Error resilient packet switched H.264 video telephony over third generation networks.
Real-time video communication over wireless networks is a challenging problem because wireless channels suffer from fading, additive noise and interference, which translate into packet loss and delay. Since modern video encoders deliver video packets with decoding dependencies, packet loss and delay can significantly degrade the video quality at the receiver. Many error resilience mechanisms have been proposed to combat packet loss in wireless networks, but only a few were specifically designed for packet switched video telephony over Third Generation (3G) networks. The first part of the thesis presents an error resilience technique for packet switched video telephony that combines application layer Forward Error Correction (FEC) with rateless codes, Reference Picture Selection (RPS) and cross layer optimization. Rateless codes have lower encoding and decoding computational complexity compared to traditional error correcting codes. One can use them on complexity constrained hand-held devices. Also, their redundancy does not need to be fixed in advance and any number of encoded symbols can be generated on the fly. Reference picture selection is used to limit the effect of spatio-temporal error propagation. Limiting the effect of spatio-temporal error propagation results in better video quality. Cross layer optimization is used to minimize the data loss at the application layer when data is lost at the data link layer. Experimental results on a High Speed Packet Access (HSPA) network simulator for H.264 compressed standard video sequences show that the proposed technique achieves significant Peak Signal to Noise Ratio (PSNR) and Percentage Degraded Video Duration (PDVD) improvements over a state of the art error resilience technique known as Interactive Error Control (IEC), which is a combination of Error Tracking and feedback based Reference Picture Selection. The improvement is obtained at a cost of higher end-to-end delay. The proposed technique is improved by making the FEC (Rateless code) redundancy channel adaptive. Automatic Repeat Request (ARQ) is used to adjust the redundancy of the Rateless codes according to the channel conditions. Experimental results show that the channel adaptive scheme achieves significant PSNR and PDVD improvements over the static scheme for a simulated Long Term Evolution (LTE) network. In the third part of the thesis, the performance of the previous two schemes is improved by making the transmitter predict when rateless decoding will fail. In this case, reference picture selection is invoked early and transmission of encoded symbols for that source block is aborted. Simulations for an LTE network show that this results in video quality improvement and bandwidth savings. In the last part of the thesis, the performance of the adaptive technique is improved by exploiting the history of the wireless channel. In a Rayleigh fading wireless channel, the RLC-PDU losses are correlated under certain conditions. This correlation is exploited to adjust the redundancy of the Rateless code and results in higher Rateless code decoding success rate and higher video quality. Simulations for an LTE network show that the improvement was significant when the packet loss rate in the two wireless links was 10%. To facilitate the implementation of the proposed error resilience techniques in practical scenarios, RTP/UDP/IP level packetization schemes are also proposed for each error resilience technique. Compared to existing work, the proposed error resilience techniques provide better video quality. Also, more emphasis is given to implementation issues in 3G networks.
- PhD