Deconvolution methodology test, sample impulses

[deLuther]

As a part of my experiments with digital signal processing, I`m tried to make my own impulses, I`m tried two principal approaches: pulse method and deconvolution using Voxengo Deconvolver. Pulse method (i.e. recording response of pulse signal) gives somewhat more natural result, but lacks top end, compared to impulses with deconvolution of sweep test signal response. While trying to understand what is happening, I`m noticed that test sweep signal in deconvolver spans from around 45Hz, i.e. low end cannot be retrieved properly.

So I`m decided to make my own sweep signal which spans from 1 Hz and also compare deconvolver to my own deconvolution.
Difference in test signals:

Difference in recorded response:

Here is some test impulses from Randall RA412XJM cab loaded with V30 speakers, Shure SM57.
1) Voxengo Deconvolver + original test signal (45-22k), MP-Transform (minimal phase transformation) enabled: http://forum.guitarplayer.ru/index.php?action=dlattach;topic=65249.0;attach=852706 (NSb_IR_vxr_vx.wav)
2) Full range sweep signal (1-22k) + my own deconvolution, MP-transformed: http://forum.guitarplayer.ru/index.php?action=dlattach;topic=65249.0;attach=852705 (NSb_IR_fr_mp.wav)
Then I`m decided to test what happens if full range response will be deconvolved with Voxengo and short range sweep response with my deconvolution algo
3) 1-22k sweep + Voxengo Deconvolver: http://forum.guitarplayer.ru/index.php?action=dlattach;topic=65249.0;attach=852707 (NSb_IR_fr_vx.wav)
4) 45-22k sweep + own deconvolution: http://forum.guitarplayer.ru/index.php?action=dlattach;topic=65249.0;attach=852708 (NSb_IR_vxr_mp.wav)

ps. Some additional impulses from my home rig (KMG 5W FET Pa into modified Randall RG 75 cab, Shure SM58):
1) Voxengo Deconvolver approach: http://forum.guitarplayer.ru/index.php?action=dlattach;topic=147691.0;attach=849043 (KMG5_RG75CB_SM58NC_vox.wav)
2) My own approach: http://forum.guitarplayer.ru/index.php?action=dlattach;topic=147691.0;attach=849044 (KMG5_RG75CB_SM58NC_mp.wav)

If links will not work, then I`m upload impulses to dropbox.

 

[Mutant]

Interesting.

 

[hsnyder]

Dunno how I missed this thread two weeks ago, but I'm going to bump it - TheWinterSnow and I are doing some experiments in this vein in an attempt to produce a more satisfactory speaker/cab simulation solution that doesn't cost thousands of dollars.

Thanks for this info, ill compare the IRs this weekend

 

[deLuther]

Sweeps (full range and deconvolver`s), if someone willing to experiment:
https://dl.dropbox.com/u/3630120/Sweeps.rar (44k so far).

 

[deLuther]

Recently I`m decided to try a new approach to impulse retrieval, based on spectrum matching idea: match line source to miced target and get matching impulse.
I`m reamped some DI, recorded with mic and also reamped to get line signal (preamp), actually these signal can be recorded at once. Then I`m perfromed frequency analysis on both signals to get average spectrums, after division of target spectrum by source (line) spectrum and transformation to impulse (using minimal phase transform to get suitable form).
Of course, I`m decided to check against conventional deconvolution method using full range sweep and own deconvolution (as in first post).
Results: https://dl.dropbox.com/u/3630120/ImpulseComparison.rar
Including samples of miced signal, line signal, convolved signals and two impulses (for description check ImpulseComparison.txt in archive).

 

[hsnyder]

oh wow! Very interested, I'll check these out when I get a chance...