Average consensus is a fundamental concept in distributed computing, where distributed agents exchange messages in order to obtain the average of their ini- tial values without relying on a centralized computing unit. However, achiev- ing average consensus in the presence of communication imperfections, such as quantization and random link or node failures, becomes more challenging. This thesis evaluates various average consensus algorithms regarding their ability to mitigate quantization effects and explores node dropout for reducing communi- cation cost per iteration. It also identifies the conditions required for achieving average consensus in both scenarios.  The first part of this thesis deals with average consensus with quantized up- dates, comparing algorithms such as quantized gossip, average preserving quan- tized gossip, and CHOCO-GOSSIP. CHOCO-GOSSIP stands out as the most ef- fective algorithm, which shows the importance of pre-compensating the quanti- zation error before transmitting the node values. Among other algorithms, aver- age preserving quantized gossip shows slightly better performance. Additionally, graphs with higher connectivity tend to perform better.  The second part of this thesis focuses on energy-efficient average consensus with random node dropout. It compares the optimized node dropout proba- bilities with heuristic designs such as the degree-based method and Metropolis- Hastings method. The degree-based method is shown to give good convergence performance despite its simplicity. Furthermore, in irregular graphs, the perfor- mance difference between optimized probabilities and heuristic designs tends to be more pronounced.