Optimizing the Energy Efficiency of Unreliable Memories for Quantized Kalman Filtering

Area of Science:

• Digital Signal Processing
• Embedded Systems Engineering
• Computer Architecture

Background:

• Quantized Kalman filters (QKFs) are crucial for state estimation in resource-constrained embedded systems.
• Unreliable memories introduce computational errors, impacting QKF performance and reliability.
• Memory energy consumption is a significant factor in embedded system design.

Purpose of the Study:

• To develop an error propagation model for QKFs implemented with unreliable memories.
• To establish a relationship between QKF performance, energy consumption, and memory noise levels.
• To introduce optimization techniques for minimizing memory energy consumption in QKFs.

Main Methods:

• Formulated updated Kalman filter equations incorporating quantization and memory errors.
• Developed an error propagation model to predict estimation error covariance.
• Introduced two optimization methods for energy allocation in memory banks.
• Validated the model and methods through simulations comparing theoretical analysis and experimental results.

Main Results:

• The proposed error model accurately predicts estimation error covariance.
• A clear relationship was established between filter performance, energy consumption, and memory noise.
• Optimization methods achieved significant reductions in memory energy consumption, exceeding 50%.
• Simulations demonstrated close agreement between theoretical predictions and experimental outcomes.

Conclusions:

• The developed error model effectively characterizes errors in quantized Kalman filters using unreliable memories.
• The proposed optimization strategies offer substantial energy savings for embedded systems.
• This work provides a framework for designing energy-efficient and reliable QKFs.