Re: LF: Re: Slow mode comparisons

[MarkusVester@aol.com]

Hi Alberto,

a good time of the night to discuss this last 0.8 dB ;-) My reasoning was that when variing the FFT bandwidth around the optimum BW, the readability for a symbol should change only very slightly - otherwise it wouldn't be an optimum in the sense of a zero gradient of merit. If the BW is made a little too large, you do get some more noise but also benefit from a fractional bit of incoherent averaging; on the other hand if it's too small you only gently start to decrease the signal more than the noise. Even at constant receive BW one gains by transmitting slower, eg. 5s instead of 3s DFCW dashes.

73
de Markus, DF6NM


In einer eMail vom 06.03.2005 01:13:58 Westeuropäische Normalzeit schreibt dibene@usa.net: 

> MarkusVester@aol.com wrote:
>
> >My half-penny's worth is that the transmitted energy per symbol is 3.3 
> >times larger, giving exactly 5.23 dB improvement ;-) That's assuming 
> >that the FFT bandwidth is somewhere near the optimum (0.3 Hz and 0.1 
> >Hz respectively).
>
> Markus,
>
>  you are correct, but in an ideal situation. Given that for performance 
> reasons the FFT works at its best when done
> on a sample length which is a power of two, in Argo, when going from 
> QRSS3 to QRSS10 I multiply the sample length
> by four, which causes a reduction of four times of the bin size. This 
> has the consequence that the noise energy that falls
> into a single bin is 6 dB lower, hence the 6 dB gain in SNR.
> There are algorithms to perform FFTs on arbitrary sizes, but they are 
> slower than when working with powers of two.
>
> 73  Alberto  I2PHD