Crossover
I'll explain it with a simpler example for splitting in 2 bands, it can be applied for 3, or 4...etc.
basically you need a filter with no resonance, I prefer to use a steep highpass, 24db minimum.
Ok, now you send the original sound to the highpass, so you will obtain the high portion of the sound.
To get the remaining part, you invert this high part and you sum it with the original signal. The inverted sound will cancel that portion in the original sound.
If you use 2 filters you can obtain 3 bands...
basically you need a filter with no resonance, I prefer to use a steep highpass, 24db minimum.
Ok, now you send the original sound to the highpass, so you will obtain the high portion of the sound.
To get the remaining part, you invert this high part and you sum it with the original signal. The inverted sound will cancel that portion in the original sound.
If you use 2 filters you can obtain 3 bands...
Joris, almost contemporary post!
I made a 96dB/octave x-over this way:
dry signal -> 48db high pass - > 48dB high pass -> this is the treble
dry signal -> 48dB low pass -> 48dB low pass -> this is the bass
After that, you can just split one wet output once more.
If I only use one 48dB filter instead of 2 in serial, the summed signal will not be frequency linear.
dry signal -> 48db high pass - > 48dB high pass -> this is the treble
dry signal -> 48dB low pass -> 48dB low pass -> this is the bass
After that, you can just split one wet output once more.
If I only use one 48dB filter instead of 2 in serial, the summed signal will not be frequency linear.
It's better but not linear too. The inversion technique is the way to make sure that the little quantity of shared freqs between the bands (96db is very steep but not vertical) when summed, match the original amplitude. This way also softer filter curves can be used correctly.On 2004-12-20 11:15, Immanuel wrote:
If I only use one 48dB filter instead of 2 in serial, the summed signal will not be frequency linear.
Alfonso
I chalenge you to make an x-over as I described and find just one frequency which doesn't come out at the same level, when the bits are summed. I really looked hard, but I found nowhere in the frequency range, where the summed output differed from the input. I used sine waves from the controll room device for testtones.
It works
Haven't tested for phase issues though.
I chalenge you to make an x-over as I described and find just one frequency which doesn't come out at the same level, when the bits are summed. I really looked hard, but I found nowhere in the frequency range, where the summed output differed from the input. I used sine waves from the controll room device for testtones.
It works
Haven't tested for phase issues though.
Well, I agree that the result is very little, but theoretically it's clear enough. If you have a Xfrequency shared between two opposite filters, there is a small area in wich the two slopes overlap. That area, no matter how small, is in fact doubled. Regarding the phasing issues, if the two filters operate with a different internal reaction time (very probable thing if they are 2 different algos, LPF and HPF) and maybe on different dsp's you will have rhat portion of the shared spectrum not only doubled, but phasing....
You are right, with 96db filters the negative effect is vey small, but it can't be considered absent.
You are right, with 96db filters the negative effect is vey small, but it can't be considered absent.
Maybe I didn't explain it well in the first place. I use 4 filters for one mono 2-way x-over. All filters are set to the same frequency. My guess is, that you can do the same with other filters than the 48dB/octave filter, I use.
If the x-over is used for speakers, then it really doesn't matter (except for a few rare speaker designs), if the trebble is a bit ahead or behind the bass. The treble will almost always hit you first in a traditional speaker design with straight horizontal front. Some horn loaded speakers may be longer than the bass units (?). In that case the bass will hit you first.
If the x-over is used for speakers, then it really doesn't matter (except for a few rare speaker designs), if the trebble is a bit ahead or behind the bass. The treble will almost always hit you first in a traditional speaker design with straight horizontal front. Some horn loaded speakers may be longer than the bass units (?). In that case the bass will hit you first.
If you have a certain LP filter frequency, say 750Hz, you must admit that after that frequency you have a "decreasing" slope, but some of the material higher than that is passed, as the slope is not neg. infinite, right? Only after an octave you have 96db reduction, but an octave higher of 750hz is 1500hz.
A corresponding frequency on a HP filter, that reduces lows, reduces 96db only after an octave, that is 350hz.
Now you have an area of 2 octaves, from 350hz to 1500hz, that is partially passed both from the LP and the HP, in a way that increases and overlaps bands.
How can this,although limited, being ignored?
It's minimal, but existing
A corresponding frequency on a HP filter, that reduces lows, reduces 96db only after an octave, that is 350hz.
Now you have an area of 2 octaves, from 350hz to 1500hz, that is partially passed both from the LP and the HP, in a way that increases and overlaps bands.
How can this,although limited, being ignored?
It's minimal, but existing
I never intended to claim, that a 96dB/octave is infinite. Also, I didn't see anything in the request, which made me think an infinite slope was required. Also, I strongly believe, that infinite slopes can lead to damaged speakers. Say you have a 444Hz filter. Now you play that A on your guitar - and vibrate the string a bit. The sound will bounce from one output to the other ... sometimes at the peak of the waveform ... forcing the diaphragm to try to go from center to ... somewhere else ... in zero milliseconds. For special effects it may have it's use, but as a x-over, I wouldn't recomend it.
By the way. Those 350 should have been 375
By the way. Those 350 should have been 375
I just wanted to interrupt the escalation of the slope-steepness contest to share an experience.
I had the chance to use an older model Ashley analog crossover which has a variable slope control (0 to 12 dB). The best sound I could dial in used slopes closer to 6 dB per octave than to 12 dB. And yes, it sounded fine.
A friend tried a DBX DriveRack with the same speaker system and used 24 dB per octave slopes, and it also sounded fine. But a look at the graphic of the curve showed considerable overlap between the lows and mids and a lesser (but still visible) overlap between the mids and highs.
The lesson - use your ears.
to share an experience.I had the chance to use an older model Ashley analog crossover which has a variable slope control (0 to 12 dB). The best sound I could dial in used slopes closer to 6 dB per octave than to 12 dB. And yes, it sounded fine.
A friend tried a DBX DriveRack with the same speaker system and used 24 dB per octave slopes, and it also sounded fine. But a look at the graphic of the curve showed considerable overlap between the lows and mids and a lesser (but still visible) overlap between the mids and highs.
The lesson - use your ears.
Sorry for any misunderstanding. The 24 dB per octave slopes I was referring to in the case of the DBX DriveRack had the overlap because the crossover point for each band was set independently. For example, say the top of sub-woofer curve was down -n dB at 500 Hz. The overlap would come from setting the bottom of the mid-range curve to be down -n dB starting at 100 Hz. That means the freq's from 100 to 500 Hz overlap even though the slopes were 24 dB per octave.
A 6 dB per octave curve using the same crossover points and the same type of filter would be expected to have more overlap. In the case of the Ashley crossover I was describing, one control sets both the top of the sub-woofer band and the bottom of the mid-range band (this is true of most discrete crossovers). Making the slopes more shallow simply provides another way to overlap the speakers. Devices capable of doing so are usually called something such as "Loudspeaker Control Systems."
By the way, the makers of high-end loud-speaker control systems (Lake, BSS, XTA, EAW (Ashley), DBX, etc) get a lot of money for their products. Could be an area of opportunity for a CreamWare developer.
A 6 dB per octave curve using the same crossover points and the same type of filter would be expected to have more overlap. In the case of the Ashley crossover I was describing, one control sets both the top of the sub-woofer band and the bottom of the mid-range band (this is true of most discrete crossovers). Making the slopes more shallow simply provides another way to overlap the speakers. Devices capable of doing so are usually called something such as "Loudspeaker Control Systems."
By the way, the makers of high-end loud-speaker control systems (Lake, BSS, XTA, EAW (Ashley), DBX, etc) get a lot of money for their products. Could be an area of opportunity for a CreamWare developer.
-
ChrisWerner
- Posts: 1738
- Joined: Fri Aug 31, 2001 4:00 pm
- Location: Germany/Bavaria
- Contact:
Yep, nice thingy.
http://www.planetz.com/forums/viewtopic ... 5&forum=15
http://www.planetz.com/forums/viewtopic ... 5&forum=15