.png){kind=icon}
.png){kind=icon}
Return
Efficient Sim2Real Deep Learning for Device-Specific OFDM Frequency Offset Calibration
.png){kind=icon}
PRE
AI
DOI：10.1109/TWC.2026.3707555.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Frequency offset estimation in Orthogonal Frequency Division Multiplexing (OFDM) systems suffers from performance degradation across heterogeneous software-defined radio (SDR) devices due to uncalibrated hardware impairments, especially when a low-cost SDR device is used as a receiver. Existing deep neural network (DNN)-based methods lack device-level adaptability, limiting their practical use in dynamic environments. This paper introduces a Sim2Real transfer learning framework for per-device frequency offset calibration, integrating simulation-based pretraining with lightweight receiver-side adaptation. A backbone DNN is first pretrained on synthetic OFDM signals emulating parametric hardware distortions such as phase noise and IQ imbalance, enabling generalized feature extraction without the need for extensive cross-device real data collection. Subsequently, only the regression layers are fine-tuned using as few as <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$1,000$ </tex-math></inline-formula> real OFDM frames per target device, thus preserving hardware-independent knowledge while efficiently adapting to device-specific impairments. Experimental validation across three SDR platforms (USRP B210, USRP N210, HackRF One) demonstrates a 48% reduction in bit error rate compared to conventional cyclic prefix-based methods under indoor multipath propagation. The proposed framework effectively bridges the simulation-to-reality gap for robust frequency estimation, facilitating scalable and cost-effective deployment in heterogeneous communication systems.
Keywords:
Frequency offset
DNN
hardware calibration
OFDM
SDR
Journal
IF:
10.7
Papers:
1.3W
Citations:
5.3W