There’s a very interesting and comprehensive paper by KEF on the design of their Reference Series speakers. It shows, I suppose, that there is a huge difference between ’boutique’ manufacturers and a world class company that can pour huge resources into optimising every aspect of a design. I can’t say I have heard these speakers, but I would be expecting something pretty impressive.
I like the idea of the ‘tangerine waveguide’ which seems to be an object whose shape was possibly ‘evolved’ in a computer using finite element analysis (FEA). This could be an example of the power of ‘dumb’ software alone to trump all human expertise. By that I mean that once a computer can be programmed to simulate a physical system effectively from first principles, it can be given the task of finding an optimal solution to a particular problem – if such a solution can be found at all. The process can be speeded up using human intuition to define the starting conditions perhaps, but the computer would eventually find the answer anyway.
In this case, the human input might involve the desire to modify the tweeter’s dispersion so that it blends perfectly with the surrounding mid-range driver, with an expectation that a small object placed in front of the tweeter might do it, but with no idea of what the shape of that object might be (that would be my situation anyway!). I imagine that a virtual model of the tweeter dome and mid-range driver cone could be provided, along with some constraints on the shape of the waveguide – it has to be manufacturable – and that the computer could start simulating the system’s performance in terms of on- and off-axis frequency response, slowly homing in on better and better solutions using genetic algorithms, perhaps. The waveguide takes shape as a complex 3D object that in the good old days of audiophile yore it might not have been possible to make, but can now be manufactured using CAM machines. At no point in this process would a human even need to understand the mathematics of the shape, but they could verify that the solution is robust by checking the virtual performance with extremes of manufacturing tolerances, surface finishes and so on. The presence of the waveguide will influence the design of the crossover, and again the computer could be programmed to come up with the optimal solution for that, also.
It is a heady brew, I think: the person who can program a computer to simulate systems from first principles has at his fingertips the power to exceed all previous human design expertise. This expertise was, in reality, just rules of thumb and short cuts, and engineers of all kinds have always engaged routinely in trial-and-error optimisation of systems; they may ‘understand’ the mathematics behind the system they are designing, but ultimately they optimise using trial-and-error anyway.
There were a couple of sections within the KEF paper that I looked forward to reading, anticipating that the engineers might have some difficulty sounding enthusiastic:
(a) the whole notion of passive crossovers vs. active. Given the KEF engineers’ apparent freedom to build an optimal speaker from the ground up, they are still constrained commercially by the convention of having to use passive crossover filters. And their arguments against linear phase filters (which would solve a lot of their problems and make more of a case for active filters) seem half-hearted.
Ultimately, the way I see it they have a space age speaker that is limited by steam-age technology. You may disagree.
(b) they admit that bass reflex (ported) speakers do not behave well in terms of time domain response and other difficulties, suggesting that 100l enclosures (about the size of my woofer enclosures) would be a ridiculous size, but that they would perform well. A more “reasonable” size of 15l requires the use of a port. The analysis is very detailed, but presumably this does not actually make the problem go away! Ditto giving the user the option of whether to use a full-blown port or an interchangeable reduced-strength version – in fact, on the larger models there are two ports per speaker, each of which can undergo this change, so there are several permutations for the user to try. Surely this means that one of them will be optimal..? I would, ultimately, say no – but you may think that my holding such a critical opinion of an aspect of this engineering tour de force is laughable, of course…