# Modulation-Recognition-for-Wireless-Signals
Algorithm for classification of Digital Modulation such as FSK, PSK, ASK, QAM ........
# Short introduction about the problem of Modulation Recognition
We begin with modulation; wich simply a process by wich the desired information to be send is encoded inside an other signal that in turn can be transmitted on a physical medium.
Due to the changing RF environment, the recognition of received signals modulations schemes is becoming challenging and traditionnal methods based on manual features extractions by experts show lack of performance due to new channel conditions and different rates of SNR. So to solve this, new approaches are taken into account, the major one is " Deep Neural Networks", this is firstly due to the high capabilities of DNN for features extractions when it comes with dealing with unsupervised learning, mainly we use CNN (Convolutionnal Neural Networks), and for more advanced architecture with better accuracy, we combine CNN with LSTM or we create a residual connection.
LSTM are types of RNN ( Reccurent Neural Network), mainly used while working with sequences input. For more details about LSTM, one of the best articles that explain this architecture with highlight on the different gates and their functionnalities can be found in the following link : http://colah.github.io/posts/2015-08-Understanding-LSTMs/
It's an amazing blog post.
So, as we explain it before the call for deep learning architectures for solving the modulation recognition problem is due to its power for feature extraction. We've choosed to work on digital modulations schemes, as we know most of them are based on keying; where values of modulated signal are always one of specific set of predetermined values at all times.
# For the input dataset used is explained here
The folder " Dataset_input" contains the different input datasets that can be used created by DeepSig.
RadioML 2016.10A : A synthetic dataset, generated with GNU Radio, consisting of 11 modulations (8 digitals and 3 analogs) at varying SNR levels.
RML2016.10B : Larger Version of RadioML 2016.10a (including AM-SSB).
RadioML 2016.04C : A synthetic dataset, generated with GNU Radio, consisting of 11 modulations. This is a variable-SNR dataset with moderate LO drift, light fading, and numerous different labeled SNR increments for use in measuring performance across different signal and noise power scenarios.
The original datasets can be found Here :
https://www.deepsig.io/datasets
For the data it was generated in a way that captures various channel imperfections that are present in real system using GNU.
Concerning the input datasets that will be used in our first experiment is RadioML 2016.10b, consisting of 160 000 samples, each input is a 2*128 vector, separating real and imaginary parts in complex time samples and labels include SNR ground Truth as well as the modulation type. SNR is between -20db and +18db.
THe folder "input_data" contains the necessary files in order to generate the input datasets. One can start from here and modify the script in order to generate datas in different conditions depending on the studies.
# Architecture of Deep Neural Networks
First, we clarify that in the beginning we will be using as framework "Keras" with Tesnorflow as back end, it's more suitable for fast developement in order to evaluate our designed architectures.
For architectures, we will focus on 3 Types :
- CNN (convolutionnal Neural Networks)
- ResNet (CNN with residual connection)
- CNN_LSTM (CNN followed by an LSTM network)
Details about the architectures, performance and comments can be found in each specific folder for this Architecture.
Better architectures will be developped along the way addressing different issues, so stay Tuned for better work !
For practicality, we would love to adapt the best and choosen architecture for researchers who are in the area of communications, by writting it in MATLAB using 'Deep learning toolbox'. But unfortunately, this toolbox is no longer maintained and updated, we will investigate this, and try to figure out if it's possible to use Neural Network Toolbox in matlab for this issue. Furthermore, we project soon to develop an entire designed architecture that solve other challenges in modulation recognition problem using tensorflow with high level API, in the following weeks.
没有合适的资源?快使用搜索试试~ 我知道了~
FSK、PSK、ASK、QAM等数字调制分类算法......_Jupyter Notebook_Python_下载.zip
共30个文件
py:12个
ipynb:5个
me:4个
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 166 浏览量
2023-04-26
11:31:29
上传
评论
收藏 3.05MB ZIP 举报
温馨提示
FSK、PSK、ASK、QAM等数字调制分类算法......_Jupyter Notebook_Python_下载.zip
资源推荐
资源详情
资源评论
收起资源包目录
FSK、PSK、ASK、QAM等数字调制分类算法......_Jupyter Notebook_Python_下载.zip (30个子文件)
modulation-recognition-for-wireless-signals-master
Residual_CNN
read.me 1B
ResNet.ipynb 80KB
ResNet_CNN.py 6KB
LICENSE 34KB
CNN_LSTM_Matrix.py 9KB
ResNet.ipynb 80KB
tensorflow_cnn_mod.py 7KB
CNN for mod.py 4KB
CNN_with_Keras
Architecture_CNN_used.png 72KB
read.me 244B
CNN Training performance.png 33KB
CNN1.ipynb 67KB
CNN for mod.py 4KB
CNN.wts.h5 3.06MB
CNN3.ipynb 63KB
CNN_LSTM
CNN_LSTM.py 4KB
read.me 1B
ccc.ipynb 64KB
README.md 4KB
input_data
source_alphabet.pyc 2KB
read.me 202B
generate_alphabet_dataset 944B
transmitters.pyc 6KB
source_alphabet.py 2KB
transmitters.py 3KB
generate_RML2016.04c.py 2KB
analyze_stats.py 2KB
generate_RML2016.10a.py 3KB
analyze_stats.pyc 2KB
timeseries_slicer.py 1KB
共 30 条
- 1
资源评论
快撑死的鱼
- 粉丝: 1w+
- 资源: 9156
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功