IR Properties

Discussion in 'Digital & Modeling Gear' started by Jay Mitchell, May 17, 2019.

  1. Dancing Fool

    Dancing Fool Member

    Messages:
    37
    Joined:
    Jul 3, 2018
    Fwiw, just played around a bit with Jays IRs. As has been said by others, B isn't worth much, if anything. Weird honky-phase-y things going on all over the place, regardless of the used amp model. Out of A and C I seem to prefer A for the time being, simply because it sounds a bit richer in the low mid area, but maybe at volume (have only been able to try at a volume that wouldn't make the neighbours call the police, but it was fairly loud...) and in a band context, I might prefer C.
    Noticed the same thing as with Jays Atomic cabs, there's very little low end affairs that need to be taken care of, something I usually have to deal with when creating my live IRs (and no, I'm not talking Fletcher Munson here, just more or less about "general balance"), in other words: These sound very balanced, which is great. Unfortunately, the speaker captured doesn't seem to be too suitable for overdriven amps, but it goes quite well with the Amplifires 65 Clean and certain pedals in front, so this should be pretty suitable for anything in the blues/fusion realm.

    On a sidenote: I have one IR that I created myself (using anything else but great source capturing and/or scientific methods to modify those source IRs, to put it carefully...) which comes pretty close in terms of playing feel - but it certainly doesn't sound like any authentic cab.

    In case I find some time, I'll record some bits later on.
     
  2. Dancing Fool

    Dancing Fool Member

    Messages:
    37
    Joined:
    Jul 3, 2018
    Btw, thanks for this thread, it has been very enlightening on a whole number of levels for me.
     
  3. Chocol8

    Chocol8 Member

    Messages:
    162
    Joined:
    Mar 2, 2019
    Are you willing to share anymore details or suggestions on these parts?

    You say there are multiple processes to deconcolve, does it matter which is used? Seems like this should just be math and the software should handle it, but I don’t know what I don’t know here.

    What do you use to trim the resulting IR down to the pre-reflection piece?
     
  4. Dancing Fool

    Dancing Fool Member

    Messages:
    37
    Joined:
    Jul 3, 2018
    I'd say any bog standard audio editor would be sufficient.
     
  5. Chocol8

    Chocol8 Member

    Messages:
    162
    Joined:
    Mar 2, 2019
    True, I guess it is just a wav file at that point so I can just zoom way in and slice out the first 20-30ms in Reaper.

    If Mother Nature would cooperate with some dry and non-threatening weather I could run some test sweeps in the driveway and get a better idea of what further questions I need to ask. I know it’s not as ideal as a warehouse, but it is close and easy for practice and testing. You never know, perhaps JBL plus passing Toyota with a touch of chirping birds is just the exact tone I am looking for.
     
    Alex Kenivel and chumbucket like this.
  6. Yek

    Yek Member

    Messages:
    1,224
    Joined:
    Dec 28, 2003
    Location:
    Netherlands
    Wait for a passing Lexus = better tone.
     
    Chocol8 likes this.
  7. Cool Hand Luke

    Cool Hand Luke Member

    Messages:
    1,196
    Joined:
    Apr 22, 2009
    Yes, but the B IR is the bad sounding one. You should aim for C :)
     
  8. Chocol8

    Chocol8 Member

    Messages:
    162
    Joined:
    Mar 2, 2019
    I am trying to be realistic for my first attempt. If I get a loadable file that sounds something like an IR of a cab I will consider that positive progress. Especially if no electrical smoke is released in the process.
     
    Alex Kenivel and Cool Hand Luke like this.
  9. AlbertA

    AlbertA Member

    Messages:
    49
    Joined:
    Mar 21, 2006
    Already discussed earlier in the thread.
     
  10. gigsup

    gigsup Supporting Member

    Messages:
    1,434
    Joined:
    Aug 28, 2016
    Hi Albert,
    I'm having trouble with this, the ceiling height being less than dmic, and taking trfz at 20ms.
    I'm wondering about a few things with regard to ellipses, can you look at this and tell me what I'm not seeing?


     
  11. AlbertA

    AlbertA Member

    Messages:
    49
    Joined:
    Mar 21, 2006
    In a room with those dimensions with the mic that far apart (20') you would indeed get reflections in the window of interest.

    The green and red parts are also incorrect, by definition since it is an ellipse, both of those reflections would arrive at the same time.

    Moving the mic further apart increases the necessary room dimensions to keep the same reflection-free window of time.

    Note you don't want the direct sound to be delayed by 20ms (or trfz) - the time of flight for the direct sound is whatever it is - The distance from the mic to the cabinet should be enough to put you in the far-field of the guitar cabinet in question.
     
    Last edited: May 25, 2019
    yeky83 likes this.
  12. AlbertA

    AlbertA Member

    Messages:
    49
    Joined:
    Mar 21, 2006
    @gigsup Imagine the top half on an ellipsoid, enclosed exactly by a rectangular box.

    If the mic and the cab are the foci points of such an ellipsoid, then any reflection will travel 2a, where 2a is the "length" of the ellipsoid. If a rectangular box of size L, W, H is enclosing the top half of the ellipsoid, then we'll make 2a = L.

    Since we want the reflection to arrive trfz seconds later than the direct sound, then the reflection must travel at minimum, the distance the direct sound traveled (i.e. dmic) + the distance traveled in trfz seconds.

    This means then we have the following inequality bounding one of the box dimensions: L > dmic + speed_of_sound*trfz.

    You can then derive the rest of the dimensions of the enclosing box, since you know the distance between the foci points (i.e. dmic) and you have a bound for 2a.

    The maximum width of the ellipsoid (i.e W of the bounding box) and the height of the ellipsoid (i.e. 2*H of the bounding box, since the bounding box only bounds the top half) can be derived with some trigonometry:

    [​IMG]

    From the diagram above:
    c = dmic/2 and a = L/2. Given that a^2 = b^2 + c^2, then b = sqrt (a^2 - c ^2) and that 2b = W and b = H,

    Then W > 2*sqrt ((L/2)^2 - (dmic/2)^2) and H > sqrt ((L/2)^2 - (dmic/2)^2)
     
  13. AlbertA

    AlbertA Member

    Messages:
    49
    Joined:
    Mar 21, 2006
    And a neutral mic pre-amp and neutral sound capture interface (or at least know what the response looks like on loopback so you can compensate for it later).

    The best stimulus is an exponentially swept sine sweep, because this has some nice properties:
    • Low crest factor - increased SNR vs other signals, much easier to keep the system-under-test in its linear mode of operation
    • Such a sweep pushes non-linear distortion artifacts (if any) into negative time after deconvolution.
    You can use Octave (https://www.gnu.org/software/octave/) which is essentially an open source version equivalent of Matlab, to generate the sweeps, deconvolve, window the ir, etc...

    I have some scripts that generate the exponentially swept sine sweep that have some additional useful properties as described by Novak and improved by Vetter and di Rosario, mainly:
    • Sweep ends and begins in sine phase - Ending in sine phase avoids ripple/ringing at the high frequencies in the frequency domain.
    • Fades in the first octave - to suppress ripple/ringing at the low end in the frequency domain.
    • Every octave ends in sine phase - so that each order of any harmonic distortion products can be extracted as an "impulse response" from negative time in the deconvoluted output, i.e. the Diagonal Volterra kernels.
    A sweep and its inverse as well as the Octave scripts that generated them are here:
    https://www.dropbox.com/sh/uvtyr8keao1a9r6/AABrN7pVYt9RFwKHCaC7J69_a?dl=0

    Any DAW or Audio editing program should do.

    You can use Octave as well.

    An example workflow would be:
    • Generate sweep and its inverse with Octave - you only need to do this once. Generate them using the same sample rate you will use for recording to avoid resampling.
    • Import Sweep into your DAW
    • Make sure the project settings are at the sample rate of your recording interface to avoid any resampling.
    • Make sure when importing the sweep into your DAW, that you remove any fade-in or fade-out it may automatically do.
    • Playback the sweep and record the microphone input in another track.
    • Make sure you remove any fade-in/fade-out your DAW may apply to the captured track.
    • Export the captured track.
    • Import the track into Octave; [capture fs] = audioread("mycapture.wav");
    • Deconvolve by convolving the inverse of the sweep (from first step) with the captured signal; in octave this would be h = fftconv(inv_sweep, capture);
    • Figure out the index where "Zero time" starts in the output sequence. The resulting IR is the sequence from that "zero time" to the end; for the convolution above, in octave you can do ir = h(length(inv_sweep):end);
    • That output IR will included recording latency + direct sound time-of-flight. Trim as necessary.
    • After trimming from the beginning, Window the IR - you could use the second half of a hanning or a tukey window of appropriate size; For example for a 20ms window at 48KHz, the number of samples should be 20*48. In octave you could do, num_samples=20*48; win=tukeywin(num_samples*2, 0.25)(num_sampels+1:end); windowed_ir=win.*trimmed_ir(1:num_samples);
    • Normalize the windowed IR to unit energy - so that all your resulting IRs have around the same loudness when applied; in octave that would be: final_ir=windowed_ir/sqrt(sum(abs(windowed_ir).^2));
    • Export the final IR - in octave: audiowrite("myir.wav", final_ir, fs, 'BitsPerSample', 32);

    This was already pointed out - ground plane measurement technique. Use a hard surface.

    Take multiple captures - in case of external environmental sounds during the capture (loud vehicles, planes, AC turning on/off, etc) - or power spikes, or temporary electrical induced noise, etc.

    Avoid ground loops.

    The required space should be empty - a room with the required dimensions but with couches, book shelves, tables, etc doesn't count - you'll get reflections from them.

    Keep the temperature and humidity as constant as you can.
     
    Last edited: May 25, 2019
  14. Chocol8

    Chocol8 Member

    Messages:
    162
    Joined:
    Mar 2, 2019
    A big THANK YOU AlbertA! The details are very helpful.

    I will try the workflow in both REW and Octave. REW does the interface calibration well enough, handles my mic calibration file, and I am familiar enough with it, so I will start there. Your octave info will help me with the window settings and it should be a good plan B if REW has any limitations I run into.

    How big of a signal to noise ratio do I need? Obviously more is better, but will 60-70 dB do?
     
  15. Will Chen

    Will Chen Member

    Messages:
    5,696
    Joined:
    May 1, 2008
    Location:
    Allen, TX
    Gave the IRs spin last night and this afternoon. I don't know which is which between A & C, I liked both. When first auditioning at lower volumes I liked A more as it seems to have a bit more lows and highs but once things were cranked up loud C seemed to work better. These worked wonderfully with cleaner to midrange stuff but at higher gains I'd need to experiment some more as things didn't come together as quick (though that could be a function of the speaker the IR was taken of). Neither requires any fixing in terms of EQ in the cab block to reduce mud or ice pick highs, just wonderfully voiced, very balanced. I'd love to have a few more speakers captured in a similar fashion, a "no fuss" IR pack like this would be a dream.

    B is something different, sounds kinda like a room miced amp but something sounds a little off. Could work for a special effect but I can't see anyone using it for everyday tones.
     
  16. gigsup

    gigsup Supporting Member

    Messages:
    1,434
    Joined:
    Aug 28, 2016
    Thanks Albert!
    I redid the numbers. I think the biggest error was not doubling b to arrive at the correct height of the room.
    This results in very large, very circular ellipse.
    Does this look approximate to what you would expect for a 20ms trfz space indoors?

    dmic = 2m ; c =1m

    L > dmic + (c*trfz) ; 2m + 6.86m = 8.86m ; rounded up to 9m. Then a = 9m/2 = 4.5m

    b^2 = a^2 -c2 = sqrt(19.25) = 4.38m ; but 2b is the minor axis not just b, so the height is 8.76m

    a = 4.5m ; b = 4.38m ; c = 1m


    [​IMG]

    From the diagram above:
    c = dmic/2 and a = L/2. Given that a^2 = b^2 + c^2, then b = sqrt (a^2 - c ^2) and that 2b = W and b = H,

    Then W > 2*sqrt ((L/2)^2 - (dmic/2)^2) and H > sqrt ((L/2)^2 - (dmic/2)^2)[/QUOTE]
     
  17. djd100

    djd100 Member

    Messages:
    2,929
    Joined:
    Nov 25, 2006
    Hi all, just heard these through my home studio's NS-10M's so far, will get to the studio next week to hear them in a better environment with better monitors at a higher SPL, but in general I agree with all who found A-C more balanced, along with a much more believable/sweeter upper-mid/top end vs my various commercial IR's (this is the area of dramatic improvement to me).

    Briefly using "A" with a Deluxe model in the Axe II brought a smile to my face, which brings this initial observation...

    ...that reflection free close(r)-mic'd IR's as well as FF would be a substantial improvement over the current crop of commercial IR's with their unavoidable comb filtering (at some degree etc), no?

    For those with mic IR's like some Fractal models have, using them to color the BK's flat response yields decent results as well, so perhaps a set of great reflection free IR's along with mic IR's for color?

    Again, thanks to Jay, and everyone else who has contributed to a genuinely intelligent thread sans the typical TGP BS issues etc, thanks!
     
    Last edited: May 25, 2019
  18. yeky83

    yeky83 Member

    Messages:
    2,154
    Joined:
    Dec 11, 2007
    Doesn't look right. FYI he did the calcs 2 pages back and gave dimensions.

    What software are you using for your graphs and 2D & 3D drawings?

    Edit: oh you defined by a, b, and c, not h, w, and d, duh. My bad, looks right.
     
    Last edited: May 25, 2019
    gigsup likes this.
  19. AlbertA

    AlbertA Member

    Messages:
    49
    Joined:
    Mar 21, 2006
    Yeah the full ellipsoid in this case will be close to a sphere since its dimensions are close to each other.
     
  20. aleclee

    aleclee TGP Tech Wrangler Staff Member

    Messages:
    11,886
    Joined:
    Jan 4, 2002
    Location:
    5868 ft above sea level
    Got some quality time with Jay's IRs yesterday. Most excellent! "A" is more clear to my ear while "C" sounds a bit more bass-heavy. Both are great but "A" is better for my purposes.
     

Share This Page

  1. This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
    By continuing to use this site, you are consenting to our use of cookies.
    Dismiss Notice