E2E Network Slicing Optimization for Control- and User-Plane Separation-based SAGINs with DRL

Existing works on space-air-ground integrated networks (SAGINs) have not sufficiently explored the synergy between control- and user-plane separation (CUPS) and end-to-end (E2E) network slicing. In this paper, we present a new paradigm of E2E network slicing for CUPS-based SAGIN (CUPS-SAGIN), where the control base stations (BSs), control unmanned aerial vehicles (UAVs) together with the low earth orbit (LEO) satellites form the control plane (CP) to provide access for users, while the traffic BSs and traffic UAVs form the user plane (UP) to provide various services for users. We develop two typical E2E network slices for the CP, i.e., massive access and wide-area access slices, as well as two typical E2E network slices for the UP, i.e., high throughput and low delay slices. The utilities of the two CP access slices are defined based on their access percentages, while the utilities of the two UP service slices are defined based on their E2E throughput and delay, respectively. To maximize the sum-utility of the four E2E network slices, a judgement network-assisted deep deterministic policy gradient algorithm is designed to jointly optimize the UAV position, users' association with BS/UAV/LEO satellite and the allocations of the transport network capacity, downlink bandwidth and transmission power. Simulation results demonstrate the advantage of the proposed E2E network slicing for CUPS-SAGIN in terms of the access percentage, E2E throughput and delay as compared with the benchmarks, including the E2E network slicing schemes for the terrestrial cellular networks, coupled SAGINs, etc.