2
\$\begingroup\$

I have a signal that should have this theoretical shape:

Theoretical shape

The signal I am receiving experimentally has the following shape:

Experimental unclipped shape

And here is a plot that better shows the shape of the waveform (I have clipped off most of the excess data):

Clipped experimental data

Notice the large amount of what seems to be high frequency noise in the second plot and third plot. I am trying to figure out how to filter out the noise of the the experimental waveform with keeping the overall shape. I have considered using a low-pass filter or a voltage regulator but I am not sure if they will have the desired affect. Basically I need to keep the shape (and ideally magnitude) of the peak while filtering out the channel noise.

EDIT: Don't be concerned about the reflection of the theoretical peak in plot 1. This is just a direction change and is on no concern.

\$\endgroup\$
  • \$\begingroup\$ The noise seems quite small compared to your signal. Are you sure it's causing problems? \$\endgroup\$ – The Photon Feb 12 '13 at 20:26
  • \$\begingroup\$ Yeah, really, what's the problem? If only all my signals were this clean... \$\endgroup\$ – Olin Lathrop Feb 12 '13 at 20:41
  • \$\begingroup\$ Could this be RFI sneaking into your device? If so: shielding, ferrite chokes, and fix any instances of a "pin 1 problem" (connector hookups which connect the shields of external cables to ground/return through the circuit board, instead of through the metal chassis, thereby bringing RF induced in cable shielding into the board.) \$\endgroup\$ – Kaz Feb 13 '13 at 2:20
4
\$\begingroup\$

Looks like you actually have pretty good signal to noise ratio, and the noise that is there seems to be largely above the major frequency components of the signature you are looking for. Looking at the noise at this level is not meaningful. You want to see how much signal to noise headroom you have after proper detection of your doublet (some sort of echo chirp?).

The obvious detection procedure is too convolve the input signal with the signature from your first plot. That should give you a nice single spike when the signature is detected. The width of the filter kernel (the fixed waveform you are convolving the input with, which is the first plot) indicates that much of the high frequency hash will be removed. This is a necessary step in detecting the signature anyway. Try that first and then see what the signal to noise ratio is. If the data you show is representative, then there will be a quite healthy signal to noise ratio. Detecting the chirp from the background and determining the chirp's center time should work very nicely.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.