One of the most widely used framework to implement True Random Number Generators (TRNG) on an electronic ship is based on free running oscillators. In order to make sure that a TRNG has a given entropy rate, it is desirable to embed with it an online test of entropy. According to [1], the only way to do achieve that, is to use a statistical model of the TRNG and to have an online measurement method of the physical parameters of the model. In the context of oscillator based TRNG a method is proposed in [2] to assess the statistical parameters of the model proposed in [3]. In this paper, we explain that this method has some shortcomings, that the choice of its parameters, not clearly explained in [2] can have dramatic effects on its precision and reliability. In this paper, we propose a recipe for choosing good parameters and present some algorithmic tweaks to improve its robustness. We have extensively tested the new method with simulations and implemented it in hardware to verify its efficiency and practicality.