Thanks James. All I can claim on the cabs (besides the design and drawings) is the finishing, including hand rounding of edges, sanding, Danish oiling (4 coats) followed by 5 coats of Mylars carnuaba wax. About 3 days off and on went into the finishing. The internals needed to be precisely cut and jointed so I entrusted a local CNC machine shop to do the work only to find the finished internal panels required hand finishing to get them mm perfect anyway. It was like a 3-D wooden puzzle to assemble!
Further testing and evaluation has resulted in several more additions and fine tuning to the circuit which is now contained on two separate boards to keep inductor spacing at controlled distances.
Continuing with the explanation of design, there is one further surprise in store which many don't consider. Look at the graph below:
...this is the frequency response of the mid-woofer in a 21 litre closed box.
Look what is happening just above 3KHz. Theres a (roughly) 2.5dB rise to a peak before the woofer trails off. This is the upper resonance point of the driver cone and represents a point of high distortion and cone break up. It is far more benign and less pronounced than most (some have up to a 20dB rise at this point!).
Whilst the crossover point is quite a bit below this, we can see two things: Firstly average sensitivity before the filter is added is about 90dB/1w/1m. Secondly, for lower order filter slopes, the summation of this peak in output will be additive to the tweeter response. The two together, close to the crossover point should result (in an ideal world) to a flat response across the crossover. This peak will do two things: it will increase output at roughly a 3100 Hz peak and it will introduce audible distortion at and just either side of that peak, so how to cure this?
Well, what was added was another resonance filter, or notch filter as it's more often known. This is made up of a capacitor, an inductor and a resistor placed between positive and negative rails (ie out of the signal path). It works by introducing shunt resistance across a narrow audio band, in this case, 2.9KHz to 3.3KHz with a dip in woofer impedance to half normal value, which in this case, squashes the resonant peak and restores the flat response above crossover. More importantly, it lowers audible distortion to a minimum without affecting overall system impedance (ie no greater amp load).
Whilst you can choose drivers which are run full range, only a tiny minority of them can do so without audible distortion at cone break-up; that's just laws of physics, so advice is to avoid such designs.
What hasn't yet been addressed is phase response. Most designers will put phase coherence at crossover as their top priority but I believe this to be a mistake. Getting audibly low driver distortion is by far and away the top goal. No-one is golden eared enough to identify something like 10 degrees of phase difference, although 90 degrees can be audible and (for example) is often treated in 1st/3rd order 2 way networks by time aligning the tweeter 10 to 12mm different to the woofer or by introducing time delay circuits (as with modern Tannoys). However, 10 degrees is largely inaudible, so the advice for good integration is to get drivers operating within their lowest distortion envelopes and to integrate sensitivity accurately plus choose drivers where no large dispersion steps take place off axis at crossover.
In summing up, a well designed crossover for a two way should contain the following elements:
Zobel to flatten woofer impedace;
Notch filter to flatten Woofer upper resonant peak;
Notch filter on tweeter to flatten lower resonance point (hence lower distortion);
use of air cored inductors for low distortion for everything below say 1mH value;
Over-specified steel laminate low DCR cored inductors for higher inductor values (reduces cross-coupling and cost);
Air cored inductors for all low value inductors, and spacing with inductor windings at 90 degree angle to each other;
High quality capacitors in the signal path;
Minimum 2nd order or higher filter slopes (anyone providing lower than this IS cutting corners and it can be a tell tale sign of someone who does not understand how to properly undertake crossover design or a failure to understand driver response). You would be surprised at the number of commerical designs using 1st order slopes. It isn't acceptable. Consider everything above and ask yourself what the problems with such a set up might be? I know of at least one manufacturer (a large British based one at that) who uses the SAME simplified 3 components crossover in ALL its speakers. Totally unacceptable.
Phase response well within 10 degrees and preferably spot on at crossover point. Its impossible to have phase flat all they way over the audio band and it can vary by up to 60 degrees depending on drivers/design chosen.
Selection of drivers with wide dispersion pattern to a wide degree off axis (INCLUDING the woofer: look at the chart above...30 degrees off axis performance matches that of many tweeters).
The ability to measure and understand what it is you are measuring (its a dirty word to some but unavoidable when designing speakers. Anyone who doesn't use measurements when designing speakers isn't designing them; they're chucking drivers into a box and hoping to get lucky).
In summing up, a well designed crossover for a two way should contain the following elements:
Zobel to flatten woofer impedace;
Notch filter to flatten Woofer upper resonant peak;
Notch filter on tweeter to flatten lower resonance point (hence lower distortion);
use of air cored inductors for low distortion for everything below say 1mH value;
Over-specified steel laminate low DCR cored inductors for higher inductor values (reduces cross-coupling and cost);
Air cored inductors for all low value inductors, and spacing with inductor windings at 90 degree angle to each other;
High quality capacitors in the signal path;
Minimum 2nd order or higher filter slopes (anyone providing lower than this IS cutting corners and it can be a tell tale sign of someone who does not understand how to properly undertake crossover design or a failure to understand driver response). You would be surprised at the number of commerical designs using 1st order slopes. It isn't acceptable. Consider everything above and ask yourself what the problems with such a set up might be? I know of at least one manufacturer (a large British based one at that) who uses the SAME simplified 3 components crossover in ALL its speakers. Totally unacceptable.
Phase response well within 10 degrees and preferably spot on at crossover point. Its impossible to have phase flat all they way over the audio band and it can vary by up to 60 degrees depending on drivers/design chosen.
Selection of drivers with wide dispersion pattern to a wide degree off axis (INCLUDING the woofer: look at the chart above...30 degrees off axis performance matches that of many tweeters).
The ability to measure and understand what it is you are measuring (its a dirty word to some but unavoidable when designing speakers. Anyone who doesn't use measurements when designing speakers isn't designing them; they're chucking drivers into a box and hoping to get lucky).
Pitifully simple when you put it like that Paul. It's remarkable that everyone isn't making speakers as incredible as yours... ;-)
Thanks Ben. I think that the better manufacturers do take this trouble but you'd be amazed at just how many don't! The stock excuses are "we keep things simple for phase accuracy" or "driver integration" , Truth is that most drive units are not a perfect match...very far from it, so for accuracy, some degree of crossover matching is almost always required and at least 2nd order filter slopes are essential for almost all drivers (some actually require 3rd or 4th order slopes). Point is, all drivers have an operating envelope within which they perform optimally in terms of low distortion. Push them a little above that envelope and the results can be pretty horrible or at least, unacceptable. There are a few exceptions where specially developed drivers are used but this really is the exception as very few "off the shelf" units have the required characteristics.
Thanks Ben. I think that the better manufacturers do take this trouble but you'd be amazed at just how many don't! The stock excuses are "we keep things simple for phase accuracy" or "driver integration" , Truth is that most drive units are not a perfect match...very far from it, so for accuracy, some degree of crossover matching is almost always required and at least 2nd order filter slopes are essential for almost all drivers (some actually require 3rd or 4th order slopes). Point is, all drivers have an operating envelope within which they perform optimally in terms of low distortion. Push them a little above that envelope and the results can be pretty horrible or at least, unacceptable. There are a few exceptions where specially developed drivers are used but this really is the exception as very few "off the shelf" units have the required characteristics.
I guess that's why Joe Ackroyd spent his time doping drive units.
Many others have taken that approach too David as damping the unit helps the damp the resonance at cone break-up frequencies but don't forget that the driver characteristics if run this way are far removed from linear or ideal. Unless you can dope the cone such that you get a known roll-off (dB/Octave) then you can't possibly integrate a tweeter properly and cannot hope to achieve anything like a flat response running the driver full range unless its natural roll off matches the crossover slope needed. The beneficial effects of doping the driver cone are to damp the resonances so that when a filter IS applied to the woofer, no notch filters are needed as it approaches resonance, so a more benign resonance point occurs which helps negate the need for things like notch filter complexity/cost. (ie its cheaper to damp the cone than to build in a notch filter).
Modern materials such as the Harbeth RADIAL drivers, SEAS Polycurve woven Polypropylene and a few others have by design a very smooth roll off, yet still benefit from a notch filter for lowest possible distortion.
The hard truth is that the claims of running drivers full range and doping to smooth them off has some benefits and some merit but a proper crossover design more often than not results in lower distortion. You have to take care with phase relationships between drivers and in some configuration, time dealy circuits are sometimes employed, but my take on this is that you are often better off with an active design (more economical and more efficient) than pursuing very complex 4th order (or higher) passive crossovers. There is no excuse really for not employing 2nd or 3rd order designs for a vast majority of speakers though imho.
Thought I'd resurrect the thread since things have moved on apace since the start of the Rhapsody project.
The use of calibrated acoustic mics and other test gear (including the Mk1 ear-'oles) have enabled significant refinement and improvement of the design.
Now building a pair in a Walnut finish for a customer and have the crossovers committed to custom designed PCB boards (which use 2Oz copper tracks and are 2.4mm thick, so fairly heavy duty).
Response is now pretty smooth and the following is the latest frequency response measurement taken using Pink Noise using impulse measurements at 2m on-axis and applying 1/3 Octave smoothing:
This is both channels showing how closely matched the speakers are. the slight ripple lower down is due to the PCBs not being fixed into their final resting places (and covered with acoustic foam) resulting in some fairly minor ripple reflections:
Veneering is something which fascinates me, I really must have a go one day. Are special tools needed?
It depends how you go about it Alan. For these, I'll need to bookmatch each front and rear panel due to width which involves cutting perfectly straight bookmatched lines along the veneer. I built a tool to help years back which was like a giant paper cutter on a frame but that's long gone so these days I use a 3 ft steel rule and a scalpel plus a steady hand!
For the actual veneering, I use a specific PVA grade (not obtainable at your local DIY store) which allows 20 minutes or for positioning and a softwood block with rounded edges for smoothing the veneer out, starting from the middle of each panel and working outwards to expell any trapped air pockets.
Finally, a press of some sort is used until the glue has set properly (24 hours). I used to use a ply sheet and sash clamps but it's a right faff so now I use vacuum bags (the exact same ones obtainable from Ebay for vacuum packing clothes) and a vacuum cleaner to suck the air out. This holds the veneer firmly in place for the gluing stage.
The finishing stage is very delicate. Excess has to be carefully trimmed, lightly sanded and bearing in mind that each veneer is just 0.6mm thick, you cant use anything other than a fine grade sandpaper and lightly sand to a smooth finish. Following that, I use either a water based (powder form) lacquer and build up the coats, sanding between each (minimum 3 coats) or Danish oil, taking care to apply only a thin initial sealing coat as the volatiles can otherwise make the veneer come unstuck! Once the first coat is on, I build up to around 3 coats before the first light sanding with a very fine grade sandpaper or wire wool, then apply another coat and repeat the process for a further 3 coats minimum. I use a satin finish for this. It takes ages doing this step but it's the part I enjoy most as it brings the grain alive and puts the stamp of the finisher ont he final product in terms of the care and attention lavished on a valuable timber. I think it's important as it shows respect for the timber to take great care and finish to the best possible standards.
Comments
;-)
:!!
Sound pretty good too!