This is a matter of achieving a desired response while navigating the inherent driver response and I don't see a difference whether you strip it back all the way before crossing, or you make the connection more directly and simply. I think you can be sure that John achieved the goal properly.
Last edited:
Let me repeat too: Reality viewpoint you prefer is not possible in reality without significant restrictions with many dsp gear (some/most miniDSP and FourAudio) why I have protested against FIR in output channels only. In addition, input FIR is much simpler and faster to configure; single coefficiens and all taps available for whole band.
Oh well, I had typed up a detailed response but when I click Submit it didn't work and it was lost. So I'll just say I tried just about every approach I could think of; passive and active speaker systems with phase correction by pre processing (LS3/5a and my NaO system); 2-way, FIR LR4 systems from scratch where each individual driver's acoustic output was switchable between linear phase or minimum phase; LR4 IIR with FIR correction to the system or individual drivers; individual LR2 FIR switchable between minimum phase and linear phase (tweeter revered polarity for min phase normal polarity for Lin phase); ....
The bottom line was that there were times, with certain passages, where sometimes it was though there was a difference, but there was never any consensus on which system was which or which sounded better. It was all pretty random. The only times they was some consistency was with bass response. There was some agreement that bass hade better impact when low frequency phase was linearized. But that is not crossover related.
I will say that for a given passage of music there might be a preference for one over the other due to some subtle difference. But for another piece the preference might be reversed. And when it comes down to comparing a 3-way using a B&O type crossover compared to the same 3-way with LR4 crossovers, as I have tried to point out, there are far bigger differences than just phase response. In fact, in previous posts I linked to my old B&O like TP 3-way, original with passive crossover, then revised with digital crossover. Today that system sounds better than it ever did, since I redesigned it with straight LR4 crossover. It doesn't reproduce a square wave for crap anymore, but it sounds good with music.
I don't fault anyone who wants to play around with this. And if they feel differently then me, if they believe TP make a difference, God bless. My experience is just different.
One other thing I would like to point out about the B&O approach. I said it before, it is just a subtractive type TP crossover. You set up a LP and an HP and sum them in phase, subtract from flat, zero phase and you get a filler. With 2nd order HP and LP the filler is a 2nd order band pass. But consider the LP and the filler sum. That is what will sum to the HP response to yield flat amplitude and zero phase. So instead of doing it that way, just set up a HP filter and subtract and you get the same response in a single LP as the B&O LP + filler. You will find that response looks like a 2nd order LP with Q>1. Not actually but it has similar appearance, a rise above flat before rolling off. That rise is over 1dB if the HP is an LR2, if a B2 the rise is about 2dB. This means that off axis can have a polar response at certain frequencies where the output is 1 or 2 dB greater than the axial response. With the B&O its' worse because you have two crossover points where the polar get mess up as opposed to one if you use a single LP.
If you look at this old page of mine you can see what the off axis vertical response looks like for the B&O approach when different Qs are chosen.
TP3-project3
Anyway, it's all for fun, isn't it.
Hi John K, Big thanks for the considerate reply!
I've been very interested to hear about your experiments as it's clear to me you've drawn a number of the same understandings and impressions that my experiments have led me to. Sounds like you've tried a bunch...
Interesting...the most common comment I hear from folks who have done extensive phase linearization comparisons (mostly from the prosound world) is that it's probably most audible with bass.
I believe that too, but can only really A/B the improvement outdoors.
My zeal for FIR and phase linearization has more to do with maximizing the probability of getting the best out of our speakers, than whether phase is audible or not.
I think my zeal here often gets misunderstood, and pooh-pooed as just another raging linear phase advocate..
The ease of using FIR is just too brain dead simple compared to staying IIR, even if using DSP for IIR, imo/ime.
Simply correct driver-by-driver with min phase mag and phase flattening, both in-band and out-band through xover region, and then add in linear phase xovers.
Then I find, when every driver is flat mag and phase, timing is simply a matter of exact physical distance offsets between acoustic centers.
And then the real beauty begins...
I can put the speaker on the spinorama and start dialing in best crossover frequency for polars, where all I have to do is change xover freq...no new timing..no new EQs..simple
(This of course assumes complementary linear phase crossovers and sufficient out of band flattening.)
Then, more beauty piles on...
I see how polars, especially vertical, improve as I increase xover order...with zero detriment to phase / group delay. (I've yet to hear pre-ringing although I've used a maximum of 6144 taps at 48kHz)
Anyway, I've been doing this on 4-ways with what I believe are extraordinary results. Transient response and clarity out the gazoo
Yes, all for fun and all the best
Last edited:
That's cool...but I can only repeat..i think input FIR leaves a lot on the table...
and tries to fix what should have been fixed on output..or tries to fix what can't be fixed..
This repeating is already quite stupid. Please understand that decent output FIR is not always possible.
yes, why not? With software like the Ultimate Equalizer you can construct a multiway, active speaker where each driver is equalize to a specified acoustic target using FIR filters which are either minimum phase or linear phase. Now, if you want to make a passive speaker I would agree that any eq and phase correction must be at the input, but that's a different issue.
Also, a few posts back it was stated that input phase correct was pretty much the same as phase correction on individual driver. That statement really only applies to 2-way crossovers using LR4 or LR8 crossovers. 2ndf and 6th order LR require inversion of the tweeter. You can linear phase of those at the input but it's not the same as making each driver linear phase because in toy do it driver by driver they must be connected with normal polarity. Or if you have a speaker with a 3rd order Butterworth crossover you can still use input FIR to make the system linear phase, but it is not the same as making each driver's filter linear phase since the driver's acoustic output sums in quadrature. If you make each driver linear phase with an order Butterworth response you get a 3dB hump at the crossover point, regardless of order.
But all these configurations, 2 and 3 way surely can be digitally equalised before D/A to have a linear FR and phase behaviour - right? The only difference would be how they handle input power and thus producing distorsion. Or is there something you can create with drivers and L/C/R that isn't correctable? I didn't think so.
//
//
Hold on there- you can only attenuate blemishes with DSP, you cannot (partially) short them out. This is where the passive and active/DSP functions greatly differ. It's also one of the reasons passive filters are not archaic and might actually be simpler than using a boat load more parts from a DSP and several amplifiers further complicated by a complex assembly and connection.
There are drivers that can only function in DSP via steep order because the issues cannot be compensated well enough at lower order via DSP to be viable. It messes with the alignment of the adjacent drivers. (Now- if using the FIR stuff this is not my cup of tea or forte, so it might be a moot point there.)
For an example- If you have a driver such as the TB W4-1798S, and it has a well known energy storage due to the membrane material, shape, and diameter that rears its ugly head and is sharp and nasty without being sorted out at 1.5kHz. Using sharp deep narrow Q passive notches for steep filtering, I could not suppress this energy storage at all, as it was always being excited. There was no way to attenuate this blemish via attenuation methods and parallel notches. Once I switched to the series notch across the driver and shorted it out, the problem was able to be minimized satisfactorily. The other problem with this related to DSP is there is no voltage divider present to make the filter work as intended, therefore, there is no way to implement this kind of conjugate for such problems when using DSP or active implementations.
When Brian was doing his Ultimate Small Speaker thread about the 'Reference Mini' design, he and I discussed this kind of thing in detail with help from others at PETT. Where he thought DSP would work and be usable in all applications, he was actually misinformed up to that point in time.
It's not all cut and dried to use DSP.
Later,
Wolf
The big difference between both passive notching methods is that one of them is shorting the driver at the critical frequency range and the other doesn't.
With an active scenario (be it DSP or analog) the driver would even be shorted by the amp over a much wider frequency range. This - combined with an appropriate notch before the amp - has the potential to be even better than your solution if done correctly.
Regards
Charles
Last edited:
Wolf, it is, because:
1) all signals start out as digital.
2) single ADC before crossover necessitates further analog steps between ADC and drivers (crossover, corrections).
3) all these further analog steps could have been performed with more precision and orders of freedom in the digital domain. Plus, performing these steps in the analog domain introduces artefacts that are absent when similar processing is done digitally.
The main problem I presently see with the DSP solution is that you need a high quality DAC for each driver, instead of one for each loudspeaker. The standard DSP solutions on the market seem to fall short in this respect, so it is soldering time. Other than that, even for the simplest 2-way, DSP wins hands down.
1) all signals start out as digital.
2) single ADC before crossover necessitates further analog steps between ADC and drivers (crossover, corrections).
3) all these further analog steps could have been performed with more precision and orders of freedom in the digital domain. Plus, performing these steps in the analog domain introduces artefacts that are absent when similar processing is done digitally.
The main problem I presently see with the DSP solution is that you need a high quality DAC for each driver, instead of one for each loudspeaker. The standard DSP solutions on the market seem to fall short in this respect, so it is soldering time. Other than that, even for the simplest 2-way, DSP wins hands down.
Reading of earlier messages could help a bit. Your questions also reflect assumption that applications such as brute fir or eq apo are always available and possible to select for the project. Reality could be different; simpler and faster to design, but not so perfect due to limited capacity.
All dsp gears do not have adequate number of taps to make "the best possible" corrections to multi-way, and the problem could be bigger if convolver is available for outputs only. Maximum taps is not necessarily "parallel" resource available for all channel separately. If you use 256 taps for the tweeter and 768 for the mid, that drops taps available for the woofer by 1024 which needs the most to get target response exactly. You are forced to optimize -> more unwanted compromises.
Filter designer application of selected gear could also have some limits related to FIR. For example if you decide to use FIR XO, you are forced to use symmetrical linear phase slopes. That is not always the easiest/fastest and most economical -> IIR EQ blocks will be wasted for nothing. Or XO of all ways are forced to FIR linear phase which is not necessarily good thing for delay requirements. Number of taps could be limited to guarantee small delay for pro applications.
So input FIR could enable better compromise, including possible few degrees phase mismatch at -60 dB below nominal sensitivity.
No. That is simply not true. First of all, FIR doesn't mean linear phase. FIR can be arbitrary phase, including minimum phase. Second, X-Os using linear phase filters need not be symmetrical. Aside from symmetrical linear phase variants of the LR type filters I wrote a paper back in 2002 where I developed asymmetric linear phase XO filters for which the HP filter had twice the slop of the LP, or vise versa. For example, the LP could have a 2nd order roll off and the HP would then have a 4th order roll off. Being linear phase, the HP and LP sum in phase and the polar response is symmetric. Third, given that the goal is a linear phase speaker, FIR can be used to implement the so called subtractive-delayed crossovers (Lipshitz and Vanderkooy) which use a minimum phase LP section as the starting point.
As for the number of taps that generally isn't an issue with PC based processing. 4096/channel with 16 channels is done with some software and proper PC hardware. .
Last edited:
@Wolf:
When Brian was doing his Ultimate Small Speaker thread about the 'Reference Mini' design, he and I discussed this kind of thing in detail with help from others at PETT. Where he thought DSP would work and be usable in all applications, he was actually misinformed up to that point in time.
The quote above made me curious: what exactly was wrongly assumed by Brian with the design assumptions in the Ultimate Small Speaker?
Eelco
When Brian was doing his Ultimate Small Speaker thread about the 'Reference Mini' design, he and I discussed this kind of thing in detail with help from others at PETT. Where he thought DSP would work and be usable in all applications, he was actually misinformed up to that point in time.
The quote above made me curious: what exactly was wrongly assumed by Brian with the design assumptions in the Ultimate Small Speaker?
Eelco
What the f... I was not saying a word about general FIR theory. Just practice with some actual dsp devices you cannot control freely e.g. with IR + convolver. Only with filter design software by the manufacturer. That sw controls everything you can do. For example linear phase XO slopes could be only choice you have (in manual mode) if you have selected FIR XOs. Possible automatic algorithms do what they do.
So earlier "not always" meant real life dsp devices with actual restrictions. Not ideal academic theories.
For an example- If you have a driver such as the TB W4-1798S, and it has a well known energy storage due to the membrane material, shape, and diameter that rears its ugly head and is sharp and nasty without being sorted out at 1.5kHz. Using sharp deep narrow Q passive notches for steep filtering, I could not suppress this energy storage at all, as it was always being excited. There was no way to attenuate this blemish via attenuation methods and parallel notches. Once I switched to the series notch across the driver and shorted it out, the problem was able to be minimized satisfactorily. The other problem with this related to DSP is there is no voltage divider present to make the filter work as intended, therefore, there is no way to implement this kind of conjugate for such problems when using DSP or active implementations.
It's not all cut and dried to use DSP.
Later,
Wolf
I will refer you to another achieved web page of mine.
Stored_energy_1
About 1/2 down the page you will see two CSD plots of raw 18cm drivers (ScanSpeak and Seas). Neither is particularly "clean" and the Seas (Fig2) shows a particularly bad metal cone breakup resonance at 5k Hz. Below those two CDS is a 3rd. The 3rd is the CDS of a purely electrical band pass filter. That CSD represents the best you can do with such a filter. Below that are two additional CSD plots of the two drivers when their acoustic response is shaped to the same band pass response as used in Fig 3 using FIR filters. Note that if the filters were "perfect" the CDSs shown in Fig 4&5 should match Fig 4. Indeed they are very similar, particular Fig 5 for the Seas driver with 5k resonance peak.
I think this should suffice to show that serious driver resonances can be controlled through DSP before the amplifier.
Now, this has nothing to do with shorting the driver at the resonance frequency. Whether done actively before the amp or passively across the driver's terminals it is about stored energy and its release. The high Q resonance of the driver stores "positive energy" when stimulated at the resonant frequency. A simple resistive load across the drivers terminals would dissipate that energy over time when the stimulus stopped. A passive notch filter, no matter how implement, simply limits the magnitude of the stimulus to control the resonance. That notch filter stores "negative" energy which is also release when the stimulus is removed. The net result is the positive energy and negative energy cancel each other. (Note the concept of positive and negative energy is just a relative term like positive voltage and negative voltage.)
With Active EQ, analog or digital, the same thing happens. The "negative" energy of the active notch filter is released when the stimulus is removed, amplified by the amplifier and cancels the positive energy released by the driver's resonance.
Or, perhaps more clearly stated, the voltage oscillations of the driver resonance are equal in magnitude but 180 degrees out of phase with the notch filter's.
So "shorting" a resonance with some finite resistance is just adding damping to the system, like dragging a weight across a floor, then increasing the weight. you can't drag the heavier weight as far before you get tired. A correctly designed notch filter is like having another person of equal strength trying to pull the weight in the opposite direction. The weight doesn't move at all.
Last edited:
Ok, don't get your panties up in a bunch. It's the internet. Things are always clear.
Sorry, I can't keep my ADC's from my DAC's, apparently. Please read 'DAC' under point two where I mixed up. Otherwise incomprehensible.