Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

ECE 553/653 Neural Networks

Project 1: Vehicle Positioning

You will design a simple machine learning algorithm for vehicle positioning. Your algorithms will be responsible for estimating vehicle position using channel state information from the users.  Your  protocol will betested on an emulated unreliable network.

1. Requirements

You will write two programs: a LoS/NLoS classification program that classifies the signals of the vehicle according to the channel state information (CSI), and a vehicle positioning program that estimates the vehicle position using CSI. It is recommended that your programs are written in python.

For each program, you will propose a machine learning (ML) method (i.e., regression, KNN) and an artificial neural network (ANN) method.

You have to design your own data preprocessing method, input format, neural network architecture, and use ML and ANNs for LoS/NLoS signal classification and vehicle positioning.

2. Dataset Introduction

The Scenario

Information of collecting locations

This CSI dataset is collected in an underground parking lot of an office building. For data collection, the vehicle is randomly placed at 476 different locations which are either line-of-sight (LoS) or non-line-of- sight (NLoS). The scenario layout and information of the 476 locations are shown in the two images above.

The CSI dataset has been divided into a training dataset with 15000 data samples and a validation dataset with 5000 data samples. Each data sample includes a complex-valued CSI matrix used as input of a model, and one label vector. The CSI matrix has a dimensionality of 4 × 1632 , where 4 is the number of antennas of the remote radio unit (RRU), and 1632 is the number of sub-carriers. The label vector has 3 dimensions, in which the first two dimensions respectively represent the x and y coordinates of the vehicle (relative to the coordinate of RRU), and the third dimension is a binary LoS/NLoS indicator (0 for LoS, 1 for NLoS).

The two datasets are named as ‘train_dl.pt’ and ‘valid_dl.pt’. To read in those two files, you may need to install and import the following libraries: torch, torch.utils.data.TensorDataset, torch.utils.data.DataLoader.

You can download the dataset using the link:

https://drive.google.com/drive/folders/14EzLvk5PsTht_kiYymW4Kmwqu0I1N9_S?usp=drive_link

3. Report

You will write a report for this project. There is no strict format of this report but below are the essential parts:

1.   How you design the data preprocessing methods, inputs, activation functions, and output for ML and ANN based methods.

2.   A screen shot of the results in a successful run of your programs.

3.   Try different configurations of ML and ANNs and report the results.

4.   Evaluate the efficiency of your algorithms (e.g. the time used for training, the size of the model). 5.   Perform a comparative study of ML and ANN methods.

There are other aspects that you can explore and report, which are up to you. More comprehensive study will receive more credits.

* Do not paste a lot of codes in the report. Focus on the design and analysis. Your codes will be submitted separately.

4. Submitting your project

You should submit your project in Blackboard. The submission includes: a project report and your source codes. Please put everything in a zip/rar file, name it “FirstName_LastName_Project1.zip” .

5. Advice

Start early! Do not wait until the last few days.