Adaptive FHSS: prove restoration on air, and draw it in the RF primer - #339
Merged
Conversation
The failsafe mode stopped at the exclusion, so the transmitter's own probe-driven restoration had never run on hardware — a merged path with unit tests and no on-air evidence, which is the shape of dead code. An autonomous exclusion that cannot be undone is worse than none: the link shrinks permanently the first time something transient sits on a channel, and with the receiver mute there is no other evidence that could bring it back. The mode now runs the same jammer-removal phase the receiver-driven modes do, and asserts the transmitter aired keyed probes on the channel it excluded, restored it from them, and that the receiver followed back to the full hopset. Measured, receiver muted throughout: exclude then restore, both attributed to the failsafe origin; probes on the excluded channel delivering; mask history gen 1 mask 0xd then gen 2 mask 0xf.
Both places the README describes hopping were written before the schedule could change: they promise a fast keyed hop and stop there. The link now also drops a channel that stops delivering, by authenticated agreement between the two ends, and revisits it later by keyed probes in case it recovers — which is the part a reader evaluating this for a contested band actually wants to know. The deeper-reading entry now says what the design article argues rather than only what it covers: which endpoint decides and why it is the one that has to decode, what standing the other has to object, and what stops an adversary who can make channels look bad from herding the link onto one it then jams.
josephnef
changed the title
The primer's hopping section stopped at the textbook result — spread the energy, a parked interferer clips one dwell in N, the outer FEC eats the erasure. Nothing in that picture is something this project does. Add the second half: the receiver scores what it decodes and proposes dropping the channel it keeps losing, the transmitter commits the new schedule at a named future slot so both ends swap together, and a keyed probe still revisits the excluded channel so the exclusion can be undone. The figure uses the same channel set and the same parked interferer as hop_pattern.gif so the two read as one story, but fills a fixed time axis left to right instead of scrolling — the whole point is the contrast between the two halves, which a moving window can't hold. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The figure draws a stationary occupant, and that is the case worth drawing — most interference is a fixed AP, another link, a microwave, and exclusion against it is a permanent win. But the section read as though the result generalized, and the measurements say it does not. Two of them. Against a blind parked jammer a keyed hop order and a sequential one deliver identically, so secrecy only pays against an adversary that reacts. And against one that follows the exclusions the link does not improve — it settles at its minimum active set and stays there, which is a pass rather than a win, because the exclusion loop is itself an attack surface: anything that can make a channel look bad can otherwise steer the schedule down to one frequency of its choosing. That is what the timid policy is for, and it is the actual advantage — not that the link learns, but that it learns without becoming teachable. Scope the figure's subtitle to the stationary case to match. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The previous commit said in prose that a following interferer changes the outcome, then illustrated only the case where it doesn't. The picture was still a victory lap, so it showed nothing the reader didn't already believe. Draw the other ending, on the four channels and the floor of three the herding run was actually measured with. The link excludes the jammed channel and recovers — the stationary case, in the first half. Then the interferer moves onto a channel that is still active, delivery collapses again, the receiver proposes a second exclusion, and the transmitter refuses it: granting it leaves two channels and the move after that leaves one. The loss counter keeps climbing to the last frame, because it does. Better to bleed on three channels than be herded onto one. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The two adaptive figures were drawn on different bands — eight channels in the first because it inherited them from the plain-hopping figure, four in the second because that is the geometry the herding run was measured on and the only one where the floor of three actually binds. Each choice was locally defensible and the pair was unreadable: you cannot compare two pictures that do not share an axis, and a "floor 3" readout next to eight active channels never does anything. Put both on the eight channels of hop_pattern.gif, with the interferer starting on the ch 52 it occupies there, so the section reads as one band observed three times. The walk-down is the better picture for it anyway. On four channels the floor binds after a single exclusion and there is nothing to see; on eight the staircase is the whole argument — five granted exclusions, each individually correct, the band closing row by row, and then the sixth refused with three channels left. Prose follows the figure, and still names the four-channel bench for the measured numbers. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
josephnef
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two small things, both closing gaps left by #265.
1. An autonomous exclusion had never been undone on hardware
MODE=failsafestopped at the exclusion, so the transmitter's own probe-driven restoration — theprobe_good_run→ProposeRestorepath inHopsetSense.h— had unit tests and no on-air evidence. That is the shape of dead code, and #265 lists "jammer removal: autonomous exclusions recover through keyed probes" as a hardware criterion.It matters beyond box-ticking: an autonomous exclusion that cannot be undone means the link shrinks permanently the first time something transient sits on a channel, and with the receiver mute there is no other evidence that could bring it back.
The mode now runs the same jammer-removal phase the receiver-driven modes already did, and asserts the transmitter aired keyed probes on the channel it excluded, restored it from them, and that the receiver followed back to the full hopset.
Measured with the receiver muted throughout:
2. The README still described a link that only hops
Both places it covers hopping were written before the schedule could change: they promise a fast keyed hop and stop there. The link now also drops a channel that stops delivering, by authenticated agreement between the two ends, and revisits it later by keyed probes — which is the part a reader evaluating this for a contested band actually wants to know. The deeper-reading entry now says what the design article argues rather than only what it covers.
No library code touched: tests and documentation only.
🤖 Generated with Claude Code