Rateless FEC for Lossy Mesh Media

Mechanism

The rateless forward-error-correction (FEC) primitive operates as follows. A source agent partitioning a message M of length K source symbols invokes an encoder that produces a potentially unbounded stream of coded symbols c₁, c₂, c₃, …. Each coded symbol is a deterministic function — typically an XOR — of a pseudorandomly selected subset of the source symbols, with the subset identifier carried in the symbol header. The encoder is rateless: it never exhausts its symbol space and never requires foreknowledge of which receivers will be served, what loss rates they will experience, or what aggregate erasure pattern the mesh will impose.

A receiver collects coded symbols from any reachable source — the originator, any intermediate relay that has cached the stream, or any peer that has received and is re-radiating fragments. When the receiver has accumulated a number of distinct symbols equal to K + ε, where ε is the small reception overhead characteristic of the chosen code family (typically a fraction of a percent for Raptor codes, a few percent for LT codes), it executes belief-propagation or Gaussian-elimination decoding to recover the original K source symbols. Because each coded symbol is independently useful, the receiver does not care which symbols arrive, in what order, or by what path. The decoder treats the channel as a pure erasure channel and absorbs the underlying physical-layer mechanisms — interference, jamming, multipath fade, structural attenuation — as undifferentiated symbol loss.

The wire format binds rateless FEC to the spatial-mesh framing layer. Each coded symbol carries a payload identifier, a symbol index, an encoder seed reference, and a hop-history annotation. Relays forward symbols opportunistically without reassembling the source payload; reassembly is strictly an endpoint operation. A relay that holds a partial stream may emit additional coded symbols on behalf of the source by re-running the encoder against the cached source symbols if it possesses them, or — in the more common opaque-relay case — by simply re-radiating the coded symbols it has already received. Receivers tolerate duplicates without penalty: a duplicate symbol contributes no new information to the decoder but costs only the receive slot.

Critically, the architecture eliminates the round-trip negotiation that defines retransmission-based transports. There is no acknowledgment timer, no selective-repeat request, no congestion window, and no head-of-line blocking. The source emits symbols at whatever rate its policy permits; receivers consume what they can; reconstruction completes when the symbol count crosses the decoding threshold. The protocol's progress is monotonic in received symbols, not in successful round trips.

Operating Parameters

The principal tunable parameters are the source-block size K, the symbol size, the code family (LT, Raptor R10, RaptorQ, or other systematic-rateless variant), the encoder seed schedule, and the relay caching depth. Source-block sizes in the range of 64 to 8192 symbols cover the operational envelope: smaller blocks reduce decoding latency at the cost of higher per-block overhead; larger blocks amortize overhead across more payload at the cost of memory pressure on constrained receivers. Symbol sizes are chosen against the underlying mesh frame size such that one coded symbol fits inside one mesh frame after framing overhead is deducted.

The reception overhead ε is bounded by the code family. RaptorQ achieves ε values below 0.02 with high probability for K in the practical range. LT codes incur larger ε but admit simpler encoders suitable for severely constrained source platforms. The architecture does not mandate a single code family; it specifies the integration points — symbol header layout, encoder seed propagation, decoder dispatch — and admits any code family that satisfies the rateless erasure-channel contract.

Source emission rate is governed by an upper bound that reflects the source's duty-cycle budget and a lower bound that reflects the longest tolerable reconstruction latency at the worst-served receiver. Between these bounds, the source may modulate emission in response to coarse channel-quality feedback, but it does not require per-receiver feedback. Relay caching depth — the number of distinct payloads, and the symbol count per payload, that a relay retains — is a policy parameter set against the relay's storage budget and the expected receiver-arrival distribution.

Alternative Embodiments

The disclosure contemplates embodiments in which the FEC layer is composed with link-layer convolutional or BCH coding, such that the rateless layer absorbs the residual erasure pattern that survives the inner code's correction capability. In other embodiments, the rateless layer is the sole coding layer, with the underlying physical layer presenting raw symbols and the rateless decoder absorbing all loss directly.

Embodiments differ in how the encoder seed is communicated. In one embodiment, every coded symbol carries a full degree-distribution descriptor and source-symbol index list. In another, symbols carry only a compact seed identifier, and receivers reconstruct the degree distribution and selection mask by replaying a shared pseudorandom generator. The compact-seed embodiment minimizes per-symbol overhead at the cost of requiring receiver-side state initialization; the full-descriptor embodiment is stateless at the receiver but consumes more wire bandwidth per symbol.

Further embodiments support multi-source emission, in which two or more distinct source agents holding identical source-symbol blocks emit independent coded streams under coordinated seed schedules. Receivers combine symbols from all sources into a single decoder instance. This embodiment is suited to expeditionary operations where redundant transmitters provide fault tolerance against transmitter loss and to satellite-link operations where multiple ground stations or multiple satellites cooperate in serving a single receiver population.

An additional embodiment integrates rateless FEC with the spatial-mesh hop-history primitive such that the hop-history descriptor of the reconstructed payload reflects the union of paths traversed by the contributing symbols. This produces an audit record of path diversity that is useful for forensic analysis, anti-jamming attribution, and operational assurance that no single path was a single point of failure.

Composition With Mesh Substrate

Rateless FEC composes naturally with the spatial-mesh routing primitives. Because each coded symbol is independently routable and independently useful, the routing layer is freed from the burden of maintaining flow affinity: there is no requirement that successive symbols traverse the same path, and there is no penalty for spreading symbols across all available paths simultaneously. This permits aggressive multipath emission and aggressive opportunistic forwarding without the reordering penalties that conventional transports incur.

The primitive composes with the substrate's caching and re-radiation policies. A relay that holds a coded-symbol cache continues to serve receivers that arrive after the source has stopped emitting, extending the temporal reach of the payload beyond the source's duty cycle. The primitive also composes with the substrate's quality-of-service classes: high-priority payloads receive larger source-block budgets and higher emission rates; bulk payloads receive smaller budgets and longer reconstruction windows.

Prior-Art Distinction

LT codes (Luby, 2002) and Raptor codes (Shokrollahi, 2006) are well established in the literature and have been deployed in 3GPP MBMS, DVB-H IPDC, and various store-and-forward systems. The novelty of the present disclosure is not the rateless code family itself but the integration architecture: the binding of the rateless primitive to a spatial-mesh wire format such that opaque relays, multipath emission, hop-history annotation, and policy-governed caching are all first-class behaviors at the FEC layer rather than overlay behaviors above a conventional transport.

Conventional rateless deployments treat the rateless layer as an application-level concern above a session-oriented or datagram transport, which forecloses the multipath and opaque-relay properties that the present integration achieves. The disclosure positions the primitive at the layer where current mesh deployment encounters the loss-pattern frontier — defense radio under jamming, satellite link in deep fade, dense-urban penetration, indoor wall-loss — that retransmission-based architectures cannot service.

Failure Modes and Mitigations

Three failure modes warrant explicit treatment. The first is decoder starvation, in which a receiver accumulates fewer than K + ε distinct symbols within its operational window. The architecture mitigates starvation by relay re-radiation, by source emission persistence beyond the nominal completion threshold, and by policy-controlled fallback to a smaller source-block size for the affected payload. A receiver that is starving signals nothing — the architecture does not depend on starvation feedback — but the source's emission policy may be configured to run a fixed multiple of K symbols before declaring the payload complete, providing the structural over-provision that absorbs worst-case starvation.

The second failure mode is degenerate symbol diversity, in which the symbols a receiver collects are linearly dependent at a rate that exceeds the code family's tolerance. This is most acute when a single relay is the dominant source of a receiver's symbols and that relay holds only a narrow slice of the original symbol stream. The architecture mitigates by relay-side seed perturbation in re-encoding embodiments and by topology-aware source emission that diversifies the seed schedule across emission slots. The result is that even when a receiver draws disproportionately from a single relay, the symbols it draws span enough of the source-symbol space to satisfy the decoder.

The third failure mode is adversarial symbol injection, in which a hostile node emits forged coded symbols that decode to corrupted source content. The architecture mitigates with per-payload integrity, in which the source binds the decoded payload's hash to a signed manifest that travels in the symbol stream. Receivers that complete decoding verify the manifest signature before accepting the payload; injected symbols that lead to a decoded artifact whose hash does not match the manifest are discarded. The integrity layer is independent of the FEC layer and composes with whatever signing infrastructure the deployment uses.

Disclosure Scope

The provisional application teaches the wire-format integration, the relay caching and re-radiation behavior, the multipath emission model, the hop-history composition, the failure-mode mitigations, and the policy parameters that govern emission rate, source-block size, and relay caching depth. The disclosure is code-family agnostic, naming LT and Raptor as exemplary rateless families and reserving the same architectural treatment for any future erasure code that satisfies the rateless contract. The scope encompasses defense mesh, satellite-link mesh, expeditionary mesh, dense-urban mesh, and indoor mesh deployments operating in the loss-pattern regime that motivates the integration. The disclosure further reserves application to disaster-recovery deployments where infrastructure is partial or destroyed, to maritime and aviation mesh where physical-layer conditions are intermittently severe, and to deep-space relay applications where round-trip latency precludes any retransmission-based transport entirely.