Hi all,
I'm trying to estimate the impulse response of the room by playing a
4 sec white noise speech file from one end of the room and recording it at the
other end. I'm told to estimate the impulse response of the room by using
Recursive Least Squares method. Please help how to get this done, any help in
this regard will be greatly appreciated.
Thanks,
Abhishek
Abhishek Agarwal
Estimating the Impulse Response of the room !!!
Started by ●May 1, 2006
Reply by ●May 2, 20062006-05-02
if you are using white noise as excitation, RLS and
NLMS are pretty close. however, 4 sec is not enough.
use 10...20.
use something like that:
xrcv = wavread();% ref (RCV) samples
xsnd = wavread();
elen = % echo len in samples. shall be at least 0.7
*RT60
h=zeros(elen,1);
shft=elen/8;
for k=elen:xrcv_length,
x=xrcv(k+1-elen:k);
err=xsnd(k-shft)-x'*h;
ss=1/(1+0.3*(k/elen));
h=h+ss*x*err/(x'*x);
end;
h=flipud(h);
q=norm(h(1:elen/10))/norm(h(elen/10+1:))
if elen is correct for tis room and alg converges, the
first 1/8th of h shall be almost 0, the end of echo
tail shall be on the same very low level, and q shall
be at most 0.01. if it isn't, look for mistakes in
your audio setup.
good luck!
michael.
--- abhishek agarwal wrote:
---------------------------------
Hi all,
I'm trying to estimate the impulse response of
the room by playing a 4 sec white noise speech file
from one end of the room and recording it at the other
end. I'm told to estimate the impulse response of the
room by using Recursive Least Squares method. Please
help how to get this done, any help in this regard
will be greatly appreciated.
Thanks,
Abhishek
Abhishek Agarwal
NLMS are pretty close. however, 4 sec is not enough.
use 10...20.
use something like that:
xrcv = wavread();% ref (RCV) samples
xsnd = wavread();
elen = % echo len in samples. shall be at least 0.7
*RT60
h=zeros(elen,1);
shft=elen/8;
for k=elen:xrcv_length,
x=xrcv(k+1-elen:k);
err=xsnd(k-shft)-x'*h;
ss=1/(1+0.3*(k/elen));
h=h+ss*x*err/(x'*x);
end;
h=flipud(h);
q=norm(h(1:elen/10))/norm(h(elen/10+1:))
if elen is correct for tis room and alg converges, the
first 1/8th of h shall be almost 0, the end of echo
tail shall be on the same very low level, and q shall
be at most 0.01. if it isn't, look for mistakes in
your audio setup.
good luck!
michael.
--- abhishek agarwal wrote:
---------------------------------
Hi all,
I'm trying to estimate the impulse response of
the room by playing a 4 sec white noise speech file
from one end of the room and recording it at the other
end. I'm told to estimate the impulse response of the
room by using Recursive Least Squares method. Please
help how to get this done, any help in this regard
will be greatly appreciated.
Thanks,
Abhishek
Abhishek Agarwal