Evaluation of Encoding Schemes on Ubiquitous Sensor Signal for Spiking Neural Network

Sizhen Bian,Elisa Donati,Michele Magno

2024-07-12

Abstract:Spiking neural networks (SNNs), a brain-inspired computing paradigm, are emerging for their inference performance, particularly in terms of energy efficiency and latency attributed to the plasticity in signal processing. To deploy SNNs in ubiquitous computing systems, signal encoding of sensors is crucial for achieving high accuracy and robustness. Using inertial sensor readings for gym activity recognition as a case study, this work comprehensively evaluates four main encoding schemes and deploys the corresponding SNN on the neuromorphic processor Loihi2 for post-deployment encoding assessment. Rate encoding, time-to-first-spike encoding, binary encoding, and delta modulation are evaluated using metrics like average fire rate, signal-to-noise ratio, classification accuracy, robustness, and inference latency and energy. In this case study, the time-to-first-spike encoding required the lowest firing rate (2%) and achieved a comparative accuracy (89%), although it was the least robust scheme against error spikes (over 20% accuracy drop with 0.1 noisy spike rate). Rate encoding with optimal value-to-probability mapping achieved the highest accuracy (91.7%). Binary encoding provided a balance between information reconstruction and noise resistance. Multi-threshold delta modulation showed the best robustness, with only a 0.7% accuracy drop at a 0.1 noisy spike rate. This work serves researchers in selecting the best encoding scheme for SNN-based ubiquitous sensor signal processing, tailored to specific performance requirements.

Signal Processing

What problem does this paper attempt to address?

The paper attempts to address the problem of encoding sensor signals in Spiking Neural Networks (SNN) and evaluating their performance. Specifically:

1. **Evaluation of Encoding Methods**: The paper aims to comprehensively evaluate four main encoding schemes (rate coding, first spike time coding, binary coding, and delta modulation) and explore the performance of these encoding methods under different metrics, including Average Firing Rate (AFR), Signal-to-Noise Ratio (SNR), classification accuracy, robustness, as well as inference latency and energy consumption.

2. **Sensor Data Processing**: The paper uses Inertial Measurement Unit (IMU) data as a case study to explore how to convert sensor data into spike sequences and deploy the corresponding SNN model on the neuromorphic processor Loihi2 to evaluate the post-deployment encoding performance.

3. **Robustness and Information Loss**: The paper focuses on the robustness of encoding schemes under noisy conditions, such as evaluating the accuracy degradation by introducing erroneous spikes, and analyzes the degree of information loss in different encoding schemes.

4. **Practical Application Guidance**: The ultimate goal is to provide researchers with a method to select the best encoding scheme to optimize the design of sensor signal processing systems based on specific application requirements (e.g., energy efficiency, low latency, etc.).