Active Subtractive XOs

[cocolino]

I understand that, but the transient response of the whole system (this means low and high acoustically summed after filtering) is at its best if the filters are phase-symmetrical, because they compensate the phases with each other. if the phase is not symmetrical , then you will just get a worse transient response.... and you'll get lobing either upwards or downwards due to the phase difference....

It is not true that You get better transient response with symetrical filters leaving aside 6 db filters (HP and LP) .
Quite the opposite.
Where did You get the information from that symmetrical filters have better transient response - have You ever seen on a scope screen what two summed (HP+LP, equal slopes) signals for example from conventional second order (or any higher order) filters look like?

It is true that asymmetrical filters (be they subtractive or not) cause lobing but this must not always be of disadvantage.
For example operation of D`Appolito systems (filters + physical arrangement of drivers on the baffle, eg. Mid-Tweet-Mid or Woof-Tweet-Woof, with certain distance between drivers) are based on lobing effects due to 3.order filters causing phase differences between drivers, which in their total effect lead to a wider and more uniform radiation pattern though.

But I just posted this, because of the contradiction I read about transient of filters and subtractive filters.... and because this is my experience with real life subtractive filters. I know that you can make the subtractive filter more complex, and compensate completely for the phaseproblem (thus transient response problem), but by the time you reached this point you'll end up using more components then with an ordinary HP/LP symmetrical setup.....

There is no contradiction of transient response and subtractive filters.
Unless (conventional) filler driver circuits and 6dB filters (and maybe unless digital filter implementation from which I have no clue), actually a subtractive filter is the only way to produce such filters.
Other than the 3 exceptions mentioned, I`ve yet to see another filter topology which can achieve transient perfection without subtraction?

Working subtractive transient perfect filters (even with driver-filter-effects taken into account) must not necessarily be more complex than ordinary HP/LP setups - see Charles`s (phase_accurate) example above.

I don`t say a subtractive filter is always a better solution. In fact due to the considerable frequency overlap and therefore particular at higher crossover frequencies, they have rather limited use IMO but there are cases in which they might perform better. A good example is Charles´s application above combining a FR-unit with a woofer (at rather low fc, ~80Hz-300Hz)

And I don't understand what you mean by the subtractive filter being steeper en having a better transient response..... im sorry but in my opninion it just isn't true.... if it is steeper it has more phaselag, and in the case of the assymetrical phase resaponse of the subtractive filter also a worse transient response....

Maybe I have expressed myself not so good.
What I`ve meant is that there simply isn`t any (symmetric or not) conventional (not based on subtraction) transient perfect higher (>6dB) order filter (and again except filler driver filters which BTW are non symmetric by their nature).

In contrary to "ordinary" filters, filters based on subtraction steeper than 6dB slope (at least in one branch, although there are even symmetrical subtractive filters too) are possible and which are transient perfect (at least in theory and leaving aside for now how to achieve to maintain this behavior in the acoustical domain also).

but I can be wrong, i'm not really overseeing the phase of the summed output right now... it is possible that with 4th order (180 degrees -3dB) it comes to a point that it is phase correct due to the negative summing.... this means that in the 4th order case it might work, if used with an allpassnetwork.... hmmmm

No offense taken but I´d suggest further studies on this topic.

[dokter dB]

by more further examining (lipshitz & vanderkooij)
it seems to work, ... with proper allpass delays, for even order filters (2nd or 4th etc).
The phase difference between drivers is near zero, with the expense of the phaselag being translated to the whole system-sum..... Every step from one driver to the next is causing a phase change if you go upwards in frequency....
quite nice, but that means that transient response is really changed from input to output... hmmmmm

well i'm still in for symmetrical phase filters, have seen no one using subtractive filters somewhere, only in a very old kustom bass-amp from the 70's years ago, this one was without allpass correction..... (and because of that the hump)

[dokter dB]

Originally posted by cocolino

It is not true that You get better transient response with symetrical filters leaving aside 6 db filters (HP and LP) .
Quite the opposite.
Where did You get the information from that symmetrical filters have better transient response - have You ever seen on a scope screen what two summed (HP+LP, equal slopes) signals for example from conventional second order (or any higher order) filters look like?

yes, and like I said, the phase of symmetrical filters is compensating each other, thats why a -3 dB butterworth point is still giving a flat amplitude line, and not +3, because of the phase vector pulling it down. and sorry but wich scope does show transient response?



It is true that asymmetrical filters (be they subtractive or not) cause lobing but this must not always be of disadvantage.
For example operation of D`Appolito systems (filters + physical arrangement of drivers on the baffle, eg. Mid-Tweet-Mid or Woof-Tweet-Woof, with certain distance between drivers) are based on lobing effects due to 3.order filters causing phase differences between drivers, which in their total effect lead to a wider and more uniform radiation pattern though.

dont think so, two drivers placed vertical like d'appolito are having a small dispersion at higher frequencies (to be precise: 90 degrees/quarter wavelength if you measure center center of the bassdrivers, going smaller towards piston upwards, and with normal tweeters crossfrequencies and hifi-filterslopes this gets quite small, thats why 2.5 way is more effective to me)

There is no contradiction of transient response and subtractive filters.
Unless (conventional) filler driver circuits and 6dB filters (and maybe unless digital filter implementation from which I have no clue), actually a subtractive filter is the only way to produce such filters.
Other than the 3 exceptions mentioned, I`ve yet to see another filter topology which can achieve transient perfection without subtraction?

give me examples, i'm curious, who uses this with succes? I know that digital fir filters are used with succes, (also by me: the hugo/k&h pocessor actually a german design) and they are really effective and more versatile than any other filtertype I know, you can just overlap driverbands without any problem

Working subtractive transient perfect filters (even with driver-filter-effects taken into account) must not necessarily be more complex than ordinary HP/LP setups - see Charles`s (phase_accurate) example above.

hmm if they are transient perfect yes but they simply aren't. its impossible to turn back time so, with the first low pass it goes wrong, the low to mid step has to be compensated at the mids side (with an all pass network) for time due to the low pass of the bass unit...., this means a step in time backwards, and so the transient is disturbed, aside from if you really hear this, and i think you don't and that way its a nice filter to me

I don`t say a subtractive filter is always a better solution. In fact due to the considerable frequency overlap and therefore particular at higher crossover frequencies, they have rather limited use IMO but there are cases in which they might perform better. A good example is Charles´s application above combining a FR-unit with a woofer (at rather low fc, ~80Hz-300Hz)

what frequency overlap? seems impossible to me because you use the same filter for generating the both slopes... so the cross frequency is always the same... but yeah yeah charles combination is nice

Maybe I have expressed myself not so good.
What I`ve meant is that there simply isn`t any (symmetric or not) conventional (not based on subtraction) transient perfect higher (>6dB) order filter (and again except filler driver filters which BTW are non symmetric by their nature).

In contrary to "ordinary" filters, filters based on subtraction steeper than 6dB slope (at least in one branch, although there are even symmetrical subtractive filters too) are possible and which are transient perfect (at least in theory and leaving aside for now how to achieve to maintain this behavior in the acoustical domain also).
okeeokee im getting all these emails of responds on this topic so im gonna place this one and see what you guys have to say.... reelax maahn


No offense taken but I´d suggest further studies on this topic.

[

[phase_accurate]

Apart from a FIR (which can be made subtractive as well quite conveniently if symmetrical coefficients are used) only the output signals of a subtractive crossover sum to the original input signal in time AND frequency domain.
It is simply so because one output signal is the difference between the input signal and the filtered signal (i.e. the one that is needed to complement the filtered signal to unity).
NO arrangement using seperate high and lowpass filters (apart from a 6 dB filter and the filler driver, which is mathematically related to a symmetrical subtractive filter) is able to do this. Period.


Regards

Charles

[cocolino]

originally posted by phase_accurate
NO arrangement using seperate high and lowpass filters (apart from a 6 dB filter and the filler driver, which is mathematically related to a symmetrical subtractive filter) is able to do this. Period.

Amen,
thank You Charles

[phase_accurate]

Here is the symmetrical subtractive 2nd order crossover. I mathematically derived it from the one I called "advanced filler driver". I once read in a book by John Watkinson that symmetrical subtractive crossovers should be feasible but failed to find any decent info, so I did it by myself.

It's output looks the same as the one from the NS APP note mentioned within post #40.
There is one difference however: This one here takes less OP AMPs. If one uses an inverting 2nd order filter topology, it could be implemented with only two OPs (input buffer not counted).

It could also be implementded with a filtered lowpass and derived highpass by using a 2nd order lowpass and a 1st order highpass.

If the pole frequency of the 1st order filter and/or the gain of the intermediate stage is/are changed, the symmetry can be varied (din't do the maths for that).

There are also 3rd order symmetrical versions possible (slightly different topology) but it comes at the cost of larger humps and increased overlap. Even 2nd/3rd asymmetrical should be possible but I haven't done the math for that so far.


Regards

Charles

[dokter dB]

this is a very elegant filter phase ac., no question about that...
1st: I don't mean to be irritating anyone with my sayings here and i'm quite open to be convinced by you guys, but i'm still not....

If I look at the HP output, it has the normal 2nd order phasechange (compared with he input) around the crossoverfreq. and under that... If you sum the LP/HP outputs together and there is a flat line (flat summing amplitudewise) this means that the LP output also has the same phasechange , otherwise its impossible to get the flat line right?.

If the 2 outputs both have the same phasechange, there is phasechange in the acoustically summed signal, or if you like after summing the signals electrically this is visible/measurable.

But it is true that the phasechange isn't disturbing the amplituderesponse, what is your goal, but a good symmetrical filter isn't either, if you stay on axis.

Only off axis of the drivers, the lobing with your subtractive filter is broader and so vertical (if the drivers are placed so) it will behave much better because the outputs doesn't have to compensate. With a normal symmetrical filter changing vertical listening-angle means changing the compensating mechanism by varying the vectors... So the vertical lobe to the listener is smaller (only around the crossoverfreq), and in that way the subtractive filter seems more right to me.
However as long as the physical distance between drivers compared with the wavelength is small, this lobing (around the cross.) of standard symmetrical filters is much less, and not a problem to me.

But I must say its a very interesting topic, and it sure changed my feelings about the subtractive filter..... it is quite elegant, but the story about the transient respone seems untrue to me, because there is a common phasechange on both the outputs. But if you change for instance the low-end eq of your amp (wonder if any of you hififreaks have any ) you'll get the same effect. they say its inaudible, and i believe it.
cheers tonite guys, happy new filter

[phase_accurate]

If you sum the LP/HP outputs together and there is a flat line (flat summing amplitudewise) this means that the LP output also has the same phasechange , otherwise its impossible to get the flat line right?.

Not at all !

The end result is the vector sum of the two signals i.e. for an LR4 where you have a phase difference of 360 degrees (i.e. NOT zero as some believe !!!) both signals add to unity at the crossover point because they are both 6 dB down.

With all the subtractive filter topologies there is a phase shift in the 100 degrees area (even for the symmetric 2nd order one) and significantly less decay than 6dB (most often even less than 3 dB) summing again to unity. The difference between the LR and the subtractive x-over is that the latter is phase_accurate (at the cost of less steep slopes and increased lobing).

Regards

Charles

[dokter dB]

Originally posted by phase_accurate


Not at all !

The end result is the vector sum of the two signals i.e. for an LR4 where you have a phase difference of 360 degrees (i.e. NOT zero as some believe !!!) both signals add to unity at the crossover point because they are both 6 dB down.

I absolutely agree with the latter, and it wasn't me who said the output would be of the same phase correlation, I stated that because of the sayings of others over here (transient response etc.) , I'm aware of the phase-relation (and thus vertors), that I mentioned before

With all the subtractive filter topologies there is a phase shift in the 100 degrees area (even for the symmetric 2nd order one) and significantly less decay than 6dB (most often even less than 3 dB) summing again to unity.

so thats the same as with what I mentioned a standard LP/HP butterworth, -3dB points summing @0dB instead of +3dB because the phase lags/vectors taking away efficiency at cross


The difference between the LR and the subtractive x-over is that the latter is phase_accurate (at the cost of less steep slopes and increased lobing).

??? so the subtractive filter has more lobing? that can only be because it is not well designed, or because of the broader overlap between the drivers, tell me what exactly is the score of subtractive filters then? i'm still not convinced now

Regards

Charles

[phase_accurate]

tell me what exactly is the score of subtractive filters then?

Flat amplitude AND phase response !

Regards

Charles