随机计算是一种新型计算范式,它使用概率来编码数值,这种表示使简单的逻辑门能够执行复杂的算术运算。这项工作提出了一种快速一元中值滤波器设计,提出的滤波器基于计数器生成随机数,使用随机相关逻辑组成最基本的排序网络单元,依据输出结果形成反馈回路,不消耗额外硬件面积,可动态实现提前结束运算,减小了巨大的电路延迟。实验结果表明,所提出的中值滤波器设计在实际比特流长度和能量消耗上均优于现有的滤波器设计,所提出的3×3窗口的中值滤波电路可以减少55.58%的能量消耗。利用中值滤波应用对加入椒盐噪声之后的图像进行了进一步验证,结果表明电路具有较好的精度。所提出的设计在16输入的排序网络应用中,在输入范围为[0,0.5]时,电路具有更低延迟,实际比特长度和能量可以减少50%。

Stochastic computing (SC), an unconventional computational paradigm, employs probabilities to represent numerical values. This representation enables complex arithmetic operations to be performed using simple logic gates. This work presents a fast unary median filtering circuit design. The proposed filter utilizes counters to generate stochastic numbers (SNs) and constructs fundamental sorting network units using stochastic correlation logic. A feedback loop, formed based on the output, dynamically terminates computations early without consuming additional hardware area, significantly reducing substantial circuit latency. The experimental results demonstrate that the proposed median filter design outperforms existing implementations in both actual bitstream length and energy consumption. Specifically, the proposed 3×3 window median filter circuit achieves a 55.58% reduction in energy. Further validation using median filtering on images corrupted by salt-and-pepper noise confirms the accuracy of the proposed circuit. For a 16-input sorting network application, the proposed design exhibits lower consumption when inputs range within [0, 0.5], achieving up to a 50% reduction in actual bitstream length and energy consumption.