FPGA Central

On 6 Feb, 22:45, adamchapman <adamchapman1...@hotmail.co.uk> wrote:
> Hello,
>
> I was reading this paper (http://www.kecl.ntt.co.jp/icl/signal/
> nakatani/papers/icslp2002f0.pdf) and thought his Power spectra looks
> ridiculously high.
>
> using 10*log10(S) for the log power of spectrum S, S would need to be
> about 10^9 to get that high dB range.
>
> Or is there some standard that I'm unaware of?

Assuming a signed 32-bit integer number format, 1e9 represents
about 50% of the available dynamic range. It could be as simple
as that the authors used the integer format without paying
attention to scaling.

Rune

On Feb 6, 10:30*pm, Rune Allnor <all...@tele.ntnu.no> wrote:
> On 6 Feb, 22:45, adamchapman <adamchapman1...@hotmail.co.uk> wrote:
>
> > Hello,
>
> > I was reading this paper (http://www.kecl.ntt.co.jp/icl/signal/
> > nakatani/papers/icslp2002f0.pdf) and thought his Power spectra looks
> > ridiculously high.
>
> > using 10*log10(S) for the log power of spectrum S, S would need to be
> > about 10^9 to get that high dB range.
>
> > Or is there some standard that I'm unaware of?
>
> Assuming a signed 32-bit integer number format, 1e9 represents
> about 50% of the available dynamic range. It could be as simple
> as that the authors used the integer format without paying
> attention to scaling.
>
> Rune

Im not sure, the author has used the same scale for different examples
in other papers. The method I used gave a similar scale, but I don't
know if that's because Im doing it right or if Im doing it almost
right, or im just lucky with my particular data.

On Feb 6, 5:40*pm, adamchapman <adamchapman1...@hotmail.co.uk> wrote:
> On Feb 6, 10:30*pm, Rune Allnor <all...@tele.ntnu.no> wrote:
>
>
>
>
>
> > On 6 Feb, 22:45, adamchapman <adamchapman1...@hotmail.co.uk> wrote:
>
> > > Hello,
>
> > > I was reading this paper (http://www.kecl.ntt.co.jp/icl/signal/
> > > nakatani/papers/icslp2002f0.pdf) and thought his Power spectra looks
> > > ridiculously high.
>
> > > using 10*log10(S) for the log power of spectrum S, S would need to be
> > > about 10^9 to get that high dB range.
>
> > > Or is there some standard that I'm unaware of?
>
> > Assuming a signed 32-bit integer number format, 1e9 represents
> > about 50% of the available dynamic range. It could be as simple
> > as that the authors used the integer format without paying
> > attention to scaling.
>
> > Rune
>
> Im not sure, the author has used the same scale for different examples
> in other papers. The method I used gave a similar scale, but I don't
> know if that's because Im doing it right or if Im doing it almost
> right, or im just lucky with my particular data.- Hide quoted text -
>
> - Show quoted text -

the Y axis is power in dB relative to what??? probably some
arbitrary small number..

Mark

On Feb 6, 10:49*pm, makol...@yahoo.com wrote:
> On Feb 6, 5:40*pm, adamchapman <adamchapman1...@hotmail.co.uk> wrote:
>
>
>
>
>
> > On Feb 6, 10:30*pm, Rune Allnor <all...@tele.ntnu.no> wrote:
>
> > > On 6 Feb, 22:45, adamchapman <adamchapman1...@hotmail.co.uk> wrote:
>
> > > > Hello,
>
> > > > I was reading this paper (http://www.kecl.ntt.co.jp/icl/signal/
> > > > nakatani/papers/icslp2002f0.pdf) and thought his Power spectra looks
> > > > ridiculously high.
>
> > > > using 10*log10(S) for the log power of spectrum S, S would need to be
> > > > about 10^9 to get that high dB range.
>
> > > > Or is there some standard that I'm unaware of?
>
> > > Assuming a signed 32-bit integer number format, 1e9 represents
> > > about 50% of the available dynamic range. It could be as simple
> > > as that the authors used the integer format without paying
> > > attention to scaling.
>
> > > Rune
>
> > Im not sure, the author has used the same scale for different examples
> > in other papers. The method I used gave a similar scale, but I don't
> > know if that's because Im doing it right or if Im doing it almost
> > right, or im just lucky with my particular data.- Hide quoted text -
>
> > - Show quoted text -
>
> the Y axis is power in dB relative to what??? * *probably some
> arbitrary small number..
>
> Mark- Hide quoted text -
>
> - Show quoted text -

That's what Im thinking, I've chosen the small number to be the
smallest value present in the original power spectrum (abs(fft(x)).^2)
that is not zero. However I don't know if this is the right value to
use. I expected there was some kind of simple rule that everyone knew
about but me, Im very new to audio processing