🚀 go-pugleaf

In article <Bz3vHr.LKK@news.cso.uiuc.edu>, ja51359@uxa.cso.uiuc.edu
(axelrod) writes:
> 
> 
> Hi.  I know this is possible because I've heard the military does it for
> helicopter headsets; and I heard that the car companies will be doing it
> too.  Does someone know of how the basics work behind active noise can-
> cellation?
> 
> What I'm interested in is a system where you have a speaker or two
> somehow produce a signal that is directly out of phase with the ambient
> noise.  This seems impossible, since a speaker radiates in all directions,
> thereby having a different phase at different points in the room.  Is 
> there some way of getting an average phase or something, in order to
> decrease the noise in the room/car?
> 
> I'd like references to articles on this subject, along with a simple
> explanation if possible.
> 
> Thanks
> 
> 

     The type of cancellation you are referring to is one that I have heard
most commonly called reference cancelling.  The idea is pretty much as you
have described it.  You have one primary channel, and 1 or more reference
channels.  The primary channel contains your desired signal corrupted by
unwanted noise, the reference channels contain noise which is (hopefully)
highly correlated with the unwanted noise in your primary channel, but
contain no trace of your desired signal.  

     A classic paper on this topic is by Bernie Widrow.  While trying to
examine the heartbeat of a fetus, the heartbeat of the mother is a strong
interfering signal.  Using a primary channel of mother+fetus and being able
to locate a position to get a "clean" but maybe slightly out of phase
refernce of the mother's heartbeat only, they were able to obtain the
desired results.  I'm sorry I don't have the exact reference.  

    I have also read the paper where they discuss reference cancelling on
military hellicopters.  I would recommend you look in the reference section
of _Adaptive Signal Processing_ by Bernie Widrow and _Adaptive Filter
Theory_ by Simon Haykin for all the references you will ever need on this
topic.  Since these concepts are about 30 or so years old, you can find
concise explanations and examples in the textbooks.  

     I also wrote my master's thesis on use of this technology applied to
car telephones.  The problems in a car a similar to those in a helicopter,
but not exactly the same. A large problem is spatial correlation of
signals.  In other words, what do the cross-correlations of certain
composite signals look like.  
The conclusion I reached in my thesis was the same as in the helicopter
paper... that the further the reference channels were from the primary
channel, the less correlated that the noises on the 2 channels were.   

     To understand this (get ready for an over-simplified explanation),
realize that the speed of sound is slow, on the order of 1000 feet per
second.  The bandwidth of the phone channel is about 4KHz which means you
sample at 8KHz.  A large majority of the noise that you get is from the
engine and the tires.  If a noise originates at a tire, and has to get 10
feet to get to the primary channel and 14 feet to get to the reference
channel.  A differnce of 4 feet, and 32 samples.  In other words, a sound
originating at a tire will appear at the reference microphone 32 samples
after it appears on the primary.  In this same scenario, it is possible
that the reference channel is closer by 4 feet to another tire, meaning
that a sound originating in the second tire will appear 32 samples earlier.
 Perhaps you can begin to see from this oversimplified explanation why
putting the 2 mics close offers a reasonable solution to this problem,
however, the reference channel is not allowed to pick up any of the desired
signal or you'll cancel that too.  

     In a hellicopter,  the pilot will wear a specail head set, and the
noise is so great that a directional mic can be placed near the mouthpiece
mic without picking up any voice.  People don't really want to wear those,
not to mention headphones are illegal in some states while driving.  The
point is that the methods that work in hellicopters for human engineering
reasons would not be reasonable to implement in cars.  

     Sorry if I rambled a bit much.

--Jeff

m22257@mitre.org (Jeff Correia) writes:

>In article <Bz3vHr.LKK@news.cso.uiuc.edu>, ja51359@uxa.cso.uiuc.edu
>(axelrod) writes:
>> 
>> 
>> Hi.  I know this is possible because I've heard the military does it for
>> helicopter headsets; and I heard that the car companies will be doing it
>> too.  Does someone know of how the basics work behind active noise can-
>> cellation?
>> 
>> What I'm interested in is a system where you have a speaker or two
>> somehow produce a signal that is directly out of phase with the ambient
>> noise.  This seems impossible, since a speaker radiates in all directions,
>> thereby having a different phase at different points in the room.  Is 
>> there some way of getting an average phase or something, in order to
>> decrease the noise in the room/car?
>> 
>> I'd like references to articles on this subject, along with a simple
>> explanation if possible.
>> 
>> Thanks
>> 
>> 

>     The type of cancellation you are referring to is one that I have heard
>most commonly called reference cancelling.  The idea is pretty much as you
>have described it.  You have one primary channel, and 1 or more reference
>channels.  The primary channel contains your desired signal corrupted by
>unwanted noise, the reference channels contain noise which is (hopefully)
>highly correlated with the unwanted noise in your primary channel, but
>contain no trace of your desired signal.  

>     A classic paper on this topic is by Bernie Widrow.  While trying to
>examine the heartbeat of a fetus, the heartbeat of the mother is a strong
>interfering signal.  Using a primary channel of mother+fetus and being able
>to locate a position to get a "clean" but maybe slightly out of phase
>refernce of the mother's heartbeat only, they were able to obtain the
>desired results.  I'm sorry I don't have the exact reference.  




There is a book "Adaptive Signal Processing," Widrow and Stearns, which deals
with this subject.  There are also many articles in the Signal Processing
and Biomedicl Engineering literature.  As Jeff has already noted the main
requirement for using the algorithms in the book is a set of independantly
measured, correlated noise signal, which may be cancelled from the reference
signal. The algorithms  also assume that the signal are linearly combined.  
If Nonlinear signal processing is needed then Neural Network models may be
needed.  Currently, here at JHU/APL, we are using Widrow's LMS algrothms 
for deriving fetal EKG signals from non-invasive measurements from the
mother.  If there is any interest, please post here or e-mail me.


-- 

_______________________________________________________________

    	John A. Cristion