A carregar...
Publicação
EM Simulation-Driven Forward Modeling for RFID Tag Localization
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 9.82 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Tag localization problems are addressed usually either through calculations where
signal features such as received signal strength indicator (RSSI), phase, and time
of flight (ToF) are derived solely from analytical expressions, or through field
experiments using commercial off-the-shelf (COTS) readers. While the first approach
often produces oversimplified and unrealistic results, the latter requires full system
deployment in the target environment, which can be costly and time-consuming.
This work explores a middle ground by leveraging electromagnetic (EM) simulations.
Such simulations make it possible to extract realistic signal features - RSSI, phase,
and ToF - directly from a modeled environment that closely resembles a realworld
indoor RFID deployment scenarios. In the context of this work, a logistics
conveyor scenario was modeled and used to assess the utility of EM simulations
for determining the locations of tagged parcels. Particular emphasis was placed on
analyzing the limitations and requirements of each positioning technique to ensure
accurate operation. In addition to traditional range-based techniques, this work
also studies range-free techniques that employ artificial intelligence for tag position
estimation, evaluating how their performance compares to range-based approaches.
Bilateration algorithm is employed with intensity-, phase-, time-based distance
estimation techniques. Achieving high accuracy, with positioning errors below 15 cm.
Phase- and, time-based techniques provided the most robust and consistent results.
Although intensity-based approach was capable of achieving comparable accuracy
under controlled conditions, it required prior knowledge of tag orientation to maintain
this performance and demonstrated higher sensitivity to antenna radiation variation,
leading to overall less consistent results. In contrast, the fingerprinting approach
was evaluated using two deep neural network (DNN) models, but both exhibited
lower performance, with error of 38 cm and 42 cm, likely due to the limited dataset
with the coarse spatial sampling.
Descrição
Palavras-chave
RFID EM Simulation RSSI Phase Time of flight Angle of Arrival S-parameters Antenna
Contexto Educativo
Citação
Editora
Licença CC
Sem licença CC