GIoTS 2022 YuukiTakagi

Accurate indoor positioning
based on
beacon weighting using RSSI
Wednesday 22 Jun 2022
Yuuki Takagi, Takayuki Kushida
Global IoT Summit 2022 Paper Presentation

Agenda
1. Background
2. Introduction and issue
3. Proposed method
4. Experiment
5. Evaluation Method
6. Results
7. Discussions
8. Conclusions
1

3
Ref: Why GPS Signal Lost Is Showing On Your GPS Unit (https://hikinggpszone.com/why-gps-signal-lost-is-showing-on-your-gps-unit/)
The global indoor location market size is expected to grow
from USD 7.0 billion (2021) to USD 19.7 billion (2026).
One of the ways…
BLE(Bluetooth Low Energy)
Advantages
• Low cost
• Low power consumption
• Ease of deployment
Challenges
• Prone to radio interference (2.4[GHz])
→ Large distance error
Background

[1] Ma, Z., Poslad, S., Bigham, J., Zhang, X., Men, L.: A ble rssi ranking based indoor positioning
system for generic smartphones. In: 2017 Wireless Telecommunications Symposium (WTS). (2017)
Introduction and issue
4
§ RSSI(Received Signal Strength indication) is a parameter for location
estimation in BLE.
§ Distance error in BLE is caused by 2.4 [GHz] radio band [1].

This proposal is defined by 2 phases.
1. Learning Phase : Calculation of weights for each beacon from RSSI distribution
2. Calculation Phase : The location is calculated by selecting the beacon based on
the weights defined as ”Confidence Level (CL).”
5
Overview – Proposed Method
Objective:
Reduce distance error from existing 3-point positioning
1. Calculation of weights
2. Beacon Selection
Reduction of errors
RSSI distribution

Proposed method – Learning Phase (step1)
Objective:
Calculation of weights (Confidence Level) by quantifying the
difference in signal output strength of each beacon
6
The Friis transmission formula : Theoretical value
Approximation by collecting RSSI : Realistic value
Steps:
1. Conversion from RSSI to distance
2. Calculation of Typical Value
3. Calculation of Confidence Level

7
RSSI measurement data
（frequency distribution）
Calculation of
Typical Value from
collected RSSI data
2. Calculation of statistically Typical Value
= RSSI values changes when the same location is measured

8
Creation of a
separator called
“split”

9
Consideration of
radio wave blurring
𝑹𝑺𝑺𝑰𝒊 =
Calculation for mean of
“split7” including mode

10
𝐓𝐲𝐩𝐢𝐜𝐚𝐥 𝐕𝐚𝐥𝐮𝐞 = 𝑅𝑂𝑈𝑁𝐷
1
𝐵𝑒𝑎𝑐𝑜𝑛𝑁
;
"#$
%&'()*+
𝑅𝑆𝑆𝐼"
Calculation of Typical Value for all other beacons
within radio range from target beacon
𝑹𝑺𝑺𝑰𝒊 = Calculation for mean of “split” including mode

= Select a beacon with fewer outliers
11
Calculation of R𝑆𝑆𝐼"
and Typical Value
Case of 2 beacons
within radio range of
target beacon

= Select a beacon with fewer outliers
12
Calculation of the
probability of Typical
Value appearing
Smallest probability
of any beacon
within radio range

Proposed method – Calculation Phase (1/3)
13
Objective:
Calculation of coordinates excluding beacons with low
Confidence Level (CL)
1 CL is calculated from the Learning Phase
2 Distance from beacon to IoT device is calculated from RSSI
Steps:
1. Selection of beacons with high CL1 and close distance 2
to IoT device
2. Calculate coordinates of IoT devices

Proposed method – Calculation Phase (step1)
14
Case of 4 devices selected
Order of sorting
①Distance from IoT device to
beacon
(Ascending Order)
②Confidence Level
(Descending Order)

15
Select 4 devices in range to
convert RSSI to distance correctly
Order of sorting
①Distance from IoT device to
beacon
(Ascending Order)
②Confidence Level
(Descending Order)

16
• Calculation of coordinates in
𝑛∁𝑟 different ways
• Calculation for center of gravity
Coordinates of IoT Device =
Calculation for mean of the
coordinates of the green point

Formula for spherical surfaces
𝑆𝑝ℎ𝑒𝑟𝑒 𝐴:
(𝑥 − 𝑎,)-+(𝑦 − 𝑎.)-+(𝑧 − 𝑎/)-= 𝑟0
-
...
17
Preconditions
• Radio waves are emitted in spherical form.
• z-coordinate of IoT device = height of user's chest
Calculation of
𝒙, 𝒚 coordinates
Calculation of the
intersection with 3 circles

§ Advantages of Learning Phase：
Statistical reading of the environmental features of the
installation place.
§ Advantages of Calculation Phase：
Selection of beacons with statistically fewer outliers and
calculation of coordinates.
18
Proposed method – Summary of advantages
This proposal is defined by Learning and Calculation Phases

§ Device used
§ IoT devices :
ESP-WROOM-32
(10 units)
§ Number of data used
§ Learning Phase : 15622 records
§ Calculation Phase : at least 3000 records for each position
Experiment
19
receive

Evaluation Method
§ Error calculation method : L2 norm error
§ The comparison target
§ Tape measure
§ Proposed method
§ Three-point positioning of existing method
§ Follow-up experiments
§ Comparison of the number of selected beacons and distance errors
20

Results - Experiments
21
• The figure shows one of the
experimental results.
• The average of the 16 times
results was calculated.
Mean of distance error
• Existing method :1.337[m]
• Proposed method :0.449[m]
→ Reduction of 66.3%

Results - Follow-up experiments
22
Trend of distance error:
Decrease, Increase, Decrease
Number of the smallest
distance errors:
• 𝑎 , 𝑑 : 5 devices
• 𝑏 , (𝑐) : 4 devices
Results obtained with more than
75% probability．

Discussions
23
Results with 25% probability of many outliers
• Increase in the number of RSSIs collected at the learning phase
• Selection of the correct “split” by search

Example of a easy choice of ”split”
24
Frequency distribution of RSSI collected

Example of a difficult choice of ”split”
25
Frequency distribution of RSSI collected

Conclusions
Strengths
§ Reduces distance error
§ Learning Phase : Exclusion of beacons including many outliers
§ Calculation Phase : Calculation of 3D coordinates using beacon
with fewer outliers
26
Future Plan
§ Proposed method of exclusion of outliers
§ More variations in locations and room sizes measured
§ Measurements with different beacon locations

GIoTS 2022 YuukiTakagi

More Related Content

Similar to GIoTS 2022 YuukiTakagi (20)

Recently uploaded (20)

GIoTS 2022 YuukiTakagi