Datasets

  1. The signal files used for the Gallery post.
  2. A MATLAB m-file used to read my binary files (rename to read_binary.m).
  3. The DSSS signals used in the DSSS post. Use read_binary.m to read the files into MATLAB.
  4. Dataset used in my tunneling CSP paper (My Papers [43]).
  5. Dataset for the Challenge for the Machine Learners on modulation recognition. Please call this CSPB.ML.2018 in published papers. Update September 2023: Use the corrected Challenge dataset.
  6. The Generalized Challenge dataset. Please call this CSPB.ML.2022 in published papers.
  7. The Cochannel Challenge dataset. Please call this CSPB.ML.2023 in published papers.

9 thoughts on “Datasets”

  1. hello sir, i want to use your LTE dataset (5MHz signal) for blind parameters estimation using cyclostationary analysis. Can you please tell me its parameters, that could be very helpful. Like number of subcarriers, Symbol time, length of cyclic prefix, modulation used etc.

    1. Thanks for reading the CSP Blog, Shahzeb, and for the comment.

      The 5-MHz LTE signal in the Data Sets archives is captured over-the-air LTE. My analysis shows it has the expected major system parameters. That analysis is external only–no attempt at demodulation, just use of statistical techniques.

  2. Hello Sir, I am currently working on wireless signal classifiers and was trying to open the deepsignal datasets of 2018 but unable to do it which is in hdf5 format.can you please guide me for the same. Additionally can you suggest any other data sets and steps to go ahead with classifying the signal using deep learning?
    Thank You!

    1. Thanks for stopping by the CSP Blog ms!

      open the deepsignal datasets of 2018 but unable to do it which is in hdf5 format.

      I analyze DeepSig’s 2018 dataset in this post. I use h5dump under linux (Ubuntu and Fedora Core) in various ways to understand the dataset archive file. Several of the exact h5dump commands are contained in the post. Did you read that post? I analyze other DeepSig datasets here and here.

      can you suggest any other data sets

      A Challenge for the Machine Learners.
      Dataset for A Challenge for the Machine Learners.

      steps to go ahead with classifying the signal using deep learning?

      Well, no. I don’t actually recommend that researchers jump into modulation recognition with deep learning (multilayer neural networks trained with extensive labeled data examples). Machine learning has not been fairly compared to alternatives, which most machine learners appear to be underprepared to understand, code, or evaluate (just as I am underprepared to analyze and select hyperparameters and DNN/CNN layer configurations). A big flaw in the published work on modulation recognition using machine learning is that the involved researchers don’t appear to understand communication signals, statistics of communication signals, communication-signal generation, or decision theory. Thus the quality of a dataset is not considered, partly I suspect for expediency, and partly because machine learners are not trained or qualified to analyze communication signals. So … I recommend that if you want to make a contribution to modulation recognition, you start by trying to understand modulation. Gaining such understanding is, of course, the purpose of the CSP Blog.

      1. Hello Sir,
        Thanks for your valuable inputs.

        I have tried to extract and analyze the Deepsignal data sets ((2018.01.OSC.0001_1024x2M.h5.tar.gz) using software Wave Metrics Igor Pro as I don’t have Linux OS.

        I was able to extract Y and Z datasets Graph, Tables and Hdf5 dump notebook. However, was unable to retrieve data from X datasets. I have tried using other tools which are HDFView, Pandas but it did not allow me to extract it.

        I have reached out to the OmniSIG team as well to guide me further on wireless signals classifications but did not receive any concrete information on this topic.

        Can you please help me guide on how to extract analyze and Simulate the datasets in windows OS environment.

        I would be highly obliged if you could help me with this.

        1. Can you please help me guide on how to extract analyze and Simulate the datasets in windows OS environment.

          No, I can’t. I don’t do any technical work using Windows. It is a PowerPoint platform as far as I am concerned. I had no idea how to ‘extract’ the signals in DeepSig’s h5 file, so I just had to learn about the available tools under linux (h5dump). ‘Analyze’ and ‘simulate’ are much more involved, and I suggest you start with something simple like making digital signals using MATLAB.

    1. Yes. Do you already have it? I had to post it as a file with a different extension because, for some reason, WordPress does not allow me to post files ending in ‘.m’. So you can go to the Downloads page, download read_binary.doc, then just rename it to read_binary.m. It will read any .tim file I’ve posted to the CSP Blog into MATLAB.

Leave a Reply to msCancel reply