# Filtering...looking for constant amplitudes

Started by April 28, 2009
```Hello,

I have a continuous spectral plot that I need to filter. The spectrum is
unique in that at a glance it looks like white noise, however if you look
closer there are bins (frequencies) that have constant amplitudes. The
problem is if I simply average the continuous spectrum, the noise averages
to the same amplitudes as the constants. What would be the best algorithm to
extract these constant amplitudes/frequencies? Would video averaging work or
does it just average the noise and I wouldn't see the constants?

Thomas

```
```On Apr 28, 1:08&#2013266080;pm, "Thomas Magma" <somewh...@overtherainbow.com>
wrote:
> Hello,
>
> I have a continuous spectral plot that I need to filter. The spectrum is
> unique in that at a glance it looks like white noise, however if you look
> closer there are bins (frequencies) that have constant amplitudes. The
> problem is if I simply average the continuous spectrum, the noise averages
> to the same amplitudes as the constants. What would be the best algorithm to
> extract these constant amplitudes/frequencies? Would video averaging work or
> does it just average the noise and I wouldn't see the constants?
>
> Thomas

Can you decrease the resolution bandwidth (larger FFT)?

John
```
```
Thomas Magma wrote:

> Hello,
>
> I have a continuous spectral plot that I need to filter. The spectrum is
> unique in that at a glance it looks like white noise, however if you look
> closer there are bins (frequencies) that have constant amplitudes. The
> problem is if I simply average the continuous spectrum, the noise averages
> to the same amplitudes as the constants. What would be the best algorithm to
> extract these constant amplitudes/frequencies? Would video averaging work or
> does it just average the noise and I wouldn't see the constants?

Obvious answer: average the complex spectrum, not the power spectrum.
If you only have the power spectrum, then compute the mean and the
deviation for every bin; estimate S/N in the bin from there.

DSP and Mixed Signal Design Consultant
http://www.abvolt.com

```
```On 28 Apr, 19:08, "Thomas Magma" <somewh...@overtherainbow.com> wrote:
> Hello,
>
> I have a continuous spectral plot that I need to filter. The spectrum is
> unique in that at a glance it looks like white noise, however if you look
> closer there are bins (frequencies) that have constant amplitudes. The
> problem is if I simply average the continuous spectrum, the noise averages
> to the same amplitudes as the constants. What would be the best algorithm to
> extract these constant amplitudes/frequencies? Would video averaging work or
> does it just average the noise and I wouldn't see the constants?

The answer depends entirely on what you attempt to do.

What kind of application is this? Exactly what is it
you try to infer from the data? The mere presence of
sinusoidals? Frequencies? Amplitudes? Bandwidths?

Rune
```
```"Vladimir Vassilevsky" <antispam_bogus@hotmail.com> wrote in message
news:_fYJl.15429\$hc1.2488@flpi150.ffdc.sbc.com...
>
>
> Thomas Magma wrote:
>
>> Hello,
>>
>> I have a continuous spectral plot that I need to filter. The spectrum is
>> unique in that at a glance it looks like white noise, however if you look
>> closer there are bins (frequencies) that have constant amplitudes. The
>> problem is if I simply average the continuous spectrum, the noise
>> averages to the same amplitudes as the constants. What would be the best
>> algorithm to extract these constant amplitudes/frequencies? Would video
>> averaging work or does it just average the noise and I wouldn't see the
>> constants?
>
> Obvious answer: average the complex spectrum, not the power spectrum.
> If you only have the power spectrum, then compute the mean and the
> deviation for every bin; estimate S/N in the bin from there.
>
>
> DSP and Mixed Signal Design Consultant
> http://www.abvolt.com
>
>

Thanks for the response, I don't have the complex spectrum, but it looks
like I ended up doing something very similar to what you suggested. I ran
the bins through a low pass filter and then determined the absolute
difference between the filtered data and each raw point. The constant
amplitude bins now stand out like sore thumbs.

Thomas

```
```On Wed, 29 Apr 2009 16:27:34 -0700, "Thomas Magma"
<somewhere@overtherainbow.com> wrote:

>"Vladimir Vassilevsky" <antispam_bogus@hotmail.com> wrote in message
>news:_fYJl.15429\$hc1.2488@flpi150.ffdc.sbc.com...
>>
>>
>> Thomas Magma wrote:
>>
>>> Hello,
>>>
>>> I have a continuous spectral plot that I need to filter. The spectrum is
>>> unique in that at a glance it looks like white noise, however if you look
>>> closer there are bins (frequencies) that have constant amplitudes. The
>>> problem is if I simply average the continuous spectrum, the noise
>>> averages to the same amplitudes as the constants. What would be the best
>>> algorithm to extract these constant amplitudes/frequencies? Would video
>>> averaging work or does it just average the noise and I wouldn't see the
>>> constants?
>>
>> Obvious answer: average the complex spectrum, not the power spectrum.
>> If you only have the power spectrum, then compute the mean and the
>> deviation for every bin; estimate S/N in the bin from there.
>>
>>
>> DSP and Mixed Signal Design Consultant
>> http://www.abvolt.com
>>
>>
>
>Thanks for the response, I don't have the complex spectrum, but it looks
>like I ended up doing something very similar to what you suggested. I ran
>the bins through a low pass filter and then determined the absolute
>difference between the filtered data and each raw point. The constant
>amplitude bins now stand out like sore thumbs.

I was thinking run each bing through a highpass filter and an
envelope (absolute value and running average) function, so the
constant bins will have zero output. But what you did is essentially
the same.

This is an intriguing signal. Is it some sort of steganography
technique?

>
>Thomas
>

```
```>>> Thomas Magma wrote:
>>>
>>>> Hello,
>>>>
>>>> I have a continuous spectral plot that I need to filter. The spectrum
>>>> is
>>>> unique in that at a glance it looks like white noise, however if you
>>>> look
>>>> closer there are bins (frequencies) that have constant amplitudes. The
>>>> problem is if I simply average the continuous spectrum, the noise
>>>> averages to the same amplitudes as the constants. What would be the
>>>> best
>>>> algorithm to extract these constant amplitudes/frequencies? Would video
>>>> averaging work or does it just average the noise and I wouldn't see the
>>>> constants?
>>>
>>> Obvious answer: average the complex spectrum, not the power spectrum.
>>> If you only have the power spectrum, then compute the mean and the
>>> deviation for every bin; estimate S/N in the bin from there.
>>>
>>>
>>> DSP and Mixed Signal Design Consultant
>>> http://www.abvolt.com
>>>
>>>
>>
>>Thanks for the response, I don't have the complex spectrum, but it looks
>>like I ended up doing something very similar to what you suggested. I ran
>>the bins through a low pass filter and then determined the absolute
>>difference between the filtered data and each raw point. The constant
>>amplitude bins now stand out like sore thumbs.
>
>   I was thinking run each bing through a highpass filter and an
> envelope (absolute value and running average) function, so the
> constant bins will have zero output. But what you did is essentially
> the same.
>
>   This is an intriguing signal. Is it some sort of steganography
> technique?
>

I had to look up definition of steganography...so no I'm not doing that.
It's part of a search algorithm used to position locate based on the Doppler
shift.

Thomas

```