Abstract
When it comes to multimedia communications, they are considerably sensitive to delay time and packet erasures. Digital Fountain codes are becoming increasingly important thanks to the ability to protect the source data in high probability. The most representative Digital Fountain codes are LT codes. With LT codes, the number of encoding packets can be decided on-the-fly and it is considerably favorable characteristic especially in broadcast channel. It is because different end-users actually suffer from distinct channel loss rates. However, LT encoder simply generates the encoding packets through randomly chosen encoding degree so that it cannot control or manage the service quality according to the distinct service demands of clients. In this thesis, we proposed a namely “service-driven approximate LT codes” that using numerous basic building structures to form the coding graph. The proposed encoder arranges the encoding degree through the basic building structure to generate encoding packets in accordance with various transmission policies and the overall degree distribution approximates to Ideal Soliton distribution. Through our investigations, we found that after transmitting a number of encoded packets in broadcast channel, the clients under low channel loss rates need more high-degree encoding packets to achieve fully-decoded, whereas others need more low-degree packets to recover more source data due to the lack of receiving sufficient encoding packets. Hence, the proposed encoder ensures to transmit low-degree encoding packets first to significantly cut down the average encoding degree then switches to generate high-degree encoding packets if the majority of clients decode a large percentage of source data to always satisfy the demands of the majority first. Furthermore, comparing to LT codes in point-to-point protocol, the proposed codes also introduce lower decoding code overhead in terms of short source data block length, better intermediate performance to achieve graceful quality degradation over a wide range of channel loss rates, and built-in unequal error protection property through appropriately arranging the placements of source data in the coding graph. Last but not least, similar to LT codes, the decoding complexity of the proposed codes is rather low so that it can be applied on real-time decoders, such as mobile phones, which only offer limited computational power and strictly delay-time constrained.