Versatile approaches of conventional channel estimation (CE) algorithms such as linear minimum mean square error (LMMSE), compressive sensing (CS), and Gaussian mixture model (GMM) have been applied in multi-input multi-output (MIMO) and Orthogonal Frequency-Division Multiplexing (OFDM) wireless communication systems such as the fifth generation new-radio (5G-NR). As such, it remains a mature engineering field while diminishing returns with regards to performance-improvement and complexity-reduction.
However, conventional CE algorithms strongly dependent on assumptions on channel such as sparsity, Gaussian-like distribution, and stationarity. In practice, uncertainties remain due to the nature of wireless channel and the complex and changeable electromagnetic propagation environments. Conventional CE approaches still have some limits, and machine learning (ML) and artificial intelligence (AI) based approaches may improve CE by exploiting potential hidden factors in the propagation environments.
Currently, researchers are exploiting CE from the perspective of model and data driven AI, but many questions remain open and yet to be answered.
In this master thesis project, entitled as “AI Boosted Channel Estimation for 5G and Beyond”, our targets are to investigate state of the art of model-driven and data-driven deep learning (DL) based channel estimation, and compare performance of different DL architectures with conventional methodologies. The student(s) will conduct researches under supervisions including the following tasks:
- Build up an OFDM system model according to 3GPP standard for channel estimation, and use 3GPP and Quadriga channel models.
- Explore and simulate CE performance of conventional methods (LMMSE, GMM) and DL-based benchmarks such as DNN, CNN, RNN and their variants.
- Analyze and understand the underlying impacts of different AI structures to the performance, find metrics to measure the accuracy of learning, measure the sensitivity of model’s mismatch to performance behavior.
- Draw observations and conclusions about how and when AI based CE can improve conventional CE.
Qualifications & Experience
- Master students in their second study-year, i.e., who are expected to graduate before the summer of 2023.
- Good knowledges of communication theory, signal processing, OFDM system, channel modeling and estimation, etc.
- Good scores of studied courses.
- Experienced in writing codes with Matlab and python. Hands on experiences with TensorFlow is a plus.
- Be able to work on site in our Lund office.
Preferred starting date
As soon as possible.
Sha Hu, 0700962540