This paper deals with an efficient secure communication system based on a speech watermarking approach. The main goal of the system is to efficiently support air traffic control (ATC) communications by improving safety and security of the existing ATC equipment embedding digital side information into speech signals in order to allow an automatic identification of the talker. An implementation based on the use of the well known LPC (linear predictive coding) principle is discussed. The LPC residual is split up into voiced and unvoiced segments; while the voiced segments cross the system unmodified, the watermarking data are embedded into the unvoiced segments, spectrally similar to a white noise. The watermark signal is decoded by a very similar scheme. We provide also a statistical analysis on the unvoiced parts of the ATC communications that demonstrates the system feasibility. In addition to this, an optimization of the entire system has been also performed to increase performance. The obtained results highlight that the watermark data are nearly imperceptible and provide a data channel with rate from 2000 to 5000 kbits/s, with high robustness to channel noise; the system requires only modular, low-cost and easy-to-install encoders and decoders, maintaining compatibility with legacy RF DSB-AM transceivers. Moreover, it is well suited both for the actual 25 kHz RF channels and for the new 8.33 kHz European RF channels.of network security attacks [1]. In particular, after the 11 September terroristic attack, these requirements have become of paramount importance for ATC (air traffic control) communications.Almost all tactical ATC communications are handled using DSB-AM radio systems. DSB-AM voice channels are allocated sector-wise and their SECURE RADIO COMMUNICATIONS BASED ON SPEECH WATERMARKING 307 ‡ Mean opinion score is an evaluation scale commonly adopted for the subjective measurement of voice quality [23,24]; the scale goes from 1 (bad quality) to 5 (best quality), and is evaluated through extensive campaigns involving a sufficiently high number of listeners, each one giving his or her opinion on the audio quality. SECURE RADIO COMMUNICATIONS BASED ON SPEECH WATERMARKING 311
Segmentation ResultsAfter completing the optimal system parameters setting and assumed a Voicing Threshold value of 0.35 we investigated the influence of the ATC channel on the system performance. The segmentation is assumed performed differently at the embedder and the decoder