Initial Listening and Gradual Refinement
Having already written the basic crossover software, and after a day's furious woodworking and wiring up, the system worked! I set the relative driver levels using the amp volume controls, doing this by ear, matching the levels at the crossover frequencies. So at this stage I could listen to CDs or any audio stream through my own active speakers, with the prospect of refining the crossovers to any level I liked.
Initial impressions were of an incredibly open, rich, precise, powerful sound with pinpoint imaging, profound bass, and of a system very tolerant of the exact choice of crossover frequencies and slopes – I couldn't hear any differences over quite a wide range. Gradually I began to realise that it wasn't quite perfect, though: too much emphasis on the upper mids and treble, and no amount of volume control adjustment would make it right. Over the next few weeks I gradually refined the software to provide the following:
Baffle step correction. Separate adjustable curves for each of the drivers. I figured this was necessary because of the different baffle widths of the enclosures. This was the silver bullet that pulled the system together into a coherent whole.
Humans and phase shifts
Driver correction derived from nearfield measurements of the drivers' impulse responses, using the free application Room EQ Wizard and a measurement microphone. Humans are reputedly unable to detect phase shifts, but without phase correction we will get a frequency response suckout at each crossover (which may be reduced in width by a sharper crossover slope). Without phase correction, to my ears the sound 'hardens' and takes on a subtle edge.
Adjustable delay for each driver, compensates for differing acoustic centre depths of the three drivers and for vertical listener position. Without this we will have deeper interference effects between the drivers and cannot get the fabled perfect impulse response.
My hearing acuity for all adjustments increased. I eventually settled on my ideal crossover frequencies and slopes, smoothed phase correction of the drivers, driver delays derived from calculation, and moderate baffle step correction with corner frequencies derived by calculation from baffle widths. I didn't attempt room correction.
A painful experience resulted from the need to upgrade from XP to Windows 7 (Microsoft are stopping support of XP which threatens virus Armageddon for that operating system, apparently). The new X-Fi sound card drivers no longer provided the facility to use the card as the audio source, and there was little I could do about it. The solution was to use two sound cards, synchronised via SPDIF, so I added a second sound card merely as the destination for audio apps that my crossover software could pick up the audio from. This was a reminder that in the world of PC-based audio, nothing is permanent, and we only get these incredibly cheap-but-high-performance systems on licence from huge corporations, and they can make our systems unusable on a whim. Maybe Linux next time...
I replaced the tweeter protection capacitors with 'audiophile grade' 80uF film caps – which are physically huge!