Silent speech interfaces (SSIs) are subject of growing interest, as they can enable speech communication even in the absence of the acoustic signal. Among sensing techniques used in SSIs, radar sensing has many desirable characteristics, such as non-invasiveness and comfort. Although promising results have been achieved with radar-based SSIs, some of its crucial parameters are yet to be investigated, e.g., the optimal type and position of the antennas. To fill this gap, this study investigated the performance of a radar-based SSI with 3 antenna types attached to 3 positions on the speaker's cheek (9 setups). A corpus of 25 phonemes uttered under co-articulation effects was recorded with the 9 setups by 2 native German speakers and then classified with respect to the phonemes. A linear mixed-effect model was fitted to the resulting recognition rates and likelihood ratio tests showed significance for the effects of antenna type and position. The two monopole-type antennas performed better than the Vivaldi-type antenna (2.7% ± 2.8% and 6.2% ± 3.0% improvement), and the two positions closer to the speaker's lips performed better than the most distant position (decrease of 2.8% ± 0.9%). This provides more solid foundation for the development of this type of SSI.