Efficient Bernoulli Probability Distribution Estimation for Arithmetic Coding

摘要

State-of-the-art arithmetic coding utilizes finite state machines to select symbol probability estimates from the predefined finite set, resulting in inherently limited precision and increased code length. This paper presents novel method for Bernoulli probability distribution estimation, based on low pass filtering with varying dominant pole. This solution uses integer-only arithmetic, without multiplication or division operations, thus providing significant reduction in required resources of IoT devices.