I have a 2 resistors of 1000 k each , I have connected them in series and biased one end with +9 V and other end with +9 V.
The output is taken from connection point of two resistors and I have given that to amplifier with Gain 20000 then to spectrum analyzer.
In log mode of operation I should get a flat signal which is independent of frequency (x axis) , but the signal is gradually decaying.
Does anyone have any idea where I am going wrong and what should I do so that the plot comes flat.
If you are using an amplifier, what you see is not just Johnson noise: actually, even if you use just a resistor you will also see interference coming from broadcasting, the 50/60 Hz grid, other devices, and the instrument itself. All these components of the noise will have a different spectrum, and what you see is the sum of those spectra.
Now, since you're also using an amplifier, you will see the other sources of the noise, like the Pink, shot, burst noise, etcetera. Some of them are inversely proportional with the frequency, some are "flat" like the thermal noise, but overall you shouldn't be able to isolate the latter, unless you make it more significant than the other sources.
Just another addition. I don't know what kind of amplifier you are using, but with such a high gain I don't expect it to be very "noise-free", as well.
