Date of Completion


Embargo Period


Major Advisor

Lei Wang

Associate Advisor

John Chandy

Associate Advisor

Shalabh Gupta

Field of Study

Electrical Engineering


Doctor of Philosophy

Open Access

Open Access


Conventional sensing techniques often acquire the signals entirely using a lot of resources and then just toss away a large portion of the obtained data during compression. This motivates an emerging research area called as Compressive Sensing (CS) that allows efficient signal acquisition under the sub-Nyquist rate while still able to promise reliable data recovery. Despite the benefits of compressive sensing, one critical issue in the practical applications of compressive sensing is how to reliably recover the original signals from only a few measurements in an efficient way. The Orthogonal Matching Pursuit (OMP) algorithm has shown a good capability for reliable recovery of compressed signals. Due to the simple geometric interpolation and good efficiency, the OMP-based greedy algorithms are often the preferable choice in hardware implementations for real-time signal recovery. However, practical applications of compressive sensing in hardware platforms are limited as signal reconstruction is still challenging due to its high computational complexity.

On the greedy algorithms for compressive signal reconstruction, we will first investigate the computation steps of the Orthogonal Matching Pursuit algorithm. In the iterative computations, intermediate signal estimates and matrix inversions can be decoupled, thereby enabling parallel processing of these two time-consuming operations in the Orthogonal Matching Pursuit algorithm. Based on the observation, the implementation technique of an algorithmic transformation technique referred to as Matrix Inversion Bypass (MIB) is proposed to improve the signal recovery efficiency of the Orthogonal Matching Pursuit based CS reconstruction. The proposed MIB naturally leads to a parallel architecture for high-speed dedicated hardware implementations and the hardware implementations will be studied for verification.

For the OMP-based signal recovery, we find out that more significant elements of the signal are likely to be recovered first in the iterative OMP algorithm. On the other hand, as iteration order goes up, the OMP algorithm still suffers from significantly increasing computational complexity despite relatively low complexity of hardware implementation. Based on this, a Soft-thresholding Orthogonal Matching Pursuit (ST-OMP) technique is proposed for efficient signal reconstruction in compressive sensing applications. The proposed ST-OMP recovers less significant signal elements using a low-complexity procedure without much degradation in reconstruction quality. The proposed ST-OMP is applied in systems powered by non-deterministic renewable energy sources. The threshold of employing the efficient reconstruction is made dynamically adjustable according to the performance requirements and energy levels. Simulation results demonstrate that the ST-OMP can achieve good recovery performance while significantly reducing the energy consumption as compared to the original OMP implementation.