1.     Don't be a dotard!
    Read these "rules" AND introduce
    yourself     before your first post
  2. A healthy approach to understanding measurements
    Being true to what the artists intended
  3. After being asked for the 308th time, Merv lists top 5 headphones
    (opinion / entertainment piece)
    Dismiss Notice
  4. $1000 to the first to design a headamp using discrete components with better measurements than Magni 3+
    Dismiss Notice
IMPORTANT: New Members Read This Don't Be a Measurement Fascist Merv's Top 5 Headphones - 2023 Edition $1000 Challenge

Burst Response! HD800, SR-207, HD650

Discussion in 'Measurement Techniques Discussion' started by purr1n, Jan 8, 2017.

Page 5 of 14
  1. Garns

    Garns Friend

    Pyrate
    Joined:
    Jul 9, 2016
    Likes Received:
    2,484
    Trophy Points:
    93
    Location:
    Sydney, AUS
    I agree with Marv's position here, the fact that the frequency domain transform of a fixed number of cycles of a sine wave contains all sorts of other frequencies is not relevant. We don't hear all sorts of other frequencies, we hear a few cycles of a sine wave. Our ears aren't FFTing the whole signal, they are doing windowed FFTs with shorter window length and some entirely time domain specific stuff. We are trying to get an objective correlate for the time domain part of this.
     
    Garns, Mar 21, 2017
    #81
  2. SSL

    SSL Friend

    Pyrate
    Joined:
    Nov 12, 2015
    Likes Received:
    1,239
    Trophy Points:
    93
    Actually, the added frequencies are audible. I had never noticed this consciously until my music tech class in college. They taught us that hard-cutting an audio clip results in a start/stop transient audible as a click. The solution was to cross-fade the tracks, which as I explained above filters out the spurious frequencies. The exact same principle is at work here.
     
    SSL, Mar 22, 2017
    #82
  3. logscool

    logscool Friend

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    69
    Trophy Points:
    28
    Hmm, could be something interesting here. I guess we'll see.

    I do think this idea of having high frequency content in the sudden on and off of the sine wave is important though. This high frequency content will excite any high frequency resonances and modes in the driver, frequencies much higher than that of the sine wave tone. It's these characteristics of the drivers mode shapes and resonances that will result in anomalies in the sine wave not seen in the .wav file.
     
    logscool, Mar 22, 2017
    #83
  4. Garns

    Garns Friend

    Pyrate
    Joined:
    Jul 9, 2016
    Likes Received:
    2,484
    Trophy Points:
    93
    Location:
    Sydney, AUS
    I agree, we hear something other than a pure sine wave when the sound starts and stops. But I think that is exactly the time domain specific stuff that our ears do, and what we are trying to ascertain by looking at the burst response. If you window it that information will be lost. Ie, we may simply discover that headphones are all equally good at playing a slowly fading-in sine wave.
     
    Garns, Mar 22, 2017
    #84
  5. SSL

    SSL Friend

    Pyrate
    Joined:
    Nov 12, 2015
    Likes Received:
    1,239
    Trophy Points:
    93
    Agreed, but then the implication is that we are not getting any more information from a "sine burst" than from an impulse or step response, and possibly less since it would contain fewer frequencies. I'm having trouble wrapping my brain around that particular issue. So far it seems like the sine burst may at least be easier to coordinate to listening impressions.
     
    SSL, Mar 22, 2017
    #85
  6. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    So, gave some further thought to this.

    As far as the step response vs the tone burst, the tone burst is more of a pulse response. However, this pulse is centered around the frequency of interest, not DC.

    In regard to what @JohnM was saying about the analytical signal, I think I get more the idea. Basically one takes the Hilbert transform of the tone burst and applies the absolute value. The result should be sort of a like a pulse response centered a the frequency of interest w/o the sinusoidal blurring things out.

    The idea is that the tone burst is a pulse multiplied by a tone. One can shape the pulse with a given window to get more resolution in time, or frequency, or a compromise. Whatever correlates better with what we hear would be the way to go of course. The signal goes through the system (headphone or whatever). Then the recorded signal goes to the Hilbert transform which modifies the sinusoidal in the recorded signal into a complex exponential which gets removed by the absolute value, and one gets only the pulse centered at the frequency of interest w/o the sinusoidal modified only by the response of the system (headphone or whatever). In other words, it erases the tone and makes the visualization of the response easier to understand by keeping only the envelope untouched. (I should have got this right away since I used it a lot, but we are only human)

    A wavelet transform as proposed could also yield a lot of information, but it would require a bit more thought.

    Anyhow. I do see a lot of value in what @Marvey and @Serious are doing. The tone burst is not just a pulse or a step response. It's a pulse centered at a given frequency, and again, applying that Hilbert transform + absolute value might yield interesting results in terms of visualizing what we hear. Different tone bursts should yield different results after things are processed.
     
    Last edited: Mar 22, 2017
    ultrabike, Mar 22, 2017
    #86
  7. Garns

    Garns Friend

    Pyrate
    Joined:
    Jul 9, 2016
    Likes Received:
    2,484
    Trophy Points:
    93
    Location:
    Sydney, AUS
    As I understand it, step or impulse responses are usually derived from sweeps and so are actually the step/impulse response of the nearest linear approximation to the system. We are explicitly interested in non-linearities.

    What might also do this is a step response actually derived from playing back a step function and recording the result.
     
    Garns, Mar 22, 2017
    #87
  8. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    It will pretty much look exactly the same.

    Ok, now that looks interesting. I will try the same thing and while I only have the HD600 and HD800 here with me, I'll try it on my three couplers just to get more data. Only headphones for now because I don't have an anechoic chamber.
    The tone burst signals do look like they correlate more to dynamics, but it could still be a thing of FR. Phones which measure with a lot of bass bleed seem to rise slower.
     
    Serious, Mar 22, 2017
    #88
  9. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    Tone bursts, CSDs, FFTs and so forth are just visualizations of the same information that may help correlate better with our observations of what we hear.

    Steps and impulse responses are not strictly speaking derived from sweeps. One can derived them that way in certain situations, but that's not necessarily the best way to think about those in many situations.

    Regardless, all these things we are talking about have little to do with non-linearities. Again, these are different visualizations of the same information (which is mostly linear).
     
    ultrabike, Mar 22, 2017
    #89
  10. Garns

    Garns Friend

    Pyrate
    Joined:
    Jul 9, 2016
    Likes Received:
    2,484
    Trophy Points:
    93
    Location:
    Sydney, AUS
    As what? The sine wave bursts? I agree with that. But possibly different from the step response derived from the FR alone. Actually, I suppose this is what one should measure: the difference between the ideal (linear) step response derived from FR and the measured response to an actual physical step.

    I disagree with this. A compressor is a non-linear system. Its main purpose is to change our impression of dynamics. You wouldn't try to calculate the frequency response of a compressor. Rather you would try to understand its envelope which you could do by feeding it a sine wave burst. That is what we are doing here.
     
    Garns, Mar 22, 2017
    #90
  11. SSL

    SSL Friend

    Pyrate
    Joined:
    Nov 12, 2015
    Likes Received:
    1,239
    Trophy Points:
    93
    If I'm not mistaken, there should be literally no difference. I'd be interested in seeing evidence either way, though.
     
    SSL, Mar 22, 2017
    #91
  12. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    Yes. I tried this with an IR (as a step response would need a FR to DC), but they looked exactly the same, at least to the eye. Not sure if there was a difference.
    You could look at the IR vs the minimum phase IR, but with single-driver headphones there will pretty much be no difference. With most speakers and their crossover networks there will be a lot of excess phase.

    Anyway, running some tests to see what happens now.
     
    Serious, Mar 22, 2017
    #92
  13. johnjen

    johnjen Doesn’t want to be here but keeps posting anyways

    Pyrate
    Joined:
    Sep 29, 2015
    Likes Received:
    505
    Trophy Points:
    93
    Location:
    Well Grounded
    Very Interesting Indeed. :)

    This should be very informative and I agree with Marv about what I grasp of all of this.
    And that we are exploring this bleeding edge tells me there is much to learn and glean from these upcoming results.

    I've been exploring a variant of this for a while and potentially will be able to 'speed up' my refinement due to some equipment for measuring my 800's I'm using now.

    Interesting Times In Audio Indeed!

    JJ
     
    johnjen, Mar 22, 2017
    #93
  14. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    I'm done testing. The 10kHz bursts don't contain any more information than the IR/Step response already does. It's really just a different way of looking at a bandlimited version of the same thing. The results will depend more on the coupler than anything else and I don't think it tells us anything new about the sound.

    Here are my results: https://www.dropbox.com/sh/rcllvrznkfc42oj/AAANgYg91gxIrk7JLmJQvBG4a?dl=0

    The recorded IR and step response plots do look a little different than the ones from REW, but that's probably just a matter of ADC/DAC behaviour and how REW looks at the data.
     
    Last edited: Mar 22, 2017
    Serious, Mar 22, 2017
    #94
  15. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,938
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    I don't think you have enough data (different headphones with varying levels of "dynamics") to make that determination yet. Changing the coupler is definitely going to change the results as it will affect damping. You should an approach which is consistent and not change the method of measurement. I used the same coupler with my own measurements, and the burst response in those seem to indicate otherwise and provide much more correlation to "dynamics" than the step-response. The step-response might give us a clue to how the burst looks like, but ultimately the burst visualization provides a bit more information.
     
    purr1n, Mar 22, 2017
    #95
  16. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,938
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    I like this abs value of the Hilbert transform idea to get an amplitude "envelope". The calculations seem computationally expensive though. Maybe I should just fake it and draw dots on the abs value of the crests and connect them together? Ha ha ha. Seriously, sometimes simple works. We're not making missile guidance systems.
     
    purr1n, Mar 22, 2017
    #96
  17. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    It's not too bad.

    I actually miss-wrote stuff a bit. One takes the absolute value of the analytic signal, not the Hilbert transform. The Hilbert transform can be used to arrive to the analytic signal. But there are other equivalent means to the end.

    All that mumble jumbo removed, this is what can be done:

    Code:
    % windowed burst
x = 1:1000;
y = sin(2*pi*x/20);
w = window(@gausswin,length(y),2.5).';
z = [zeros(1,1000), y.*w, zeros(1,1000)];

% analytic signal envelope
Z = fft(z);
A = zeros(size(Z));
A(1:length(Z)/2) = 2*Z(1:length(Z)/2);
a = ifft(A);
e = abs(a);

% plots
plot(z);
hold on;
plot(e,'r');
grid on;
axis tight;
    Where w is whatever window you like. Above I used Gaussian, just cuz it came up in this thread. Basically one does the following:

    1) Produce the windowed burst and pass it through the headphone or amp or DAC or whatever.
    2) Get the analytic signal absolute value (i.e. envelope):
    a. Get the FFT of the windowed burst.
    b. Zero out the negative frequencies in the FFT above (this is the FFT of the analytic signal).
    c. Scale up by 2 (since we took out about half the power).
    d. Do an inverse FFT (the output will be complex).
    e. Take the absolute value and there you go... envelope.
    3) Enjoy pretty plots.

    Here are some with different windows (burst is blue, envelope is red - using Matlab code stuff above):

    Gaussian

    Gaussian.png

    Rectangular

    Rectangular.png

    Blackman-Harris

    BlackmanHarris.png

    Now, the plots above have not gone through the headphone system. Once they go through them each envelope will be different depending on both the frequency selected for the burst, and the headphone response.

    If one already has FFT and IFFT libraries for CSDs, this should not be too bad IMO.

    Obviously one could just eye out the burst response with the "carrier" there and that might work just as well which indeed would be simpler.
     
    Last edited: Mar 22, 2017
    ultrabike, Mar 22, 2017
    #97
  18. Serious

    Serious Inquisitive Frequency Response Plot

    Pyrate BWC MZR
    Joined:
    Sep 28, 2015
    Likes Received:
    2,594
    Trophy Points:
    113
    Location:
    near Munich, Germany
    That is true but I don't really see anything in the 10kHz bursts that I don't see in the square wave and impulse response. I do really think that we're analyzing FR (and phase) more than "dynamics", at least from looking at your plots with the step responses and the 10kHz bursts. I think it's more of a coincidence that they correlate more with dynamics, especially considering all the variables involved which are fundamentally very different from how we listen to music (especially the coupler).
    The point for me was to show that this affects the measurements more than the headphones being measured and that the results would be consistent across different couplers. Really only to produce more data from the two headphones that I have here.
    In this case Coupler 1 is definitely more lossy than coupler 3 (creatology foam), which shows in that the measurements generally have less bass, but it also tends to have less treble ringing. The head shows a more accurate, less warm FR, which can be seen in the better shape of the step response, but the outer ear also creates resonances that aren't originially there. The skin itself seemed to be better damped than the other materials.

    Again, I do think that we're mainly just showing parts of the IR and FR in a different way and as long as the couplers fundamentally aren't accurate I don't think it makes too much sense to find new visualizations for the inherently flawed data. In other words I think that neither flat plate couplers, nor dummy heads or in-ear measurements really work for headphone measurements.

    That's not to say that the results aren't interesting or else I wouldn't have tried it myself, but I think it's too FR dependent. Maybe a different kind of tone test can give us a better idea.

    @ultrabike doesn't the REW option for the filtered IRs produce similar results to those?
     
    Last edited: Mar 22, 2017
    Serious, Mar 22, 2017
    #98
  19. logscool

    logscool Friend

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    69
    Trophy Points:
    28
    @ultrabike I can probably try this with some actual headphones later today and see what I get.
    I should be able to just export z as a .wav record on a headphone and re-input into matlab in the analytic signal envelope correct?
     
    logscool, Mar 22, 2017
    #99
  20. ultrabike

    ultrabike Measurbator - Admin

    Staff Member Pyrate MZR
    Joined:
    Sep 25, 2015
    Likes Received:
    8,960
    Trophy Points:
    113
    Location:
    Irvine CA
    I think flat plate couplers, dummy heads and in-ear measurements work for headphones, but one has to be careful about how to interpret the results.

    As far as REW, I don't think filtered IR produce similar results to what we are showing above.

    Yup. Can do that. Try different frequencies for the "carrier" like @Marvey did. And try different windows as well. See which one correlates best to what we are hearing.
     
    ultrabike, Mar 22, 2017
    #100
Page 5 of 14

Share This Page