ROCCO has been implemented together with CRTP in a testbed environment for real-time IP traffic over wireless channels. Focus has so far been on voice traffic because for voice there is a challenge to meet the spectrum efficiency and speech quality of current circuit switched solutions. ROCCO is initially also designed with profiles optimized for voice traffic. However, also new profiles optimized for conversational video and general compression profiles will be implemented and evaluated in the testbed. The picture below shows a block diagram with the testbed components.
Randomly taken audio snapshots from the testbed can be found below with four different speakers. They show examples of performance differences between CRTP and ROCCO for equal conditions, and are provided in WAV and AU format. Listen to one speaker at a time for the CRTP and ROCCO cases and focus on quality differences due to different packet loss characteristics introduced by header compression. CRTP, which is completely intolerant to packet losses, produces some glitches in the audio while ROCCO, which has a high degree of robustness against packet losses, degrades the speech quality less.
| Speaker A | BER = 1e-3 | CRTP | CRTP_e3A.WAV | CRTP_e3A.AU |
| ROCCO | ROCCO_e3A.WAV | ROCCO_e3A.AU | ||
| BER = 1e-2 | CRTP | CRTP_e2A.WAV | CRTP_e2A.AU | |
| ROCCO | ROCCO_e2A.WAV | ROCCO_e2A.AU | ||
| Speaker B | BER = 1e-3 | CRTP | CRTP_e3B.WAV | CRTP_e3B.AU |
| ROCCO | ROCCO_e3B.WAV | ROCCO_e3B.AU | ||
| BER = 1e-2 | CRTP | CRTP_e2B.WAV | CRTP_e2B.AU | |
| ROCCO | ROCCO_e2B.WAV | ROCCO_e2B.AU | ||
| Speaker C | BER = 1e-3 | CRTP | CRTP_e3C.WAV | CRTP_e3C.AU |
| ROCCO | ROCCO_e3C.WAV | ROCCO_e3C.AU | ||
| BER = 1e-2 | CRTP | CRTP_e2C.WAV | CRTP_e2C.AU | |
| ROCCO | ROCCO_e2C.WAV | ROCCO_e2C.AU | ||
| Speaker D | BER = 1e-3 | CRTP | CRTP_e3D.WAV | CRTP_e3D.AU |
| ROCCO | ROCCO_e3D.WAV | ROCCO_e3D.AU | ||
| BER = 1e-2 | CRTP | CRTP_e2D.WAV | CRTP_e2D.AU | |
| ROCCO | ROCCO_e2D.WAV | ROCCO_e2D.AU |
Note: These audio files are only produced to show the difference in audio quality depending on used header compression scheme. Thus, they should NOT be used for something else. Any other usage is prohibited. Any commercial usage or distribution of the audio files above is also prohibited.
The cellular link used is a WCDMA channel simulated with the fading model in [WCDMA]. No link layer ARQ (Automatic Re-transmission Request) is used (e.g. no low-level fast re-transmission is used over the radio link) in order to minimize delay for the real-time voice service. Instead the bit error rate is reduced by channel encoding. The reported bit error rates, BER, are the BERs seen by the link layer and is thus the BER after channel coding. The reverse channel used for context request messages is an error-free channel. The RTT is set to 100 ms and the BER's of the cellular link are 1e-3 and 1e-2.
The VoIP applications uses a GSM Enhanced Full Rate (GSM-EFR) codec, which has a bit rate of 12.2 kb/s and a packet interval of 20 millisecond.