Trying to understand hi lo frequency in FFT

I the frequency spectrum the lower numbers are low freq. and high numbers
hi freq.


If if sample at a higher rate. Ie. double the sample speed I get high
frequencies?

that is correct?

So if i keep the same number of samples but double the rate of capture I
will end up with 2 faster samples than when the don't overlap make a more
detailed view of the slower sample.

Could you then compare the frequencies that overlap between the fast and
slow capture rates, I am trying to find a stable method of captureing hi an
lo freq.

My FFT is very biased towards low frequencies

Thanks

On May 4, 8:08&#4294967295;pm, "tinkerz" <nwoodhamuk@n_o_s_p_a_m.googlemail.com>
wrote:
> I the frequency spectrum the lower numbers are low freq. and high numbers
> hi freq.
>
> If if sample at a higher rate. Ie. double the sample speed I get high
> frequencies?
>
> that is correct?
>
> So if i keep the same number of samples but double the rate of capture I
> will end up with 2 faster samples than when the don't overlap make a more
> detailed view of the slower sample.
>
> Could you then compare the frequencies that overlap between the fast and
> slow capture rates, I am trying to find a stable method of captureing hi an
> lo freq.
>
> My FFT is very biased towards low frequencies

Frequency is the reciprocal of time. Since an FFT is for discrete
(sampled) signals, the unit of time is the sampling interval. With
higher sampling rates, eavh bin represents higher actual frequencies.

Jerry
--
Engineering is the art of making what you want from things you can get.

On 5/4/2011 5:08 PM, tinkerz wrote:
> I the frequency spectrum the lower numbers are low freq. and high numbers
> hi freq.
>
>
> If if sample at a higher rate. Ie. double the sample speed I get high
> frequencies?
>
> that is correct?
>
> So if i keep the same number of samples but double the rate of capture I
> will end up with 2 faster samples than when the don't overlap make a more
> detailed view of the slower sample.
>
> Could you then compare the frequencies that overlap between the fast and
> slow capture rates, I am trying to find a stable method of captureing hi an
> lo freq.
>
> My FFT is very biased towards low frequencies
>
> Thanks
>
>

Jerry already gave you a good answer.   I'll just try to address each 
question as you wrote them:

If you sample at a higher rate then the highest frequency bin out of the 
FFT is 1 frequency sample interval less than that.
The frequency sample interval is equal to the reciprocal of the time 
window / length of the temporal sequence that you transformed .. plus 
one time sample interval.

If you keep the number of samples the same and change the sample rate, 
two things happen.  Taking the "increased sample rate" change:

1) the temporal window gets smaller.  This makes the frequency sample 
interval larger.  So, frequency resolution is reduced.

2) Think of it as: the frequency range is fs and you are splitting it by 
N.  You say that N is fixed so the N frequency samples are further apart 
when you increase fs.

Of course, you wouldn't sample with fs not greater than the highest 
frequency in the signal.  So, if the signal character doesn't change 
then increasing fs may not provide any advantage - although it's maybe 
easier to look at.

By "overlap" it sounds like you're taking N samples and then taking the 
next N samples, etc.  Let's call this N and 2N, OK?

If you take 2N samples at fs then the temporal window is 2N/fs and the 
frequency resolution is fs/2N.

If you take N samples at fs then the temporal window is N/fs and the 
frequency resolution is fs/N.

[by "frequency resolution" here I mean fs/N or fs/2N - the distance 
between the frequency samples].

If you want more information then generally you need to grab more 
incoming data.
- If you keep the time window fixed and increase the sample rate then 
you improve the temporal resolution and increase the frequency span.
- If you keep the sample rate fixed and increase the temporal window 
then you improve the frequency resolution and increase the time span.
- If you increase both the temporal window and the sample rate then you 
improve both the temporal resolution and the frequency resolution.
It's an engineering tradeoff to select the numbers you want/need.

Fred

On May 4, 8:08&#4294967295;pm, "tinkerz" <nwoodhamuk@n_o_s_p_a_m.googlemail.com>
wrote:
> I the frequency spectrum the lower numbers are low freq. and high numbers
> hi freq.
>
> If if sample at a higher rate. Ie. double the sample speed I get high
> frequencies?
>
> that is correct?
>
> So if i keep the same number of samples but double the rate of capture I
> will end up with 2 faster samples than when the don't overlap make a more
> detailed view of the slower sample.
>
> Could you then compare the frequencies that overlap between the fast and
> slow capture rates, I am trying to find a stable method of captureing hi an
> lo freq.
>
> My FFT is very biased towards low frequencies
>
> Thanks

the frequencies you get are seperated by sample_rate/sample_points Hz

so for 1000Hz sample rate and 1000 samples you get 1 Hz wide spectral
estimates

increasing the sample rate to 2000Hz means each bin is now 2 Hz wide

if you want low frequency accuracy you have to increase the samples

OK thanks, that is clearer

My frequency spectrum is showing that the highest frequency is the low part
of the spectrum, it is constantly close to the first bin, the lowest
frequency.

I would like to have a spectrum where high and low frequencies are almost
balanced in power.

So if I had 64 bins, then power would be almost equal for bins 1 to 32 and
33-64,

I would have to increase my sample rate, or sample rate? the data will be
fairly consistent.

Thanks

On May 5, 3:33&#4294967295;pm, "tinkerz" <nwoodhamuk@n_o_s_p_a_m.googlemail.com>
wrote:
> OK thanks, that is clearer
>
> My frequency spectrum is showing that the highest frequency is the low part
> of the spectrum, it is constantly close to the first bin, the lowest
> frequency.
>
> I would like to have a spectrum where high and low frequencies are almost
> balanced in power.
>
> So if I had 64 bins, then power would be almost equal for bins 1 to 32 and
> 33-64,

then you need an impulse input, set one of the input samples to 1, the
rest to 0, then you will have identical power everywhere in the
spectrum

>then you need an impulse input, set one of the input samples to 1, the
>rest to 0, then you will have identical power everywhere in the
>spectrum

Impulse input? I am a but lost on this, can you explain a bit more.

I should have said:

In my power spectrum, the low frequencies carry the most power, and its
always in the first one or two bins.

This is what i am trying to find out, I think either i am not sampling
enough or the data capture is too slow.

Or something else :)

I dont think it should be that skewed all the time

On May 5, 4:53&#4294967295;pm, "tinkerz" <nwoodhamuk@n_o_s_p_a_m.googlemail.com>
wrote:
> I should have said:
>
> In my power spectrum, the low frequencies carry the most power, and its
> always in the first one or two bins.
>
> This is what i am trying to find out, I think either i am not sampling
> enough or the data capture is too slow.
>
> Or something else :)
>
> I dont think it should be that skewed all the time

Hard to say without more info (what kind of setup you have), I assume
you don't know the bandwidth of the input signal and aliasing may be
occurring resulting in the FFT erroneously reporting higher powerer at
lower frequencies than their really is. In that case increasing the
sampling rate is necessary, increasing the sampling time won't help.

>
>Hard to say without more info (what kind of setup you have), I assume
>you don't know the bandwidth of the input signal and aliasing may be
>occurring resulting in the FFT erroneously reporting higher powerer at
>lower frequencies than their really is. In that case increasing the
>sampling rate is necessary, increasing the sampling time won't help.
>

OK thanks, then i need to find a method to gauge the bandwidth on the input
signal

What sort of code/formula area can i research to gain an understanding of
bandwidth?