The overall phase response and group-delay is the same for both. But the Duelund allows a steeper lowpass and highpass function. LR2 3 way OTOH allows a more flexible choice of crossover frequencies.
Regards
Charles
FIR doesn't sound better. In fact ( if correctly implemented*) they doesn't sound at all. And this is the point imo.
So why sacrify latency for using them? Because they help to manage acoustic behaviour in a transparent manner. And they offer things which are not possible with IIR like very very steep slope.
In some very specific case it can be useful. It's another tool with it's own pro and cons, nothing more.
If you want to determine the difference between IIR and FIR ( their own sound) process an audio file an apply a xover in the 3khz ( if possible same slope for both). If possible the source material should be a 'complex' signal, a voice you know, some big classical ensemble,... you split the signal in lp and hp way and then mix them back ( thru the mixing desk of your daw).
Listen to the result though headphones and spot the difference (if any). As you took away your loudspeaker and room you should really hear the 'sound' of the filters by themself.
I won't tell you what i listen to ( within the signals used as source) as i don't want to bias you, but at 3khz area you should be able to easily spot the difference. Then try at different freq. But don't expext something wow. It is subtle but there.
Once you reintroduce the acoustic anomaly caused by loudspeaker ( destructive interference and overall directivity pattern) difference have the potential to be much less subtle... with vertically aligned (d'appolito) it can really solve things...
Have you thoughts about including time alignement when you made your comparison? My first trys (when i didn't had a dsp, i did this to try FIR) i forgeted them... it took me some time too spot something weird with the FIR. Once implemented it was like a level of focus was gained. Nothing dramatic but an upgrade to me.
* i compared processors when i acquired mine and there was difference in sound between brands. I don't know if it was because implementation of calcul, difference in hardware or just expectation bias. Anyway this was almost 15 years ago so things may have changed, i'm in no way up to date with gear this days ( my own satisfy my needs). Only loudspeakers i want to try different ones.
Last edited:
Unfortunately I have not learned FIR fully yet, the recording is from miniSHARC filter. I later found out that the test does not reflect reality because I have a baffle step slope of 0.5Q, which should drive the phase wrong further? and also a standing wave cut.
One should be able to enter the settings in FIR without measuring? type in from whatever settings is in SHARC and flaten the phase (does not have a good mic and room for measuring).
Group delay I usually measure with DCX2496 and a mic that I connect temporarily but had no xover delay compensation in the recording.
When i talk about FIR i talk about complementary filters, iow xovers. There is not really more things to learn than for IIR. They just don't induce phase shift by their own. They require drivers to be all same absolute polarity and to be linearised as previously described.
The bsc and target curve ( if you use one) doesn't need to be FIR, a regular eq is all you need. As you correct some minimal phase effects, a typical eq will induce phase shifts which compensate what you correct as well as freq ( frequency/phase tracks themselfs, in fact their are the same phenomena observed a different way).
Yes you can dial them by ear if you know what you are doing. Iow don't induce acoustic anomaly by trying to mix drivers which can't work seemlesly togethers from a directivity pov, don't do stupid things if you use steep slope ( the knee can induce more membrane displacement than softer slope so be carefull).
But without measurement there is no way to be sure it is what you expect by design choice.
Here are my measurements of the SH-50 phase (total and excess) and group delay (total and excess), showing that the effect of bass reflex excess phase growth below ~120 Hz, and woofer-midrange crossover network excess phase growth at 500-800 Hz:The phase-rotation of the DSL SH-50 over it's rated -3dB pass band is 720 degrees, equivalent to 8 "poles" (6dB crossover filters) or two "trips around the globe" .
Some of that phase rotation is inherent in the bass-reflex ("phase inversion") alignment, Fb (box tuning) at around 90Hz.
Reproducing a reasonable facsimile of a square-wave-signal requires smooth, not flat phase response
Note the magnified vertical scale on the group delay plot showing 500 µs resolution. The excess group delay shown here says that there is almost no all-pass behavior of the two crossovers in the SH-50 that is traceable to the passive network (all "IIR" filters).
For a comparison to typical store-bought loudspeakers, here are the plots from Toole ("Subjective Measurements of Loudspeaker Sound Quality and Listener Preferences", JAES 1985) showing the relative levels of phase growth that is tolerated in most passive networks:
The red curve in each plot represents the measured phase response of the TAD TD-4002 Jubilees in my listening room that use only IIR filters to flatten SPL response. Note the lower frequency limit of 200 Hz as plotted, not showing the dramatic rise in phase response due to bass reflex ports.
Here is a plot of the Jubilee vs. the SH-50 for reference:
This has been my experience also. As long as the response is all minimum-phase, I can't hear a difference (except as bass frequencies below 100 Hz) which is, I believe, the punch line to all this. Just avoid all the all-pass behavior of the crossover filters and physical alignment mismatches of the drivers to each other (time/phase alignment at the crossover interference bands within 90 degrees of phase), and more than 99% of the desired sound quality will due to these factors will be achieved.
Chris
Attachments
here's my 2c on why i think FIR sounds better....
the first reason revolves around the potential of phase audibility.
There are a number of credible sources that contend phase audibility at low frequencies is real.
Michael Gerzon, who seems to have been a true listening / spatial replication genius, made the following quote that i've heard others i respect from today's era, more of less repeat.
“ The subjective effect of phase compensation of the bass from loudspeakers is very marked, giving a much tighter and more 'punchy' quality, with greater transparency, and interestingly a subjective extension of bass response of at least half an octave. The improvement is audible even on loudspeakers with a very high cut-off frequency, such as Quad electrostatic designs. . . . The benefits of bass phase equalisation are considered, by those who have heard it, to be a substantial improvement over what was hitherto possible with analog technology, and digital equalisation provides a way of improving bass performance without going to ridiculously large giant space-consuming power-hungry monster speakers, and is certainly a much cheaper route “.
I share that opinion with flat phase down to the very bottom of low end response, before natural IIR roll-off begins.
Particularly so when i set my rig up outdoors.
But we all know how highly debated this topic is.... and the improvement i think i hear varies depending on how a particular recording was made (I guess).
This really isn't the primary reason I think FIR sounds better.
Pragmatically, I think FIR dramatically increases the probability of achieving good tuning.
Especially so on speakers that have what i consider to be realistic SPL and dynamic capability. IOW, for multi-ways beyond 2-ways without a sub.
It is so relatively easy to achieve flat mag and phase with FIR, using linear phase xovers..as .compared to IIR.
(I've played with both a lot, because i was getting involved in renting my speakers, amps, and processors for live sound. IIR xovers were needed to keep latency low.)
It is so easy to get spot on time alignments between driver sections throughout their summation regions, when phase trace overlay alignment means getting one flat horizontal trace to lie on top of another flat horizontal trace......as opposed to getting phase rotations to lie on top of each other.
It is so easy to adjust xover frequencies to achieve best pattern width matching between driver sections, using steep complementary xovers.
Bottom line, i simply think i get better sound from achieving more certain and consistent, more accurate, tuning..
I don't see digital as having IIR properties, other than maybe the region DACs can't produce from their very lowest response down to DC...iow, DAC bottom end rolloff.
My understanding is that DAC's anti-aliasing low pass filters, are almost universally FIR.
Doesn't seem correct to characterize digital either way imo.
The 20-20K digital passband simply has flat mag and phase....like it's not a filter at all.
Yes, I was referring to filters. In my attempt to give a straightforward answer I used an oversimplification.
Concerning Rephase, can i just type in my values from miniSharc and press generate? It feels like risky waters when zoom is called views and one have to handle IEEE 754, floats, wraps and so on 😎
And why does not miniSharc offer linear phase eq and LR? it costs money and Rephase can offer it for free.
Cheers!
And why does not miniSharc offer linear phase eq and LR? it costs money and Rephase can offer it for free.
Cheers!
Attachments
Last edited:
This is a great analysis. Testing/tuning my SH-50 clone lead to having a very narrow freq band for the mids and a lot of overlap with the bass and highs. I did get better square waves using higher order cross overs on the bass LP and High HP. The Danley synergy design is a work of art.
I had been advocating long step FIR many years, but now I only use IIR. One important thing is, some digital filters sound better than the others, they are not equal at all. Also, until you start hearing pre-ringing, FIR linear phase filter sounds fine, but once you start noticing it, it is always there.
I think what sounds better is simply a function of how well they were constructed and implemented.
Case in point to previous statement......
If pre-ringing can be heard, the FIR filter(s) aren't well constructed/implemented.
Global FIR with phase linearization, be it an overlay to correct existing speaker tuning, or for room correction, or for both,....certainly has the greatest potential for misuse and pre-ring.
On a driver by driver basis, phase only equalization past the use of minumum phase EQs, has a bit of potential for pre-ring.
System high pass, and low pass, if linear phase, have potential for pre-ring. Best use IIR.
Complementary linear phase xovers, no matter their order, have very little potential for pre-ring. Pure win-win, ime/imo.
My Symetrix DSP got fixed finally. And I had all the electrolytic capacitors replaced. So I will soon dig into all this again.
Reading a thread on ASR, there are basically two approaches. I will try to sum up, starting with a 2 way xo only for my Kallax Synergy.
1) Choose target acoustic curves and match the measured ones with EQ. Here I still do not understand how do I protect the tweeter from LF without a HP - or should I e.g. use a 2nd order HP plus EQ to get 4th order acoustic slopes? How about the polarity/phase/delay this will introduce. Can this be fixed (FR, phase) with a global stereo FIR filter? Not sure how low can I get with 1024 taps and 48 kHz, need to check.
2) Make the drivers flat at least one octave around crossover point and then use a steep (48 dB/oct) LR FIR crossover. I am not sure 1024 taps will be sufficient as well. Maybe 800 Hz crossover will be possible.
All this should be done ideally slightly off axis at expected listening distance outside with dual channel measurement.
Did I forget anything in the summary?
Reading a thread on ASR, there are basically two approaches. I will try to sum up, starting with a 2 way xo only for my Kallax Synergy.
1) Choose target acoustic curves and match the measured ones with EQ. Here I still do not understand how do I protect the tweeter from LF without a HP - or should I e.g. use a 2nd order HP plus EQ to get 4th order acoustic slopes? How about the polarity/phase/delay this will introduce. Can this be fixed (FR, phase) with a global stereo FIR filter? Not sure how low can I get with 1024 taps and 48 kHz, need to check.
2) Make the drivers flat at least one octave around crossover point and then use a steep (48 dB/oct) LR FIR crossover. I am not sure 1024 taps will be sufficient as well. Maybe 800 Hz crossover will be possible.
All this should be done ideally slightly off axis at expected listening distance outside with dual channel measurement.
Did I forget anything in the summary?
Apart from discussing which of those approaches to use at this point, I suspect you have a limitation in your capabilities with your software. Can you create and follow target slopes? Doing so would allow you to add the effect of a capacitor to the target. It would also bypass the need to make anything flat. So this seems to be a practical issue.
If you chose to add a capacitor to protect the tweeter, you could sim it onto your target slope, or you could measure with it in place from the start.
If you chose to add a capacitor to protect the tweeter, you could sim it onto your target slope, or you could measure with it in place from the start.
I create target slopes as IR in rePhase and import them to REW. My tweeter is a relatively large planar, which plays down to 150 Hz. A capacitor will be fine, since the impedance is very flat.
I don't get what you describe as it seems to imply FIR and IIR without complementary filters ( xovers).
Whatever for IIR filters implementation : either your 2)* either you use a 'template textbook' in your dsp ( 2pole but, 4pole lr,...) and eq to reach an acoustic target as this is what matter.
You could use FIR profile to implement IIR complementary filters but i don't see the point...
*one octave linearising is ok with 4pole filters ( 24db/octave), if you target 2pole expect to linearise for 2octaves. If you use steeper FIR complementary filters you can lower range: i use half an octave linearisation with 48db/octave filters ( that said i still eq down all the coarse peak that might be present but i'm not as accurate as in the 'passband'.
The way i implement FIR complementary filter is as your 2).
Last edited:
The simplest paint by numbers approach is to flatten the drivers an octave or two outside their passbands with EQ, (the equalization can be worked out virtually so there is no risk to the drivers) then apply whatever order and type of crossover that you want electrically. The EQ and electrical crossover will combine to make the acoustic crossover the same as the chosen electrical one. If the filters are IIR then the delay will need to be set to get the correct summed response, if the crossover is linear phase the delay will not be needed to compensate for the crossover group delay but can still be useful to ensure the best time alignment.
If you use IIR filters in this way and want to linearize the phase of them afterwards it is then just a matter of using the compensate function in rephase for the type and order of filters you used.
With 1024 taps any filter greater than 4th order below 500Hz will deviate from ideal. So for 48dB 1K is the lower limit before the filter will not be quite what you wanted.
With a multiple entry horn the band pass nature of the mid will make the above approach use a lot of boosting to only be cut again afterwards. It makes more sense in that situation to set a target and use PEQ to reach it.
Ultimately it is the acoustic slopes that matter not so much how you get there.
If you use IIR filters in this way and want to linearize the phase of them afterwards it is then just a matter of using the compensate function in rephase for the type and order of filters you used.
With 1024 taps any filter greater than 4th order below 500Hz will deviate from ideal. So for 48dB 1K is the lower limit before the filter will not be quite what you wanted.
With a multiple entry horn the band pass nature of the mid will make the above approach use a lot of boosting to only be cut again afterwards. It makes more sense in that situation to set a target and use PEQ to reach it.
Ultimately it is the acoustic slopes that matter not so much how you get there.
Why do you not use the target slope feature in the EQ section of REW ?
Last edited: