I'm currently buying parts for my next speaker project, a three-way with a widerange driver in the middle, I'll be crossing it over to a large woofer and a tweeter to fill in both the top and low ends, but my problem is this, since it's a widerange driver, I'll be crossing over high for the tweeter and low for the woofer, so I'm sure a single-order slope would work (maybe 2nd for the tweeter, just to be safe), but would it sound good? I remember reading somewhere that odd-order slopes sounded better than even-order, why is this? If so, I think I'd lean towards a third-order, since it'll be an active x-over. But are there any drawbacks to higher order slopes? Should I keep it low if possible?
I've always built the crossovers to suite the drivers (high-order on ribbon tweeters, because I had to), now that I've got the choice, I'm not sure which is preferable. Oh, and this begs another question, should I look into subtractive x-overs or not? Hmmm... So many questions today.
I know in the end I'll have to tweak it to the loudspeaker, I would just like to start off as well as possible.
I've always built the crossovers to suite the drivers (high-order on ribbon tweeters, because I had to), now that I've got the choice, I'm not sure which is preferable. Oh, and this begs another question, should I look into subtractive x-overs or not? Hmmm... So many questions today.
I know in the end I'll have to tweak it to the loudspeaker, I would just like to start off as well as possible.
XO slopes required depend totally on the measured response of the drivers in the enclosures, along with cone break-up and response peaks/nulls. Once you know all of that then you can design a crossover, not before.
So, my advice is to build the boxes first, then test and experiment to see what suits your taste and measures well.
Doesn't really help that much, I know, but it is a realistic answer...
So, my advice is to build the boxes first, then test and experiment to see what suits your taste and measures well.
Doesn't really help that much, I know, but it is a realistic answer...
JoeBob said:
This is the kind of system i tend to build (think about building). Often on the top end i just let the FR run to its limit and bring in a superT with a single cap. On the bottom, i've been partial to subtractive -- just been reading Nelson's paper on the subject over & over. His use of a buffered PLLXO for the filter stage is facinating. The droopy respnse of a 2nd or 3rd order section helps kill the bump in the derived leg of the XO. What i'm having a bit of trouble with is how the levels add up right going into the adder.
In a fully active system i'd have 2 or 3rd order on the bottom, same on the top, with the FR having the derived 1st order bandpass. (i keep looking a a set of little teeny EL94 OPTs that should make a killer HF amp)
I've a set of Arun Cantus 2si -- i haven't decided whether to pursue a system with multiple FE103A or a JX125 (my JX150s look like they will morph into 125s) in the middle -- with 4 Foster 12s or 4 Peerless 8s on the bottom)
dave
Low- vs. High- Order Active Crossovers. High- Order Active Crossover
Hmmm....
I tried to question the same in this thread:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=21184
Right now I'm using 4th order LR active filter - which I like
After my surfing'n'reading on the internet and discussing with other DIY-HiFi people
My conclusion is: 4th order active
Well that is just my opinion!
- You can find people prefering 1st order passive x-overs
- You can find people prefering 32th order active
Regards, Ask
Hmmm....
I tried to question the same in this thread:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=21184
Right now I'm using 4th order LR active filter - which I like
After my surfing'n'reading on the internet and discussing with other DIY-HiFi people
My conclusion is: 4th order active
Well that is just my opinion!
- You can find people prefering 1st order passive x-overs
- You can find people prefering 32th order active
Regards, Ask
Something that's been hinted at but not stated outright is to go for the lowest order crossover that will get the job done. Sure, it's easy enough to say that, well, a six (1st order) crossover will do, but I think I'll go to 12 or 18. Unh unh. Stop at six if it does the job. Or if it takes 12, stop there, etc. Odd order crossovers have some advantages, but it's seldom, if ever, the make-or-break kind of thing that you might expect.
Your goals are:
1) Protect drivers from frequencies that could damage them
2) Don't ask drivers to reproduce frequencies that they can't do well.
3) Sound good.
4) Avoid--to the extent possible--having to tailor the frequency response.
1 is an absolute--if you're burning drivers, you're going to run out of money quickly. 2 & 3 are related. 4 is loosely related to 1, 2, & 3, but keep in mind that frequency tailoring involves phase shift, which is bad. The arguable exception here is the very lowest frequencies. It's very difficult to get flat response below about 30 or 40Hz without having to tweak things a bit. Related point: Don't believe Thiele-Small simulations--they are poor approximations of reality (don't say I didn't warn you).
Grey
Your goals are:
1) Protect drivers from frequencies that could damage them
2) Don't ask drivers to reproduce frequencies that they can't do well.
3) Sound good.
4) Avoid--to the extent possible--having to tailor the frequency response.
1 is an absolute--if you're burning drivers, you're going to run out of money quickly. 2 & 3 are related. 4 is loosely related to 1, 2, & 3, but keep in mind that frequency tailoring involves phase shift, which is bad. The arguable exception here is the very lowest frequencies. It's very difficult to get flat response below about 30 or 40Hz without having to tweak things a bit. Related point: Don't believe Thiele-Small simulations--they are poor approximations of reality (don't say I didn't warn you).
Grey
Since you want to go active anyway I'd recommend a subtractive crossover.
Since the derived slopes are only 1st oder usually, the feasibility depends heavily on the drivers used of course.
You can do symmetric and asymmetric constant-voltage subtractive crossovers of higher orders like 2nd/2nd or 3rd/2nd but that comes at the price of increased overlap and higher bumps (i.e. incrased power needs).
Regards
Charles
Since the derived slopes are only 1st oder usually, the feasibility depends heavily on the drivers used of course.
You can do symmetric and asymmetric constant-voltage subtractive crossovers of higher orders like 2nd/2nd or 3rd/2nd but that comes at the price of increased overlap and higher bumps (i.e. incrased power needs).
Regards
Charles
I think it depends on the variables
Running the Azuras full-range and crossing to a sealed sub at 134hz, it always integrates better with the higher orders -
When using the Behringer - I migrated to 48DB slope.
When I sold the Behringer and started using the BSS FDS366, it sounded better with the 52dbm NTM slope.
One major advantage of using a digital crossover has been the ability to try the various slopes, in addition to being able to dial in delay, phase and EQ.
Does that mean I think everybody will be better with 48db or 52db slopes at 130-135 hz? Nope! Not at all.
I believe you will find differences in whether it is done digitally, passive or active.
Saying one slope or approach is better - overlooks the differences in the various frequencies and approaches.
Soooo, I don't think you can say any one slope is best for all situations - it's going to depend on the drivers, the hz, etc., and what method you're using to achieve the crossover function
Regards
Ken L
Running the Azuras full-range and crossing to a sealed sub at 134hz, it always integrates better with the higher orders -
When using the Behringer - I migrated to 48DB slope.
When I sold the Behringer and started using the BSS FDS366, it sounded better with the 52dbm NTM slope.
One major advantage of using a digital crossover has been the ability to try the various slopes, in addition to being able to dial in delay, phase and EQ.
Does that mean I think everybody will be better with 48db or 52db slopes at 130-135 hz? Nope! Not at all.
I believe you will find differences in whether it is done digitally, passive or active.
Saying one slope or approach is better - overlooks the differences in the various frequencies and approaches.
Soooo, I don't think you can say any one slope is best for all situations - it's going to depend on the drivers, the hz, etc., and what method you're using to achieve the crossover function
Regards
Ken L
It depends on the variables
I agree in this statement
Please don't forget the subject for this thread: "What's your favorite crossover slope?"
It could be interesting to hear what DIY people has as thier favorite crossover slope.
Be free to post what you don't like, but also remember to be constructive!
Hence what about posting your own favorite crossover slope...
Which slope is your favorite crossover slope? And Why?
Regards, Ask
Ken L said:
I agree in this statement
Please don't forget the subject for this thread: "What's your favorite crossover slope?"
It could be interesting to hear what DIY people has as thier favorite crossover slope.
Be free to post what you don't like, but also remember to be constructive!
Hence what about posting your own favorite crossover slope...
Which slope is your favorite crossover slope? And Why?
Regards, Ask
Re: It depends on the variables
My favorite crossover slope starts at about 6 dB/oct, then gradually increases slope until it ends up at 90 dB/oct. It's a Finite Impulse Response (FIR) filter, and the other band is derived by subtracting the filtered signal from a delayed version of the input. I use these for $DAYJOB, and they work marvelously well, especially if you try to keep the filter as short as possible. Longer filters tend to have pre-ringing, which on axis isn't a problem since everything adds up to 1.0 if your drivers are matched at xover frequency, but off axis the pre-ringing creeps in again. The shorter FIRs I design allow about an octave of overlap between the drivers from -1dB to -20dB, but with decent drivers that isn't really an issue. Typically it isn't what your driver does within half an octave of the crossover which kills you, it's what the things do an octave or two into their rolloffs which trash the passband.
For example, a 2500 Hz FIR crossover lowpass filter could be -1dB at 1800 Hz, -6dB at 2500 Hz, -20dB at 3200 Hz, and greater than -60dB at more than 4000Hz. Its HPF dual is -1dB at 3200 Hz, -6dB at 2500 Hz, -20dB at 1800 Hz, and greater than -60dB at less than 1000 Hz. Even though the filters allow some overlap between drivers, they're essentially out of the picture less than an octave into their stopband.
For woofers, you avoid exciting code breakup modes, and excursion is considerably reduced for tweeters. The implications for metal-cone woofers and ribbon tweeters are obvious. You also get phase linearity as a side benefit; it literally drops out of the equation.
Cheers,
Francois.
askbojesen said:
I agree in this statement
Please don't forget the subject for this thread: "What's your favorite crossover slope?"
Which slope is your favorite crossover slope? And Why?
Regards, Ask
My favorite crossover slope starts at about 6 dB/oct, then gradually increases slope until it ends up at 90 dB/oct. It's a Finite Impulse Response (FIR) filter, and the other band is derived by subtracting the filtered signal from a delayed version of the input. I use these for $DAYJOB, and they work marvelously well, especially if you try to keep the filter as short as possible. Longer filters tend to have pre-ringing, which on axis isn't a problem since everything adds up to 1.0 if your drivers are matched at xover frequency, but off axis the pre-ringing creeps in again. The shorter FIRs I design allow about an octave of overlap between the drivers from -1dB to -20dB, but with decent drivers that isn't really an issue. Typically it isn't what your driver does within half an octave of the crossover which kills you, it's what the things do an octave or two into their rolloffs which trash the passband.
For example, a 2500 Hz FIR crossover lowpass filter could be -1dB at 1800 Hz, -6dB at 2500 Hz, -20dB at 3200 Hz, and greater than -60dB at more than 4000Hz. Its HPF dual is -1dB at 3200 Hz, -6dB at 2500 Hz, -20dB at 1800 Hz, and greater than -60dB at less than 1000 Hz. Even though the filters allow some overlap between drivers, they're essentially out of the picture less than an octave into their stopband.
For woofers, you avoid exciting code breakup modes, and excursion is considerably reduced for tweeters. The implications for metal-cone woofers and ribbon tweeters are obvious. You also get phase linearity as a side benefit; it literally drops out of the equation.
Cheers,
Francois.
mhelin said:
Well, like John Curl, I can't give *everything* away, because I do need to keep some magic from which I might eventually get product happening, but try looking for a FIR design program which lets you specify the number of coefficients: that'll take you a long way there. Specify an odd number of coefficients, and derive the highpass by subtracting the original signal delayed by (N+1)/2 samples. For example, if your filter is of length 71, then delay the original by 36 samples before subtracting it from the LPF to generate the HPF.
Word length isn't quite as critical for FIRs as it is for IIRs, but you still want 24 bit math at least for nice stopbands.
That should keep you busy for a while....
Cheers,
Francois.
Thanks,
If I design a FIR filter using the window method there will be ripple in stopband. If I then use the subtractive method to calculate the HPF the ripple will be in passband, right (or not, is the subtraced response mirrored vertically or horizontally after all)?
Using the Remez method there will be ripple in both pass and stop band, is it any better then? Also which windowing method is best (Kaiser, Hamming or Blackmann, the other are not good I think, Kaiser was recommended somewhere, how about Chebyshev, I think it also looks nice)?
If I design a FIR filter using the window method there will be ripple in stopband. If I then use the subtractive method to calculate the HPF the ripple will be in passband, right (or not, is the subtraced response mirrored vertically or horizontally after all)?
Using the Remez method there will be ripple in both pass and stop band, is it any better then? Also which windowing method is best (Kaiser, Hamming or Blackmann, the other are not good I think, Kaiser was recommended somewhere, how about Chebyshev, I think it also looks nice)?
I don't remember Francois mentioning that he hasn't any ripple in the stopband !?
I for myself would definitely go for a filter derived by using the windowing method because I want a flat passband.
The stopband ripple looks large on a diagram showing the response in dB but measuerd in volts or what ever it is not much. It will therefore cause a passband ripple in the derived branch that would be best expressed in milli-dB !
The usual way to get FIR higpass parameters using the windowing method is the subtraction of the coefficients of a lowpass from the coefficients of an allpass anyway ! Not much different than the subtractive method mentioned by Francois.
I have to admit thopugh that I am a fan of analogue solutions and I would therefore go for an analog subtractive crossover anyway.
Regards
Charles
I for myself would definitely go for a filter derived by using the windowing method because I want a flat passband.
The stopband ripple looks large on a diagram showing the response in dB but measuerd in volts or what ever it is not much. It will therefore cause a passband ripple in the derived branch that would be best expressed in milli-dB !
The usual way to get FIR higpass parameters using the windowing method is the subtraction of the coefficients of a lowpass from the coefficients of an allpass anyway ! Not much different than the subtractive method mentioned by Francois.
I have to admit thopugh that I am a fan of analogue solutions and I would therefore go for an analog subtractive crossover anyway.
Regards
Charles
So the ripple is there in "subtracted" high pass but it is in the stopband, not in the passband.
So I really need to use convolution only for the LPF, and the HPF output is got by subtracting the delayed input sample from the LPF output sample? In case of 71 point LP kernel it's 71 multiplications and 72 additions (71 MAC's + one add in DSP).
So I really need to use convolution only for the LPF, and the HPF output is got by subtracting the delayed input sample from the LPF output sample? In case of 71 point LP kernel it's 71 multiplications and 72 additions (71 MAC's + one add in DSP).
mhelin said:
Most filter CAD programs let you specify passband and stopband ripple. 1dB passband ripple (not what you want to use, really), translates to the subtracted filter having about 20 dB stopband, 0.1dB ripple translates to 40dB stopband, 0.01dB ripple means 60dB stopband, and so on.
Of course the LPF stopband ripple also turns into HPF passband ripple, but I wouldn't worry about 0.01 dB ripple in either case.
And yes, convolving for the LPF and subtracting to generate the HPF is exactly right. You can also do it the other way, if you want, convolving to generate the HPF and subtracting to generate the LPF. That one might be a bit easier if your tool doesn't allow you to plot response from inputted coefficients; you need to do this with the subtracted set to make sure the stopband is behaving, and HPF stopband behaviour is more of an issue if you wish to avoid blowing up tweeters.
Francois.
Re: It depends on the variables
Well, I guess I took the liberty of enlarging the thread somewhat, hopefully others won't feel I'm threadjacking _grin_
Most certainly, and why also.
Actually, I thought my post was constructive in addition to being relevant to the thread.
I don't have a favorite as such but since I'm using NTM 52 db right now, I guess it is my favorite at the moment _big grin_
The Neville Thiele Method slopes are pretty new so I'll post a couple of links
http://www.bss.co.uk/includes/product_sheet_ntmv2_include.aspx
In the first one above you'll notice they incorporate a notched response for steeper rolloff.
In this next link, the graphics indicate that a 4th order NTM has steeper roll-off than 4th order L-R.
http://www.fmsystems.net/pdf/cutsheet/fds334t.pdf
while I know that I prefer the NTM 52db slope to the LR 48 DB slope in my current application, I haven't actually done any listening tests to compare the NTM 48DB slope to the LR 48DB.
Regards
Ken L
askbojesen said:
Well, I guess I took the liberty of enlarging the thread somewhat, hopefully others won't feel I'm threadjacking _grin_
askbojesen said:
Most certainly, and why also.
askbojesen said:
Actually, I thought my post was constructive in addition to being relevant to the thread.
askbojesen said:
I don't have a favorite as such but since I'm using NTM 52 db right now, I guess it is my favorite at the moment _big grin_
The Neville Thiele Method slopes are pretty new so I'll post a couple of links
http://www.bss.co.uk/includes/product_sheet_ntmv2_include.aspx
In the first one above you'll notice they incorporate a notched response for steeper rolloff.
In this next link, the graphics indicate that a 4th order NTM has steeper roll-off than 4th order L-R.
http://www.fmsystems.net/pdf/cutsheet/fds334t.pdf
while I know that I prefer the NTM 52db slope to the LR 48 DB slope in my current application, I haven't actually done any listening tests to compare the NTM 48DB slope to the LR 48DB.
Regards
Ken L
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.