"What's in a name? That which we call a RoSE by any other name would smell as sweet..."

RoSE Project

Test Results-2

Transmission with Network Impairments

In this part, we provide the experimental results under emulated network impairments on RoSE testbed. In the first scenario, a minimal distortion on video stream is realized using a single RTP packet drop. The Remux output is encoded at 1536 kbps and streamed to Postmux with a modified network API call implementing packet loss (packet-hijacking). The 73th packet during the transmission is dropped and the results are examined. All the streamed frames could be reconstructed at the Postmux, i.e. number of reconstructed frames is 300. The PSNR performance is given in Figure 1. In summary, the effect of impairment can be distinguished in the graph and also in the overall average PSNRs. There is a quality loss of 0.43 dB in terms of average PSNRs. For error-free case, PSNR1 and PSNR2 are 36.1 and 36.51 dB, respectively. For the packet-loss case, these values fall down to 35.67 and 36.08 dB.

Fig. 1.   PSNR performance for the entire sequence with encoding bitrate 1536 kbps. The raw video is Foremansequence with 300 frames and CIF size. A single packet is dropped.

In the upper part, the effect of minor channel degradation was studied as the first scenario. As the second scenario, we provide the experimental results obtained using RoSE testbed under packet erasure channel. The linux bridge in the system is configured to generate 10-1, 10-2, and 10-3 packet loss rates over the streaming traffic randomly.

The rate-distortion curves are shown in Figure 4. As expected, in general increasing packet loss incurs a larger PSNR penalty. The average PSNRs are larger for higher bitrates since streamed data are more interleaved and thus recovery is better. However, since the reconstructed frames are fewer than original, there is a coherency problem during evaluation, i.e. the corresponding frames are not compared all the time. For instance, there are 151 PSNRs for 8192k in Figure 2. This may be the reason why the system shows erratic behavior for varying bitrates.

For some bitrates, all the frames could be constructed. This is especially valid for PER = 10-3. The number of reconstructed frames are plotted in Figure 3. However, there is strange behavior where recovery at 10-1 is better than 10-2. This pattern needs more investigation.

All this information is summarized in Table I. We also employ offline post-processing by FFMPEG to replace missing frames with the adjacent available frames. This is a simple typical error concealment technique. More advanced reconstruction techniques such as averaging or macro-block level decomposition may also be used.

Fig. 2.   PSNR performance for the entire sequence with encoding bitrates 96 kbps (red), 768 kbps (blue), and 8192 kbps (green). Solid lines are for PSNR1, whereas dashed and dotted lines represent PSNR2 and PSNR3, respectively.The raw video is Foreman sequence with 300 frames and CIF size. PER = 10-3

Fig. 3.   Number of reconstructed frames at the Postmux for varying PER.

Average PSNR Values for Foreman sequence on the testbed.

Packet loss PSNR type 48k 96k 192k 384k 768k 1536k 2048k 4096k 8192k

1 28.77 28.96 31.10 33.46 35.87 38.63 39.90 42.30 42.30

2 27.34 39.82 28.93 31.13 33.42 12.11 36.87 40.13 12.59

3 28.99 28.96 30.35 32.24 34.19 12.16 37.16 40.00 12.64

number of reconstructed frames
300 300 300 300 300 149 300 300 151

2 11.75 12.07 11.46 30.72 11.76 11.66 31.41 36.59 33.77

3 12.11 12.48 11.68 31.79 11.86 11.71 31.60 36.53 33.64

number of reconstructed frames
150 149 148 300 150 149 294 300 295

2 22.80 17.88 24.86 16.59 13.57 14.26 17.55 17.77 17.36

3 23.84 18.32 25.85 16.78 13.64 14.31 17.60 17.81 17.41

number of reconstructed frames
289 273 289 271 260 266 284 291 288

(a) PER=10-3, random
(b) PER=10-2, random
(c) PER=10-1, random
Fig. 4.   Rate-distortion curves at Postmux module for varying PER. Raw Foreman video (CIF, 300 frames) input and MPEG4-coded output at the Remux module.

[Home] [Research] [Released Software] [Documentation] [Contact]