Project 1: Vehicle Positioning
• Radio frequency based positioning
– Use wireless signals to calculate the distance
between the transmitter and the receiver
distance 𝑑
Transmitter Receiver
1
Project 1: Vehicle Positioning
• Radio frequency based positioning
– Use wireless signals to calculate the distance
between the transmitter and the receiver
• The transmitter transmits a signal at time 𝑡!
• The receiver receives the signal at time 𝑡"
• Signal transmission speed is the light speed 𝑐
distance 𝑑
!# "!$
=
#
Transmitter Receiver 2
Project 1: Vehicle Positioning
• Radio frequency based positioning
– Use distances between transmitters and
receivers to localize a device (transmitter or
receiver)
3
Project 1: Vehicle Positioning
• Radio frequency based positioning
– Applications: GPS
4
Project 1: Vehicle Positioning
• Radio frequency based positioning
– Not all of transmission links are line-of-sight
5
Project 1: Vehicle Positioning
• Radio frequency based positioning
– Not all of transmission links are line-of-sight
– Cannot calculate the distance between the
transmitter and the receiver correctly
6
Scenario Layout
7
Dataset
• Original Dataset: 𝒟 = 𝑯! , 𝒑! , 𝑐! , … , 𝑯" , 𝒑" , 𝑐"
– Input Data: 𝑯# ∈ ℂ$!×$" , channel state information
(CSI) matrix (received signals)
– 𝑁& : Number of antennas (𝑁& = 4)
– 𝑁' : Number of sub-carriers (𝑁' = 1632)
8
Dataset
• Original Dataset: 𝒟 = 𝑯! , 𝒑! , 𝑐! , … , 𝑯" , 𝒑" , 𝑐"
• Position Coordinate: 𝒑! ∈ ℝ"×$ , the rectangular
coordinate of a position
$
𝒑# = 𝑥# , 𝑦# , 𝑧#
• LoS/NLoS Indicator: 𝑐$ ∈ 0,1
– 𝑐# = 0 à an LoS position
– 𝑐# = 1 à an NLoS position
9
Data Preprocessing
• Step I: dataset split
– There are 20000 samples in the dataset
– 15000 samples are randomly selected for model
training
– The rest of 5000 data samples are used for test
15000 5000
Training Test 10
Dataset Dataset
Data Preprocessing
• Step II: Convert the complex-valued CSI matrices
to real-valued matrices
– 𝑯! is a complex-valued matrix -> cannot be directly
used as the input of a general neural network.
– One method
𝑁! 𝑁!
Take Modulus
Value
𝑁% 𝑁%
𝑯" ∈ ℂ#!×#" ' " ∈ ℝ#!×#"
𝑯
11
Data Preprocessing
• Step III: Normalize the values of CSI matrices to
[0,1]
– Normalization
– Standardization
– Others
12
Task 1
• Task 1: LoS/NLoS position classification
– Dataset: 𝑯 2 $ , 𝑐$ , … , 𝑯
2 ! , 𝑐! , … , 𝑯
2 % , 𝑐%
– Input CSI format design
• Take Modulus of complex CSI matrices
• Use FFT to change frequency CSI to time domain CSI
• Calculate the total Power of the CSI signal
– Neural networks, i.e., FNN, CNN
13
Task 1
• The simplest method
– Determine LoS/NLoS positions using the received
signal power
– Calculate power of 𝑯2!
– Plot cumulative distribution function (CDF) of LoS
positions and complementary cumulative distribution
function (CCDF) of NLoS positions
14
Task 1
• The simplest method (baseline)
15
Task 2 (optinal)
• Task 2: Vehicle position estimation
– Dataset: 2 $ , 𝒑$ , … , 𝑯
𝑯 2 ! , 𝒑! , … , 𝑯
2 % , 𝒑%
– Neural networks, i.e., FNN, CNN
16
Evaluations
• How you design the data preprocessing methods, inputs,
activation functions, and output for ML and ANN based
methods.
• A screen shot of the result
– use the power of CSI as input for LoS/NLoS classifications.
• A screen shot of the result
– use the time domain CSI as input for LoS/NLoS classifications.
• A screen shot of the result
– use the frequency domain CSI as input for LoS/NLoS classifications.
• Evaluate the efficiency of your algorithms when you use
different CSI inputs.
• Report writing
– Do not copy code
– Maximum page number: 5
17