IIR Filter Dithering to Prevent Fixed-Point Self-Oscillation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Infinite impulse response (IIR) filters are prone to inaccuracy and divergence, especially when implemented with fixed-point processors, due to their precision sensitivity, which can lead to self-oscillation and instability, even with mute inputs.
Innovation Solution
The implementation of dithering by applying a random noise value to the output of higher order IIR filters before quantization, using a single random number generator circuit to minimize computational cost and enhance accuracy and stability, by reducing correlation between signals and quantization errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If IIR filters are implemented with fixed-point processors, then computational cost is reduced, but precision sensitivity increases leading to inaccuracy and divergence
Solution Approach 1:
A dither signal is introduced as an intermediary element between the filter computation and quantization. This random noise signal mediates the interaction between the fixed-point arithmetic and the filter output, preventing the direct correlation that causes self-oscillation and instability while maintaining the computational efficiency of fixed-point implementation
Solution Approach 2:
The system changes the parameter of the signal being processed by adding a dither component. By modifying the signal parameters through dithering before quantization, the system transforms the deterministic quantization error into a stochastic process, thereby eliminating self-oscillation and improving stability without increasing computational complexity
2Reliability
If dithering is applied to reduce self-oscillation, then stability improves, but computational load increases
Solution Approach 1:
The patent applies partial dithering by using a simple random noise generator that adds minimal computational overhead. Rather than implementing complex dithering schemes, the system uses a straightforward approach of adding random values at critical points in the filter chain, achieving sufficient stability improvement with minimal additional computational load
Solution Approach 2:
The dither signal is generated using simple random number generation that can be implemented with minimal hardware resources. The system essentially serves itself by using readily available random noise sources (such as thermal noise or simple LFSR generators) that require negligible computational resources compared to the main filter operations
Data Source
AI summary
A method of filtering includes generating a random value by a random number generator circuit, filtering a first signal by a first filter to form a filtered first signal, dithering the filtered first signal using the random value to form a dithered first signal, filtering a second signal by a second filter to form a filtered second signal, and dithering the filtered second signal using the random value to form a dithered second signal.