PEMNet: Towards Autonomous and Enhanced Environment-Aware Mobile Networks

TL;DR

PEMNet addresses the need for environment-aware mobile networks operating under strict latency and energy constraints in evolving 5G/6G contexts. It introduces PEM, a site-specific perception embedding map that jointly encodes grid-level channel statistics and spatial-temporal traffic demand using standard measurements, yielding outputs such as $[H\_hat, T\_hat] = \mathcal{M}(z,t)$. PEM is constructed via environment perception (MR/MDT, drive tests) and information embedding (STF feature $z$, CKM for APS/DPS from RSRP, DTM with models like GPR/ConvLSTM/GraphNets, plus data interpolation). The PEM outputs feed PEMNet optimization (problem (P1)) across PHY, MAC, and network layers with long-term variables $x_s$, short-term variables $x_r$, and objective $U(\cdot)$ under feasibility sets $\mathcal{F}_s(\cdot), \mathcal{F}_r(\cdot)$. Case studies on multi-cell beamforming and receive-beam design demonstrate reduced signaling overhead and improved hotspot-oriented performance, validating the practicality of standard-measurement–driven environment perception for autonomous wireless networks. The work enables practical, cross-layer optimization and points to future directions such as multi-modal data fusion, collaborative DTM construction, and autonomous PEM updates to sustain performance in dynamic deployments.

Abstract

With 5G deployment and the evolution toward 6G, mobile networks must make decisions in highly dynamic environments under strict latency, energy, and spectrum constraints. Achieving this goal, however, depends on prior knowledge of spatial-temporal variations in wireless channels and traffic demands. This motivates a joint, site-specific representation of radio propagation and user demand that is queryable at low online overhead. In this work, we propose the perception embedding map (PEM), a localized framework that embeds fine-grained channel statistics together with grid-level spatial-temporal traffic patterns over a base station's coverage. PEM is built from standard-compliant measurements -- such as measurement report and scheduling/quality-of-service logs -- so it can be deployed and maintained at scale with low cost. Integrated into PEM, this joint knowledge supports enhanced environment-aware optimization across PHY, MAC, and network layers while substantially reducing training overhead and signaling. Compared with existing site-specific channel maps and digital-twin replicas, PEM distinctively emphasizes (i) joint channel-traffic embedding, which is essential for network optimization, and (ii) practical construction using standard measurements, enabling network autonomy while striking a favorable fidelity-cost balance.

Figures (4)

• Figure 1: The PEM is a site-specific database that embeds the perceived environment information to provide statistical channel knowledge $\hat{{\bm H}}$ and the predicted data traffic demand $\hat{{\bm T}}$. The overall coverage area of the site (BS) is divided into multiple spatial grids. With the STFs $\{({\bm z}, t)\}$ of UEs as inputs, the PEM outputs their corresponding $\{(\hat{{\bm H}}, \hat{{\bm T}})\}$ to support the environment-aware communication design of mobile networks. In the PEM-enabled open radio access network (O-RAN), each RAN consists of one or several radio units (RUs), a distributed unit (DU), a centralized unit (CU), a communication design unit (CDU) and a PEM. The PEM is connected to DU for data collection and STF inputs, and its output is fed into the CDU to conduct environment-aware optimization.
• Figure 2: A multi-cell network including 3 BSs and 36 UEs.
• Figure 3: ESR achieved by different BF schemes versus $N_t$, $n_0 = 500$.
• Figure 4: PEMNet-based beam intensity map aligned with user distribution hotspots.