Hacker News new | past | comments | ask | show | jobs | submit login
Acoustic Prism Invented at EPFL (epfl.ch)
79 points by aethertap on Aug 11, 2016 | hide | past | favorite | 22 comments



The actual paper is here: http://scitation.aip.org/content/asa/journal/jasa/139/6/10.1...

With pictures of the apparatus.


Isn’t this what the cochlea already does in a human ear?


Yes, essentially! One key difference is that frequency and intensity coding in the cochlea is facilitated by a a built-in active physiological component (outer hair cells) that enhances frequency selectivity and provides compression of the structure that vibrates inside the cochlea (basilar membrane). The outcome is that listening acuity is preserved across a large dynamic range and variable background noise levels.


What do you mean when you say the hair cells are "active"? I can see how the hairs enhance certain frequencies, but does their tuning change dynamically somehow?


Great question, and sorry for not being clear. It's not the frequency tuning of a given outer hair cell (OHC) that that changes, it is their response to the sound level (what I meant by dynamic range). OHCs are activated by basilar membrane motion (the basilar membrane mechanically vibrates due to the physical sound). At low sound levels, the basilar membrane itself does not have a large displacement. The theory is that when OHCs depolarize (fire) due to basilar membrane displacement, resulting transduction currents activate motor proteins that change the length of the cell so that it "jumps" to displace the vibrating membrane (basilar membrane) more.

http://i.makeagif.com/media/10-06-2015/yxvqkm.gif

So at low sound levels, OHCs create a sharper displacement (gain) where vibration is maximal on the basilar membrane, which enhances the frequency coding onto auditory nerve fibers (via inner hair cells).

High sound levels displace the basilar membrane more, which can lead to broad patterns of excitement of nerve fibers coding for adjacent frequencies. At these levels, the OHCs do not change length as much, which compresses the membrane movement. The compression helps to maintain sharper acuity at the frequency that is physically present, and less so at adjacent frequencies. The function describing OHC length changes against level of sound is thus nonlinear, with gain at low levels and compression at high levels. This compression is not seen when the organism is dead, when OHCs are damaged, or when OHCs are genetically knocked out.



It looks like a Flute. Air pressure enters and each hike let's a specific frequency go out.


I wonder how much of an improvement this will be over microphone arrays and determining direction in software.


I've marveled at the difficulty of determining the direction and origin of unknown sounds, esp. low pitch sounds.

For example, the people of Windsor, Ontario, Canada, have been hearing a rumbling noise for six years and have been unable to locate it, although they finally have a theory that it's coming from a blast furnace[1].

We're talking about a city of 210,000 where a large segment of the city have regularly heard it over the span of 6 years: "In 2012, more than 22,000 people dialled in to a local teleconference about the hum [to voice their concerns]."

I'm surprised that there isn't an instrument that you can flip on and have it reliably locate a continuous sound that lasts for hours or days at a time. Is there something about low pitch sounds and vibrations that make them particularly hard for direction finding tools?

[1] https://www.theguardian.com/world/2016/jun/07/windsor-hum-ca...


Don't know about the specifics of that case, but I can imagine that for low-frequency pervasive sounds, they bounce everywhere and thus any such equipment would measure them as "coming from everywhere", because they literally are. There's a source, but it's masked completely by reverberation effects. At some point with enough diffusion there is very little signal in the noise, if that helps..


I still don't understand why they could not locate it.

Sounds falls off with the square of the distance. A single, localized measurement can have a hard time pinning a direction, but a series of sample over an area will give you a 3D curve of intensity with a maxima near the source...


Random thoughts about the difficulties from someone who definitely isn't an expert:

Low frequency noise is harder to block with materials, so you can't put a microphone in a foam box with one side open then swing it around looking for the direction the sound is loudest.

If it travels further (I guess it should if it's harder to block) then there's a greater chance anything you can hear is coming from further away.

If it's far away, it's harder to triangulate (because you need to go further between measurements).

If it's shaking the ground, does that mean the strength only falls off linearly? Do you get surface 'waves' that are then only spreading out linearly?

edit - This is a really interesting question though, I guess it is fairly hard, but it does feel like it should be simple.


If it's shaking the ground, then every surface near you (which is sufficiently rigidly attached to the ground) will be producing sound waves. Hence even if you can identify a direction to the sound, it's probably a red herring.


Low frequency sounds are more difficult to localize than high frequency sounds, for a variety of reasons. I would be surprised, though, if a widely-spread grid of detectors wouldn't be able to find it.

https://en.wikipedia.org/wiki/Sound_localization#Evaluation_...


Human hearing can't really localise sounds below 100hz-ish, but it should certainly be possible to triangulate this 35hz hum with microphones.

Probably even easier ways to rule out the cause - turn the furnace off and see if the hum stops? Seems like the cause is known in this case but they just don't want to address it


Often blast furnaces and smelters can't be turned off.


True but they can be turned "up" or "down". Reducing the air flow to the lowest value that can be sustained for, say, 12 hours would give an opportunity to find out if the intensity and type of noise has changed. Do that a few times and see if it is repeatable.


YouTube link from the article : https://www.youtube.com/watch?v=6sSBPxAv2qk


It's not loading for me, so here's a text-only version from Google. https://webcache.googleusercontent.com/search?strip=1&q=cach... Sounds like a cool idea, but being able to load the pictures would probably help.


I wonder whether this has any real-world applications, since the Fourier Transform is a well understood and easy to implement method for analyzing spectra digitally.


The question is: when can you not use the Fourier Transform (or it is really bad)?

I can think of situations when one frequency drowns the other in your transducer

Also this was built for the audible range (apparently), but it might have more interesting uses in different ranges or fluids


They mention using it for sonar in the paper




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: