Physical layer security mechanisms have drawn increasing research interest recently along with the development of software defined radio (SDR) techniques. This paper proposes a physical layer watermarking technique named Watermarked Direct Sequence Spread Spectrum (DSSS) or WDSSS technique, which embeds authentication information into pseudonoise (PN) sequences of a DSSS system. The design and implementation of the WDSSS prototype system on the GNU Radio/USRP SDR platform are discussed, as well as two embedding methods, the maximized minimum distance method and the sub-sequence method. Theoretical analysis and experimental results on the WDSSS prototype system are presented to evaluate the performances of both the content signal and the watermark signal. Results show that, for the 11-chip PN sequence, the impact of artificial chip alteration to the content signal is quantitatively predictable, with 2 dB extra signal-to-noise ratio (SNR) required to maintain an acceptable packet error rate for one additional flipped chip. The properties of embedding methods are also analyzed and compared.