Deep Learning for Synchronization and Channel Estimation in NB-IoT Random Access Channel
Author
Jespersen, Mads Helge
Term
4. term
Education
Publication year
2019
Submitted on
2019-06-28
Pages
51
Abstract
Effektiv dekodning af trådløse signaler kræver, at modtageren kan: opdage hvilke brugere der er aktive, synkronisere i tid og frekvens, estimere radiokanalen og udligne kanalens påvirkning. I traditionelle systemer estimeres disse ukendte kanalparametre hver for sig. Denne afhandling foreslår en ny, fælles estimeringsmetode baseret på dyb læring, som udleder flere parametre direkte fra den modtagne blanding af transmissioner. Metoden er rettet mod basestationen i scenarier med massiv konnektivitet, hvor mange lavkompleksitetsenheder sender samtidig. Den kan afgøre antallet af kollisioner, håndtere synkronisering og kanalestimering ved overlappende, ikke-ortogonale transmissioner, og den kan finde bedre præambelsekvenser via en auto-encoder. Vi demonstrerer fremragende ydeevne på NB‑IoT-standardens random-access-præambelstruktur og estimerer Time-of-Arrival (ToA, ankomsttid), Carrier-Frequency Offset (CFO, bærefrekvensafvigelse), kanalforstærkning og kollisionsmultiplicitet. Den foreslåede CNN-baserede estimator opnår en ToA-RMSE på 2,88 µs og en CFO-RMSE på 3,44 Hz ved 10 dB SNR, mens en konventionel totrins-estimator har RMSE-værdier på henholdsvis 16,20 µs og 7,98 Hz.
Decoding wireless signals effectively requires several steps: detecting which users are active, synchronizing timing and frequency, estimating the radio channel, and correcting its effects (equalization). Traditional receivers estimate these unknown channel parameters separately. This thesis proposes a deep learning–based joint estimation method that infers multiple parameters directly from the received mixture of transmissions. It targets base-station decoding in massive connectivity settings with many low-complexity devices operating at once. The method can determine the number of colliding signals (collision multiplicity), handle synchronization and channel estimation under overlapping, non-orthogonal transmissions, and discover improved preamble sequences using an auto-encoder. We demonstrate strong performance on the NB-IoT random access preamble structure, estimating Time-of-Arrival (ToA; signal arrival time), Carrier-Frequency Offset (CFO; difference between transmitter and receiver carrier), channel gain, and collision multiplicity. The CNN-based estimator achieves a ToA RMSE of 2.88 µs and a CFO RMSE of 3.44 Hz at 10 dB SNR, whereas a conventional two-stage estimator yields 16.20 µs and 7.98 Hz RMSE, respectively.
[This abstract was generated with the help of AI]
Documents