Validation of uncertainty model for reverberation chamber measurements

The reverberation chamber (RC) is basically a metal cavity, which is mechanically stirred to emulate a Rayleigh fading environment. It has been traditionally used for electromagnetic compatibility (EMC) [1]. For the past decade, it has found new applications for over-the-air (OTA) testing because of its capability to emulate Rich Isotropic Multipath (RIMP) [2]. Due to the complicated and time-varying boundary conditions, RC measurements are studied from a statistical point of view for both EMC and OTA applications. A statistical RC measurement is incomplete without the analysis of measurement uncertainty. As a result, the RC measurement uncertainty has been a popular research topic. Different approaches have been adopted in tackling the problem. Most of the studies (that are based on independent sample number e.g., [3]) offer limited insight into how to improve the measurement uncertainty. The authors have proposed an uncertainty model (referred to as Chalmers model hereafter) [4] that, for the first time, separates the stirred component and the unstirred one (i.e., the K-factor [5]), with the later being the residual error, that is, a small measurement uncertainty requires not only a large number of independent samples but also a small K-factor [4]. Based on this insight the measurement accuracy of the Bluetest RCs was significantly improved (see [4]) by introducing a metallic shield to block the line-of-sight (LOS) path. In order to validate the proposed uncertainty model, we proposed a nine-case-measurement uncertainty assessment procedure. That is, we repeat the same measurement sequence nine times, each time with a different height/orientation of the reference antenna on the turntable platform, i.e., the antenna on the platform is placed with three different heights and at each height it is placed with three different orientations. The height separation should be large enough (e.g., half-wavelength at the lowest frequency) to ensure independent measurements. (The half-wavelength is a rule-of-thumb for the correlation length [6], beyond which the correlation becomes negligible, in a well-stirred RC.) In order to validate the uncertainty model, we performed extensive measurement campaigns in four RCs (see Table 1) with different sizes and stirring mechanisms. Fig. 1 shows the estimated standard deviations (STDs) of the average power transfer functions from nine-case-measurement uncertainty assessment procedure together with the modeled STDs for the four RCs, respectively. The solid curves represent measurements and the dotted curves correspond to the model. Note that the measurement campaigns were performed with a 5-year span with the main intention to validate the uncertainty model in different RCs. Hence, different loading conditions, sample numbers, and frequency ranges were used. However, it is safe to conclude that the RTS90 has the best measurement accuracy, even though it is operating in a lower frequency range. Anyway, as can be seen from Fig. 1, the model agrees with the measurement for all cases. This validates the uncertainty model. At the conference we will present the Chalmers uncertainty model and describe the stirring sequences and chamber configurations of different RCs in details.