I might as well jump into the deep end, and pick up a conversation I had with John a few months ago, when I was saying that bi- and multi-amped systems were not easy to pull off, with overall system integration being a big problem.

I was thinking about the known weak points of the Ariel - with only a pair of 5.5″ drivers, it’s not really strong in the 500 Hz and below region (due to cone excursion). Nothing wrong with 500 Hz and above - indeed, that’s the Ariel’s strong area, with very clean impulse response and low stored energy.

The classic solution is to use a big driver - 12 or 15-inch - and cross over somewhere between 200 and 500 Hz. This relieves the mid drivers of needing to move any distance, dramatically reducing IM distortion, and lets the big driver do the real air moving.

All very good - in principle. Just select a nice low distortion prosound driver, do a serious job on the box construction (not trivial for a large box) or better, try a dipole construction, which should get down to 60 Hz with too much of a struggle.

But now we have a problem. How do we design the crossover, specifically the high-pass section for the mid drivers? What’s awkward about doing this passively is that large capacitors are required, in the 40 to 100uF range. This is not good. The bigger capacitors are, the worse they sound. You are not going to find a Teflon in these sizes, although I guess some old oil caps could be scrounged up. These large caps - and inductors - are reserved for the mid driver, with the tweeter having its own independent crossover, which fortunately can use much smaller (and higher-quality) parts.

This is where the active-vs-passive debate enters the picture, when the passive parts available aren’t particularly good. One point in favor of the active approach is that a crossover this low isn’t going to need any exotic equalization, which isn’t fun to do actively. Both the big woofer and the 5.5″ midbass drivers are in the piston band, which means they are intrinsically flat. This is to be expected; the big woofer is at least an octave or two away from HF breakup, and the 5.5″ midbass drivers are several octaves away from their 60Hz free-air resonance. This makes the crossover easy - you can use either a 6dB/octave (1st-order) or a low-Q 12dB/octave crossover, without any requirement for equalization. (Not true for a high-frequency crossover, where equalization is almost always needed.)

All well and good. So let’s consider an active approach, omitting the noxious large capacitor for the midrange driver. The quality amplifier can be reserved for the mid and tweeter, and a big-bruiser solid-state amp used for the prosound woofer. I’d pick a UcD or Class T amp over a generic Class AB amplifier; the new amps are less sensitive to load variation than the Class AB amplifiers, and as Gary Pimm has found, have quite low upper-harmonic distortion once you get below 500 Hz.

This is starting to look good; we’ve gotten rid of some nasty parts, each amplifier is working in the frequency range where it is happiest, and there are nice overload benefits too, since amp clipping in one frequency range has no effect on the other amp.

But how to do the crossover? Don’t buy an off-the-shelf crossover - to be real blunt, these things are junk, even at very high price points. You certainly don’t want cheesy 25-year-old opamps (5532 family) that run in Class AB (almost all opamps do, new or old).

You also don’t want an incompetently engineered vacuum-tube crossover that runs 12AX7’s at miniscule currents and into plate or cathode loads that cause gross distortion. Unfortunately, I’ve described just about all the crossovers you can buy in the commercial market. (Yo, John, you hearing this? There’s a gap in the market!)

The all-transformer coupled amps I’ve been designing don’t lend themselves to cap-coupling, making it a bit awkward to use the power amps themselves as part of the crossover. Yes, there is the parallel-feed transformer-coupled approach, but one of the dirty little secrets of parafeed is if you make the cap too small, you get a nasty - and quite large - LF resonance. In the real world, parafeed caps are typically 2 to 5uF, not the little bitty 0.047uF Teflon we’d like to use for the highpass part of the circuit.

So a little bit of cleverness is called for here. Basically, you want to throw away the linestage you have now, and replace it with a linestage/crossover that has enough moxie to drive 5 to 10 meters of cable (no wussy 12AX7’s in the output stage, thank you).

You see why I said there are deep waters here? We started by trying to get away from those crummy 40 to 100uF caps, and now we have to design a whole damn linestage! Nothing simple, is there?

Still, it can be done. Use low-distortion tubes that have enough standing current to drive the capacitive (or inductive) elements of the circuit, and use other tubes that aren’t upset at driving 300 to 1000pF of cable capacitance. This combines elements of a phono preamp and linestage, although we don’t have to fight noise issues, thank goodness.

We can also cheat a little and use high-quality opamps with low DC offsets for the low-frequency part of the circuit, which is also where the fancy room equalization can be placed (shelf filters and parametric equalizers). In the high-frequency part of the circuit (above 500Hz) room equalization is inadvisable anyway, and if we’re designing a speaker, we’re smart enough to pick drivers that don’t need a lot of tricky equalization.

So in a round-robin way I’ve shown how the overall system interacts. A decision to avoid low-quality parts in the low-to-mid crossover has the consequent effect of requiring a new linestage/crossover element - a device that doesn’t really exist in the current market, no less.

I should also mention that I’ve been discussing a relatively easy (!) low-to-mid crossover between two drivers that are flat and don’t need complex equalization (room EQ is a separate issue). If we’re crossing between a mid/high horn and a big bad 12 to 15-inch driver, all bets are off. This is a completely different story.

Now everything is much harder. Why? We don’t have the luxury of generous overlap regions any more, like we did crossing between big and little cone drivers.

With the horn, the impedance curve tells the story. You don’t want to cross anywhere near the huge impedance peak at the bottom of the working range - the horn is trying to tell you distortion is very high in this region, and below that peak, output is falling like a stone and the output is almost entirely distortion. You do not want to put any power in this region at all.

So what we thought was a so-called 300 Hz horn turns out to be more like a 800 Hz horn. Ugh. You see, it’s only really a 300 Hz horn if we treat the way pros do - use a 24dB/octave (4th-order) crossover and equalize the hell out of it. That’s standard professional practice - even for one-night stands, the sound man shoots the system with a MLS stimulus, and uses automagic digital equalization to make it all nice and flat. Digital crossovers with brickwall slopes take care of all the power-handling issues, which is a big deal with a gig where down-time costs big money (ticket refunds, angry musicians, etc.)

But we don’t want to take this approach, do we? We’re talking either about a zillion op-amps (cheap ones, too) or just shoving all the problems into the digital domain, and hoping the vendor’s algorithms don’t suffer too badly from truncation and rounding-off errors. (You thought digital-domain EQ was distortionless? Uh, not quite.)

So returning back to the world of quality sound, we have some difficult choices with a 2 or more way horn system. There is the problem of extensive edge-of-band equalization, in order to approximate the smooth crossover slopes we want - any little bumps and ripples in the transition-between-drivers region automatically results in violent shifts in the polar pattern, a major source of listening fatigue. I know it’s traditional - going back to the Fifties - to just blow it off and use casual, out-of-the-textbook 1st or 2nd filters, but folks, that’s a recipe for crap sound.

Sometimes tradition is just that, crap. Not everything back then was so great - and filter design was an area where tremendous progress was made during the late Fifties going through to the early Seventies. Please, stop copying all of these stale old crossover designs from the bad old days. Filter design for a speaker is every bit is critical as it is for a Stereo FM tuner (think Marantz 10B here) or a NTSC or PAL color TV. You wouldn’t build a HDTV around mid-Fifties filter theory, would you? Well, it’s the same for crossover design.

This is an area where things have to measured using modern MLS test systems, and the physical realization is probably going to be fairly complex - whether it is done at the speaker level, with line-level equalization, digitally, or a combination. The woofer, at least, is going to be fairly well-behaved, since it is more or less in the piston band. It’s the horn that’s going to need serious equalization.

Since polar patterns of horns have a lot a rough spots, especially at the lower end of the working frequency range, you’re going to need to measure not just on-axis, but at 15 degrees, 30 degrees, 45 degrees, and 60 degrees off-axis, vertically and horizontally, and come up with composite curve.

I know that most horn designers blow this off - it’s one of the less charming parts of the horn tradition. Everybody loves to talk about horn profiles and exotic phase plugs, and nobody gives a damn about crossover design, rehashing tired old crossovers that are at least 30 years out of date. Seriously, Laurie Fincham of KEF came up with Target Function Design back in 1974, and horn designers have been blissfully ignorant of the progress made in direct-radiatior speaker designs. There’s no excuse for this - horns are harder to design and integrate than direct-radiators, not easier, so it’s about time the horn guys got with the program.

Yes, you have to use a computer.

Yes, you have to get a MLS-capable measuring system, and know how to use it.

But this is what it takes if you want a horn system that is anything close to flat, and has well-defined crossover slopes.

The pros, who are by far the majority of the horn users out there, just take the easy approach, slap on a digital 24 dB/octave crossover (or one with even higher slopes), ignore the rapid phase deviations in the crossover region, and equalize the overall result. That’s what you hear when you go to any movie theatre, or any modern concert. You can be sure it measures flat - it just doesn’t sound like it, though, does it?

That’s where the more sophisticated approach of equalizing each driver to the ideal response pays off. System integration is much better, since the phase relationships between the drivers are much better defined, and the audiophile knows where to EQ the room (500 Hz and below) and where to stop, and just accept the drivers as-is.

So what would a modern high-quality horn system look like, aside from the pretty wood horns and the exotic diaphragms? That’s a good question!

I would guess we’re talking about a quite complex crossover either at the speaker-driver level or an equally complex active-crossover/linestage. That offhand comment I threw out for John Atwood earlier would have to be revised to accomodate a set of plug-in EQ/crossover boards - tubes do the work of amplifying and buffering, and the plug-in board accomodates the required capacitors, inductors (if necessary), and resistors.

The alternative is the all-digital route, which pushes the quality problem onto the ADC/DAC conversion section, where again we want to avoid the usual crap op-amps and 100uF electrolyic coupling caps (almost invariably seen in pro-grade digital equalizer/crossovers). So just apply the usual analog modification techniques that you’d use in a cheapie CD player to the prosound digital gizmo - but be sure to apply it the ADC (input) end as well, where parts quality can be pretty dreadful.