MLS versus swept sine waves

Discussion in 'Studio Construction & Acoustics Forum' started by Ethan Winer, Nov 29, 2007.

  1. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    In another thread I told someone:

    Andre Brito stepped in to say:

    I repeated the value of a tracking filter:

    Andre didn't address the use of a tracking filter, but said MLS is better because it has a lower crest factor, and linked to a PDF document having more formulas than a graduate-level physics test:

    Eric Desart then stepped in to back up Andre, also ignoring my point about a tracking filter:

    So yet again I brought up tracking filters:

    This is followed by more back and forths that dodged the core issue by talking about crest factors rather than tracking filters. Then Andre posted this link:

    http://www.lds-group.com/docs/knowledge_base.php?id=92

    Okay, now we're getting somewhere! I replied:

    And that's where it ended. But I'm still trying to get to the bottom of this, and we ain't there yet!

    It seems to me that the crest factor of MLS offers at most a 1 to 2 dB s/n improvement over a swept sine wave. Versus a tracking filter that, according to the last article linked above, can increase s/n as much as 60 dB.

    Assembled here in this forum are some of the finest minds in the field of small-room acoustics. So please tell me guys, is my original statement about tracking filters used with swept sine waves indeed true, or is there a way that MLS can achieve the same high s/n ratio?

    Thanks!

    --Ethan
  2. andrebrito

    andrebrito

    Joined:
    Jun 28, 2004
    Likes Received:
    0
    Location:
    Almada (near Lisbon), Portugal
    Home page:
    Ethan, there was no need to make a separate post from the already existence.

    I'm trying to find literature on this at the moment also to see the bottom of this as well, since some sites say one thing and others say another thing. For instance on the ETFAcoustic it is stated "MLS tests are better for mid frequency / high frequency use while exponential sweep signals are better suited to low frequency measurements." So now I'm must say I'm also confused.

    I understand the tracking filter concept altough find the statement of "can increase s/n as much as 60 dB" a bit too much.

    I have personally never tested measurements in noisy conditions. Maybe someone else here has...

    Also I don't think there is much of a difference in small room acoustics. But in large room acoustics or other situations it can make a difference.
  3. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    Well, that thread had nothing to do with the topic at hand, so I figured it's more sensible to start a new thread devoted just to this.

    The R+D software from ETF normally uses a combination of MLS noise and an LF (20 - 200 Hz) swept sine wave at the same time. You can hear the sine wave sweeping up and down repeatedly under the noise. But when you look at the s/n ratio for the measurement it's good below 200 Hz, and above 200 Hz the s/n falls off suddenly and substantially. In fact, after I complained to Doug Plumb who wrote the software that the s/n is unusably low for measuring RT60, he added an option for only a full range sweep with no MLS. Using this newer option gives a consistently high s/n ratio across the whole band.

    Hence my assumption that a swept sine wave is better. As I recall, I first learned about tracking filters from Doug Plumb.

    I agree. It seems to me that in practice 20 dB or maybe 30 is more realistic. A filter would have to be very steep to achieve 60 dB of rejection even a third octave away, let alone closer.

    I see this as mostly a theoretical discussion. I can't see why the size of the room would matter, but maybe I'm missing something. Again, it boils down to one issue - can MLS be rigged to use a tracking filter too? Or is there a similar technique that can increase the native s/n ratio by a similar amount? Otherwise the s/n is basically dominated by the acoustic noise floor of the room being tested. As in air conditioning noise, outside traffic, etc. I don't know why that would differ with room size.

    Thanks Andre, we'll get to the bottom of this yet.

    So where's Jeff Szymanski when we need him? :)

    --Ethan
  4. avare

    avare

    Joined:
    Feb 12, 2004
    Likes Received:
    0
    Location:
    Hamilton, Ontario, Canada
    Ethan:

    The short is no. MLS methods can acheive usable results with S/N ratios as low as -20 dB. :)

    For a detailed explanation, and test results see BBC RD 1994-05(I love these BBC RD reports). Rather appropriately for your question, the title of the report is "Maximal-length sequence methods: Acoustic measurements in the presence of high levels of extraneous noise."

    Andre
  5. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    Okay, I'm reading now. :)

    Me too. I have a LOT of them. But I never had this one.

    Thanks Andre!

    --Ethan
  6. andrebrito

    andrebrito

    Joined:
    Jun 28, 2004
    Likes Received:
    0
    Location:
    Almada (near Lisbon), Portugal
    Home page:
    Ok I'm adding more PDFs to the confusion haha These state in fact sweep sines have a better S/N than MLS. BUT oddly they all been tested so it seems in quiet environments. Actually on the first PDF it is stated

    "Stan et al. compare four different room impulse response measurement techniques: pseudo-random noise (MLS and IRS), time-stretched pulses, and logarithmic sine sweep. Since the randomized phase of psuedo-random sequences makes them immune to background
    noise, MLS and IRS techniques are preferred in noisy environments. However, parameter optimization is required for high SNR because of nonlinear distortion. Nevertheless, the achieved SNR is only 60.5 dB with an MLS order of 16 and single measurement.
    Time-stretched pulses and sine sweep methods produce a higher SNR than the pseudo-random noise techniques, but they require a quiet environment. The SNR of the time-stretched pulses technique is 77 dB after precise calibration. The logarithmic sine sweep method has 80.1 dB SNR. The benefit of the sine sweep is that unlike the previous methods,
    it produces a high SNR without any calibration"

    So I guess you are correct Ethan but honestly if it is required to have a quiet environment to use sweeps I personally don't see any advantages on using MLS or sweep on quiet environments since the decay is strong enough to extrapolate T20 and T30, god knows even T60. What I wanted to check was te use of MLS vs sweep on noisy environments (without time variant components since for those it is stated Sweep is way better than MLS).

    Like I said I never measured in noisy conditions and when I tried to use MLS vs sweep I got the exact/very close results.

    Here are the PDFs

    http://zx81.isl.uiuc.edu/camilleg/jasa03.pdf

    http://pcfarina.eng.unipr.it/Public/Papers/208-UAM2005.pdf

    http://www.acoustics.net/objects/pdf/article_farina03.pdf

    http://pcfarina.eng.unipr.it/Public/Papers/218-WESPAC9.pdf
  7. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    Excellent, Andre, thanks! It's quarter to Bourbon so I'm knocking off for the day and I'll check out your links tomorrow. But I did learn some great stuff from the BBC paper Andre Vare linked. Well, actually, I'm still plowing through it. :)

    It seems that MLS and tracking filters can both be used to increase the s/n ratio over raw test signals. I suspect the concept is similar for both too - knowing the test signal's exact content allows filtering out everything else. So it may well be that the answer is they're both about the same, since such filtering always has a practical limit.

    I knew there had to be more to this! We'll get there yet. :eek:

    --Ethan
  8. avare

    avare

    Joined:
    Feb 12, 2004
    Likes Received:
    0
    Location:
    Hamilton, Ontario, Canada
    It was a little bit confusing reading this this initially with two Andres involved. I should be used to this by now. One of the things I do in my spare time (don't ask what that spare time means) is volunteer at a shelter. There is a another Andre there who makes me look short (I am 6'2"). I am the little Andre there.

    The one in Canada:
    Andre
  9. andrebrito

    andrebrito

    Joined:
    Jun 28, 2004
    Likes Received:
    0
    Location:
    Almada (near Lisbon), Portugal
    Home page:
    hahaha ! I know what you mean particulary since in my country André is not a common name so I also find it a bit odd when I met another André
  10. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    André ^2,

    I had a discussion last night with my expert friend Bill Eppler about this stuff, and also emailed Chris Liscio who wrote FuzzMeasure, and also did a little more snooping around on the web this morning.

    First, it's clear I was wrong about MLS not being as good in theory as a swept sine wave, since MLS can use correlation to improve the s/n ratio in the same way a tracking filter can improve s/n with swept sine Waves. So I apologize to André (you know which André I mean). But in my defense :) I offer the following points:

    * As best I can tell, MLSSA is the only commercial measurement system that takes advantage of correlation to increase s/n. I know MLSSA is expensive, but there's no prices on their site and so far they haven't answered my email asking for the price.

    * According to Chris Liscio and the ETF web site, swept sine waves are used to increase s/n over the way most software uses MLS. This is from the R+D page at www.etfacoustic.com:

    The key here is "standard methods" meaning how MLS is routinely used in popularly priced commercial room analysis software.

    This is from Chris Liscio (FuzzMeasure author):

    Chris also mentioned that he's not aware of any software other than MLSSA that uses correlation to realize the theoretical advantage of MLS. I'll be interested to hear if anyone knows of other software that does this.

    * My friend Bill developed some of the hardware for Goldline's TEF system (though not the software), and he said he's pretty sure they do not use correlation. When I've seen TEF used it played a swept sine wave but it may also offer MLS noise.

    Bill also explained that correlation can be used with any signal source. It doesn't have to be MLS noise. He said it can be used with swept sines or even plain pink noise, as long as there's a digital representation of the original signal available for the correlation math to act on.

    So it appears all of us were correct. :eek:

    The s/n with static or swept sine waves and a simple SPL meter (or VU meter) is limited by the ambient noise in the room and the noise floor of the electronics. However, a tracking filter can be employed in software to increase the s/n ratio. Likewise, plain pink noise, or MLS noise, or any other noise is also limited by the ambient room noise and electronics unless correlation is used to increase it.

    Does this about sum it up?

    --Ethan
  11. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    Sorry, one more thing I forgot to mention. This is also from Chris Liscio:

  12. andrebrito

    andrebrito

    Joined:
    Jun 28, 2004
    Likes Received:
    0
    Location:
    Almada (near Lisbon), Portugal
    Home page:
    I forgot to add that one !

    I have tried different acoustical programs and my favorite is Easera or winMLS. I had some problems with ETF regarding large acoustics measurements.
  13. avare

    avare

    Joined:
    Feb 12, 2004
    Likes Received:
    0
    Location:
    Hamilton, Ontario, Canada
    Yes. :D


    No accent on my e:
    Andre
  14. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    Heh, so instead of André ^2 from now on I'll use:

    André + Andre = 2 smart guys.

    :)
  15. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    FYI, this just in: $3,500.
  16. avare

    avare

    Joined:
    Feb 12, 2004
    Likes Received:
    0
    Location:
    Hamilton, Ontario, Canada
    Thanks for the price update.

    Very reasonable for commercial apps, but way too high for most.

    A great value:
    Andre
  17. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    Heh, I'd hire you in a minute. :)
  18. andrebrito

    andrebrito

    Joined:
    Jun 28, 2004
    Likes Received:
    0
    Location:
    Almada (near Lisbon), Portugal
    Home page:
    Easera is a nice software and their intermmediate package costs only $900

    http://www.rh.com/easera/EASERA_Prices_12-1-07.pdf
  19. Ethan Winer

    Ethan Winer

    Joined:
    Mar 19, 2001
    Likes Received:
    0
    Location:
    New Milford, CT USA
    Home page:
    But does it take advantage of correlation to increase the s/n of MLS noise?
  20. JohnPM

    JohnPM Guest

    In essence just about every application using MLS to measure response will be using a process that is mathematically equivalent to correlation, as will applications using swept sine signals. It would be possible to use MLS as the excitation with an RTA, which would be equivalent to using white noise, but I can't think of any other instance where correlation or an equivalent process would not be used. A correlator carries out a convolution of the input signal with a reference signal. Convolution in the time domain is enacted in the frequency domain by multiplication/division, response analysers essentially use the ratio of the frequency spectra of the excitation signal and the captured response to determine the transfer function in the frequency domain then inverse transform that to get back to the time domain impulse response.

    Generally an FFT is used to get from the time to frequency domains. With MLS the nature of the test signal allows an alternative transform to be used (the Fast Hadamard Transform), with lower computational cost, but this offers no useful benefit on modern computing platforms and MLS has several disadvantages as set out in the paper by Muller and Massarani.
Similar Threads: versus swept
Forum Title Date
Studio Construction & Acoustics Forum Skylights versus Windows Jun 6, 2008
Studio Construction & Acoustics Forum Room modes, calculated versus measured Mar 7, 2005
Studio Construction & Acoustics Forum LF performance versus density Jun 2, 2004
Studio Construction & Acoustics Forum rockwool versus fiberglass ? Jan 27, 2004

Share This Page