This thesis considers zero-delay multiple-descriptions (ZDMD) coding of stationary scalar AR(p) sources and mean square error (MSE) distortion constraints. Specifically, the symmetric two-descriptions case is considered for transmission over unreliable channels. A novel operational ZDMD coding scheme is developed based on index assignment and differential pulse code modulation (DPCM) quantization assuming feedback from the decoder. A prediction error sample is quantized and mapped to two descriptions using a non-linear one-to-many map. The decoder produces a reconstruction based on the received packets, where the fidelity is higher when both packets are received.The developed scheme can be applied to any assumption on the model order of the source. In the DPCM quantization, the MMSE predictors are used. The theoretical performance of the codes are assessed, and their practical performances are demonstrated in a simulation study.It is shown that the novel ZDMD scheme achieves a smaller MSE distortion compared to existing ZDMD coding schemes. Furthermore, the robustness against packet losses is demonstrated.