Re: Constant baseline between FFTs

[Fons Adriaensen]

On Sat, Dec 30, 2023 at 02:54:27PM +0100, Dr Cdiff wrote:

> Here is an example (without division by median!): https://ibb.co/FmxZ3ZK (x
> is time in h and y is frequency in weird unit)

If that is kHz, the sample rate would be 48 kHz. Is your input an audio
file ?

Also, if the X-axis is in hours, then each pixel would represent quite
a lot of input vectors and FFTs. How are these combined for display 
(average, maximum, ...) ?

The level of the discrete signals seems essentially the same in 
the left and right halves, suggesting that the gain is indeed 
constant. Could it be that you are just seeing the correct result ?
For radio signals it is quite normal for noise levels to change
over the course of a day. 

And it is not even clear that what you see between 16h and 22h is
continuous noise. It could be impulsive, as suggested by the occasional
vertical lines in the left part. To find out you should process a much
smaller time range, so each individual FFT output is displayed.

> no AGC slows down the flowgraph (yes, that is weird).

That shouldn't happen. There must be something wrong there. 

Could you share the input file, or some parts of it ?

Ciao,

-- 
FA