Open carlosrivera22 opened 5 years ago
For what it's worth, I concur that many of the modulation mappings appear to be incorrect, when you take the positions of the Y variable and overlay them on the positions of the classes.txt list. I have not yet figured out what the true mappings are, however.
Same problem. This is an example QPSK (supposedly) from the dataset with the following plots: spectrum, constellation plot of even samples, constellation plot of odd samples, real-part of IQ samples, imaginary-part of IQ samples.
it makes no sense to me. This looks like OOK Also, I think all the signals are oversampled by a factor of 8
This is what I get for a BPSK
This is what i get for OOK:
So 'OOK' or label 24 is definitely either PSK16 or PSK32
what the true mappings?
@pfeatherstone @carlosrivera22 @carlosrivera22 Has someone figured this out?
Don't know. I'm not trusting this dataset
The best thing to do is to buy some SDRs and generate your own dataset
@pfeatherstone
I got some raw IQ data, use model.predict to predict the model, does the incoming prediction data need to be normalized? "If necessary, how is this normalization achieved?
Is there any other better way to do signal modulation recognition
Well this is a problem that goes back decades and using neural networks is gaining interest. But I don’t recommend using this dataset for training unless the correct mapping is known and the data is checked again.
In my opinion, the modulation classification list attached to this data set is out of order. According to my exploration, I found that the correct sequence was mentioned in the author's 2018 paper
@hardware-master Thank you! Good spot. I didn't think to check the order of the list in the paper.
The best thing to do is to buy some SDRs and generate your own dataset
Could you please give me some ideas on how can I generate own dataset? Actually I was looking for BFSK and 4 MFSK datasets. Which I cant find here as well. Thanks!
I also think there is some problem with the labels of this dataset
The classes.txt file in the dataset is in the following order: classes = ['32PSK', '16APSK', '32QAM', 'FM', 'GMSK', '32APSK', 'OQPSK', '8ASK', 'BPSK', '8PSK', 'AM-SSB-SC', '4ASK', '16PSK', '64APSK', '128QAM', '128APSK', 'AM-DSB-SC', 'AM-SSB-WC', '64QAM', 'QPSK', '256QAM', 'AM-DSB-WC', 'OOK', '16QAM']
I'm not sure if that is the same order as in the dataset.
Has someone figured this out?