input referred noise for a bandpass filter - integration limits

[Winny_Puuh]

Hi guys,
just a short question: I have a bandpass with a lower cutoff frequency at about 1 Hz and a high cutoff frequency at 15kHz. When simulating the input reffered noise, I integrate from 0.01 to 20kHz but somehow I'm not sure whether those limits are justified. I could theoretically integrate up to 100kHz and my noise will increase... what upper limit is a reasonable choice?
Thanks for the help,
Winny

 

[crutschow]

It depends upon the rolloff of the filter. What is the filter order (number of poles)?

 

[FvM]

I don't see a resonable purpose to calculate an integral input referred noise quantity at all. What should be its meaning, technically?

In so far I think, the question is actually referring to a non-problem.

 

[Winny_Puuh]

@ crutschow: the rolloff frequency is at 15kHz (actually it's tunable). The system has a second pole somewhere in the Mhz range. So how does this help? Unfortunately I don't see how to reasonably choose the upper integration Limit (is it 1 or 2 or 10 times the rolloff frequency?)
@FvM: I want the input referred noise RMS value, so obviously I have to integrate (even if I had no clue - my tutor says so and papers say so, they just dont give me the limits) oh and the technical meaning: how small may an input signal be so that is distinguishable from the noise floor

 

[FvM]

technical meaning: how small may an input signal be so that is distinguishable from the noise floor.

The "detectable signal" will be different for each input frequency respectively spectrum of a wideband signal. That's why I think it's a meaningless number.

Meaningful quantities are input and output referred noise density and - for some measurement problems - output referred integral values, e.g. total noise power or RMS voltage.

 

[Winny_Puuh]

Actually, now I get what you mean and I have asked myself as well, why I would calculate the input referred noise as rms value... but some papers do that and my tutor wants me to reach a certain value as specification and he is quite experienced. I'm sure he knows what he does... I know this is quite an unsatisfying reason but it's the only one I have, so I just need to find out what limits I have to choose >__<

 

[FvM]

You may want to see the impossibility to find a reasoning for a particular cut-off frequency as a proof that it's useless. Or just follow your tutor's suggestions. :smile:

 

[crutschow]

I don't see a resonable purpose to calculate an integral input referred noise quantity at all. What should be its meaning, technically?

In so far I think, the question is actually referring to a non-problem.

I believe it means you integrate the noise over the bandwidth, given an input noise spectrum, to get the total RMS noise at the output (as would be measured by a true RMS wideband meter.

A reasonable upper frequency limit for this integration would be where the filter rolloff response is below the minimum signal level.

 

[FvM]

I believe it means you integrate the noise over the bandwidth, given an input noise spectrum, to get the total RMS noise at the output (as would be measured by a true RMS wideband meter.

A reasonable upper frequency limit for this integration would be where the filter rolloff response is below the minimum signal level.

That's in fact a classical noise measurement setup. You have a noise spectral distribution (e.g. white noise) at the amplifier input and calculate the expectable output noise voltage. Integration limits would be chosen so that no relevant noise components are missed.

But Winny_Puuh clearly describes the calculation of an input referred integral quantity and also the paradoxon that the result depends on the chosen band limits and doesn't converge againts a final value. Unfortunately this contradiction can't be resolved.