diff --git a/README.md b/README.md
index 2f1d4ef..a4b446e 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,54 @@
-# graph_embedding_link_scheduling
\ No newline at end of file
+# graph_embedding_link_scheduling
+
+This repository contains the entire code for our TWC work "Graph Embedding based Wireless Link Scheduling with Few Training Samples", available at https://ieeexplore.ieee.org/document/9285223.
+
+For any reproduce, further research or development, please kindly cite our TWC Journal paper:
+
+M. Lee, G. Yu, and G. Y. Li, "Graph embedding based wireless link scheduling with few training samples," IEEE Trans. Wireless
+Commun., vol. 20, no. 4, pp. 2282-2294, Apr. 2021.
+
+
+# How to use?
+
+## Requirements
+
+The following versions have been tested. But newer versions should also be fine. 
+
+- rdkit : [Q3 2017 Release](https://github.com/rdkit/rdkit/releases/tag/Release_2017_09_1, Release_2017_09_2)
+- boost : Boost 1.61.0, 1.65.1
+
+## Setup
+
+Go to "s2v_lib". 
+
+Build the c++ backend of s2v_lib and you are all set.
+
+```
+cd s2v_lib
+make -j4  
+```
+
+Note: We utilize existing open-source code for the structure2vec architecture (https://github.com/Hanjun-Dai/pytorch_structure2vec/tree/master/s2v_lib) and add the batch nomarlization function in it.
+
+
+## Data Prepareation
+
+Go to  "FPLinQ".
+
+Run "generate_main.m". The output data is saved in "/mat".
+
+Copy the output data into  "D2D_qua/mat".
+
+## Run
+
+Go to "D2D_qua".
+
+Run the main program.
+
+```
+./run.sh
+```
+You can also use the following command.
+```
+python main.py
+```