There are no pre echoes if you are listing at the design point or if you are doing earphone tests. On the other hand, if you use an LR4 crossover you could argue that the entire tweeter response of a 2-way is pre echeo as it arrives way before the response from the woofer as anyone who has looked at the impulse response of such a system would see. 🙂
By the way, just for the heck of it I made some simulations of 3 system reproducing a 100 Hz square wave. The system has a 1k Hz crossover and a 40 Hz B2 low frequency cut off. System 1 is fully linear phase. System 2 is minimum phase, that is, there is no crossover induced phase shift but the minimum phase associated with the low frequency roll off is included. System 3 is similar to system 2 buy with the Harsch crossover.
An externally hosted image should be here but it was not working when we last tested it.
wesayso, nice speakers 🙂
I have good 3 way WMTMW with 2nd order filters W-M and 4th order M-T without FIR. Only IIR with a little delay of about 0,4 ms.
Is that good result?
More info here - https://www.minidsp.com/forum/diy-hifi-projects/11942-wmmtmmw-3-way-active-speakers
Sounds much better than trivial 3 ways with LR12.
P.S. John K big thanks for your articles and ideas. Best regards.
I have good 3 way WMTMW with 2nd order filters W-M and 4th order M-T without FIR. Only IIR with a little delay of about 0,4 ms.
Is that good result?
More info here - https://www.minidsp.com/forum/diy-hifi-projects/11942-wmmtmmw-3-way-active-speakers
Sounds much better than trivial 3 ways with LR12.
P.S. John K big thanks for your articles and ideas. Best regards.
Attachments
Imho, a seamless xover is a lie or an utopia. There will always be some kind of flaw. I understand btw that some forgive their own flaws to full range drivers to get rid of an xover.
Last edited:
I think you are confusing propagation delay and the resulting phase shifts and coherent reproduction. If you want to maintain the correct temporal relationship between different frequency components in a speaker then it must have linear or zero phase shift. The phase difference between two signals of different frequency which both start at T = 0 will of course differ in time as the phase rotation is faster for the higher frequency signal. That has nothing to do with coherent reproduction though. A speaker with nonlinear phase introduces a one time incremental shift of different magnitude to each signal. That is, what ever the phase relationship is between signal A and signal B in time, say Phi(A) - Phi(B) = F(t), F(t) will be altered after the signals pass through a speaker which is not linear phase.
Anyway, while there is nothing wrong with chasing these ghosts. I simply don't find they add any sense of enjoyment when listening to music, and in fact, they are more of a distraction. I find too many audiophiles spend more time listening to how music is reproduced as opposed to listening to the music for what it is, IMO. I used to do that. Not any more. I have designed too many speakers to know that I can always make a speaker that sounds different, but different is not better. It becomes a colossal waste of time chasing the rainbow. I have better things to do.
Last edited:
YES! A basic concept/premise that still seems to fly over the head of most users evaluating linear and non-linear phase response differences in speaker systems.
Dave.
Yes!
Nowadays i just listen to the music. I have a system similar to wesayso's. Frequency response is king. Fixing the phase is now possible with FIR filters without altering anything else and my own tests showed me that i can't hear it. Toying with it while keeping the amplitude response the same is the way to go, which makes the harsch crossover particularly bad because it adds amplitude ripples, something we can definitely hear, to avert the normal
phase warp in a crossover, something we likely cannot hear.
As far as what makes the LX521 special, it's the details like carefully chosen crossover points, sculpted rolloffs, and great frequency and polar responses. I haven't heard the NaO incarnations but bet they sound fantastic as well. There is no magic here, just a great understanding of what's important and how pyschoacoustics work.
To the original poster Mistertaz, I've heard Sig's LX521, LXmini, and LXstudio, all great stuff. The LX studio - is really a fantastic system. Mr. Linkwitz does a lot of shaping in his filtering and while other driver or crossover ideas might be fun to pursue I'm not sure you'd end up with anything better but just different. To my thinking if you want to look at alternatives but stay with the basic design sticking with the LXmini and using alternatives for subwoofing would be what I would do. Just my two cents.
Looking at the time domain response is just bunk.
Your brain does not interpret the time domain response as what it is "hearing", but rather something like the Fourier Transform of it, e.g. the frequency response.
This is why the high frequency audio sampled at 44.1kHz are not utterly unlistenable and why sampled audio is completely suitable for capture and playback of audio signals for human listening. How in the world could you come to the conclusion that a 15kHz sampled signal would be "hi-fi" if you look at the time domain response of its output? It's almost a square wave with about 3 samples per period @ 44.1kHz SR. Yet we perceive a "pure sine wave" tone. Let's see the TP crowd explain that one...
A very good example of this is when you dither a very low level signal that is near or even BELOW the bit depth resolution of the sampled audio. The result looks TERRIBLE in the time domain (see plot below) but you perceive a pure tone even though (in this example) the frequency is smack dab in the middle of the most sensitive region of human hearing.
Note that the above is from a dithered 16-bit system and the level of -102dB is BELOW 16 bits!
Source: Dither and ecasound | Richard's Stuff
In my opinion, going to extremes to put TP over other design criteria is just chasing something that is so unimportant that is is almost irrelevant to what your brain perceives when you "hear" the loudspeaker.
Last edited:
No, that's not what I'm talking about. I'm talking about trying to avoid applying correction for phase changes due to the room you're in. Using frequency dependent views to look at an earlier point in time of the wave front. You can't avoid the room. I think my current FDW for phase correction is 3 cycles at 20 Hz up to 500 Hz. About half the window size I use for FR.
To get it right you need to study the room you're in, as long as you don't have an anechoic room anyway. Do not try to fix everything. Even though the tools allow you that freedom, it's not going to work.
Now why do I do all of this? Because it gets me closer (in)to the music.
I have no real shot at analyzing anymore once the music takes over. You're just enjoying listening from that point on.
But it's cool to learn something too. I see many people claim: do this for this type of music and do that if you like that genre. Know what? The better the speaker plus room behaves as a whole, the better every piece of music and every genre sounds. Phase is just a tiny piece of that puzzle.
This quote above, while well written, has nothing to do with what I was suggesting or getting at. Which should be obvious if you look at the APL_TDA plot I showed.
A pure loop of my DAC to APL_TDA looks like this:
Compared to my speakers+room at the listening position:
By the way, I tried cutting off at 40 Hz a while ago, I'm loosing way too much atmosphere if I do that. So my -3 dB point is 17 Hz, while I cannot really sustain any true notes at high dB at that point, at least not 110 dB peak. This hasn't presented itself as a problem while listening to music, but for home theatre at high levels I should use subwoofers.
While the actual effect of the transient response of a speaker is definitely arguable - I can't see what the effects of a speaker's transient response and dithering could have in common. 😕
Regards
Charles
Look at this time domain response (the top panel) of a 3-way system using LR4 type crossover filters. Wow, that must sound horrible, right?
An externally hosted image should be here but it was not working when we last tested it.
A main objection that I have heard from the TP crowd is that this MUST SOUND BAD because of the waveform is not really like the input signal, a nice square wave. The objectors then identify "non linear phase" which is something that they can quantify with "must sound bad" based solely on observations of the signal's time domain response. Forming this kind of relationship in that way is just wrong because it is based upon the false pretense that you "hear" the time domain signal.
My dither analogy was presented to show a waveform that looks "bad" in the time domain but is known to sound pure as an example of why you cannot look at the time domain behavior (alone) to decide how a signal will sound. In this case of the dithered signal it's not frequency dependent delay or non-linear phase that is the source of the odd looking time domain signal, but the conclusion is the same: the time domain is not what your brain is perceiving as "hearing".
Last edited:
Well, you are presenting a steady state tone argument. Sure that will sound like a tone but what TP designs provide is the transient attack that is clear and sharp. I know many people say that they can't hear phase, but how phase impacts the attack of a sound from a piano, plucked instruments like classical guitar or stand-up bass, tom toms, rim shots, all sound more realistic to me. I can tell the difference. Maybe I should set up a virtual blind test with sound clips from a TP vs a 4th order LR for percussion and see if people hear the difference with headphones?
So if we look at what the transient sound of the attack portion of a percussive strike, we are no longer looking at a continuous wave train where the brain can pick up on the FFT and distill the frequency. Really, I do agree that the ear works as a bank of about 32 FFT processors (the tuned cilia which vibrate a particular frequency). But they are also designed to hear a sharp crack of a rim shot or pool cue ball hitting the rack differently if the sound originates from a signal like the sharp little spike inside the red box vs a solid square wave edge like the lower trace. There is a perceptible difference.
So if we look at what the transient sound of the attack portion of a percussive strike, we are no longer looking at a continuous wave train where the brain can pick up on the FFT and distill the frequency. Really, I do agree that the ear works as a bank of about 32 FFT processors (the tuned cilia which vibrate a particular frequency). But they are also designed to hear a sharp crack of a rim shot or pool cue ball hitting the rack differently if the sound originates from a signal like the sharp little spike inside the red box vs a solid square wave edge like the lower trace. There is a perceptible difference.
Attachments
Last edited:
It's a form of distortion indeed, audible, or not, or under some circumsatances...🙄
The FD squad is also well known to overlook what they consider irrelevant.
Mandingo fights as usual, and both need to ask their psy...
Anyway, it's very easy to build with FIRs a real min phase LR24 and its linear phase equivalent and compare. Frequecy reponse, polar response, and power response will be the same the only difference being TP, and of course for the FD squad, off axis pre response, veeeery bad thing, no doubt.... No problemo for me, though...
Last edited:
Well, you are presenting a steady state tone argument. Sure that will sound like a tone but what TP designs provide is the transient attack that is clear and sharp. I know many people say that they can't hear phase, but how phase impacts the attack of a sound from a piano, plucked instruments like classical guitar or stand-up bass, tom toms, rim shots, all sound more realistic to me. I can tell the difference. Maybe I should set up a virtual blind test with sound clips from a TP vs a 4th order LR for percussion and see if people hear the difference with headphones?
So if we look at what the transient sound of the attack portion of a percussive strike, we are no longer looking at a continuous wave train where the brain can pick up on the FFT and distill the frequency. Really, I do agree that the ear works as a bank of about 32 FFT processors (the tuned cilia which vibrate a particular frequency). But they are also designed to hear a sharp crack of a rim shot or pool cue ball hitting the rack differently if the sound originates from a signal like the sharp little spike inside the red box vs a solid square wave edge like the lower trace. There is a perceptible difference.
I agree that the 3 way, with that horrible looking square wave won't sound that bad. But it doesn't give the bite of a more time aligned system. The drum hits as X mentions have way more to it than that sound alone. At lower frequencies we feel it as much as we hear it. Getting that in time is much like getting sound and picture in sync.
When a simple triangle is hit in front of you and you get the pressure wave at the same time (you can feel it on your eye lids) that's what you get with time coherency. The closer you get to it, the more "real" that presentation becomes.
I used to think it would do marvels for imaging, in things like depth etc... but it doesn't.
There is good work of Manger.
I was thinking also that phase matters much but some tests make me believe that polar and energy is important more. FIR filters with fast role off was not sounding better than IIR version with XO like harsch.
I think there is some things that are before TP like FR, thd, polar resp. But if all OK with them tp types of filters will be better.
Also such filters may eliminate baffle effects due to other sum out off axis.
I was thinking also that phase matters much but some tests make me believe that polar and energy is important more. FIR filters with fast role off was not sounding better than IIR version with XO like harsch.
I think there is some things that are before TP like FR, thd, polar resp. But if all OK with them tp types of filters will be better.
Also such filters may eliminate baffle effects due to other sum out off axis.
As always, "it depends". Sure, at some point if you take a signal, decompose it into some frequency bands, and then start to move those apart in time (introduce group delay differently for different bands) at some point you will DEFINITELY hear it!
What I am arguing is that, in the above example, the time differences are too small for you brain to notice. Now a "purist" would say "well if I can hear large delays then I don't want ANY delay" and they would pursue the TP approach because they would feel better that there are no group delay differences. Taking an engineering approach, you only need to do what is necessary to achieve your goal and anything more is added expense or frivolity. In that case you can use the fact that they brain lumps together the first couple of milliseconds of the time domain and is continually doing this as part of the hearing process. This explains why there is a threshold of audibility to group delay, which at it lowest is on the order of 1 msec at 2kHz, and this grows significantly towards the low and high ends of the audio spectrum. See this page for more info on that:
https://www.trueaudio.com/post_010.htm
So, my assertion is that the group delay that is typically introduced by the types and orders of crossovers that are presented by the TP supporters as evidence of why you need to use TP just is not enough group delay to be significant to the hearing process. You certainly can design high order crossovers having peaks in the group delay that cross into the audible realm for group delay but this type of offense is not typically what I observed being discussed.
Correcting for group delay has its usefulness in some (rare) circumstances, but doing overall correction so that the system is TP is really overkill, and IMHO is trying to solve a non-problem.
I can see on the plot that some of the energy is sooner than other parts, it doesn't really flow just yet. What does the DFR plot look like? And the accompanying FR plot? I'd try to get it to look more even, letting each driver hand off to the next. You'll get a bigger sense of rhythm if you get that part right. As any musician if they think timing is important 🙂.
Something like this would have my preference:
See how it gets gradually wider the lower in frequency you go? But smoothly without obvious quirks?
Frequency response is king, I can't argue with that. Power response is very important too, no argue there either. But I'm convinced phase plays a bigger role than many give it credit for. This has become clear in many experiments I have done so far. I did not spend months trying to figure out how to get the best "looking" IR or FR response. You can get that in a day. Sadly that doesn't work and won't sound any good either. It took me way longer to figure out WHEN you want the FR and phase to be correct. Frequency dependent windows are key to look a bit earlier in time. Correct it there and the response in the usual windows will start to look better too. No avoiding the room though, so work with it.
Now I know I "look at things" a bit different from most who have been at it longer than I am. That doesn't mean I did not pay attention to what you guys (you know who you are, and of coarse that includes John K) had to say.
Ask yourself when you want the frequency response to be right. To make the prettiest plot or when you (get to) hear that wave front with your ears?
Every signal arriving within the Haas limit will add to the perceived tonal balance, so that makes power response a pretty important factor for tonal balance. You can even use it (the Haas limit) to your advantage for things like depth perception.
You can easily try and test that yourself.
But the way we humans perceive direction and distance we is all done within a very small time frame. FR isn't the only player there. We can actually analyze extremely short time differences to get that info. Give your brain more credit!
Though we seem to be better at direction vs distance with our ears. At least I am.
Play with the concept of how we hear and you'll get ahead.
Last edited:
No human, musician or otherwise, can resolve 1 milllisecond time differences unless those time differences result in frequency response changes (for example because the delay causes interference between two or more sources that leads to peaks and dips in the FR). This is just a fact of human hearing physiology...
Within a couple of milliseconds, everything is at the same "time" for humans.
I suggest that you might try to re-color your plot (if that is even possible), shown below, to highlight where the audibility threshold for group delay is exceeded, and by how much. That kind of presentation is what is important, not the value of group delay alone.
- Status
- Not open for further replies.