The cellular paradigm allocates spectrum, infrastructure, and identity through a centralized operator. This is a product of a 1980s regulatory environment that no longer reflects the computational capacity of end-user devices or the threat model of a connected society. Modern devices ship with multi-core SoCs, neural accelerators, multi-band radios, and hardware secure enclaves — a surface sufficient to participate as a first-class network element. We argue that the operator-centric design is no longer technically necessary, and that a properly engineered peer-to-peer mesh is strictly superior on resilience, privacy, and cost.
Contributions. We contribute (1) an on-device PPO routing agent with federated weight aggregation; (2) a unified multi-radio scheduler minimizing a single energy–latency–reliability Lagrangian; (3) a hardware-anchored decentralized identity with post-quantum social recovery; (4) a relay incentive market that does not require a synchronous blockchain consensus layer; (5) simulation evidence at planetary scale.
AETHER builds on the lineage of MANET protocols (AODV [1], OLSR [2], BATMAN), opportunistic networks (Haggle, Bundle Protocol), decentralized identity (DIDs [8], SSI), and decentralized storage (IPFS [7]). It differs from prior mesh systems on five axes simultaneously: per-node RL routing rather than hand-tuned heuristics; multi-radio joint scheduling; hardware-anchored post-quantum identity; federated learning across the mesh itself; and a relay incentive market without synchronous global consensus.
AETHER decomposes into four layers: a discovery layer (BLE/UWB/Wi-Fi/LoRa beaconing with signed capability advertisements), a topology layer (link-state gossip with bloom-filter hop summaries), a routing layer (per-node PPO agent producing a softmax distribution over top-K neighbors), and a transport layer (multi-path delivery with end-to-end FEC). Global naming is provided by a Kademlia DHT [3] shardable along Google S2 geographic cells, yielding O(log N) lookup with O(1) regional locality.
The routing agent is a small (≈ 40 KB) PPO [6] model with an 8-dimensional state — signal quality, battery headroom, trust, available bandwidth, predicted latency, mobility alignment, congestion, Sybil risk — and a 64-unit MLP. Weights are federated hourly via a privacy-preserving aggregation channel: only encrypted gradient summaries leave the device. The result is a routing policy that is globally informed yet locally executed, with no controller and no single point of failure.
Adversary model: Dolev–Yao with quantum capability and up to f = ⌊(n−1)/3⌋ compromised neighbors per node. Confidentiality is provided by hybrid Kyber-768 [4] + X25519 key exchange; integrity by Dilithium [5] signatures over Merkle-DAG receipts; anonymity by onion routing with cover traffic; Sybil resistance by proof-of-useful-work bound to neighbor attestations. Eclipse attacks are mitigated by enforcing diverse peer selection across S2 cells.
A reference device pairs an 8-core 3 nm ARM SoC with a 30 TOPS NPU dedicated to routing and DSP, a FIPS 140-3 L3 secure enclave for keygen and attestation, a multi-radio front-end (BLE 5.4, UWB, Wi-Fi 7, LoRa 868/915, S-band sat), a 4×4 MIMO beamforming antenna array, and a 6500 mAh Si-anode solid-state battery with a solar trickle backplate. This is not a smartphone; it is a mesh node.
We evaluate AETHER on ns-3 with a custom mesh extension, federated across 5,000-node clusters to emulate planetary topologies. Latency degrades sub-linearly with scale; throughput is sustained above 700 Mbps aggregated; loss stays below 0.31% at 10 B nodes. Under a 20% adversarial failure rate, delivery remains at 99.7% with sub-second self-healing.
| Scale | Median latency | Aggregate throughput | Loss | Self-heal time |
|---|---|---|---|---|
| 10 M | 45 ms | 1.2 Gbps | 0.04% | 180 ms |
| 100 M | 62 ms | 1.0 Gbps | 0.09% | 240 ms |
| 1 B | 84 ms | 850 Mbps | 0.18% | 360 ms |
| 10 B | 118 ms | 700 Mbps | 0.31% | 520 ms |
Devices earn AETHER credits per verified relayed byte, weighted by reputation. Verification requires neighbor attestations and proof-of-useful-work signatures; double-relay and wash-relay attacks are detected by checking attestation consistency across non-collusive peers. Credits redeem for premium QoS, hardware, or fiat off-ramp via federated exchanges, without requiring a synchronous global consensus layer.
AETHER is not a replacement for fibre backbones — it is a replacement for the last-mile cellular access network and the SIM-based identity layer above it. The proposal raises legitimate regulatory questions around spectrum allocation, lawful access, and emergency services. Authority-signed emergency broadcast and verifiable identity attestations are stronger primitives than the operator-mediated equivalents they replace; spectrum policy can adopt a tiered access model already proven in CBRS-style deployments.
AETHER demonstrates that a peer-to-peer, AI-routed, post-quantum mesh can replace the cellular access layer on every measurable axis: latency, throughput, coverage, resilience, privacy, cost, and crypto. The remaining work is engineering, not science.