• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

Dr. Klaus Heinz of HEDD Audio (ex ADAM Audio) - measuring speakers, in particular speaker dynamics

mitchco

Active Member
Joined
May 24, 2016
Messages
257
Likes
655
I have tried many preringing experiments using linear phase DSP on speakers with music playing, and like @Cosmik have not (knowingly) heard it.

Wrt to transient response of a speaker system, I have a mix of bass reflex cabs and sealed subs. Step response of my system at the listening position:

JBL 4722 with Dual F18 subs Step.jpg


I use linear phase digital XO's but the target response is mixed phase, mostly minimum phase with a dash of linear phase at the low end which produces the slight drop at the beginning of the step. Most advanced audio DSP software products comes with preringing compensation control and I have AB'd with and without while listening to music and can't really hear a difference. Makes a substantial difference in the measurement, but not audible to my ears.

Here is an example where the preringing comes into view when measuring off axis. My speakers are set up in about a 9ft equilateral triangle. I moved the measurement mic 3 feet to the left of center from the LP, and captured both the left and right speakers and again moving the mic 3 feet to the right of center and measured both left and right speakers:

JBL 4722 w Dual Rythmik F18 subs across 6ft at LP step.jpg


One can see preringing and the traces with the most preringing are across the longest distance opposite. E.g. measuring the left speaker where the mic is placed 3 feet to the right of center at the LP. I can reduce the amount of preringing by reducing the amount of bass correction in the DSP and widen the sweet spot, but given I can't knowingly hear it, I am happy with the result of the time alignment, transient response and low frequency control over time.

The only experiment I have come across about the audibility of group delay in vented speaker boxes is: http://speakerdesignworks.com/group_delay.html
 

Krunok

Major Contributor
Joined
Mar 25, 2018
Messages
3,863
Likes
1,988
Location
Zg, Cro
Heh, my step response graph looks quite different! :)
Step.jpg


Beige is taken at app 30cm from the speaker to avoid reflections. Green is taken at LP which is app 4 meters from the speakers, also arranged in equilateral triangle. I guess you can call my speakers "half-dipole" because of upward firing 2nd woofer which plays up to 1800Hz (XO point with tweeter) which probably causes all those reflections all over the room. But on the other hand, I'm quite happy with the sound.

P.S. Isn't there something strange with x-axis of your step response graph? I find it hard to believe that it takes 40ms to get to the first dip..
 

mitchco

Active Member
Joined
May 24, 2016
Messages
257
Likes
655
P.S. Isn't there something strange with x-axis of your step response graph? I find it hard to believe that it takes 40ms to get to the first dip..
Nope, the measured in-room frequency response from dual 18" subs is -3 dB at 6 Hz in my room. Looking at that from a step response perspective yields the charts posted.
 

Krunok

Major Contributor
Joined
Mar 25, 2018
Messages
3,863
Likes
1,988
Location
Zg, Cro
Nope, the measured in-room frequency response from dual 18" subs is -3 dB at 6 Hz in my room. Looking at that from a step response perspective yields the charts posted.
But the midrange response (which comes after tweeter) should drop below zero after 5ms or so.

Also, isn't it strange that after 355ms the response still didn't settle?
 

dc655321

Senior Member
Joined
Mar 4, 2018
Messages
445
Likes
373
the measured in-room frequency response from dual 18" subs is -3 dB at 6 Hz in my room
Wow.

Let's test how "textbook" that step response is:

Python:
import scipy.signal as sig
import numpy as np
import matplotlib.pyplot as plt


num, den = sig.butter(4, 6/24000, btype='highpass')
t, y = sig.dstep((num, den, 1/48000), n=16000)

plt.plot(t, np.squeeze(y))
plt.grid()
plt.xlabel('Time [s]')
plt.show()

Figure_1.png



Again, wow...
 

Arnandsway

Active Member
Joined
Sep 13, 2018
Messages
205
Likes
144
I believe the frequency response can be extrapolated from the impuls response, right? And because of this almost perfect impulse response, means your FR is extremely flat too?
What factor(s) play the biggest roll in achieving this kind of response?
 

dc655321

Senior Member
Joined
Mar 4, 2018
Messages
445
Likes
373

andreasmaaan

Major Contributor
Joined
Jun 19, 2018
Messages
3,517
Likes
2,759
I'm not suggesting for a moment that the FIR filters being discussed here generate audible ringing, but I think it should be pointed out, just in case it's not clear, that the the y-axis (amplitude) of an impulse response does not visually correlate at all well to its SPL (magnitude) or perceived loudness.

To illustrate, here is an impulse response of an FIR filter using the standard amplitude scale (FWIW, this is a 16th order linear phase high-pass filter centred at 500Hz):

1560373282741.png


It looks incredibly benign. Yet here is exactly the same impulse response using a dB scale instead (i.e. same time scale, same zoom, same everything else):

1560373223851.png


It may well be benign. Indeed I'm not suggesting it isn't. I'm just suggesting that an impulse response that looks clean at first glance may well be hiding things one may want to know about.

NB: knowing what @mitchco's filters are actually doing, I strongly doubt there is anything going on that is not benign - so this is certainly not intended as a dig at his step response, which is spectacular!

EDIT: step responses of the same impulse generated in REW:

1560374862369.png


and using a dB scale (all other settings unchanged):

1560374796163.png
 
Last edited:

Arnandsway

Active Member
Joined
Sep 13, 2018
Messages
205
Likes
144
DSP, and system capable of physically realizing what it's being commanded to
As I thought :). Every DSP software would handle it differently too, right?
@mitchco is it possible to use your uncorrected impulse response and test it with other (open source) DSP's? I am curious to what extend this is achievable with freely available software, and where the difference will be most noticeable.
 

PierreV

Senior Member
Joined
Nov 6, 2018
Messages
349
Likes
771
DSP, and system capable of physically realizing what it's being commanded to do.
I'll second that.

A lot of subwoofer manuals will tell you "put it there and play with the phase until the sound is the loudest". Measured results are sub-par to say the least. Without the pro software and skills Mitch has, I settled with the SB-16 Ultra DSP and REW and the results are quite good to my ear, with a narrow 4-5dB drop I can't seem to fix (but from what I read I probably shouldn't worry too much about it). Some subs are auto-tuning, but they are a bit on the expensive side for subs imho.

Maybe measuring and building a filter for Roon could work as well, no idea how well... I only used that path to "fix" poor speakers at this point.
 

dc655321

Senior Member
Joined
Mar 4, 2018
Messages
445
Likes
373
I'm not suggesting for a moment that the FIR filters being discussed here generate audible ringing, but I think it should be pointed out, just in case it's not clear, that the the y-axis (amplitude) of an impulse response does not visually correlate at all well to its SPL (magnitude) or perceived loudness.
I understand your point here (it's a good one!), but have you looked at what your 16th order, 500 Hz high-pass filter looks like in the frequency domain?

I think ringing will be near the bottom of any list of concerns.
 

andreasmaaan

Major Contributor
Joined
Jun 19, 2018
Messages
3,517
Likes
2,759
I understand your point here (it's a good one!), but have you looked at what your 16th order, 500 Hz high-pass filter looks like in the frequency domain?

I think ringing will be near the bottom of any list of concerns.
Ha, I'm not sure I follow! This is intended as an example of a linear phase crossover filter. It behaves virtually perfectly in the frequency domain. Or am I misunderstanding you?
 

mitchco

Active Member
Joined
May 24, 2016
Messages
257
Likes
655
I believe the frequency response can be extrapolated from the impuls response, right? And because of this almost perfect impulse response, means your FR is extremely flat too?
What factor(s) play the biggest roll in achieving this kind of response?
The frequency response is indeed flat, but tilted as per: The Science of Preferred Frequency Responses for Headphones and Loudspeakers:

JBL 4722 w F18 dual subs.jpg


And yes to @dc655321 with a few additions, which I don’t think would be new to most folks.
  • Purchase well-designed speakers (and subs if you are serious :)) that have both smooth off and on axis response as per Toole.
  • Go active if you can and use linear phase digital XO’s because: Thoughts about Crossovers
  • Purchase state of the art DSP like Audiolense or Acourate that have both frequency and time domain correction capabilities. Not all DSP software are created equal…
  • Use digital XO to crossover to sub(s) at a point that makes sense for your mains and in-between room modes. In my case, the big JBL’s easily play to 40 Hz, but I have a room mode there, so I digital XO’d at 45 Hz to my subs between two modes. Don’t blindly follow the XO at 80 Hz and leave mains capability on the table. Find the measured F3 of your mains in the room and hand-off or offload to subs around that point.
  • Only eq below Schroeder unless a) you need to add constant directivity horn equalisation as an example or b) you are using the DSP as a "tilting tone control" to tilt the high frequency response to taste. Personally, I find most speakers are way too bright for my ears.
  • Don't blow a fortune on passive room treatments. Most so called bass traps do next to nothing below 100 Hz. Source: commercially designed, manufactured and installed bass traps, diffusers, and wide band absorbers in many control rooms and critical listening rooms.
To answer @Arnandsway (edit and @andreasmaaan ) question, it is a bit complicated, as I am not using a linear phase target, but a minimum phase target. Actually mixed phase with 20% linear phase in the low frequencies. So it is not like what you are showing in post 212. While I am using linear phase digital XO's because they sum correctly in both the frequency and time domain as per the "thoughts on crossover" article above, and using a linear phase FIR filter (as a container), I am actually targeting a minimum phase response, as loudspeakers (and microphones) are minimum phase devices:

Modern DSP takes one or more measurements in one's room. The DSP software splits the result into two parts, the minimum phase portion in which the magnitude response (i.e. amplitude and phase) is corrected, and then excess phase (i.e. timing response) is corrected independently. When one corrects or eq's the frequency response, the phase response is also corrected as loudspeakers are minimum phase devices. But loudspeakers in a room, with passive XO's, etc., there is excess phase (timing) that needs to be corrected independently. @JohnPM's REW's paper on Minimum Phase is a really good read.

Here is an example tutorial of performing the above using shareware like REW and rePhase: https://audiophilestyle.com/forums/...nerate-amplitude-and-time-domain-corrections/ And the results look pretty good. Another freeware DSP is http://drc-fir.sourceforge.net/ with a big thread on diyAudio on how to use it as we get into complicated topics like frequency dependent windowing and pyschoacoustic filtering, etc. I have gone very deep on the audio DSP rabbit hole and one DSP stands above them all for ease of use plus the capability that one can get a pretty good, towards almost perfect correction in less than an hour.
 
Last edited:

dc655321

Senior Member
Joined
Mar 4, 2018
Messages
445
Likes
373
Ha, I'm not sure I follow! This is intended as an example of a linear phase crossover filter. It behaves virtually perfectly in the frequency domain.
I just meant that, depending on how that filter was constructed, the frequency response can range from quite good to WTF happened?
Sorry for the confusion and noise...
 

andreasmaaan

Major Contributor
Joined
Jun 19, 2018
Messages
3,517
Likes
2,759
I just meant that, depending on how that filter was constructed, the frequency response can range from quite good to WTF happened?
Sorry for the confusion and noise...
Ok I see! I have lost the data now, but in this case the amplitude was extremely accurate (it was the FIR filter itself, not the measured response of a loudspeaker to which the filter had been applied).
 

Krunok

Major Contributor
Joined
Mar 25, 2018
Messages
3,863
Likes
1,988
Location
Zg, Cro
I believe the frequency response can be extrapolated from the impuls response, right? And because of this almost perfect impulse response, means your FR is extremely flat too?
Actually, no. FR can, and indeed is, calculated from IR, but you can't possibly conclude anything about FR by looking at step response. Micthco's perfect step response actually means that phase is linear but says nothing about FR.

Take, for example, step response of Performa3 F208:

714R208fig7.jpg


As you can see, not only that it is nowhere as perfect as @mitcho 's but it is totally time unaligned. First peak is from tweeter, second from midwoofer and third from woofer(s). However, this speaker has very linear FR and is considered to be one of the best (if not the best) speaker in it's price range.

So, the answer to your question is this: IR response contains both, frequency and phase information so everything is derived from it - frequency response, phase response, step response, GD, distortion, .. But looking at step response you can't conclude anything about FR. As step response is happening in time domain so you can conclude only about phase (time coherence) of the speaker drivers. Let's also not forget that, as mighty as step response looks, it's significance on SQ is actually small to none. It is the FR that matters, esepecially when measured not only on-axis but on a wide range of horizontal and vertical angles, as is shown on famous spinorama graphs.

But, yes, @mitchco speakers indeed have extremely linear FR. :)
 
Last edited:
Thread starter Similar threads Forum Replies Date
Purité Audio Desperate Dealers Forum 2

Similar threads

Top Bottom