Originally Posted by BDP24
Originally Posted by DonH50
That depends on the crossover slopes in the AVR and the speaker. There are a couple of things going on in your scenario:
1. You want the crossovers to be set so there is no peak or valley in the frequency response through the crossover region. Some crossover designs (e.g. Linkwitz-Riley) do that automatically, or try to, but of course there's always some outlier to muck things up. If you are not using an active crossover, or using different HPF and LPF slopes, life can get complicated.
2. Where the speaker is 6 dB down is probably well below where you should crossover. Speaker manufacturers do not always spec the -6 dB point (and sometimes not even the -3 dB point), and rolloff is generally unknown to the user, so it takes additional knowledge, measurements, and/or trial and error to get a smooth FR. The other main reason I would say to use a high crossover point is because speakers rarely perform well at frequency extremes. Distortion rises, and particularly for low frequencies you give up a lot of dynamic range/headroom by crossing over low since signals down there tend to be very large. My rule of thumb has been from 1/2 to 1 octave (factor of about 1.5 - 2) above the -3 dB speaker roll-off is a reasonable place to start.
Right, both of you have stated what I understood to be the way to cross-over. The 6 dB down I got recently from Danny Richie, Brian's collaborator at GR Research. It was in regards to the Rythmik and GR subs with one of Danny's Open Baffle speakers. Maybe he was talking about that specific pairing, not in general. I do know that Danny prefers and recommends lower cross-over frequencies.
To really explain this, we need to examine Linkwtiz-Riely a bit more. Before Linkwitz published their paper, there were disagreement in engineering community in terms of what is the best crossover design. On one extreme, we have companies like Thiel saying their first order filter design (with the most gradual filtering slope) is best, and on the other hand, there were companies selling speakers with "infinite slope" crossover using filter with extremely sharp slope (also very complicated design). And there were others with anything in between these two extremes. The designers can pick any slope (determined by the order) and different family of filters (Butterworth, Bessel, Chebyschev,..etc), symmetric (both sides of crossover use same filtering slope) vs asymmetric. Which one is best? Everyone has his opinion (just like what we debate here in terms of which sub is best). Then came along the Linkwitz-Riely that just shut everyone up. Finally, we have a clear crossover design that achieves uniform frequency response (mathematically proven the amplitude response is constant 1) and both sides of the crossover maintain "in-phase" (360 degrees or multiples of 360 degrees phase difference) between two sides of the crossover at all frequencies. Most importantly, there is only one class of filters that can do that: a cascade of two Butterworth filters. And we know the crossover point response of a standard Butterworth is -3db. A cascade of two Butterworth, as you can imagine, is -6db at the same crossover point. -6db is 1/2 in magnitude (1/4 in energy). And two 1/2 response give us a 1 in response (so superimposition in speakers is on magnitude, not energy or the premise of L-R filter design is wrong). So in short, all "in-phase" crossover design will see -6db at the crossover point. The contribution of R-L is significant. It is based on elegent math with the assumption that both sides of crossover have ideal flat frequency response. But in reality no speakers are like that. For instance, a standard 4th order R-L requires both sides of crossover to use 4th order filters. But typical AVR bass management uses 2nd order on HPF (even though they advertise they use R-L filter) because they assume the front speakers has 2nd order natural roll-off with any filtering.
Now after R-L paper, we now understand the deficiency of 1st order design: the two sides of crossover are not in-phase. As a matter of fact, the phase shift is a constant 90 degrees everywhere which creates a non-symmetric dispersion pattern.
The magnitude superimposition rule I mention above is like adding two vectors on a piece of paper. If the two vectors are in the same direction (which means 0 degree phase difference), the net vector is two magnitude (or length) added together. If the two vectors are in the opposite direction (which means 180 degrees phase difference), the net is the difference of the vector length. If two vectors with identical length, but with 90 degrees in between, then the net is square root of 2 times the length. Square root of 2 is 1.44 and that is 3db. So the -3db at crossover point is good when the two sides of crossover have 90 degrees phase difference. In real world, you can have other factors affecting the actual phase difference, such as unit to unit variation, or different angle to the speakers (like dispersion pattern). It takes only 90 degrees for this design to go to out-of-phase. That is in the phase (difference) increasing direction. On the opposite phase changing direction, it takes 90 degrees for the two to become completely in-phase. That means the new magnitude is +3db higher than what we intend the design to be. In this context, you can see L-R also suffers similar problem, that is, the composite result will deviate from its goal when the phase difference is no longer in-phase. But the advantage is it will take 180 degrees phase deviation to make two sides of crossover become out-of-phase in both the positive and the negative phase deviation directions, making it simply a more robust design.
-Edited by Rythmik - 11/1/13 at 10:27am