Rational Acoustics



pilarek
March 19th, 2013, 06:11 AM
Hello, this is my first post here.
Could someone can explain to me what about the two peaks before the strongest impulse?

679

Don C
March 21st, 2013, 09:51 AM
I'm not sure I can be of much help here, still pretty new to this s/w myself. From looking at your measured delay of 28.02 ms (next to your meters) it appears that the 2nd of those early peaks lines up with the direct distance from speaker to microphone (about 33 feet). Does that seem reasonable for your setup? Your highest value seems in excess of 40 ms (47 feet). I have "played" with Smaart in a well treated theatre from that distance, but most of the serious stuff for me so far has been in the range of 7 to 10 ms (including electronic delay of 3 ms or so).

I llook forward to hearing from someone with more expertise on this subject. Still trying to get my thoughts together on how IR works without pulsing the program material.

Rasmus Rosenberg
March 21st, 2013, 12:49 PM
I could take a guess, but please describe a little more what (and how) your trying to measure? Is it a constant thing or?
Hq is proberly having more snow than here, but im sure will be happy to help, but a little more info is truly appreciated.
/R

Langston Holland
June 6th, 2013, 09:21 PM
Hello, this is my first post here.
Could someone can explain to me what about the two peaks before the strongest impulse?


Maybe the best first post question ever. :)

You've got a gain structure problem - your swept sine stimulus was in hard clipping somewhere in your system. You should also use longer FFT's that allow the noise floor to show (a flat region that doesn't go any lower as the IR tail extends in time to the right). There are S/N advantages to longer FFT's as well. Short may work, but if in doubt - even if you're not in doubt, go long.

The earliest "pre-arrival" you see is the 3rd harmonic product and the closest "pre-arrival" to the largest peak (the real first arrival) is the 2nd harmonic distortion product.

From my archives of greatness, may I present the best explanation in history:

From: Calvert Dayton, SIA
E-mail: support(a)siasoft.com
Date: 21 Jul 2004
Time: 17:26:58

Comments

If you're using Pink Sweep, I'm pretty sure what you're seeing is probably just the distortion component falling out of the measurement. Since the signal sweeps from low to high frequencies, each frequency happens only once in the sweep and the sine wave has no harmonic content itself, harmonics from lower frequencies appearing before the stimulus reaches a given fundamental frequency appear to be arriving in "negative time," before the measurement takes place. The techniques we use to measure impulse responses are circular in nature, so something that arrives "before" the beginning of the time record has no place to go but back around to the end, which is a good place for it in this case.

There's a school that holds you should run an impulse response measured using swept sine through a window function to get rid of that bump at the end, but I think it's kind of cool to be able to see how much distortion you're getting through a system. Either way, having that garbage neatly stacked up for you at the end of the record is arguably better than having it scattered around randomly throughout your measurement. You end up with a better S/N ratio as a result than you can get with a random/pseudo-random stimulus, which is nice.

Langston Holland
November 26th, 2013, 01:30 PM
More notes from my little journey that you don't want to know about:

I've finally gotten around to studying Heyser's work, which has forced me to learn more about the measurement techniques that contrast with his time delay spectrometry, such as the FFT methods employed by Smaart.

So... You'll notice that Calvert's explanation talks about the distortion products going to the end of the IR in relation to a FFT using a swept sine stimulus. This is true with the FFT procedures called (forgive me) circular deconvolution, which applied to the older versions of Smaart that he was referring to at the time. Smaart v7 does things a bit differently, thus the swept sine distortion components, if any, are located to the left of the actual first arrival.

Convolve means to mix together and deconvolve means to separate. Deconvolution referred to what we do involves separating the room's IR characteristic from the IR of the original stimulus that was used to energize the room. In this thread, everything in that IR has to do with the room at that mic position. The math behind this is still sufficiently advanced relative to my mental technology that I consider it magic. :)

Calculating the IR with something called linear deconvolution places the distortion components to the left of the real first arrival as we actually see in the measurement. This type of deconvolution is generally used for non-periodic stimuli like pink noise. My guess is that Smaart treats the swept sine like pink noise (uses pre-FFT windowing) and just takes advantage of the S/N improvement inherent with sinusoids.

Another Thought:

When discussing distortion harmonics falling out of a swept sine measurement in "negative time", that term correctly applies to whether you see these components to the left of the first real arrival or at the end of the IR. Where they are displayed is merely a function of the magic flavor that was chosen to implement the FFT. Where the harmonics are displayed is not where the term comes from, it comes from the fact that with a typical LF to HF sine sweep, the HF distortion harmonics appear to the measurement input before those frequencies were scheduled to appear. Thus the math dutifully records these HF components as showing up early (in negative time) relative to the swept fundamentals.

Example:

You're using Smaart v7 with a swept sine stimulus that sweeps from LF to HF and you were a big dummy and hard clipped the input of the audio interface because you didn't keep an eye on the input meters. I, of course, have never done this.* Even though the real stimulus is a log sweep (sweeps much more slowly at LF than HF), we'll assume a linear sweep to keep things simple. Say it takes 10ms from the start of the sweep to get to 100Hz and 20ms to get to 200Hz. The goal of the math is to move all of the frequencies to a single impulse at "time zero". Thus, when the math sees 100Hz, it moves it -10ms, when it sees 200Hz, it moves it -20ms, etc. This works perfectly when only the fundamental frequencies show up as expected.

Now say the 100Hz fundamental clips and arrives with the 2nd distortion harmonic of 200Hz, the math sees that 200Hz and moves it -20ms. The problem is that this instance of 200Hz showed up at the same time as the 100Hz fundamental (at 10ms). Thus, the 2nd harmonic distortion component is placed at -10ms (10ms -20ms) in "negative time". The 3rd harmonic, if it exists, will be placed at -20ms (10ms -30ms) in "negative time", etc.

Bottom Line:

When using the IR module, use the longest swept sine stimulus available and drive the system as hard as possible without clipping anything. But you already knew that. :)

Another Thought:

All of us real-time measurement guys tend to ignore the impulse response (IR) because it seems like something mainly for the install guys. This is very wrong - there is much to learn by looking at what the loudspeaker system does in the time domain. Make the IR part of what you look at - even if just a glance - before and after you get involved with EQ'ing the transfer function or whatever. Consider using a swept sine in the IR module displaying the ETC as a first step. Look for distortion components in negative time to the left of the main arrival spike - turn things down if you see them. Eventually, get comfortable with some of the intelligibility type calculations that can be done on the IR and compare the before/after effects of your EQ and tuning decisions. Heyser reminded us that there's no difference between the time and frequency domains, it's just a different way of looking at the same event. Try a different path to get to the places you normally go - it's fun and the view is enlightening. :)

---

* That, of course, is a lie. A great feature of another measurement program I have is that if you clip the input of the audio interface, the whole screen surrounding the measurement flashes red. You can't miss it. Some of the Sennheiser wireless receivers do the same thing - the whole screen turns red instead of just some little LED's on the top of the meter.

Arthur Skudra
November 26th, 2013, 04:18 PM
A great feature of another measurement program I have is that if you clip the input of the audio interface, the whole screen surrounding the measurement flashes red. You can't miss it. Some of the Sennheiser wireless receivers do the same thing - the whole screen turns red instead of just some little LED's on the top of the meter.[/i]

A subtle hint for a feature request Mr. Langston? I'll second the motion!! :D