Switched Radio Architecture With Non-Linear Frequency-Domain Estimation for In-Band Full-Duplex Communication

Enabled by Self-interference (SI) cancellation, In-Band Full-duplex (IBFD) communication is promising for wireless systems as it can double spectral efficiency. At high power levels, SI cancellation drops abruptly due to transceivers' non-linearity, as estimation/cancellation of SI signal with...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on wireless communications 2024-02, Vol.23 (2), p.1651-1668
Main Authors: Ayar, Hayrettin, Gurbuz, Ozgur
Format: Article
Language:English
Subjects:
Algorithms
Cancellation
Channel estimation
digital self-interference cancellation
Estimation
Frequency domain analysis
Hardware
In-band full-duplex
neural networks
Noise levels
nonlinear self-interference cancellation
Nonlinearity
OFDM
self-interference
Software radio
Time-domain analysis
Training
Wireless communication
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Enabled by Self-interference (SI) cancellation, In-Band Full-duplex (IBFD) communication is promising for wireless systems as it can double spectral efficiency. At high power levels, SI cancellation drops abruptly due to transceivers' non-linearity, as estimation/cancellation of SI signal with non-linearity is particularly challenging in the presence of the linear SI channel. To resolve this problem, we propose switched IBFD radio architecture along with a frequency-domain non-linear estimation algorithm. In this architecture, during the training phase, proposed non-linear estimation is performed on the SI signal, isolated from linear SI channel. In the cancellation phase, the estimated non-linear signal is provided as a reference to linear SI estimation, followed by linear cancellation. Experiments on a software defined IBFD radio demonstrate that the proposed solution improves linear cancellation by up to 16 dB, existing non-linear cancellation algorithms by up to 13 dB, and SI is reduced to the noise level for almost all power levels. Since the non-linear model needs to be trained once, at power-up, our IBFD solution is not affected from changes in multi-path, while existing schemes require (re)optimization and (re)training of the model. With the same reasoning, proposed non-linear solution has no estimation overhead and complexity is only for non-linear reconstruction.
ISSN:1536-1276
1558-2248
DOI:10.1109/TWC.2023.3293740