In traditional applications of coding, it has been used to create, synthetically, out of symbols whose reliability is individually poor, a link that has low error probability while maintaining throughput to the extent possible. We propose to create systems, based on coding, that are synthesized algebraically from heterogeneous networks that do not individually have the performance characteristics we seek. Recent developments in video transmission have indeed shown that the current approach based on single networks (say 3G or 4G alone) may not suffice to satisfy demands.
Our results show that the judicious use of a more expensive network only in case where we run the risk of an interruption in viewing video provides essentially the same performance of having the best of networks, while only incurring essentially the same cost as the cheaper network. Approaches based on combining heterogeneous networks without coding suffer from the difficulty of having to coordinate across separate flows, which may need retransmission of specific packets across one or more networks. Instead, coded approaches allow degrees of freedom to be gathered in a flexible way across different networks, as shown below. We consider the case where a WiFi system is free but unreliable, while the LTE system is expensive but reliable. For off-line policies (Queue-length not observable), the optimal policy is greedy and we use the costly network only for a certain time.  For online policies (Queue-length observable), we can consider a safe and a risky policy. For the safe policy, we start with the costly network until the queue-length hits a threshold. Once we hit the threshold, we never switch back.  In the risky policy, we use the costly network only if the queue-length falls below a threshold. The trade-off between cost and delay is strongly dependent on the choice of policies.