Signal Design and Processing for Grant-Free Access
Abstract
Abstract
Grant-free access is an attractive approach to enable spectrum-efficient low-latency access
for systems with massive number of users. In this dissertation, we investigate the pilot
design and beam alignment techniques for grant-free access systems. Non-orthogonal pilot
designs play a crucial role for grant-free access as it needs to provide a large number of access
codes with fast collision detection capability and good channel estimation performance. We
propose novel non-orthogonal pilot designs for non-sparse, sparse and block-sparse channels
in grant-free access systems.
Recently, a nice solution based on an on-off type non-orthogonal pilot design with collision
detection capability has been proposed. It can serve more users than the orthogonal pilot
design but at the cost of degraded channel estimation performance compared to the orthogonal optimum pilot design. First, we propose a new non-orthogonal pilot design with collision
detection capability and improved channel estimation performance for non-sparse channels.
Next, we investigate non-orthogonal pilot designs for fractional bandwidth allocation. We
propose two new non-orthogonal pilot designs for physical resource block (PRB) based resource allocation. Both of the new designs support fast collision detection at the receiver.
Performance evaluation results show that the proposed schemes provide equivalent or better
channel estimation performances and support much more users than the orthogonal pilots
defined in the current 4G standards.
Next, we explore sparse channel estimation with compressed sensing technique. We prove
several propositions regarding compressed sensing (CS) based estimation performance. Using these propositions two novel orthogonal pilot designs are developed for sparse channel
estimation with optimized performances . Utilizing these orthogonal pilot sets we propose a
non-orthogonal pilot design with collision detection capability for sparse channel estimation.
We also explore the non-orthogonal pilot design for highly sparse and block-sparse channels.
We present several propositions related to the performances of the CS based sparse and blocksparse channel estimation. Utilizing these propositions, we develop a novel non-orthogonal
pilot design with fast collision detection capability for grant-free access in block-sparse channels. We also propose two methods to optimize the Peak-to-Average Power Ratio (PAPR)
of the time domain signals of the proposed non-orthogonal pilot codes. The simulation results illustrate that the proposed design provides similar or better channel estimation and
collision detection performances with much better pilot resource efficiency when compared
to the existing designs.
Finally, we investigate a millimeter wave multi-user uplink system with a large antenna array
at the base station. In such a system, it is important to know the angle of arrival (AoA) of the
strongest path of each user to achieve high beamforming gain. We propose a low complexity
deep neural network based beam alignment method (LCNNBA) to estimate the AoAs of
all the users simultaneously. We also propose three novel analog combiner configurations.
Using these combiners, the LCNNBA achieves large performance improvements with lower
complexity and memory requirements compared to a recently developed machine learning
based method.