Question about doing modulation recognition using spectograms

[Maxwell20]

I think you lose phase information with spectograms how does the model pick up modulation type without knowing phase over time?

[MattCarrickPL]

Many signals can be distinguished from one another visually within the spectrogram. We drop the phase information but differences can still be seen through the magnitude as shown in this spectrogram from our GNU Radio paper:

[MattCarrickPL]

You also have the option to use the narrowband model, which is what the first paper focuses on: https://arxiv.org/pdf/2207.09918

[Maxwell20]

Understood. It makes since to me that modulations that varies in power such as qam may be distinguished but what about PSK where only phase is varied? Excuse me if this is an ignorant question I am relatively new to DSP but experienced with programming and ML. Thanks

…

On Mon, Jan 20, 2025, 8:16 AM MattCarrickPL ***@***.***> wrote: Many signals can be distinguished from one another visually within the spectrogram. We drop the phase information but differences can still be seen through the magnitude as shown in this spectrogram from our GNU Radio paper: Screenshot.from.2025-01-20.09-09-51.png (view on web) <https://github.com/user-attachments/assets/789cfbf1-0cc1-49f2-a8e9-aa9bc714491b> — Reply to this email directly, view it on GitHub <#263 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACS3FTYJC2QCESY6RWJWGBD2LUAMHAVCNFSM6AAAAABVMTDFGSVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCOBZGEYTKMI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[MattCarrickPL]

It's a good question. From our 2022 paper you can see that we are able to distinguish between signal families well:

But distinguishing within a family is more difficult, for the reasons you have mentioned. You can see in the dispersion in the confusion matrix around the QAM/PSK/ASK signals, and also the MSK/GMSK/FSK/GFSK signals:

[ereoh]

Feel free to reopen if you have more questions.