Correlation-controlled stochastic computing for low-power FIR and IIR filters in edge DSP

Conventional digital filters in edge devices rely on fixed-point multipliers and adders, which dominate power consumption and hardware cost, limiting their deployment in low-power biomedical and IoT applications. This work proposes a correlation-controlled stochastic computing (SC) framework for finite impulse response (FIR) and infinite impulse response (IIR) filters, in which arithmetic operations are performed on stochastic bitstreams rather than on fixed-point units. Correlation-controlled bitstream generation and optimized encoder-decoder architectures are introduced to suppress correlation-induced bias and reduce hardware overhead. All results reported in this paper are obtained via software simulation on ECG-like, speech-like audio, photoplethysmography (PPG), and industrial IoT vibration signals representative of edge DSP workloads. The proposed design achieves a mean squared error (MSE) below 1% and output signal-to-noise ratios of 23 dB (ECG), 45 dB (audio), 49 dB (PPG), and 47 dB (IoT vibration) at a bitstream length of [Formula: see text], while reducing switching activity by up to 10 times compared with a 16-bit fixed-point DSP baseline. Correlation control alone reduces bias from 3.8% to below 0.5%, as confirmed by ablation experiments. These results demonstrate that correlation-controlled SC provides a favorable trade-off between computational efficiency and signal fidelity, establishing a practical route toward low-power stochastic DSP for resource-constrained biomedical and IoT sensor nodes.