[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [linrad] RE: Silly question or is it?

Thanks for the info Leif, I now understand how it works. Just what I needed
on the eve of a Maths exam at University hi! Anyway I need not have worried,
it went very well, only 2 more exams to go and then I can return to the land
of the living for another 3-4 months.



-----Original Message-----
From: owner-linrad@xxxxxxxxxxxxxxxxxxxxxx
[mailto:owner-linrad@xxxxxxxxxxxxxxxxxxxxxx]On Behalf Of Leif Asbrink
Sent: 14 January 2003 12:35
To: linrad@xxxxxxxxxxxxxxxxxxxxxx
Subject: [linrad] RE: Silly question or is it?

Hi Roger and All,

> Yes, I just did it again.  With [1] as the choice for 2nd forward FFT
> routine  my total delay is:  6.96 seconds and the fft2 delay is 2.73
> sseconds.
> I go into the parameter screen and change it to [2].  I change nothing
> else and directly leave the parameter screens by repeatedly hitting
> enter and keeping my mouse out of trouble.  Now my total delay is 17.22
> sec and the fft2 delay is 10.92 seconds.
> I go back to [1] as the choice and the delay is again 6.97 seconds and
> the fft delay is 2.73 seconds.
> Here are my parameters:

> First FFT bandwidth (Hz) [50]
> Second FFT bandwidth factor in powers of 2 [6]
> Second FFT window (power of sin) [2]
OK. This is the explanation:

You ask for a bandwidth of 50Hz divided by 64 = 0.78Hz.
With a sine squared window the fft must span 2.5 seconds
which means that the transform size has to be 246000.
Now the sizes come in powers of two so you will get
an fft2 size of 262144 which spans 2.730667 seconds.
These transforms are interleaved so a new one arrives
every 1.365 second. An fft2 delay of 10.92 seconds
means that you have 8 completed transforms in the
buffer. This is because you have specified delay=8
in the AFC window. There is an additional delay of
2.73 seconds to allow for the collecting of the next
complete transform so there is 3.57 seconds for
everything else.

These settings are for coherent averaging of stable signals.
I do not think the combination of narrow bandwidth and
long delay time is useful for ordinary CW.

Now, when you select fft2 mode 0 or 1, these algorithms
are limited to max size 65536. The unscramble tables
use 16 bit words to save some time and therefore the
table can not go higher than 65536.

With the fft2 transforms arriving four times more often
the fft2 related delays will go down by a factor of four.
so the eight transforms are now 2.73 seconds and the
incoming one 0.68 for a total of 3.41. You observe a total
of 6.97 so in this case there is 3.56 seconds for
everything else. Presumably you have a baseband filter
with excessive delay.

The parameter select screens for fft1 and for fft2 are followed
by test routines that give information about what the
parameter selection resulted in. The parameters represent
what you ask for, the test screen tells you what you really
got. Look carefully at the test screen. There are many situations
where what you actually get deviates from what you ask for.

If you change to
> Second FFT bandwidth factor in powers of 2 [4]
you should get identical results with the different
fft2 versions.

By the way, First FFT amplitude [300] is an indication to me
that something is wrong. Too much gain somewhere. As long as
the AD converter does not saturate it does not matter but
should you ever notice A/D margin=0 when running the
amplitude test (A from keyboard) there is an obvious reason;)