Rateless Forward-Error-Correction for Lossy Mesh Media

Mechanism

The transmitter encodes the source message — a sequence of N input symbols of fixed length — into an unbounded stream of output symbols, each of which is a randomized linear combination of input symbols selected by a degree distribution and a seed that the receiver can reproduce from a small header. The fountain property is that any subset of N plus epsilon output symbols (where epsilon is a small overhead determined by the code family — single-digit percentage for Raptor codes, modest constant for LT codes) suffices to reconstruct the original message with overwhelming probability, regardless of which specific output symbols the receiver actually obtained. The transmitter does not need to know in advance which symbols any given receiver will collect; the receiver does not need to negotiate which symbols the transmitter should send.

The encoding is rateless. The transmitter can produce arbitrarily many distinct output symbols from the same source message without coordinating with any receiver. A transmitter operating in a one-to-many or many-to-many topology emits a stream that any receiver, at any point along its reception path, can sample from until it has accumulated enough symbols to reconstruct. There is no retransmission negotiation, no selective-acknowledgment loop, no congestion-window-controlled feedback path on the message-recovery axis.

The receiver collects fragments through any combination of sources — direct line-of-sight, multi-hop relay, alternate-path forwarding, opportunistic store-and-forward from passing units. Each fragment carries the header information required to identify the source message it belongs to and to reconstruct the linear combination it represents. Once the receiver has accumulated N plus epsilon symbols whose combinations are linearly independent over the appropriate field, the message reconstructs through standard linear-algebra decoding (Gaussian elimination for small messages; belief-propagation decoding accelerated by precoding for Raptor codes at large message sizes).

The architecture treats the FEC descriptor as a structural element of the wire format. The message header carries the source-message identifier, the input-symbol count N, the fountain-code family identifier, the degree-distribution parameters, and the seed material required to reproduce per-symbol combinations. Receivers do not negotiate FEC parameters with transmitters; they read parameters from the header and decode accordingly.

Operating Parameters

Source-block size N is chosen against the message-size distribution and the channel's loss profile. Small messages — single-observation broadcasts, status pings, attestation tokens — use N in the tens to small hundreds with belief-propagation decoding latency dominating. Medium messages — multi-observation bundles, attestation aggregates, configuration updates — use N in the thousands with Raptor decoding's near-linear time complexity. Large messages — firmware updates, model deployments, archive transfers — use N in the tens of thousands with Raptor's precoded structure preserving feasibility.

The overhead epsilon is determined by the chosen code family. Raptor codes (RFC 5053, RaptorQ in RFC 6330) achieve epsilon below 1 percent for typical N, which means a receiver collecting roughly 1.01 times N symbols decodes with probability greater than 0.999. LT codes achieve worst-case epsilon around 5 to 10 percent depending on the degree distribution, which is acceptable in deployments where decode complexity is the binding constraint. Random linear fountain codes achieve information-theoretically optimal epsilon (effectively zero) at the cost of decode complexity that scales as N-cubed; they are used only at very small N.

Fragment size is matched to the link-layer maximum transmission unit minus protocol overhead — typically 200 to 1500 bytes depending on the underlying medium (LoRa, sub-GHz mesh, IEEE 802.11, satellite uplink, optical free-space). The transmitter generates fragments at the line rate of its outbound interface; the receiver accepts fragments at whatever rate it observes them. There is no rate-matching feedback loop because the rateless property eliminates the need for one.

Reconstruction-stop signaling is local to the receiver. Once a receiver decodes the source message, it ceases to be interested in further fragments of that message and may suppress further forwarding of those fragments through its own relay path. The signaling is a local-state change, not a network-wide notification, which keeps the architecture's control-plane footprint small. Transmitters continue to emit fragments for a configured duration after first transmission, then stop; receivers that arrive late in the emission window may still reconstruct from the cumulative population of fragments still in flight or buffered at relays.

Loss-rate operating range extends from near-zero to greater than 90 percent end-to-end loss. At low loss, the rateless overhead is a few percent of total bytes — a small efficiency cost compared to retransmission-free operation. At high loss, the architecture continues to deliver where retransmission-based protocols stall completely; the receiver simply waits longer to accumulate the required N plus epsilon symbols. The upper bound on tolerable loss is determined by the relationship between transmission duration, fragment generation rate, and minimum acceptable reconstruction latency, not by the protocol itself.

Alternative Embodiments

Hybrid systematic embodiments transmit the original input symbols first, followed by rateless repair symbols. Receivers that experience low loss may decode directly from received input symbols without invoking the linear-algebra decoder, saving compute. Receivers that experience high loss fall back to the full rateless decode using whatever combination of input and repair symbols they collected. The variant is the basis of the RaptorQ standard and is preferred in deployments where most receivers see low loss but a long-tail subset experiences deep loss.

Network-coded relay embodiments allow intermediate relays to combine fragments they receive into new randomized combinations that are themselves valid rateless symbols of the same source message. A receiver downstream of such a relay decodes from the combined symbols as if they had been generated at the original transmitter. The variant amplifies fragment diversity in multi-hop topologies and increases the probability that any given receiver collects a linearly-independent set quickly. It composes naturally with hop-history because each combined fragment carries the merged hop history of its constituent fragments.

Layered-priority embodiments split the source message into priority tiers, each rateless-encoded separately. A receiver with limited reception capacity decodes the high-priority tier first and obtains a usable reduced version of the message; receivers with greater capacity decode subsequent tiers and obtain progressively more complete versions. The variant supports graceful degradation in deeply heterogeneous mesh populations where some receivers are deep in fade and others have good reception.

Authenticated-fragment embodiments bind each fragment to the originator's credential through a per-fragment signature or aggregated signature scheme. A receiver verifies fragment authenticity before contributing it to the decode pool, which prevents adversarial injection of corrupted fragments that would cause the linear-algebra decoder to produce invalid output. The variant is required in defense and safety-critical deployments where the threat model includes active fragment-injection attacks.

Coded-caching embodiments treat relays as opportunistic caches of rateless symbols. A relay that observes fragment traffic for a popular source message retains a subset of fragments and serves them to later-arriving receivers on demand. The mechanism extends rateless transport into a content-distribution layer for the mesh and is particularly effective for periodic broadcasts (mapping updates, configuration distribution, public-key directory updates) that many receivers consume independently.

Streaming-fountain embodiments adapt the rateless mechanism to continuous data streams rather than discrete messages. Source data is partitioned into overlapping windows, each rateless-encoded with bounded decode latency. The variant supports telemetry, video, and continuous attestation streams in lossy mesh while preserving the no-retransmission property.

Composition With Mesh Routing and Memory-Native Transport

The rateless fragment is a first-class memory-native transport unit. Each fragment travels through the mesh independently — relays forward fragments without reassembling the source message, multiple paths carry different fragments simultaneously, and receivers anywhere in the mesh begin reconstruction as soon as their accumulated fragment set crosses the N-plus-epsilon threshold. The architecture composes naturally with hop-history: each fragment carries its own hop record, and the reconstructed source message's effective hop history reflects the diversity of paths the constituent fragments traveled. A receiver gains insight not just into where the message originated but into the topology of the mesh between origin and self.

Composition with the governed-mesh credentialing layer places fragment authentication on the same credential infrastructure as observation authentication. The originator signs at the source-message level (or, in authenticated-fragment embodiments, at the per-fragment level); receivers verify against the dynamic device hash chain that the credentialing authority has issued for the originator. The composite effect is a transport that is simultaneously loss-tolerant and credential-bound, which is the combination that defense, V2X, and regulated-industrial deployments require.

Composition with capability-awareness composite admissibility places the reconstructed source message into the same admissibility pipeline that fully-received messages enter. The TTL, class membership, credentialing, and freshness checks apply identically to a message that arrived through a single transmission and a message that reconstructed from fragments collected over many hops and many seconds. The fountain decode is invisible to layers above transport; from the perspective of capability-awareness, a successfully reconstructed message is simply a message.

Multi-source emission is a natural deployment pattern. Multiple transmitters, each holding the same source message — perhaps because a coordination layer above transport has selected several relays to amplify a critical broadcast — emit independent rateless streams using different seeds. A receiver collects fragments from all of them indistinguishably and reconstructs once it has N plus epsilon linearly independent symbols across the combined population. The pattern is the mesh analog of a content-distribution network's edge-server fan-out, but without the centralized coordination that CDN architectures impose.

Prior-Art Context

Fountain coding has a substantial cryptographic and information-theoretic literature. Luby's LT codes (2002) introduced the practical fountain construction; Shokrollahi's Raptor codes (2006) added precoding to achieve linear-time decode; RFC 5053 standardized Raptor for IETF transport; RFC 6330 standardized the more efficient RaptorQ variant; 3GPP MBMS uses Raptor codes for cellular broadcast. The prior art establishes the coding mechanism, its decode complexity, and its overhead bounds in detail.

What the prior art does not establish is the structural integration of rateless FEC into a credentialed governed-mesh transport that is simultaneously memory-native (fragments are first-class units of the protocol's memory model rather than payload of a separate transport sublayer), credential-bound (fragments verify against the same dynamic device hash chain that observations verify against), and composable with capability-awareness admissibility above the transport layer. Existing rateless-FEC deployments treat the code as a transport-internal mechanism; the disclosed architecture treats it as a first-class element of the protocol's structural model.

Retransmission-based transport — TCP, QUIC, and the broader family — depends on round-trip negotiation between sender and receiver. The pattern is well-suited to the low-loss, low-asymmetry network conditions for which it was designed; it fails in deeply lossy mesh because retransmission negotiation requires a feedback channel of comparable quality to the forward channel, which mesh environments cannot guarantee. Forward-error-correction overlays on TCP exist but are constrained by TCP's reliable-ordered-stream abstraction in ways that prevent effective deployment in one-to-many or intermittent-connectivity contexts. The disclosed architecture replaces the retransmission abstraction outright at the layer where mesh deployment encounters the loss-pattern frontier.

Application-layer rateless deployments (file-transfer overlays, content-distribution protocols, cellular broadcast) demonstrate the engineering feasibility of fountain coding at scale but do not address the credentialing, audit, or composite-admissibility integration that decision-grade mesh deployment requires. The disclosed architecture closes the integration gap.

Disclosure Scope

Provisional Application 64/050,895 discloses rateless FEC as a primitive of the memory-native protocol. The disclosure includes (a) the integration of fountain coding (LT, Raptor, RaptorQ, and equivalent rateless erasure codes) as a structural wire-format element rather than as an application-layer overlay; (b) the per-fragment header structure carrying source identifier, input-symbol count, code-family identifier, and seed material; (c) the no-retransmission decode model in which receivers reconstruct after collecting N plus epsilon symbols from any combination of sources; (d) the composition with mesh routing in which fragments traverse independently and reconstruction occurs at any topologically-reachable receiver; (e) the alternative embodiments covering hybrid systematic encoding, network-coded relay, layered priority, authenticated fragments, coded caching, and streaming fountains; and (f) the composition with credentialing and capability-awareness admissibility that places rateless transport in the same governance and audit framework as the observation layer above it.

The disclosure scope contemplates application across defense mesh radios operating in jammed conditions, satellite-link mesh in deep-fade conditions, expeditionary mesh in temporarily-disrupted environments, dense-urban wireless mesh in heavy-multipath conditions, indoor mesh through structural attenuation, and cellular-V2X in non-line-of-sight environments. The patent positions the primitive at the layer where current mesh deployment encounters the loss-pattern frontier that retransmission-based protocols cannot effectively address. By replacing the round-trip-negotiation abstraction with rateless reconstruction, the architecture eliminates the structural fragility that has constrained mesh deployment in challenging media for decades, and does so in a form that integrates with the credentialing and admissibility layers above transport rather than operating in isolation from them.