HOLMImpulse: Measurements in practice

"Suspicious" first measurement

I took the attached measurement earlier of this (not complete) speaker:

http://www.diyaudio.com/forums/showthread.php?s=&threadid=139174

As my first measurement ever. I built recently a mic with WM-61A capsule (I have just realised, I actually forgot to apply the .cal file :eek: ) but to me it doesn't seem right.

The peaks in the upper ranges are explainable, I haven't set the levels perfectly yet, it's a small room with walls near the speaker and the mid and tweeter arn't recessed (and no cal). But what I find odd is basically how perfect that bass response is :confused: . It's a ported box woofer, higher than 0.7Q, tuned around 37Hz and in a small bedroom; how is it reaching 20Hz flat and with no "boom" at all? I admit, the bass sounds suprisingly good to the ear, but I'm sure it isn't reaching 20Hz. What error could cause this?

Also the phase inversion at 2k is odd, the crossovers are 820Hz and 4.65K.

I think it is feeding back at the moment too, not actually ringning or self oscillating (of course!) but the mic picks up sounds that are reaching the speaker. I think if I pan the mic to one channel only I can avoid this.
 

Attachments

  • speaker response closeup.png
    speaker response closeup.png
    12.2 KB · Views: 1,292
Comparison with LAUD

dlr said:
The problem I have is that below 1K, I don't get good agreement. It likely has to do with the window as this area changes more with the window selected. I believe that I used a half Blackman-Harris in LAUD. It would be good to have options for type of window (or have I missed that?).
Overlay of LAUD and HOLMImpulse (using Praxis for display only):
sb_acoustics_LAUD_vs_HOLM.gif

Dave

I actually think that the agreement is quite good.

Yes the time-window is essential - especial for low frequencies. I have added your feature request for symmetric classical time windows like Blackman-Harris to my list. I don't use them, because I want further flexibility.

Sorry for my late reply...
 
Re: "Suspicious" first measurement

Dr.EM said:
I took the attached measurement earlier of this (not complete) speaker:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=139174
As my first measurement ever. I built recently a mic with WM-61A capsule (I have just realised, I actually forgot to apply the .cal file :eek: ) but to me it doesn't seem right.
I think it is feeding back at the moment too, not actually ringning or self oscillating (of course!) but the mic picks up sounds that are reaching the speaker. I think if I pan the mic to one channel only I can avoid this.

Let me see your impulse response with time window also.
Easily save both graphs to one image (file.png) by:

File > 'Save Frequency- & Impulse-Response'

Also try to measure with and without microphone calibration enabled
And keep both measurements in the graph - this way we can see what goes wrong/right with the calibration
 
Thanks for replying! I've made new measurements but frankly they are even more confusing!

Blue is with cal, red is without. If I check the "invert recorded signal" box, the frequency data appears unchanged but I get 2 phase inversions. I think the way I have it here is "correct" (I don't have an absolute reference, my mixer used for connecting the mic could invert many times for all I know).

This just doesn't seem right still. The perfect flatness below 500Hz, and that large peak and lack of highs, it doesn't correlate to what I'm hearing
:confused:
 

Attachments

  • speaker response new.png
    speaker response new.png
    24.2 KB · Views: 1,224
Sorry to double post, but I think this may help. I prsent measurements of the drivers isolated, but connected through the crossover network.

An externally hosted image should be here but it was not working when we last tested it.


It makes a little more sense, driver levels arn't set accurately yet (it's one of several things this software can help me with). The woofer response is a complete mess, which could very well be the box it's in (subwoofer box, no material inside), but isn't it odd how on the overall system response the woofer region shows as perfect :confused:

The mid also appears to roll off somewhat early for a 4.65kHz crossover point. The driver naturally should be flat to 10k+

I also need to measure in a much larger room. Should this make a big difference? At the moment walls are unavoidably nearer to the speaker than the mic, a bad situation. Mic is about 1m away, pointing at tweeter centre.

All taken with cal on. If these are accurate, clearly the crossover needs some adjusting to correct those peaks. Again, hopefully if I can rely on the measurements this software will help me to do it :)
 
Sorry, I cannot help much.
I think you had this result in the low frequencies because of your window gating. It was quite high (about 1KHz).
If you move your cursor away from the IR peak, you can get more "normal" results.
But you will also include some room effect...
I don't know exactly what is a good compromise...
Your last measurements seems believable.


I also have my questions:
I am trying to time-align the tweeter with the woofer in a 2 way speaker.
The crossover is temporary (1st order high and low pass).
Here is what I think should be the best alignment:
"when both IR peak from the woofer and from the tweeter have a top at the same time"
But I am really not sure about this assumption...
I took the 0 from the Bass as a reference and then moved the tweeter until I had this:

An externally hosted image should be here but it was not working when we last tested it.


Woofer=blue
Tweeter=red
Total=green


Here is what I don't understand about this measurement:
1- Is my assumption about time aligning drivers right ? (matching IR's peaks)
2- Why is the woofer IR's peak so low compared to the one from the tweeter ?
3- Shouldn't be the phase of both drivers closer if they are correctly time aligned?

I have much more questions but if you can help answer these three and understand the reason, that would be great...
Thank you for your comments!
 
It's pretty hard to eyeball alignment of impulses. How would one align the impulses so that they are within 0.01ms? Normally, I shift the window so that the measured phase is as close to the calculated minimum phase as much as possible to determine distance of mic to acoustic source, then calculate where the acoustic center is in reference with the driver mounting surface. This information is used as the bassis for physical alignment.
 
Dr.EM said:
Sorry to double post, but I think this may help. I prsent measurements of the drivers isolated, but connected through the crossover network.
An externally hosted image should be here but it was not working when we last tested it.


It makes a little more sense, driver levels arn't set accurately yet (it's one of several things this software can help me with). The woofer response is a complete mess, which could very well be the box it's in (subwoofer box, no material inside), but isn't it odd how on the overall system response the woofer region shows as perfect :confused:

When you show the impulse, then please zoom, so that we can see the impulses with reflections/echoes and the timewindow in detail. Try Auto Zoom or zoom in yourself using the mouse. Also show the phase.

Yes the LF response is not valid (See the gating frequency)
 
domtw said:
I also have my questions:
I am trying to time-align the tweeter with the woofer in a 2 way speaker.
The crossover is temporary (1st order high and low pass).
Here is what I think should be the best alignment:
"when both IR peak from the woofer and from the tweeter have a top at the same time"
But I am really not sure about this assumption...
I took the 0 from the Bass as a reference and then moved the tweeter until I had this:
An externally hosted image should be here but it was not working when we last tested it.

Here is what I don't understand about this measurement:
The measurements are a fair beginning :up:

1- Is my assumption about time aligning drivers right ? (matching IR's peaks)
Yes, but be aware that you have to use the same time-offeset in HOLMImpulse. Since version 1.2.0.0 there is a "lock time zero" functionality.
The Phase will also show if the driver are time-aligned
2- Why is the woofer IR's peak so low compared to the one from the tweeter ?
Because there is no high frequencies... The area under the curve is though large due to the long periods. That is also why you don't hear much woofer when just playing a Dirac sample. So we need to make the logsweep... If you don't plot the mid and the tweeter and click Auto Zoom you will see the impulse of your woofer.
3- Shouldn't be the phase of both drivers closer if they are correctly time aligned?
Depends... Usually the mid, woofer must be put in front of the tweeter.

I'll post a how-to-time-align-measure a 3-way system soon...
 
askbojesen said:




Yes, but be aware that you have to use the same time-offeset in HOLMImpulse. Since version 1.2.0.0 there is a "lock time zero" functionality.
The Phase will also show if the driver are time-aligned

...

I'll post a how-to-time-align-measure a 3-way system soon...
I'm sure lot's of DIYers are going to be confused, because time alignment could be off quite a bit, but the phase would not look too bad unless it's compared with a minimum phase plot, and the phase of a system cannot correctly show time alignment.
 
askbojesen said:

Yes, but be aware that you have to use the same time-offeset in HOLMImpulse. Since version 1.2.0.0 there is a "lock time zero" functionality.
The Phase will also show if the driver are time-aligned
Yes, the measurement I posted uses the lock time zero function, the reference being the woofer.
What does the phase should show ?

Because there is no high frequencies... The area under the curve is though large due to the long periods. That is also why you don't hear much woofer when just playing a Dirac sample. So we need to make the logsweep... If you don't plot the mid and the tweeter and click Auto Zoom you will see the impulse of your woofer.
Ok, got it...


I'll post a how-to-time-align-measure a 3-way system soon...
That would be really nice...
Thanks for your answer and your software!
 
Ask

Two days of intense measurement - love the software, still has some bugs. It always writes headers in the impulse files for example (and I'm still running the "debug" version).

It can be a real challenge to get the right sound drivers, sample rates and measurement lengths to avoid getting "bad" data. I had to play arround quite a bit to get rid of some real bad distortion. But I did get it to work and the impulse responses are very clean, easy to see real data from the noise that I used to get from my system.
 
soongsc said:

I'm sure lot's of DIYers are going to be confused, because time alignment could be off quite a bit, but the phase would not look too bad unless it's compared with a minimum phase plot, and the phase of a system cannot correctly show time alignment.

I think I really need to read that book to really understand that. ;)
But the matching IR's peak seem to be a good start...
 
Phase linearisation of a multiway speaker

A HOLMImpulse application: Phase linearisation of a multiway speaker.

Measure you speaker in slot A, it's nice to have your impulse zero properly aligned (adjust the first visible peak to time 0) but that is not critical (it is a bit critical when you correct two speakers not only one, with different time alignemt you will get a slight time offset between speakers -- not a big deal if you finally re-center your sweetspot with pink or white noise when all is done).

Export impulse response from -10 to 65525 (65535 samples total).

Re-import impulse response to slot A (move original measurement if needed), re-adjust zero if needed to match the original one).

Copy slot A to an empty slot B.

Apply smoothing at about 1/1-octave to slot B.

Export frequency response, custom format, tab delimited, no header, all FFT frequencies, to a text file. Open it with Windows Editor or Wordpad.

Open Excel or Open Office Calc and paste textfile (that's why it must not be longer than 65535 rows). Open office still moans but imports everthing correctly.

Set all magnitudes to 0 (st 1st on to zero, copy it (CTRL-C), go down one cell, press CTRL-SHIFT-CURSORDOWN, then CTRL-V)
Invert all phases: copy column C to D, edit C1 to "=-D1", copy it, etc (continue as above).

Mark first three rows, copy and paste into textfile & save (different name, preferably).

Import that textfile as frequency response in slot C.


Then you should have what is shown in the top plot
blue: original response (here: 1/20oct smothing), those are my 3way JBL floorstanders.
brown: smothed response, to get the overall phase curvature
green: inverted phase response, note magnitude is 0dB everywhere.

Next export this green response into a .WAV file, containing the impulse reponse needed for correction, to be used with a realtime-convolver like foobar has one. You have to merge the left and right correction files into a stereo file.

To show the effect of the correction, set slot A to the original response and slot B to the phase correction, and select an empty slot C.

Apply the processing C = A * B.

Then (after disabling slot A and swapping B with C in the view) you should get the picture as shown in the bottom plot:
green: phase correction, as above
brown: phase linearized response.
Note that the phase still contains the minute deviations as in the original response but the overall curvature, from the allpass response of a typical box, is factored out. As is the bottom roll-off, to some extend (my measurement there was crude, no good data at LF). Note: if you plan to have good correction at LF, you'd need more than those 10 trailing samples when exporting the phase correction IR to a .WAV. Check view first how much trailing is needed, can quickly be a few hundred).

If you don't smooth the FR used for correction, you would get a linphase-equalized response but this usually doesn't sound as good as when the inherent minute phase ripple (usually minphase part of the minute response details) remains untouched.

In a similar way the magnitude response can be smoothed/flattened, too, and if you happen to get pretty equal impulse responses for both speakers in the first 20ms or so this will bring the stereo qualities of your system a great step forward. Best results are achieved when one applies a sliding lowpass to the IR first, to lessen of the influence of strong reflections, room-modes and noise. Audition's "Dynamic EQ" does it pretty well as long as Ask is still about to implement it (or a similar functionality). All this together is sort of a crude digital room equalisation like DRC or Acourate do it, way more sophisticated but still the same in base principles.

With this simple example (in no way optimized, just a quick first try, eg the 64k FFT export is ovekill. Maybe log-spaced to 100pts/oct would do, haven't tried so far) my goal was to show the power of HOLMImpulse as a toolbox, the FR/IR export/import features being the most importent parts, together with the smothing and the FFT/IFFT and FFT/IR manipulation options. Big thanks again, Ask!

- Klaus
 

Attachments

  • phase_linearisation.gif
    phase_linearisation.gif
    41.8 KB · Views: 1,296