Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
56 changes: 46 additions & 10 deletions docs/cell-m6.md
Original file line number Diff line number Diff line change
Expand Up @@ -1634,13 +1634,19 @@ measure and the CU could hand over to. Two checks make a re-aired copy
useless unless it is relayed live and on time (`src/cell/MeasGap.h`,
`examples/cell/main.cpp`).

- **A signed window.** The NS IE (version 2) carries `win`, the network
superframe / `kNlWindowSf` (64, ~6.5 s at 100 TU), inside the signature
(`'S' || NL IE || win`). The DU re-signs at every window change, and signs
only once it knows the network superframe. The UE takes its serving
cell's window as "now" and refuses a serving window that goes back. It
takes a neighbour's beacon only if the neighbour's window is within one of
the serving window, i.e. only if it was signed in the last ~13 s.
- **A signed window.** The NS IE (version 3) carries `win`, the network
superframe / `win_sf`, AND `win_sf` itself, both inside the signature
(`'S' || NL IE || win || win_sf`). `win_sf` is the DU's
`CELL_NL_WINDOW_SF` (default `kNlWindowSf`, 64, ~6.5 s at 100 TU; down to
1 superframe, ~102 ms). Signing the length too stops a replayer relabelling
a 64-superframe signature as a 1-superframe one or back. The DU re-signs
at every window change, and signs only once it knows the network
superframe. The UE takes its serving cell's window as "now" and refuses a
serving window that goes back, or one of a different length. It takes a
neighbour's beacon only if the neighbour's window is within one of the
serving window, i.e. only if it was signed in the last ~2·`win_sf`
superframes. The smaller `win_sf` is, the shorter a re-aired copy stays
fresh — at the cost of an Ed25519 signature every `win_sf` superframes.
- **The phase.** All DUs share the PTP timebase, and the serving list names
each neighbour's TBTT offset on the serving grid. The UE keeps the phase
where its serving beacons land (on its fitted DU clock) and refuses a
Expand Down Expand Up @@ -1670,9 +1676,39 @@ Refusals are logged per reason (`cell.ue.nbr_refused`, counted in
`cell.ue.done`). `CELL_UE_NBR_FRESH=0` turns the checks off, which the rig
uses for the counterfactual.

**What remains** is a real-time repeater placed on the right phase: it
relays the live beacon within the tolerance, so it is the real cell, only
louder. Only distance bounding would stop that.
**What remains, and how far.** Two replay shapes survive the signature:
- **A store-and-forward relay** — a device that decodes a genuine beacon and
re-airs it. It is beaten by the phase check: it cannot re-air a beacon
before it has received it, so its copy lands about a beacon's airtime
after the cell's own TBTT, well outside the tolerance. Measured below.
- **An on-phase copy** — a cut-through repeater or reflector that re-radiates
the beacon with almost no delay, so it lands within the tolerance, the
real cell only louder. The phase check cannot tell it from the cell. What
bounds it is the freshness window: the UE accepts a neighbour signature
within one window of the serving one, so a copy more than ~2·`win_sf`
superframes old is refused (stale), and `CELL_NL_WINDOW_SF=1` caps its
useful life at ~205 ms rather than ~13 s. A copy fresher than that, on
phase, is
physically the cell's own signal re-radiated; only distance bounding (or a
per-beacon nonce the UE challenges, not on this one-way broadcast path)
would stop it.

**Rig result, the live relay** (testrig `scripts/m6_nlrelay.sh`): 2 runs,
all pass. Per-cell keys and signed lists, both cells on one channel, DU2
alive throughout. The witness runs the replayer in relay mode: it re-aired
every genuine cell-2 beacon it heard (1267–1310 in a run), carrying a
current signature.
- **Refused on phase, not freshness.** The relay's copies landed a median
636 µs off the cell's TBTT and were refused on phase (895–896 per run);
**0** were refused as stale — their signatures were fresh. A
store-and-forward relay cannot be on time.
- **The genuine cell is undisturbed.** UE1 measured genuine cell 2 in 100 %
of its reports throughout.
- **The phase check's margin.** Genuine neighbour beacons landed within
6–7 µs (p99) and 8 µs (max) of their predicted phase — against the 300 µs
tolerance, a ~40× margin. The tolerance could be tightened a long way
before genuine beacons were at risk, which narrows the window a near-zero
latency repeater has.

**Costs.** A neighbour is measured only once the serving phase is known. On
the rig that is about the first 16 neighbour beacons after the UE starts or
Expand Down
Loading
Loading