0
$\begingroup$

I would like to validate a trained model based on RML2016.10a. My idea is to capture a certain amount of data (20 Msps) for approximately 0.3 seconds. Then, I identify all the signals in this bandwidth (I know their frequency and bandwidth). Next, I would like to feed these signals one by one into the model and classify their modulation. What preprocessing do I need to perform on each signal for my model to correctly interpret it? I have data in both the time and frequency domains

$\endgroup$

1 Answer 1

0
$\begingroup$

You should preprocess it so that it "looks like" the training data. If you did trained using raw baseband signal then use that as input, if you did some filtering or other operations to the raw signal beforehand then you should match that before passing to your model. In my experience, things can be finicky if you start giving signals that had undergone different preprocessing steps.

$\endgroup$
3
  • $\begingroup$ This is a dataset with artificially generated signals deepsig.ai/datasets. With a bandwidth of 1 Msps. I receive a signal with the same sample rate. And I filter them according to the bandwidth of the input signal $\endgroup$
    – Ivan
    Commented Oct 30, 2023 at 21:34
  • $\begingroup$ . But I constantly detect the same modulation. I think that the problem is in data normalization. And I don't understand how to normalize them. $\endgroup$
    – Ivan
    Commented Oct 30, 2023 at 21:34
  • $\begingroup$ Different normalization would fall under my experience of things being finicky when different preprocessing had been applied $\endgroup$
    – Engineer
    Commented Oct 31, 2023 at 12:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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