(A) self-attention based I/Q imbalance estimation and compensation for sub-terahertz communications

Abstract

5th generation mobile communications (5G) have been commercialized, and new research on 6th generation mobile communications (6G) is in progress by various research groups. The standard for 6G mobile communications requires a higher data rate, better stability, and shorter delay time. Terahertz (THz) spectrum such as 0.1 - 10 THz frequency is considered for 6G. THz spectrum is higher than the traditional electronic-based wireless communication system and lower than the traditional photonics-based communication system. Therefore, hardware devices for the wireless communication front-end are not sufficiently developed for THz frequencies. There are various hardware impairments for radio-frequency front-end and in-phase and quadrature-phase (I/Q) imbalance is one of the most critical impairments. Since I/Q imbalance causes severe degradation in the communication quality, estimation and compensation of I/Q imbalance are necessary to satisfy the high data rate and stability of the 6G mobile communications. The I/Q imbalance value varies with the ambient temperature of the communication hardware, which means that operation heat causes I/Q imbalance to change with time. To satisfy high data rate and short delay time, real-time estimation and compensation of time-variant I/Q imbalance are mandatory. Various studies on estimating and compensating the I/Q imbalance have been conducted. Using a compensation circuit is one of the conventional methods for I/Q imbalance compensation, but this method increases the system implementation cost. Since complementary circuits may leave some amount of I/Q imbalance not compensated perfectly, digital I/Q imbalance estimation and compensation schemes at baseband also have been studied. The data-aided method and blind method are the two main types of baseband digital I/Q imbalance estimation and compensation. Data-aided methods use pilot train signal or preamble signal transmitted from the transmitter, and the receiver estimates the I/Q imbalance based on the received pilot signal. This scheme estimates I/Q imbalance with a short delay time but reduces the spectral efficiency since data cannot be transmitted during train signal. On the other hand, blind estimation methods only use statistics of the received data signal. The blind estimation method does not need known train signals from the transmitter and does not affect the spectral efficiency. However, the blind method requires a relatively long delay time so that a sufficient amount of received data signal is observed for accurate statistics. In this thesis, a machine learning-based blind I/Q imbalance estimation scheme at the receiver baseband is proposed. Machine learning-based schemes can estimate the I/Q imbalance with a small number of received data signals and decrease the delay time of I/Q imbalance estimation. The proposed estimation method uses the attention mechanism to reflect the relation between the data symbols on the I/Q imbalance estimation. The attention-based neural network maximizes the estimation performance by weighting the relevance of each data symbol and attending to more important data symbols from the given data. Therefore, the attention-based scheme can be applied to 6G mobile communications since it reduces the estimation delay while maintaining good estimation performance. The proposed attention-based neural network was trained using datasets with various I/Q imbalance levels, signal-to-noise ratio (SNR) levels, and modulation levels. To verify the estimation performance, the mean absolute error between the estimated and true phase values for various SNRs was analyzed. The proposed scheme showed a smaller estimation error compared to the conventional scheme across all SNR ranges when the length of the received data signal is the same. Furthermore, for the high SNR regime, the proposed scheme outperformed the conventional scheme even using the shorter length of the received signal. Therefore, the proposed attention-based I/Q imbalance estimation scheme can improve the estimation performance while reducing the time delay at the receiver. Short wavelength and low diffraction rate of the THz wave causes line-of-sight dominant channel making THz links highly susceptible to the blockage effect. To avoid the blockage effect, 6G mobile communications will use the sub-6GHz spectrum of a legacy network as well as the THz spectrum. Therefore, this thesis proposes an I/Q imbalance estimation method for a multi-standard receiver where THz frequency and sub-6GHz frequency are used together. The multi-head attention mechanism was applied to estimate the I/Q imbalance using the received data signal from multiple frequency bands. The multi-head attention mechanism is an extended version of the self-attention mechanism to observe the various aspects of the data signal. Therefore, the I/Q imbalance of multiple frequency bands can be estimated. The training dataset for various situations is generated to train the multi-head attention-based neural network, and the estimation performance of the proposed neural network was evaluated. A mean absolute error of the estimated I/Q imbalance value of the multi-standard receiver was compared with the original results of the single-standard I/Q imbalance estimator. The proposed multi-head attention-based scheme showed a similar estimation error compared with the original self-attention-based scheme. The proposed self-attention-based and multi-head attention-based I/Q imbalance estimation methods can be applied to the 6G mobile communication system for real-time estimation and compensation of the I/Q imbalance. The proposed schemes show good estimation performance only using a short length of the received data signals. The proposed scheme will improve the communication quality by compensating the hardware impairment of the transceivers for 6G communications.