The phase of the La Scala folded horn shifts rapidly at its Fc, (horn cutoff frequency) of ~58Hz.
I don't recall you mentioning how the subwoofer is combined with the La Scala, but if they are in driven in parallel, you will probably be stuck with the phase cancellation in the acoustical overlap range.
The passive radiator output behaves like a port in a bass reflex cabinet (AKA phase inversion cabinet) and it's output at Fb (box tuning frequency) lags ~180 degrees behind the driver output.
The group delay reflects the phase inversion.
http://sound-au.com/ptd.htm
Regarding the fan noise in your FIR DSP 408, you might find draping it with some breathable fabric (cotton tee shirts or similar) will knock the noise down to an acceptable level while still allowing adequate air flow.
Worked for me on amplifiers..
You wrote the noise floor of your inner city Philly home is 37dBA, the fan 44 dBA at .5 meter, which should be ~38dBA at one meter, so it shouldn't take too much fabric to bring it down to ambient.
At any rate, the fabric will make it sound more like pink noise than white noise
Art
There’s no subwoofer, I did my take on the late DJK’s ported LaScala, except I increased the upper cabinet volume by 95%, coupled it to the doghouse and instead of ports, I used a 15” passive radiator, which in junction with the woofers I selected, extend to 20 Hz. I can live with the group delay as it’s not in the audible spectrum of most music
t’s getting close. There’s still some phase issues at the crossover points, at 400 Hz and 5kHz, but it’s much better than previously.
alignment is within .05ms or less through most of the range, but it gets whacky in the bass bin. I suspect that's partly due to the passive radiator and partly due to reflections, I did't bother to cover the floor when I did my measurements tonight. Next season, I'll see if that makes a difference.
Naturally, group delay in the bass bin is not great.
next up is to do the same to the left channel. The right side is in a corner and there's a substantial gain disparity between the two, about 6 db judging by ear. Next session, I'll flatten the eq on the left side and generate a separate PEQ curve for that side. It's already better sounding than I imagined possible. Thank you for your generosity, @Cask05 !
I'll also look into replacing the fan with a quieter model this week; If I can get the fan quiet, the Thomann will be a keeper, with Xilica level features at a fraction of the cost.
alignment is within .05ms or less through most of the range, but it gets whacky in the bass bin. I suspect that's partly due to the passive radiator and partly due to reflections, I did't bother to cover the floor when I did my measurements tonight. Next season, I'll see if that makes a difference.
Naturally, group delay in the bass bin is not great.
next up is to do the same to the left channel. The right side is in a corner and there's a substantial gain disparity between the two, about 6 db judging by ear. Next session, I'll flatten the eq on the left side and generate a separate PEQ curve for that side. It's already better sounding than I imagined possible. Thank you for your generosity, @Cask05 !
I'll also look into replacing the fan with a quieter model this week; If I can get the fan quiet, the Thomann will be a keeper, with Xilica level features at a fraction of the cost.
t’s getting close. There’s still some phase issues at the crossover points, at 400 Hz and 5kHz, but it’s much better than previously.
alignment is within .05ms or less through most of the range, but it gets whacky in the bass bin. I suspect that's partly due to the passive radiator and partly due to reflections, I did't bother to cover the floor when I did my measurements tonight. Next season, I'll see if that makes a difference.
Naturally, group delay in the bass bin is not great.
next up is to do the same to the left channel. The right side is in a corner and there's a substantial gain disparity between the two, about 6 db judging by ear. Next session, I'll flatten the eq on the left side and generate a separate PEQ curve for that side. It's already better sounding than I imagined possible. Thank you for your generosity, @Cask05 !
I'll also look into replacing the fan with a quieter model this week; If I can get the fan quiet, the Thomann will be a keeper, with Xilica level features at a fraction of the cost.
One thing that you'll find within REW is the "Generate minimum phase..." button under the Actions menu on the upper right side of the SPL/phase plot:
When you select that option, REW will calculate the Hilbert transform of the amplitude (SPL) response, then compare that (minimum phase) curve with the total phase curve that was measured (i.e., "excess phase"). The excess phase curve shows you the all-pass contributions of the crossover filters and any movement toward or away from the microphone of the effective acoustic centers of the drivers themselves vs. frequency. For instance:
When using IIR filters only, you can't really do better than a zero excess phase curve as shown by the white line plotted above for a stock La Scala II (tri-amped).
The same thing is true for the numeric first derivative of the excess phase curve--i.e., excess group delay (also generated by pressing the "Generate minimum phase..." button within the group delay plot window, then selecting the excess group delay curve plot button below the plot):
Since numeric derivatives are noisy processes (as compared to numeric integration - like step response from time-based impulse response), you have to take some care during in-room measurement to minimize early reflections, and use curve smoothing to average out the resulting calculated group delay curve to sweep away generated numeric noise.
In the plot above, excess group delay curve tells you how well the dialing-in process was accomplished. Note that group delay audibility is sensitive to frequency, so at midrange frequencies (~300 to ~2 kHz), group delay audibility threshold is on the order of 1 millisecond. For lower frequencies, the audibility threshold is something like a function of a wavelength period:
Chris
Last edited:
Always good to recall that audibility limits are estimates of an average for a population. IOW, they are an estimate of the center of a bell curve. What they are not, is they are not hard limits.
Therefore it might be good to aim for a couple of standard deviations below the limit. That way you cover well more than the limit for only 50% of the population.
Therefore it might be good to aim for a couple of standard deviations below the limit. That way you cover well more than the limit for only 50% of the population.
Last edited:
So for bass frequencies well below 100 Hz, as long as the group delay is below something like ~50 ms, you're probably not going to hear the effects of it. That's a key insight. Excess group delay tells you how much time misalignment you have between drivers in a multiway loudspeaker. You can read this misalignment in milliseconds as a non-zero group values (as I've discussed and illustrated before, on this forum and on the K-forum).
But group delay in general also tells you about the performance of individual drivers (and horns), and also about the effects of the cut-off frequency of the bass driver(s). If you have a high cutoff frequency in a bass bin--that alone cause high values of group delay at bass frequencies, due to the changing slope of the phase curve vs. frequency near cutoff.
So if we can hear the rate of change of phase vs. frequency in loudspeakers (i.e., group delay/excess group delay), do we also hear shifts in relative phase vs. frequency in loudspeakers? The answer is "yes", we do. So what is the threshold of audibility of phase distortion vs. frequency (i.e., relative phase distortion)? It turns out that this subject is not very well documented. The best examples I've found are actually from Dave Griesinger, who is actually talking about clarity in listening spaces, like concert halls. The absence of early reflections in-room is equated with clarity. For me, it isn't much of a stretch to convert these clarity discussions into audibility of relative phase shifts. Here is a two-slide excerpt from that .ppt presentation that illustrates the point:
Relative phase shifts in loudspeakers are audible. Large excess phase values destroy the "spikiness" (i.e., clarity) of waveforms due to the shifts in relative phase of the harmonics being reproduced by the loudspeakers.
What is the threshold of audibility of these phase shifts in loudspeakers? Certainly, it's limited by the early reflections in-room that we allow to clutter our hi-fi reproduction. If you clean up the early reflections via better full-range loudspeaker directivity (down to the room's Schroeder frequency in small rooms) and treat the areas just around the loudspeakers and the listener's ears with absorption to decrease their relative levels, then you have cleaned up the listening space for hearing loudspeaker phase shifts.
So what is the absolute threshold of audibility of phase shifts in loudspeakers/small listening rooms? I strongly believe that an approximate threshold of ±90 degrees is a good approximation (i.e., a loss of -3 dB of amplitude in two summed simple-frequency sources of 90-degree differing phase).
The uncertainty in this value is the root of the problems that we have communicating with each other about this subject. Until this is psychoacoustically measured, I believe that the debate on the subject will continue to fester. Clearly, the debates on why there is an audible difference between linear phase loudspeakers and nonlinear phase loudspeakers rest on this subject, along with how well treated the listening room is for early reflections, and the full-range directivity control of the loudspeakers happen to achieve in-room.
Until the definitive research paper is published on this subject (which I haven't seen yet), I think that the threshold of audibility will continue to be a bit anecdotal in nature. This doesn't mean that we should ignore these phase shifts altogether, or take on the opposite extreme of saying that "any phase shifts are audible". Both of these extreme positions are not warranted.
Perhaps we could settle of a conservative value of ±55 degrees over a decade of frequencies as a reasonable gentleman's agreement of what is just barely audible phase shifts...and below which are not audible.
YMMV.
Chris
Last edited:
You may not mean an "absolute" threshold of hearing, since they are estimates of an average for a population too. Absolute limits are an estimated limit for single tone detection by the average ear, which would make no sense in case of group delay.
Over at ASR they define something they call "strict" limits, which they hope is a hard limit for every human on earth. Of course there is no way to measure the whole population of the earth to get a snapshot of that at some point in time. Therefore, so far as I am aware there are no published studies on so-called strict thresholds.
Also, one of the things that makes group delay audible is that summing two or more sine waves produces beat notes, which are not new frequencies. Rather they are amplitude-envelope modulations. When a change in the amplitude envelope of a signal due to phase shifts is noticeable, its an example of audible group delay. According to published findings a few people can hear level changes of .1dB or less, so it may not take much phase shift to be audible for very sensitive people. It is also possible for frequency modulation to become audible with particular phases shifts of summed sine waves.
Last edited:
The desire for linearity in an audio system is kind of good, but up to a certain level. The higher the linearity of the speakers, the easier it is for the brain to hear the kitchen of recording and mastering a musical composition. Roughly speaking, the sound from the speakers becomes analytical, you can hear everything that the sound engineer wanted to mask. The music becomes dry and very analytical, the intelligibility of sounds is so high that if desired, it is not difficult to distinguish one track from another track of the recording. Those compositions that were previously catchy become neutral, and so on and so forth. So if there is a need to enjoy listening to your favorite compositions, you have to reduce the linearity of the speakers by one method or another.
If that's the case then you're doing something wrong. A system can be highly resolving, low distortion, a pleasure to listen to, and not dry nor analytical. That's because it isn't only about SINAD and other things an AP can measure at the touch of a button.
For example, here are some things found to audibly affect the sound of a professional digital mixing console:
In addition, a lot of work by @KSTR over at ASR showing the time-domain waveform of Hump Distortion in ESS dacs. Many people complained about the sound of such dacs and were told they were imagining something that isn't real. Now even ESS admits the new Hyperstream 4 modulator sounds significantly better than the old Hyperstream 2, that despite the fact that AP measurements are pretty similar in terms of noise and distortion (both modulator types measure very well in standard measurements).
Last edited:
Well, yes, it is, I leveled the frequency response of the speakers in the listening area at +/-1dB with 1/24 octave smoothing with a flat phase, and heard what I could not hear before. It should be understood that I am not saying that a linear system sounds bad, I am saying that the more linear the system, the easier it is to hear the shortcomings of the recording if there are any in it. Modern compositions are much better made, and they don’t hurt the ear as much as the old ones. Also, on a linear system, all sorts of software sound colors become cloyingly sweet and some Internet radio stations begin to irritate with their improved sound. In general, the higher the linearity of the system, the higher the requirements for the quality of the source musical material.
That can be easy to fix when listening to lesser quality recordings. Try changing one coupling cap in the system from something very linear like polystyrene or polypropylene to mylar instead. Often that's enough to fix the problem. Its also easy to switch back when listening to better recordings. BTW its an old trick; nothing new.
I have a three-way digitally active speaker, a powerful DAC directly powers the speaker, in my case it is much easier to introduce “nonlinearity” on the side of digital sound processing.
I just shared my observation - in pursuit of linearity of the frequency response, the requirements for the source recording are increasing.
Chris,
I've got it sounding really, really good now, but following the rubric, "perfect is good enough," there are still measurable phase/alignment anomalies at roughly the woofer/mid crossover point:
SPL+Phase
SPL:
group delay:
Spectrogram:
this is what my crossover looks like:
Here is my woofer response:
factory measured mid response:
and lastly, mid response with PRV WG45-50 18"w x 10"horn, in red, courtesy of @THD+N :
@Cask05 , what can I do to reduce the anomaly between 375 Hz and 600Hz? is it worth trying and will I be able to hear the difference? presumably, this is caused by driver overlap?
The spectrogram shows that the woofer channel could be a little too advanced in time alignment...by perhaps something like 2 ms.
The minimum phase curve should be somewhat growing from high frequency to low. You will also find that the excess group delay curve, when zoomed-in to no more than 1 ms minor subdivisions vertically, will tell you the same thing--only with a bit more fidelity than the spectrogram view.
I would recommend trying reduced midrange and tweeter channel delay settings, by equal increments, to push the bass bin time alignment back by ~2 ms.
When attempting to flatten the phase response across all channels in a loudspeaker using DSP, it's not always clear how far one can go to reduce the lower frequency driver phase and excess group delay growth. If you go a little too far, it starts to show up in the step response plot, with the low frequency drivers leading the higher frequency drivers slightly. Usually when this occurs, there are also disturbances in the SPL response curve smoothness at the crossover transition band.
To find out if this is occurring, just reduce the effective advancement of the lower frequency drivers a bit, then look to see if the SPL, phase, and group delay responses smooth out a bit.
Also, you can try flopping the woofer channel polarity (thus bringing the woofer and midrange channels 180 degrees closer together) and then readjusting the fine-tuning on the upper frequency channel delays to match the SPL, phase and group delay values through the crossover transition band.
You can also move the microphone position slightly to see if there is a better location to measure the combined output of the loudspeaker.
Chris
The minimum phase curve should be somewhat growing from high frequency to low. You will also find that the excess group delay curve, when zoomed-in to no more than 1 ms minor subdivisions vertically, will tell you the same thing--only with a bit more fidelity than the spectrogram view.
I would recommend trying reduced midrange and tweeter channel delay settings, by equal increments, to push the bass bin time alignment back by ~2 ms.
When attempting to flatten the phase response across all channels in a loudspeaker using DSP, it's not always clear how far one can go to reduce the lower frequency driver phase and excess group delay growth. If you go a little too far, it starts to show up in the step response plot, with the low frequency drivers leading the higher frequency drivers slightly. Usually when this occurs, there are also disturbances in the SPL response curve smoothness at the crossover transition band.
To find out if this is occurring, just reduce the effective advancement of the lower frequency drivers a bit, then look to see if the SPL, phase, and group delay responses smooth out a bit.
Also, you can try flopping the woofer channel polarity (thus bringing the woofer and midrange channels 180 degrees closer together) and then readjusting the fine-tuning on the upper frequency channel delays to match the SPL, phase and group delay values through the crossover transition band.
You can also move the microphone position slightly to see if there is a better location to measure the combined output of the loudspeaker.
Chris
Chris, @Cask05, thanks. 2ms was a bit much, but 1ms made the sound somewhat more coherent, but threw off the eq and driver gain. More measurements with flat eq will be required to tune it in before I move to the left channel, but boosting the non-corner loaded left side by 6 db has put it in the ballpark. Incidentally, I googled after the fact and found that corner loading adds about 6dB of gain, so my intuition was not misplaced. More measurements this weekend and hopefully I’ll be a bit closer.
While having my morning coffee, I pulled the unit and removed the cover. There doesn't seem to be much need for a fan for home use, so I've pulled it entirely.
In the coming weeks, I'll also construct a 1U rack enclosure for it, with a vented grille on top for optimum passive cooling.
In the coming weeks, I'll also construct a 1U rack enclosure for it, with a vented grille on top for optimum passive cooling.