sound elevation detection algorithms

On Thu, 21 Jun 2007 17:09:07 GMT, Richard Dobson
<richarddobson@blueyonder.co.uk> wrote:

>Sylvia wrote:
>
>> I meant two microphones that are inline horizontally(not vertically).i can
>> ofcourse do azimuth detection by using ITD etc.if elevation detection
>> cannot be done with this scenario,what additional info is required to do
>> elevation detection?
>
>Well, if you use a Soundfield microphone, you can do it with one of 
>them! Its actually four capsules internally - one omni (W) fort the 
>pressure component, three fig-8s for velocity (XYZ, where Z is height 
>component), in a tetrahedral configuration. the idea is to capture the 
>full periphonic soundfield at a point, so the capsules have to be as 
>close together as possible. I suppose if you are not interested in X or 
>Y (which are in effect a "mid+side" pair), you just need capsules for W 
>and Z ("mid+height"?). There is a lot of interest these days in 
>higher-order microphones (4 capsules =  first-order), which in theory 
>give much better localisation. Making them is another matter, needless 
>to say.
>
>Worth asking on the sursound list, lots of experts in Ambisonics, etc.
>
>Richard Dobson

That sounds like the acoustic equivalent of a monopulse radar.   It's
still a multi-sensor array, it's just arranged efficiently so that
simple algorithms can be used.

In radar, the idea is that a flat-plate antenna (or similar) is
divided up into four quadrants, so that there are top and bottom
halves, and left and right halves, that can be summed and differenced
independently.

The sum of all four quadrants makes the usual directive beam with a
sinx/x sidelobe pattern in both azimuth and elevation.   A
"difference" beam, with a cosine response across azimuth, i.e., a
"positive" lobe on one side and a "negative" lobe on the other
(phase-wise), is created by taking the difference of the signal in the
two received halves.   Then taking the ratio of what is received in
the sum and difference beams gives a nice mono-pulse imaging system in
essentially three dimensions, since the radar can also provide range
information and can discrimate location between the left and right
halves and the top and bottom.

Since the cosine response crosses zero at the boresight, the ratio of
the sum beam to the difference beams peaks there.   This makes it
great for fire-control radar where you really just want to be able to
steer so that the target stays in the boresight.

I don't know why doing the same wouldn't work acoustically, assuming
the sensor mechanics work out reasonably well.
Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

On 22 Jun, 00:37, Eric Jacobsen <eric.jacob...@ieee.org> wrote:
> On Thu, 21 Jun 2007 17:09:07 GMT, Richard Dobson
>
>
>
>
>
> <richarddob...@blueyonder.co.uk> wrote:
> >Sylvia wrote:
>
> >> I meant two microphones that are inline horizontally(not vertically).i can
> >> ofcourse do azimuth detection by using ITD etc.if elevation detection
> >> cannot be done with this scenario,what additional info is required to do
> >> elevation detection?
>
> >Well, if you use a Soundfield microphone, you can do it with one of
> >them! Its actually four capsules internally - one omni (W) fort the
> >pressure component, three fig-8s for velocity (XYZ, where Z is height
> >component), in a tetrahedral configuration. the idea is to capture the
> >full periphonic soundfield at a point, so the capsules have to be as
> >close together as possible. I suppose if you are not interested in X or
> >Y (which are in effect a "mid+side" pair), you just need capsules for W
> >and Z ("mid+height"?). There is a lot of interest these days in
> >higher-order microphones (4 capsules =  first-order), which in theory
> >give much better localisation. Making them is another matter, needless
> >to say.
>
> >Worth asking on the sursound list, lots of experts in Ambisonics, etc.
>
> >Richard Dobson
>
> That sounds like the acoustic equivalent of a monopulse radar.   It's
> still a multi-sensor array, it's just arranged efficiently so that
> simple algorithms can be used.

Don't know about monopulse radar, but the sensor is a 4C vector
sensor:
The three spatial gradients plus pressure. I've playd a little with
data
from thise sorts of things in the past, and I don't find them simple
at all.
...
> I don't know why doing the same wouldn't work acoustically, assuming
> the sensor mechanics work out reasonably well.
...

The main issue I faced when playing with these things, was the
coupling between the gradients (or particle velocities) and the
sound pressure. Another problem was waves wich showed up in
one or two components, but not the others. And, of course,
differentiating between shear waves and pressure waves. Ah,
and Scholte, Rayleigh, Stoneley waves. Oh, and the Love waves.

Acoustics is a completely different game than EM, mainly because
of the physics, but also because of the slow propagation speeds
involved. Noise is an interesting factor, as it is anisotropic and
might affect different components differently. Which makes
useful signal processing a very interesting excercise.

Rune

Eric Jacobsen wrote:
> On Thu, 21 Jun 2007 17:09:07 GMT, Richard Dobson
> <richarddobson@blueyonder.co.uk> wrote:
> 
>> Sylvia wrote:
>>
>>> I meant two microphones that are inline horizontally(not vertically).i can
>>> ofcourse do azimuth detection by using ITD etc.if elevation detection
>>> cannot be done with this scenario,what additional info is required to do
>>> elevation detection?
>> Well, if you use a Soundfield microphone, you can do it with one of 
>> them! Its actually four capsules internally - one omni (W) fort the 
>> pressure component, three fig-8s for velocity (XYZ, where Z is height 
>> component), in a tetrahedral configuration. the idea is to capture the 
>> full periphonic soundfield at a point, so the capsules have to be as 
>> close together as possible. I suppose if you are not interested in X or 
>> Y (which are in effect a "mid+side" pair), you just need capsules for W 
>> and Z ("mid+height"?). There is a lot of interest these days in 
>> higher-order microphones (4 capsules =  first-order), which in theory 
>> give much better localisation. Making them is another matter, needless 
>> to say.
>>
>> Worth asking on the sursound list, lots of experts in Ambisonics, etc.
>>
>> Richard Dobson
> 
> That sounds like the acoustic equivalent of a monopulse radar.   It's
> still a multi-sensor array, it's just arranged efficiently so that
> simple algorithms can be used.
> 
> In radar, the idea is that a flat-plate antenna (or similar) is
> divided up into four quadrants, so that there are top and bottom
> halves, and left and right halves, that can be summed and differenced
> independently.
> 
> The sum of all four quadrants makes the usual directive beam with a
> sinx/x sidelobe pattern in both azimuth and elevation.   A
> "difference" beam, with a cosine response across azimuth, i.e., a
> "positive" lobe on one side and a "negative" lobe on the other
> (phase-wise), is created by taking the difference of the signal in the
> two received halves.   Then taking the ratio of what is received in
> the sum and difference beams gives a nice mono-pulse imaging system in
> essentially three dimensions, since the radar can also provide range
> information and can discrimate location between the left and right
> halves and the top and bottom.
> 
> Since the cosine response crosses zero at the boresight, the ratio of
> the sum beam to the difference beams peaks there.   This makes it
> great for fire-control radar where you really just want to be able to
> steer so that the target stays in the boresight.
> 
> I don't know why doing the same wouldn't work acoustically, assuming
> the sensor mechanics work out reasonably well.

Sylvia asked for two microphones, not two microphone stands. Can I aim a 
laser beam to center on a photocell? Sure, if I use a four-quadrant 
device. Does that meet the problem statement? Ask Sylvia.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;&macr;

On Thu, 21 Jun 2007 22:10:56 -0400, Jerry Avins <jya@ieee.org> wrote:

>Eric Jacobsen wrote:
>> On Thu, 21 Jun 2007 17:09:07 GMT, Richard Dobson
>> <richarddobson@blueyonder.co.uk> wrote:
>> 
>>> Sylvia wrote:
>>>
>>>> I meant two microphones that are inline horizontally(not vertically).i can
>>>> ofcourse do azimuth detection by using ITD etc.if elevation detection
>>>> cannot be done with this scenario,what additional info is required to do
>>>> elevation detection?
>>> Well, if you use a Soundfield microphone, you can do it with one of 
>>> them! Its actually four capsules internally - one omni (W) fort the 
>>> pressure component, three fig-8s for velocity (XYZ, where Z is height 
>>> component), in a tetrahedral configuration. the idea is to capture the 
>>> full periphonic soundfield at a point, so the capsules have to be as 
>>> close together as possible. I suppose if you are not interested in X or 
>>> Y (which are in effect a "mid+side" pair), you just need capsules for W 
>>> and Z ("mid+height"?). There is a lot of interest these days in 
>>> higher-order microphones (4 capsules =  first-order), which in theory 
>>> give much better localisation. Making them is another matter, needless 
>>> to say.
>>>
>>> Worth asking on the sursound list, lots of experts in Ambisonics, etc.
>>>
>>> Richard Dobson
>> 
>> That sounds like the acoustic equivalent of a monopulse radar.   It's
>> still a multi-sensor array, it's just arranged efficiently so that
>> simple algorithms can be used.
>> 
>> In radar, the idea is that a flat-plate antenna (or similar) is
>> divided up into four quadrants, so that there are top and bottom
>> halves, and left and right halves, that can be summed and differenced
>> independently.
>> 
>> The sum of all four quadrants makes the usual directive beam with a
>> sinx/x sidelobe pattern in both azimuth and elevation.   A
>> "difference" beam, with a cosine response across azimuth, i.e., a
>> "positive" lobe on one side and a "negative" lobe on the other
>> (phase-wise), is created by taking the difference of the signal in the
>> two received halves.   Then taking the ratio of what is received in
>> the sum and difference beams gives a nice mono-pulse imaging system in
>> essentially three dimensions, since the radar can also provide range
>> information and can discrimate location between the left and right
>> halves and the top and bottom.
>> 
>> Since the cosine response crosses zero at the boresight, the ratio of
>> the sum beam to the difference beams peaks there.   This makes it
>> great for fire-control radar where you really just want to be able to
>> steer so that the target stays in the boresight.
>> 
>> I don't know why doing the same wouldn't work acoustically, assuming
>> the sensor mechanics work out reasonably well.
>
>Sylvia asked for two microphones, not two microphone stands. Can I aim a 
>laser beam to center on a photocell? Sure, if I use a four-quadrant 
>device. Does that meet the problem statement? Ask Sylvia.
>
>Jerry

But he clarified that the two microphones are on a horizontal line, so
some non-spatial technique must be used to get elevation resolution.

Richard described a possible technique using an unusual microphone and
I was just responding that there may be a related, fairly simple,
signal processing technique to go with it.

It's kind of related to the mention of the penna in the ear, I think,
i.e., using a co-located sensor system to discriminate direction.

Eric Jacobsen
Minister of Algorithms
Abineau Communications
http://www.ericjacobsen.org

"Paul Russell" schrieb
>> i require links about sound elevation detection algorithms using
only
>> two microphones.i will appreciate any
references,papers,algorithms etc.
>
> This is why humans and other animals have pinnae - they enable us
to
> localise sound from all directions.
>
In what respect do the earlaps enable elevation detection?
I think that (at least mammal) ears are geared towards detection of
sound
in a plane (earth surface) and are slightly suboptimal for detection
of sound
coming from above (birds) or below (moles).
AFAICT, Human ears have a reception diagram  that is maximized in
the
bore sight of vision. Sidewards and behind, there is detection, but
not
accurate  localization. In order to localize better, you'll have to
turn your head.
Having turned the head, you may also see what the sounds represents.
You may also turn your head upwards to get a better estimation of
the heght
of the dragon approaching you.

It is interesting to note that some animals can turn their ears.
Turning the ear
(a) amplifies the sound coming from that direction, enhancing the
sensitivity
and (b) enables better direction finding due to (probably, I guess
here)
already knowing the angle of the ear and using the enhanced
sensitivity.

But I believe that this sensitivity is essentially planar.

IMHO
Martin

"Rune Allnor" schrieb
>
> Oh, and the Love waves.
> 
Love waves, sounds great. 
We should have more of those.

SCNR
Martin

On 22 Jun, 11:16, "Martin Blume" <mbl...@socha.net> wrote:
> "Rune Allnor" schrieb
>
> > Oh, and the Love waves.
>
> Love waves, sounds great.
> We should have more of those.
>
> SCNR
> Martin

No, one would probably not want any more of those around
than are already there.

The Love waves makes the earth move beneath you. These are
transversal shear surface waves that belong to the class of
waves which makes people see stars -- mainly because the
houses fall down around (or on top) of them.

Rune

Martin Blume wrote:
..
> 
> In what respect do the earlaps enable elevation detection?
> I think that (at least mammal) ears are geared towards detection of
> sound
> in a plane (earth surface) and are slightly suboptimal for detection
> of sound
> coming from above (birds) or below (moles).

I think "slightyly suboptimal" is probably about right - there 
definnitely is a degree of elevation detection, at least in the 
relatively near field. It is based on each brain learning the behaviour 
of the particular ears connected to it. This is all covered by the topic 
of Head-Related Transfer Functions (HRTF, q.v.). Some success has been 
had synthesizing general HRTFs such that listening to binaural 
recordings over headphones successfully gives the impression of elevated 
positions. Typcically the main problem in these cases is front/rear 
ambiguity. However, when in-ear recordings are made, and then played 
back via headphones on those same ears (so the HRTF's correspond 
exactly) the illusion is, apparently, virtually total. So there is a 
general form of HRTF that works ~quite~ well for lots of people, and a 
personal HRTF (ostensibly as unique as a fingerprint) that works a whole 
lot better for that individual.

Of course, microphones such as the Soundfield do not deal with HRTFs at 
all, they try to record the whole 3D soundfield at that point. The 
recording has to be played back ("decoded" from B-format) using a 
precisely arranged (preferably regular) array of speakers (e.g minimum 
square for horizontal, cube for periphonic, but more is better), in 
order to reconstruct the original soundfield which we can then listen to 
as if we were there.  The tricky part is getting the "sweet spot" where 
this all works, to be of a reasonable size.

We now have a wikipedia on Ambisonics:

http://en.wikipedia.org/wiki/Ambisonics


Richard Dobson

Martin Blume wrote:
> "Paul Russell" schrieb
>>> i require links about sound elevation detection algorithms using
> only
>>> two microphones.i will appreciate any
> references,papers,algorithms etc.
>> This is why humans and other animals have pinnae - they enable us
> to
>> localise sound from all directions.
>>
> In what respect do the earlaps enable elevation detection?
> I think that (at least mammal) ears are geared towards detection of
> sound
> in a plane (earth surface) and are slightly suboptimal for detection
> of sound
> coming from above (birds) or below (moles).
> AFAICT, Human ears have a reception diagram  that is maximized in
> the
> bore sight of vision. Sidewards and behind, there is detection, but
> not
> accurate  localization. In order to localize better, you'll have to
> turn your head.
> Having turned the head, you may also see what the sounds represents.
> You may also turn your head upwards to get a better estimation of
> the heght
> of the dragon approaching you.
> 
> It is interesting to note that some animals can turn their ears.
> Turning the ear
> (a) amplifies the sound coming from that direction, enhancing the
> sensitivity
> and (b) enables better direction finding due to (probably, I guess
> here)
> already knowing the angle of the ear and using the enhanced
> sensitivity.
> 
> But I believe that this sensitivity is essentially planar.
> 

Try closing your eyes and keeping your head still and listen to the 
sounds around you - you should be able to localise sounds both 
front-back and up-down, in addition to basic localisation in a 
horizontal plane. If you then amputate your pinnae you will find that 
you will no longer be able to do this.

Paul

"Martin Blume" <mblume@socha.net> wrote in news:467b92b5$0$3783
$5402220f@news.sunrise.ch:

> In what respect do the earlaps enable elevation detection?
> I think that (at least mammal) ears are geared towards detection of
> sound

The pinna creates notches in the spectrum.  The notches move around with 
the elevation of the noise source, so a notch detector will correlate with 
elavation.

-- 
Scott
Reverse name to reply