Abdulaziz, A. (ECE) – Learning-Based Channel Estimation for Next-Generation Wireless Communications

Coherent wireless reception depends on accurate channel state information
(CSI), yet acquiring CSI through pilot transmission consumes
time–frequency resources that could otherwise carry data. This tradeoff
becomes more pronounced in time-varying and frequency-selective channels
and in short-packet communications, where accurate channel estimation is
challenging and pilot overhead can be significant. This dissertation
investigates how CSI can be estimated, exploited, and refreshed more
efficiently under limited pilot resources.

First, an adaptive deep neural network (ADNN) is developed for
non-stationary channels. When a received packet is accepted by the cyclic
redundancy check (CRC), its detected data symbols are reused to obtain an
improved channel estimate for online retraining. This allows the
estimator to respond to changes in channel statistics without requiring
externally supplied CSI labels or additional pilot transmissions.

Second, an LSTM-based OFDM estimator is developed to jointly exploit
temporal and frequency-domain correlation. Pilot-based LS estimates and
received samples at non-pilot subcarriers are processed over a temporal
window using an encoder–decoder architecture. The estimator achieves
lower MSE than conventional LS interpolation and frequency-domain LMMSE
across the evaluated pilot configurations, and generalizes without
retraining to an unseen channel profile.

Third, the LSTM model is extended to short-packet systems where the time
between channel observations varies. The model accounts for the age of
previously acquired CSI, allowing it to distinguish recent channel
information from stale information as the channel evolves. The resulting
age-aware model improves channel-estimation accuracy with only a small
increase in model size and inference cost.

Finally, the dissertation studies how pilot transmissions should be
scheduled to minimize the long-term average age of information (AoI).
The problem is formulated as a constrained Markov decision process that
jointly accounts for AoI and the age of CSI (AoCSI). The resulting policy
determines when to idle, send a data-only update, or transmit a
pilot-aided update under long-term pilot and transmission constraints.

Event Host: Alatawi Abdulaziz, Ph.D. Candidate, Electrical and Computer Engineering 

Advisor: Hamid Sadjadpour

Co-Advisor : Zouheir Rezki

Zoom: https://ucsc.zoom.us/j/98815064356?pwd=MfibjluajcETfqcpKYm9bp7ZIsJb8N.1

Passcode: 477141