedicated power amplifier. An active DSP with adjustable low-pass filters is used match the subwoofer to the main loudspeakers. In addition, the DSP provides a Linkwitz pole-zero equalisation, which is tuned yield a flat acoustic frequency response down to a -3 dB point of 21 Hz. The quality factor was set to Q = 0.65, which results in a very well-controlled impulse response. Peerless_Subwoofer_drawing

SUBWOOFER_PEERLESS_IMPEDANCE

Impedance curves of the two Peerless XXLS-308-8 woofers mounted in the box, and theoretical impedance of a woofer-box system with f0=45.5 Hz, Q=0.94, and RDC=5.8 Ω.

PEERLESS_SUB_NEARFIELD

Frequency response of the Peerless Subwoofer, measured in the nearfield to gate out room echoes. Linkwitz pole-zero equalisation from f0=45.5 Hz, Q=0.94 to f0=21 Hz, Q=0.65. High-pass filter (18 dB/oct) is a cascade of a second order filter (fc=47 Hz, Q=0.7) and a first order filter with f0=118 Hz (determined experimentally for use with the AOS S24).

THEL pre-amplifier

I built this pre-amplifier using the THEL VX-D modules for line-level amplification. The enclosure is made of a 90 mm thick chunk of layered birch ply. The bottom and rear plates are aluminum plates that were cut and engraved using a CNC machine. The custom-made source and volume control knobs were machined using a lathe.

THEL_PREThe THEL VX-D electronics use a discrete MosFET output stage with a very low output impedance. This prevents high-capacitance cables and low-impedance power-amp inputs from affecting the sound signals.

thel_pre_volknobThe power supply uses rechargeable batteries instead of the conventional DC conversion from AC voltage derived from the wall outlet. This results in the cleanest possible DC voltage supply that is free of any hum or noise, which might affect the sound quality. The batteries are recharged automatically when the pre-amp is not in use to make sure the pre-amp is always ready to play.

thel_pre_layout

300B parallel single-ended tube amp

Update: now availble on HiFi-Bau Brennwald!

Overview

300B_PSEThe 300B is an unusual single-ended tube amp with 300B power tubes. The amp uses two tubes in parallel, which makes it easy to produce 30 Watts of clean and wonderful single-ended 300B power! The input stage also relies on two paralleled triodes (ECC88), which drive the 300Bs through a step-up transformer. All this makes for a very simple and direct signal path with no capacitors. The 300B tubes can be replaced with stronger tubes, such as the 32B. In this case, this single-ended amplifier can deliver more than 40 Watts!

Description of the Amplifier

The aim in building this amplifier was to design a single-ended amplifier which has enough power to drive “normal” loudspeakers with sensitivities of 90 dB/W/m or less. I decided this design should work with down-to-earth voltages lower than 400 V and it should use easily available tubes. The concept of this amplifier is based on proven designs by Electra Print and Andrea Ciuffoli: The power stage uses two 300B tubes in parallel to produce enough power to drive “normal” speakers, while the venerable sonics and qualities of the 300B are retained. The input and driver stage (Based on a ECC88 / 6922 / 6DJ8 tube) is coupled to the power stage through a high-quality transformer. For simplicity and for best sondic results, no feedback loops exist in the amplifier. Only triode tubes are used throughout the entire amplifier. The power tubes are operated in fixed bias mode. Their operating point can therefore easiliy be adjusted to accomodate for non-matched tube pairs. This also allows the use of other power tubes, such as the AVVT 302B / 32B or the Emission Labs XLS tubes, which provide even more power than the standard 300B (e.g. more than 40 Watt with the 32B). This amp sounds very nice and clean. Yet, it has enough power to drive speakers like the ScanSpeakers without difficulties. With standard 300B tubes, the output power is 30 W.

The schematic

300B_ECC88_PSE

The power supply

HV_PSUThe B+ power supply design is, in principle, nothing fancy. In contrast to “traditional” designs, however, it uses two cascaded LC filters with massive inductors (L) and capacitors (C) to produce an extremely clean the B+ voltage. The energy stored in the last capacitor stage alone is enough to play music for almost half a minute without getting power from the mains (or to lift an average person by 1 metre into the air)! The rectifier is solid state (i.e. not a vacuum tube, which are not capable to deliver enough current for this amp). The rectifier consists of four high-speed and soft-switching diodes to avoid the negative sonic effects of “normal” solid-state rectifiers.

For the filaments, regulated DC supplies are used.  The DC voltage ramps up slowly after turning on the power switch. This increases tube life by avoiding excessive currents through the filaments when they are still cold.heater_PSU_slow_turn_on

Tube rolling

Driver stage

Following Andrea Ciuffoli’s design, I started with a 5842 (417A) tube in the input/driver stage. To make a long story short: this tube sounded thin and harsh in this design. I don’t think that this is a problem with my specific design, as it is very similar to that by Andrea. Maybe Andrea has a different taste concerning the sound of his designs. So, I was looking for another input triode which provides the following characteristics: A relatively high cathode-current rating. A cathode current of Ik=20 mA running through the tube is needed to make sure the grid current for the driver tubes does not affect the performance of the driver stage (see also the VT52.com site for an explanation of this issue; and don’t forget we have an additional transformer between the driver and the output tube, which further increases the minmum value of Ik by a factor of 2.25). A relatively high gain to assure a good input sensitivity. The gain factor of the 5845 / 417A is µ=43, which, in combination with Ik=20 mA, is tough to beat. I had some beautiful ECC33 tubes roaming around my table when I experimented with the input tube. The gain of this tube is µ=33 and it can easily handle the current, even more so if the two triode systems of this double triode are operated in parallel (note that the ECC33 is often but wrongly used as an equivalent to the 6SN7, but the 6SN7 has lower gain, and probably differs in other aspects, too). So I gave the ECC33 tubes a try, and they worked much better than the 5842 / 417A. After a while, however, I got annoyed by the overly sweet and somewhat mellow sound resulting from the ECC33. In the search for another candidate I flipped open my tube handbook. The choice was obvious: the ECC88 (6DJ8) is the way to go! I had some very good experience with this tube in the output stage of my DAC, and if the two triode systems are operated in parallel (again…), the ECC88 can easily drive the 300Bs through the transformer. This choice was further supported by the fact that I have a box full of high-quality, new-old-stock E88CC tubes made by Simens and Philips. I put a pair of the Siemens tubes in, and there they stayed… the sound is more dynamic and open, the bass is much more detailed and has more control than with the ECC33. A further advantage of the ECC88 is that good-quality tubes (both new-old-stock and newly manufactured ones) are easily available at down-to-earth prices, which is not at all the case for the ECC33. Power stage So far, I’ve tried the following power tubes in this amp, with different results (in alphabetical order):

  • AVVT 32B (price: a lot): this tube has the same specs like a standard 300B, but can be biased to deliver more power due to its larger electrodes and stronger filament. The 32B is bigger than all other 300B, and it is built like a tank. The glass envelope feels more like a bottle of beer than a fragile electron tube, the base is made of ceramic and has gold-plated pins. Two of these tubes in parallel easily deliver more than 40 Watts, which is nice (but completely unnecessary, I think). They sound like what you expect from a 300B: neutral and sweet. Despite the fact that the 32Bs feel like a bottle of beer, the sockets are prone to get loose. Also, after about one year of frequent music listening, one of the tubes died. [Note: AVVT has been replaced by Emission Labs]
  • Electro Harmonix 300B gold (US$ 60 each): this tube shows standard 300B specs. The ‘gold’ version of the EHX 300B has a gold grid, which is supposed to improve the sound of the tube. As far as I understand, Electro Harmonix tubes are produced by the same people and machines as Sovtek tubes. As I’ve never seen a Sovtek tube working longer than a few minutes (!) before, I did not expect much of the EHX 300Bs. To my surprise, however, they survived several weeks, before they were replaced with Svetlanas (not because they were failed nor because they were bad in any way – but the Svetlanas just sound better). The EHX 300B are nicely built, with a beautiful white ceramic base. They have a powerful sound, but not as sweet, refined and detailed as the other 300B variants in this round (this is not necessarily bad, as, for instance, the Fullmusic 300Bs are way too sweet for my taste).
  • Fullmusic / TJ 300B mesh plate (US$ 155 each): this tube shows standard 300B specs, but looks a bit differnt with its globe-shaped top. I like the looks, but others don’t. While it is not heavy or especially rugged, it is well built. There has been quite some buzz about the Fullmusic tubes, especially the mesh-plate version of the 300B. When I first listened to these tubes, I was blown away by the ease of how tiny little details in the sound were presented, while the sound stayed extremely sweet and some extra “quiteness” or “coolness” was added. After a while, however, I felt that these tubes lacked the punch to really enjoy the music (bass was deep, but lame) and that they made the music sound too sweet.
  • Svetlana / SED 300B (€ 119 each): a tube with standard 300B specs and no fancy gimmicks (a rare thing these days!). Build quality is ok, although, unlike the other tubes in this list, their sockets are not made of ceramics. The folks at Svetlana don’t waste their money in the advertisement business (I’m sure they could have come up with something that sounds as fancy as extra large electrodes, gold grids, or mesh plates). Similarly to previous experience with Svetlanas KT88, their 300B convinced me with great sound: neutral, detailed, and dynamic. The sound has the same strength and “punch” as the AVVT32B, but is more refined and soundstage imaging is slightly better. Also, unlike the Fullmusic, the Svetlana sounds very neutral.

So, who’s the winner? It depends. The AVVT / Emission Labs tube is expensive, but it’s construction can compete with a bottle of beer, sounds great and has enough power for a trance party. The Electro Harmnoix is the least expensive and sound ok. The Fullmusic is for those who crave for (overly) sweet sound. The Svetlana (my personal favourite) isn’t cheap but in the 300B PSE amp it sounds really good.

My test equipment

Overview

Testing and analysing the audio performance of speakers, amplifiers and other equipment requires different pieces of hardware and software. My main system is centered around a computer equipped a high-quality audio interface and a dedicated audio analysis software, and a measurement microphone for acoustic tests. Of course, I also use an oscilloscope and a multimeter.

audio_testing_computer_setup

Schematic of a typical testing setup (from the MATAA manual). In this example, the device under test (DUT) is a loudspeaker, which is fed with a test signal from a computer equipped with a sound card or external audio interface via a power amplifier (A). The test signal reproduced by the loudspeaker is recorded with a microphone (B) via a suitable suitable pre-amplifier (C) and an anti-aliasing filter (D). The signal is then digitized using the computer hardware, and is then analyzed using the different computer algorithms.

speaker_testing_photo

Acoustic analysis of a loudspeaker using a microphone. The loudspeaker is placed on a chair to increase the distance to the floor, which results in a better separation of the signal reproduced by the speaker and its echo from the floor.

Audio Interface

The audio interface provides the digital/analog and analog/digital converters that link the computer with the test devices. In many cases, the sound card built into most computers is all you need. If you want more, or if you want to protect the computer from funky test equipment frying the motherboard of your computer, you may use an external audio interface. I mostly use an external M-Audio FireWire 410 Audio Interface (FW-410) for my audio measurements. This interface was designed with music production in mind. However, the FW-410 is very well suited for audio testing purposes due to it’s high-precision audio hardware. The following are some useful application notes when using the FW-410 for audio testing (these notes may also be useful with other audio interfaces):

  • Signal input is either microphone at front panel or line level at rear of the FW-410 (there is no line-level input at the front). Choose input using the “Mic/Line” switch on the front panel.
  • Headphone output on the front panel is for monitoring during sound recording. Don’t use this as signal output for audio measurements. Use the line level ports at the rear of the FW-410 instead.
  • Sample rates higher than 96 kHz may not work well with asynchronous sound I/O. Just use 96 kHz or less if possible.
  • The driver software comes with an elaborate software mixer program. As far as I can tell, this is used to route the different audio ports to the headphones / monitoring port on the front panel of the FW-410. For audio measurements, just use the default settings of the software (click «reset» if you screwed up the settings). The software also displays the signal levels at the different sound I/O ports, which is useful with audio measurements.
  • M-Audio tends to be slow with maintaining the driver software. Let’s hope the driver still works with future releases of Mac OS X. Linux support for the FW-410 may be better (I have not tested this yet).

Software

I mostly use MATAA, which I prefer over ARTA, FuzzMeasure Pro, HBX, etc, due to its versatility and functionality.

Microphone

I use a Behringer ECM-8000 measurement microphone. This microphone does not cost an arm and a leg, but it still provides decent measurement performance. Like many other measurement microphones, the ECM-8000 requires a phantom voltage and a microphone pre-amplifier. I use the Stage Line MPA-102 microphone amplifier, or the built-in microphone amplifier of the audio interface.

Mangerbox

mangerbox_3This is my implementation of a loudspeaker with the unique and famous Manger MSW driver. Unlike with conventional loudspeaker drivers with rigid membranes, the Manger MSW is a bending-wave driver. The movement of the voice coil at the center of the highly flexible membrane induces a wave that travels radially towards the edge of the membrane. This is similar to the surface waves that result from throwing a stone into a pond. The Manger MSW has a flat frequency response from 300 Hz and higher, so it’s highly suitable for a very high-quality FAST (full range and subwoofer technology) loudspeaker system. I designed this small floor-standing loudspeaker around a Manger MSW WO-5 and a Visaton TIW 200 XS woofer mounted on the side of the enclosure. Both the MSW and the woofer are driven by a dedicated power amplifier using DSP filters. The crossover-frequency between the MSW and the woofer is about 300 Hz. Detailed notes and measurements taken during the development of the Mangerbox are available in a separate PDF document.mangerbox_1 mangerbox_2 mangerbox_plan

Tube Speaker

tubespeakerThis is a small speaker which I designed as an improved clone of a Bang&Olufsen designer speaker. Unlike the B&O model, however, the Tube Speaker is made of real metal and sounds really good. It uses two 11-cm Seas-Excel mid-woofers (yes, the ones with the magnesium membranes) and a Scan-Speak D2905/930000 soft-dome tweeter mounted in a d’Appolito arrangement. The metallic outside is backed with a 2-cm thick layer of sand-filled epoxy, which results in an extremely «dead» enclosure.

Drivers

Since these speakers are only 15cm wide I had to use very small woofers. The only good woofers in this size (about 12cm max. outer diameter) I could find at the time were the Seas Excel drivers. Their outer diamter is about 11cm and there’s even a model with the famous magnesium-membrane by Seas. Because these woofers are so small I had to use at least two of them in each speaker to get reasonable bass-response – so why not mounting them in d’Appolito style? I never built a d’Appolito speakers before so I decided to try it. Well, it was worth it: the sound qualtiy and the 3D imaging of the Tube-Speakers turned out the be very good!

The magnesium membranes have a very strong resonance peak at at about 11 kHz. This is way higher thant the cross-over frequency, so this resonance is not a problem in this design. The tweeter should be of the same good quality as the mid-woofers. I ended up using the ScanSpeak 9300 tweeter, befcaus it is a very good tweeter and it’s not too difficult to design a suitable cross-over filter network for it.

Enclosure

The design of the metal enclosure is where the Tube Speakers are different from every other speaker I have come across so far. But when you get off the beaten track, things can get rough…

The first major problem was to get a suitable metal tube. I called a metal-working companies, who told me that a suitable tube with cut-outs and a flat face plate to mount the drivers would cost a fortune. That means two fortunes for a stereo pair. Finally, a friend who works as a metal-working teacher took up the project and built the tubes for me. Thank you Erwin!

The second major problem with a metallic enclosure is that it would resonate like a church bell, very bad for a speaker cabinet. My first idea to solve this problem was to fill the metallic tube with concrete from the inside, but I was afraid the concrete wouldn’t stick to the metal. So, what else? Epoxy is really sticky… so I mixed epoxy with as much sand as possible. This gives a very heavy and solid material after the epoxy is hard. I put about 2 cm of my epoxy-sand mixture on the inside of the metal-tube. To do this I put a carton-tube inside the metal-tube. The carton-tube’s outer radius was about 2 cm less than the radius of the outer metal-tube which left the required gap to pour my mixture in. Epoxy has some disadvantages, though. Firstly, it is expensive. And secondly, it gets very hot when hardening and after becoming hard it cools down again which also means it contracts a little. This contraction caused the epoxy-sand-mix to loose contact to the metal-tube so I had to pour some more epoxy into these gaps. The result of my epoxy treatment is a very rigid and perfectly air-tight speaker-cabinet with virtually no vibrational resonances.

Crossover filter network

I used a 3rd order low-pass filter for the woofers (both woofers in parallel) and a 2nd order high-pass filter for the tweeter. Together with the natural low-frequency roll of the tweeter, this results in a 3rd order slopes in the frequency responses of both the woofers and tweeter, as required for a school-book d’Appolito design. The cross over frequency is about 2 kHz. The figure below shows the frequency response (measured with MacSpeaker using an MLS-signal), which is very flat (±1.5 dB between 500–8000 Hz, and ±3 dB throughout the entire frequency range of the measurement).

tuberesponse

Anechoic frequency response of the Tube Speaker

FuzzMeasure – some praise, comments, and rants

FuzzMeasure is an audio measurement software for Mac OS X. The operational princple behind FuzzMeasure is the same as with most other computer based audio analyzers (e.g., MLSSA, CLIO, ARTA, or my own MATAA). FuzzMeasure sends a test signal (sine sweep) to the speaker, records the signal produced by the speaker, and calculates the impulse response by deconvolution of the original and the recorded signals. The impulse response measurements are stored in a project file, and each measurement can be evaluated using different analyses either in the time domain or in the frequency domain (this is very similar to what I did with MacSpeaker in the good old times of Mac OS 7). For instance, FuzzMeasure can filter the impulse response using different window functions, calculates step or frequency response, and produces waterfall plots.

I decided to try FuzzMeasure Pro 3 to see how I get along with it when I did some work on my AOS S24 speakers recently. After playing around with the free demo, I pulled the trigger and payed 120 EUR for the full version.

Here are some random notes from my first days with FuzzMeasure:

  • Using FuzzMeasure is very easy and straightforward most of the time. I didn’t need a manual while getting started with the first few measurements. This is not only because I know how this stuff works from my experience with programming MacSpeaker and MATAA, but also because the user interface is designed very carefully. However, there are a few things that didn’t work, most likely because I couldn’t figure out how to achieve them (see below).
  • FuzzMeasure is very nicely integrated into the whole Mac OS environment. Exporting plots or configuring the audio devices is a piece of cake (except if you’re left with an M-Audio FW410 interface, which starves the microphone with poor phantom voltage and comes with the worst driver software, which tends to crash your computer every few minutes even in the year 2014 – but I can’t blame FuzzMeasure for that).
  • Repeating measurements that need to be evaluated in the same way over and over can be cumbersome. FuzzMeasure just applies it’s (hard-wired?) default time window to every new measurement. When I optimized the crossover filter of my speakers and measured the effects of my modifications, I had to adjust the window manually for every new measurement, which takes a minute or two to get everything right. My life would have been much easier if new measurements would just inherit the window settings from the previous measurement, or if a default window setting could be specified. With MATAA, I’d just put the window function into the script that handles the measurement and the subsequent processing and analysis – no tweaking and fiddling necessary.
  • FuzzMeasure exports frequency domain data to *.frd data files, a file format that is commonly used by other audio analysis tools. Time domain data, however, are exported to *.aiff sound files, which are pretty useless for most audio analysis tools. Why is there no *.tmd export, or even a *.tmd import?
  • The only test signal available with FuzzMeasure is a sine sweep. Sometimes it would be nicer to have other signals, too. I like using noise signals (red, pink or white), maximum length signals (MLS), sine bursts, square or sawtooth waves, etc. Such signals can provide a wealth of information that is hard to get with a sine sweep.
  • How on earth do electrical impedance analyses work? I was able to dig out a very brief description of an impedance jig for impedance measurements. But how do I tell FuzzMeasure Pro 3 to actually do the impedance analysis? This functionality seems to be hidden in a quirky «plugin», but it’s use is a mystery to me.
  • Frequency plots extend to ranges below the minimum frequency equal to the inverse of  length of time window. If an impulse response (or step response) is cropped to length T, the result does not contain any sensible information about frequencies lower than 1/T. Plotting data at lower frequencies is misleading and just wrong.
  • Measurements seem to go very wrong if the sound devices for input and output are set to different sampling rates. I can’t see a need for different rates, but users that are similarly stupid as me will have a hard time to understand what’s wrong. FuzzMeasure should check for equal sample rate settings of the input and output to avoid wrong measurements.
  • Newbies (and everyone else) would probably love to see a list of different audio interfaces, microphones and other hardware with a brief description of their strengths and weaknesses and their suitability for specific applications.
  • I couldn’t find a manual for FuzzMeasure Pro 3! Yes, there is a manual for version 2 floating around the internet in the form of a PDF file. And yes, I was able to get me started without the manual. But no, that’s not enough. There are a few quirky operations that are not obvious to me, as mentioned in the points above. And here are a few more: What are the different curves in the distortion plots? Why can I set colors for some plots, but not for others? Why do the measurement notes sometimes apply to the currently selected measurement, and sometimes they go into the next (new) measurement? FuzzMeasure costs a lot of money, so a manual is mandatory!

Overall, FuzzMeasure is a shiny and convenient software tool for audio engineers and DIY enthusiasts like me. While FuzzMeasure provides a number of elaborate and useful analyses,  its use is mostly straightforward (if you know the theory behind its operation and how to use the software), but does not work as expected in some other cases (even if you know what you’re doing). The functionality of FuzzMeasure Pro is not unlimited, however.

My conclusion is that FuzzMeasure may be SuperMegaUltraGroovy and stuff, but I still prefer MATAA for serious work. At the end of the day, MATAA is a workhorse that does everything I want and need, the way I like it. If I need more, it’s easy to add new functionality due to the GNU Octave / Matlab foundation. MATAA is way more powerful and convenient for me due to the its seemingly endless scripting, data processing, and plotting possibilities.

After my somewhat disappointing conclusion, I asked Chris Liscio for a refund which he allowed with no issues. Nice touch, Chris!

MacSpeaker

MacSpeaker is was a program for audio measurements, above all, loudspeaker measurements. However, I stopped maintaining MacSpeaker a long time ago. I lost all the program-code because of a disk crash (thank you Quantum). MacSpeaker was designed for the ‘Classic’ Mac OS 7–9 and does not work anymore with current Macs. As a replacement for MacSpeaker I recommend MATAA, which is a much more powerful tool than MacSpeaker.

This is how I described MacSpeaker in in the good old times of Classic MacOS 7: MacSpeaker makes use of the Macintosh sound hardware so that you do not need an expensive Analog-Digital conversion card. You only need a measuring microphone and perhaps a microphone amplifier. You will get a professional measuring system at a fraction of the price of comparable systems! Because the Macintosh has excellent sound hardware, the measurement accuracy of MacSpeaker is comparable to professional systems: 44.1 kHz 16-bit sampling; that is, CD quality. MacSpeaker controls all important loudspeaker measurements: From the measurement of the impulse response, a Fourier transform computes the frequency response of the loudspeaker (absolute and phase). Measurement errors due to unavoidable sound reflection from the walls can be filtered out. Thus, with MacSpeaker, you can measure the «frequency response in an anechoic chamber». Additionally, MacSpeaker can compute a so-called «waterfall diagram». This reveals the resonance behavior of the loudspeaker plotting time over frequency and amplitude. As a special feature, MacSpeaker can use an optional user-defined signal as measurement signal. Thus, for example, you can make burst or distortional measurements. With a little bit of imagination, you can fulfill your various desires.

macspeaker_IR_windowing

AOS Studio 24

The AOS Studio 24 BE (S24) is a closed box MTM loudspeaker (Photo 1). The S24 is sold as a kit by Axel Oberhage Starnberg (AOS) in Germany. The original AOS design is a “bookshelf” format box. However, free-standing operation of the S24 on a stand is acoustically better than stowing it away in a bookshelf. The S24 uses high-quality driver units made by ScanSpeak.

AOS_S24

Photo 1: My AOS Studio 24 BE, which I re-desinged as a full-blown floor-standing loudspeaker

A unique feature of the S24 is that the woofers protrude by 18 mm from the front baffle, whereas the tweeter sits flush with the baffle. Most speaker designs lack this offset, which, due to the recessed woofer cone(s), causes a lag of the sound wave(s) emitted by the woofers relative to that of the tweeter. The additional offset used in the S24 therefore avoids the typical misalignment of the different drivers and therefore allows time-coherent sound emission from the woofers and the tweeter.

AOS_S24_TWEETER_ALURINGS

Photo 2: Close up of the tweeter and the midwoofers, which are mounted on aluminum rings to achieve an offset relative to the front baffle

I decided to modify the original AOS design a little when I built my S24. In particular, I used a full-height floor-standing enclosure for the S24, revised the crossover filters a bit, and added an active subwoofer. A detailed description of my tests, modifications, and extensions to the S24 are available as a separate PDF document.