Reply by kundan burnwal March 10, 20052005-03-10

hi shawn,
shawn,i must thank you for your timely help in giving
me a direction,which i had been searching for so long.

ya i'm comfortable with C language,and to some extent:
Matlab n Scilab.i do have some idea about the basic
building blocks like FFTs,Filters etc(since i had
taken the DSP course in engineering),but not very good
at them. the reason i wrote i'm a newbie is that i'm
not familiar with the concepts of Adaptive/Advanced
DSP,which i thought would be required to solve this
problem of acoustic feedback calcellation.Apart from
that,i don't have a good grasp of DSP
implementations,in spite of the courses.

i'm really glad for your direction towards
implementation.The method you described is quite
similar to the one described in the patent document by
Yamaha Corporation on "Apparatus detecting howling by
decay profile of impulse response in sound system",
which i just finished reading,n yes they seem to be
realistic.

after reading the patent document by yamaha,seems i'm
going to face some issues in implementing few blocks
of the system and will get back to you soon.

thanks a lot for your help n looking forward for more
very soon!! :)
regards,
kundan

--- sks_dsp <> wrote:

>
>
> You have found logical descriptions of the
> functionality needed
> within these patents, you have described in verbally
> the
> functionality needed, the next step is to translate
> those functional
> block diagrams and descriptions into Software. Do
> you know anything
> about C language programming? Do you know anything
> about DSP related
> things like FFTs, IIR filters, FIR Filters etc. I
> would suspect you
> do
>
> The feedback supression problem has two main
> components
> (1) Determine feedback is occuring
> (2) Create a notch filter at the frequency in which
> the feedback is
> occuring.
>
> On (1): These patents all describes observations you
> can make to
> determine that feedback is occuring and what
> frequency that it is
> occuring at. Translate these into DSP Software
> Blocks (i.e. an FFT
> describes energy at different frequencies, a simple
> count variable
> incremented at a sample rate each time a condition
> is true can be
> used to monitor how long a condition is existing.
> etc.)
>
> On (2): A notch filter is a 2nd order IIR filter,
> the stuff DSP were
> designed to do. This step is easy.
>
> I would agree with the previous post that this
> problem in general is
> not a "newbie" type of problem.
>
> Good luck and have fun.
>
> -Shawn
>
> --- In , kundan burnwal
> <kburnwal@y...>
> wrote:
> > hi Mike,
> >
> > thanks for replying.
> >
> > i can understand that i'm a newbie in the area of
> > Adaptive DSP..but the hard fact is that i need to
> > solve this problem,otherwise i can't complete my
> > Engg!!
> >
> > i've gone thru lots of literture for this. Sound
> > Equipment manufacturers like
> SHURE,YAMAHA,Mitsubishi,
> > SONY etc have made such a system commercially n
> have
> > numerous US patents for the same. They have used
> two
> > approaches.
> >
> > one is an automatically tunable notch filter for
> > suppression of accoustical feedback.The apparatus
> > includes a selectively tunable notch filter having
> a
> > center frequency which is variable over at least a
> > substantial portion of the audio frequency
> spectrum.
> > The apparatus receives an audio signal which is
> > substantially non-periodic in the absence of
> > acoustical feedback and substantially periodic
> with an
> > instantaneous dominant frequency in the presence
> of
> > the same. The duration of successive periods are
> > monitored and compared by an up/down counter to
> > determine whether the audio input signal is
> > substantially periodic and to determine the
> > instantaneous dominant frequency of such audio
> signal.
> > Upon detection of an audio signal which is
> > substantially periodic, the notch filter is tuned
> to
> > the instantaneous dominant frequency so as to
> suppress
> > the acoustical feedback.this method exploits the
> fact
> > that howling(due to feedback) occurs due to a very
> > narrow band of offending frequencies.(This is US
> > Patent No. 4091236 assigned to University of
> > Akron,Ohio. URL:
> > http://www.freepatentsonline.com/4091236.html)
> >
> > another approach is to have a measuring section
> that
> > measures an impulse response of the sound system
> to
> > determine a time length of a decay portion of the
> > impulse response. A detecting section detects an
> > occurrence of the howling when the determined time
> > length is longer than a predetermined reference
> time
> > length, and further analyzes a frequency spectrum
> of
> > the decay portion of the impulse response to
> determine
> > a frequency point at which the howling occurs. An
> > attenuating section attenuates a frequency
> component
> > of the sound around the determined frequency point
> so
> > as to cancel the howling.
> > (This is US Patent No. 6442280 assigned to Yamaha
> > Corportation.URL:
> > http://www.freepatentsonline.com/6442280.html)
> >
> > I've attached a rar file containing most of the
> patent
> > abstracts related to this area.
> >
> > but the problem with these approaches is that they
> > done the full thing in hardware.our task is to
> write
> > code for this in C language,implement in CCS n
> > transfer onto DSP Processor.Unfortunately,i'm not
> able
> > to understand the implementations much in
> > hardware,since the implementation is not very
> > understandable to me.i'm making efforts for
> that.but i
> > would be really glad if someone who has done a
> similar
> > project or has good knowledge can help this
> newbie.
> >
> > thanks a lot.
> > kundan
> > --- mike ts <mikets42@y...> wrote:
> > > Dear Kundan,
> > >
> > > a typical room reverberation time 400...800ms,
> > > dictates a long filter (200...400ms), what does
> not
> > > agree with people moving dynamics and short
> sound
> > > wave
> > > length (only 33cm on 1kHz). for a
> quazi-statinary
> > > conditions, it is doable with subband and mixing
> > > noise
> > > into reference signal below audibility theshold
> in
> > > critical bands - but even that is not a task for
> > > newbie, rather for a team of skilled pros and
> > > scientists.
> > >
> > > regards,
> > > michael.
> > >
> > > --- Kundan Burnwal <kburnwal@y...> wrote:
> > >
> > > ---------------------------------
> > >
> > >
> > > hi,
> > > i'm a newbie,currently doing "Acoustic feedback
> > > cancelation in Public
> > > Address System for a Given Room" (in our
> case,its
> > > Lecture Theatre) as
> > > my final year engineering project.The system is
> such
> > > that the
> > > professor uses wireless collar microphone to
> speak
> > > in
> > > the Lecture
> > > Theatre to move freely and therefore the impulse
> > > response of the
> > > system is time-varying.
> > >
> > > I'm absolutely clueless in this project now. My
> > > faculty guide had
> > > suggested to use Apaptive filtering (like NLMS
> > > algorithm)in time domain.
> > > The major problem is that we don't have a
> > > clean-speech
> > > signal
> > > available to us a the reference signal. Also,the
> > > feedback received
> > > from the loudspeaker to microphone is a
> perfectly
> > > correlated
> > > noise(i.e. it is the time delayed version of
> input
> > > signal due to
> > > multipath effects of the room),making the task
> more
> > > challanging.
> > >
> > > Please guide me how to go ahead in this
> > > project.Should
> > > a time domain
> > > approach be used or frequency domain approach?
> I've
> > > heard that in
> > > systems like Lecture Rooms, the length of an
> > > adaptive
> > > filter has to be
> > > kept quite high (~40 or more),which i dont think
> can
> > > be processed in
> > > real time by the TMSC55 DSP board in any case.
> > >
> > > Looking for your valuable guidance
> > >
> > > thanks
> > > kundan burnwal
> > > final yr engg student
> > > DA-IICT
> > > Gandhinagar
> > > Gujarat
> > > India
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > To
> > >
> > >
> > >
> >
>
______________________________________________________________________
> > >
> > > Post your free ad now! http://personals.yahoo.ca
> > >
> >
> >
> >
> >
> >
> >
> > __________________________________
> > Celebrate Yahoo!'s 10th Birthday!
> > Yahoo! Netrospective: 100 Moments of the Web
> > http://birthday.yahoo.com/netrospective/ >
>

__________________________________



Reply by sks_dsp March 7, 20052005-03-07


You have found logical descriptions of the functionality needed
within these patents, you have described in verbally the
functionality needed, the next step is to translate those functional
block diagrams and descriptions into Software. Do you know anything
about C language programming? Do you know anything about DSP related
things like FFTs, IIR filters, FIR Filters etc. I would suspect you
do

The feedback supression problem has two main components
(1) Determine feedback is occuring
(2) Create a notch filter at the frequency in which the feedback is
occuring.

On (1): These patents all describes observations you can make to
determine that feedback is occuring and what frequency that it is
occuring at. Translate these into DSP Software Blocks (i.e. an FFT
describes energy at different frequencies, a simple count variable
incremented at a sample rate each time a condition is true can be
used to monitor how long a condition is existing. etc.)

On (2): A notch filter is a 2nd order IIR filter, the stuff DSP were
designed to do. This step is easy.

I would agree with the previous post that this problem in general is
not a "newbie" type of problem.

Good luck and have fun.

-Shawn

--- In , kundan burnwal <kburnwal@y...>
wrote:
> hi Mike,
>
> thanks for replying.
>
> i can understand that i'm a newbie in the area of
> Adaptive DSP..but the hard fact is that i need to
> solve this problem,otherwise i can't complete my
> Engg!!
>
> i've gone thru lots of literture for this. Sound
> Equipment manufacturers like SHURE,YAMAHA,Mitsubishi,
> SONY etc have made such a system commercially n have
> numerous US patents for the same. They have used two
> approaches.
>
> one is an automatically tunable notch filter for
> suppression of accoustical feedback.The apparatus
> includes a selectively tunable notch filter having a
> center frequency which is variable over at least a
> substantial portion of the audio frequency spectrum.
> The apparatus receives an audio signal which is
> substantially non-periodic in the absence of
> acoustical feedback and substantially periodic with an
> instantaneous dominant frequency in the presence of
> the same. The duration of successive periods are
> monitored and compared by an up/down counter to
> determine whether the audio input signal is
> substantially periodic and to determine the
> instantaneous dominant frequency of such audio signal.
> Upon detection of an audio signal which is
> substantially periodic, the notch filter is tuned to
> the instantaneous dominant frequency so as to suppress
> the acoustical feedback.this method exploits the fact
> that howling(due to feedback) occurs due to a very
> narrow band of offending frequencies.(This is US
> Patent No. 4091236 assigned to University of
> Akron,Ohio. URL:
> http://www.freepatentsonline.com/4091236.html)
>
> another approach is to have a measuring section that
> measures an impulse response of the sound system to
> determine a time length of a decay portion of the
> impulse response. A detecting section detects an
> occurrence of the howling when the determined time
> length is longer than a predetermined reference time
> length, and further analyzes a frequency spectrum of
> the decay portion of the impulse response to determine
> a frequency point at which the howling occurs. An
> attenuating section attenuates a frequency component
> of the sound around the determined frequency point so
> as to cancel the howling.
> (This is US Patent No. 6442280 assigned to Yamaha
> Corportation.URL:
> http://www.freepatentsonline.com/6442280.html)
>
> I've attached a rar file containing most of the patent
> abstracts related to this area.
>
> but the problem with these approaches is that they
> done the full thing in hardware.our task is to write
> code for this in C language,implement in CCS n
> transfer onto DSP Processor.Unfortunately,i'm not able
> to understand the implementations much in
> hardware,since the implementation is not very
> understandable to me.i'm making efforts for that.but i
> would be really glad if someone who has done a similar
> project or has good knowledge can help this newbie.
>
> thanks a lot.
> kundan
> --- mike ts <mikets42@y...> wrote:
> > Dear Kundan,
> >
> > a typical room reverberation time 400...800ms,
> > dictates a long filter (200...400ms), what does not
> > agree with people moving dynamics and short sound
> > wave
> > length (only 33cm on 1kHz). for a quazi-statinary
> > conditions, it is doable with subband and mixing
> > noise
> > into reference signal below audibility theshold in
> > critical bands - but even that is not a task for
> > newbie, rather for a team of skilled pros and
> > scientists.
> >
> > regards,
> > michael.
> >
> > --- Kundan Burnwal <kburnwal@y...> wrote:
> >
> > ---------------------------------
> >
> >
> > hi,
> > i'm a newbie,currently doing "Acoustic feedback
> > cancelation in Public
> > Address System for a Given Room" (in our case,its
> > Lecture Theatre) as
> > my final year engineering project.The system is such
> > that the
> > professor uses wireless collar microphone to speak
> > in
> > the Lecture
> > Theatre to move freely and therefore the impulse
> > response of the
> > system is time-varying.
> >
> > I'm absolutely clueless in this project now. My
> > faculty guide had
> > suggested to use Apaptive filtering (like NLMS
> > algorithm)in time domain.
> > The major problem is that we don't have a
> > clean-speech
> > signal
> > available to us a the reference signal. Also,the
> > feedback received
> > from the loudspeaker to microphone is a perfectly
> > correlated
> > noise(i.e. it is the time delayed version of input
> > signal due to
> > multipath effects of the room),making the task more
> > challanging.
> >
> > Please guide me how to go ahead in this
> > project.Should
> > a time domain
> > approach be used or frequency domain approach? I've
> > heard that in
> > systems like Lecture Rooms, the length of an
> > adaptive
> > filter has to be
> > kept quite high (~40 or more),which i dont think can
> > be processed in
> > real time by the TMSC55 DSP board in any case.
> >
> > Looking for your valuable guidance
> >
> > thanks
> > kundan burnwal
> > final yr engg student
> > DA-IICT
> > Gandhinagar
> > Gujarat
> > India
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > To
> >
> >
> >
>
______________________________________________________________________
> >
> > Post your free ad now! http://personals.yahoo.ca
> >
> __________________________________
> Celebrate Yahoo!'s 10th Birthday!
> Yahoo! Netrospective: 100 Moments of the Web
> http://birthday.yahoo.com/netrospective/



Reply by kundan burnwal March 7, 20052005-03-07
hi Mike,

thanks for replying.

i can understand that i'm a newbie in the area of
Adaptive DSP..but the hard fact is that i need to
solve this problem,otherwise i can't complete my
Engg!!

i've gone thru lots of literture for this. Sound
Equipment manufacturers like SHURE,YAMAHA,Mitsubishi,
SONY etc have made such a system commercially n have
numerous US patents for the same. They have used two
approaches.

one is an automatically tunable notch filter for
suppression of accoustical feedback.The apparatus
includes a selectively tunable notch filter having a
center frequency which is variable over at least a
substantial portion of the audio frequency spectrum.
The apparatus receives an audio signal which is
substantially non-periodic in the absence of
acoustical feedback and substantially periodic with an
instantaneous dominant frequency in the presence of
the same. The duration of successive periods are
monitored and compared by an up/down counter to
determine whether the audio input signal is
substantially periodic and to determine the
instantaneous dominant frequency of such audio signal.
Upon detection of an audio signal which is
substantially periodic, the notch filter is tuned to
the instantaneous dominant frequency so as to suppress
the acoustical feedback.this method exploits the fact
that howling(due to feedback) occurs due to a very
narrow band of offending frequencies.(This is US
Patent No. 4091236 assigned to University of
Akron,Ohio. URL:
http://www.freepatentsonline.com/4091236.html)

another approach is to have a measuring section that
measures an impulse response of the sound system to
determine a time length of a decay portion of the
impulse response. A detecting section detects an
occurrence of the howling when the determined time
length is longer than a predetermined reference time
length, and further analyzes a frequency spectrum of
the decay portion of the impulse response to determine
a frequency point at which the howling occurs. An
attenuating section attenuates a frequency component
of the sound around the determined frequency point so
as to cancel the howling.
(This is US Patent No. 6442280 assigned to Yamaha
Corportation.URL:
http://www.freepatentsonline.com/6442280.html)

I've attached a rar file containing most of the patent
abstracts related to this area.

but the problem with these approaches is that they
done the full thing in hardware.our task is to write
code for this in C language,implement in CCS n
transfer onto DSP Processor.Unfortunately,i'm not able
to understand the implementations much in
hardware,since the implementation is not very
understandable to me.i'm making efforts for that.but i
would be really glad if someone who has done a similar
project or has good knowledge can help this newbie.

thanks a lot.
kundan
--- mike ts <> wrote:
> Dear Kundan,
>
> a typical room reverberation time 400...800ms,
> dictates a long filter (200...400ms), what does not
> agree with people moving dynamics and short sound
> wave
> length (only 33cm on 1kHz). for a quazi-statinary
> conditions, it is doable with subband and mixing
> noise
> into reference signal below audibility theshold in
> critical bands - but even that is not a task for
> newbie, rather for a team of skilled pros and
> scientists.
>
> regards,
> michael.
>
> --- Kundan Burnwal <> wrote:
>
> --------------------------------- > hi,
> i'm a newbie,currently doing "Acoustic feedback
> cancelation in Public
> Address System for a Given Room" (in our case,its
> Lecture Theatre) as
> my final year engineering project.The system is such
> that the
> professor uses wireless collar microphone to speak
> in
> the Lecture
> Theatre to move freely and therefore the impulse
> response of the
> system is time-varying.
>
> I'm absolutely clueless in this project now. My
> faculty guide had
> suggested to use Apaptive filtering (like NLMS
> algorithm)in time domain.
> The major problem is that we don't have a
> clean-speech
> signal
> available to us a the reference signal. Also,the
> feedback received
> from the loudspeaker to microphone is a perfectly
> correlated
> noise(i.e. it is the time delayed version of input
> signal due to
> multipath effects of the room),making the task more
> challanging.
>
> Please guide me how to go ahead in this
> project.Should
> a time domain
> approach be used or frequency domain approach? I've
> heard that in
> systems like Lecture Rooms, the length of an
> adaptive
> filter has to be
> kept quite high (~40 or more),which i dont think can
> be processed in
> real time by the TMSC55 DSP board in any case.
>
> Looking for your valuable guidance
>
> thanks
> kundan burnwal
> final yr engg student
> DA-IICT
> Gandhinagar
> Gujarat
> India >
>
> To >
______________________________________________________________________
>
> Post your free ad now! http://personals.yahoo.ca
>
__________________________________
Celebrate Yahoo!'s 10th Birthday!
Yahoo! Netrospective: 100 Moments of the Web
http://birthday.yahoo.com/netrospective/


Attachment (not stored)
patents.rar
Type: application/octet-stream

Reply by Kundan Burnwal March 3, 20052005-03-03


hi,
i'm a newbie,currently doing "Acoustic feedback cancelation in Public
Address System for a Given Room" (in our case,its Lecture Theatre) as
my final year engineering project.The system is such that the
professor uses wireless collar microphone to speak in the Lecture
Theatre to move freely and therefore the impulse response of the
system is time-varying.

I'm absolutely clueless in this project now. My faculty guide had
suggested to use Apaptive filtering (like NLMS algorithm)in time domain.
The major problem is that we don't have a clean-speech signal
available to us a the reference signal. Also,the feedback received
from the loudspeaker to microphone is a perfectly correlated
noise(i.e. it is the time delayed version of input signal due to
multipath effects of the room),making the task more challanging.

Please guide me how to go ahead in this project.Should a time domain
approach be used or frequency domain approach? I've heard that in
systems like Lecture Rooms, the length of an adaptive filter has to be
kept quite high (~40 or more),which i dont think can be processed in
real time by the TMSC55 DSP board in any case.

Looking for your valuable guidance

thanks
kundan burnwal
final yr engg student
DA-IICT
Gandhinagar
Gujarat
India