Hi Alexandre,

Can you be a bit more specific as to what you mean by "narrow band noise"?=
=20
White noise is, by definition, not narrow band. I just tried a little=20
experiment with white noise and even if there is a large difference between=
=20
the sample at the end of the block and the sample at the beginning of the=20
block, you cannot hear any click unless the magnitude of the noise is=20
considerably less than the maximum range of the samples (which, it seems to=
=20
me, would be a silly way to do things, why make your noise only a tiny=20
fraction of the dynamic range of your system if the noise is the only thing=
=20
you are interested in).

So, I suspect that the problem may lie in that you are using non-white=20
noise. White noise essentially CONSISTS of large step after large step from=
=20
sample to sample, so an additional large step being added periodically adds=
=20
a slightly noticeable periodic quality to the noise, but doesn't change the=
=20
overall character of it.

One thing which might work in a variety of cases (even fairly narrow band=20
signals) would be to play the block of samples forwards and then backwards=
=20
and then forwards, etc., alternating like that. This would ensure that the=
=20
curve which the output stream followed was continuous, since the boundary=20
between two blocks would just be a repetition of the same sample. For=20
example, if your block of "noise" was
[1 5 7 9 9 4] then instead of playing [1 5 7 9 9 4 1 5 7 9 9 4 1 5 7 9 9=20
4]  play [1 5 7 9 9 4 4 9 9 7 5 1 1 5 7 9 9 4]. This still does nothing=20
about phase, though, so there would be clicks with, say, a sine wave, since=
=20
the first derivative is now discontinuous.

What is your application? You should be aware that repeating the block like=
=20
this will have an effect on the spectrum, so it may not be usable for some=
=20
purposes. Also, I think you will have trouble making the human ear not=20
notice at least slightly that it is repetitive. Alternatively, you could=20
implement a random number generator very easily in PIC code which would not=
=20
repeat for hours or days. Even if you needed a non-white spectrum, you=20
could easily use a modified version which supplies the kind of noise=20
needed, or use a white noise type generator and follow it with a simple=20
digital filter.

Hope this helps :-)

Sean

At 12:53 AM 10/16/01 -0300, Alexandre Guimar=E3es wrote:
>Hi,
>
>     This question goes to the more math enlightened of the list :-)
>
>     I have to make an audio noise generator that has to repeat a narrow=
 band
>noise over and over again. I have enough memory to hold 0.5 segs of noise.
>My problem is that when I get from the end to the begining of the cycle I
>always get a click on the sound output. I have already tried to apply a
>number of "windowing" functions ( Hamming, Kayser and etc.)  to try to make
>the begin and end of the noise to be as close to each other as possible but
>it has not worked ! I also have tried to match it by hand with wave editors
>and have not been able to get good results !
>
>     The question is : Is there any way to generate white noise, filter=
 with
>3 db by octave filters and make it so that the begin and end of the wave=
 can
>be tied together without generating and audible click ??? I need 16000
>samples at 32K sample rate. I never imagined that human audition would be=
 so
>sensitive to phase differences !!
>
>     I have really tried all the ways I can imagine and would really
>apreciate any help you guys could give me.
>
>Best regards,
>Alexandre Guimaraes
>
>--
>http://www.piclist.com hint: The list server can filter out subtopics
>(like ads or off topics) for you. See http://www.piclist.com/#topics

----------------------------------------------------
Sign Up for NetZero Platinum Today
Only $9.95 per month!
http://my.netzero.net/s/signup?r=platinum&refcd=PT97

--
http://www.piclist.com hint: The list server can filter out subtopics
(like ads or off topics) for you. See http://www.piclist.com/#topics