Low-complexity hardware-aware high dimensional channel estimator

Date

2024-3-27

Author

Aydoğdu, Ahmet Çağrı

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

15
views

0
downloads

Cite This

Deep generative models have recently garnered much interest beyond natural language processing (NLP) and computer vision (CV). This is mainly because generative models could be utilized in any domain where synthetic data generation is applicable. Motivated by this fact, this thesis aims to exploit deep generative models for channel estimation by utilizing the spatial and frequency correlations of MIMO-OFDM systems. Specifically, it is first shown that a properly trained generative adversarial network (GAN) according to channel data -- which can be termed as channel GAN -- can provide realistic channel samples for both narrowband MIMO and MIMO-OFDM channels compared to the ones created by a simulation tool, e.g., MATLAB. Then, the information within the generative model that outputs realistic channel samples is utilized for channel estimation. The high computation complexity of employing deep generative models at low complexity devices tasked with channel estimation is tackled by developing novel algorithms that leverage channel sparsity and some linearization. The results illustrate that it is possible to considerably decrease the computational complexity and speed up inference, e.g., 6 times without sacrificing the estimation performance in terms of normalized mean square error (NMSE) over a wide range of signal-to-noise ratios (SNRs). The proposed methods are compared with state-of-the-art compressed sensing algorithms and naively shrinking the model size before training, and their superiority is explicitly presented in the lack of pilot symbols. The efficiency of the generative channel estimator is studied under non-linear hardware impairments as well. More precisely, the generative estimator is subject to an aggressive clipping policy that significantly distorts the received signal. Such a clipping can facilitate employing low-resolution quantization and enables the use of a low-complexity analog-to-digital converter (ADC) at a receiver. Despite this non-linear distortion, the performance of the channel estimator is heavily preserved.

Subject Keywords

Channel estimation, Generative adversarial networks, MIMO OFDM

URI

https://hdl.handle.net/11511/109216

Collections

Graduate School of Natural and Applied Sciences, Thesis