🚆 iSTRAIN Lassiter Creek · — · railroad VHF 160–162 MHz

Signal Flow how it fits together — the air ▸ the dongles ▸ meaning

iSTRAIN — SIGNAL FLOW        the air ▸ the dongles ▸ the answer                    as of 2026-08-30
mission: is a train blocking the NS Amity Rd crossing (Galloway, OH)?
legend:  [serial] = dongle   ★ = TCXO (precise clock)   ─▶ data path

THE ANSWER COMES FROM TELEMETRY. Voice confirms it; voice never decides it.
────────────────────────────────────────────────────────────────────

1 ▸ TELEMETRY  —  UHF, the train's own devices talking to each other (1200-baud FFSK)
    dongle [1001] hops between the two ends of every train, retuning in place:
      457.9375  EOT / tail    dwell 14 s ── DECODED live: unit ID · brake-pipe psi · moving/stopped ─▶ eot.jsonl
      452.9375  BOT / head    dwell  7 s ── head-end frame recovered + unit address decoded      ─▶ bot.jsonl
      ±12.5 kHz MID / DPU     heard inside both dwells: mid-train locomotives                    ─▶ midtrain.jsonl
    a head frame that addresses a card's own EOT = one confirmed complete train (head↔tail join)
    also on 457.925: a fixed wayside EOT BOOSTER — a repeater, not a train; its floor is subtracted everywhere
    antenna: ¼-wave whip at the band                    (all UHF is presence + decode; no audio)

2 ▸ VOICE  —  VHF rail radio, 160–162 MHz, unencrypted narrow FM
    dongle [1002 ★] demodulates five channels at once (160.86–161.19 MHz); squelch-gated clips
      160.980  NS Dayton District road — the dispatcher and the defect detectors
               (CJ at MP 144.7 / 157.8 talks to every passing train: "no defects, 45 axles")
      160.920 · 161.190 · 160.860 · 161.085 — the neighbouring NS / CSX channels
    clips ─▶ captures/*.mp3 ─▶ heard.jsonl (when each clip was heard, on which channel)
    clips > 5 KB ─▶ ffmpeg comms filter ─▶ faster-whisper (small.en + VAD) ─▶ transcripts.jsonl (the words)
    antenna: tuned railroad-band vertical, indoors    (a rooftop mast is the next range lever)

3 ▸ THE BANNER  —  /api/istrain, the one line most visitors read
    decoded EOT frames become one card per train. The AIR CURVE is the state:
      psi ~85–90, moving                        PASSING through
      psi held ~55–65, stopped                  DWELLING — brakes set; may be on the crossing
      psi bled to 0 and held, head still polling CUT & STANDING — crews split the train; crossing likely open
      psi recharging, motion 0→1                DEPARTED — the roll-out is the event time
    a stopped train's PLACE comes from geometry, not loudness: each end's signal CRESTS as it passes
    the antenna (≈ the crossing). Head past its crest + tail still at its crest + brakes held + strong
    signal = a train SPANNING the crossing → YES. Both ends at crest, air held → YES. Otherwise MAYBE.
    voice/speech only ever corroborate (a burst of clips while a train is carded upgrades the detail).

4 ▸ CORRELATE  —  for the reader who wants the evidence behind the line
    one row per band, 15-minute bins, one week: read DOWN a column. BOT + MID + EOT lit together in one
    bin is a train even when nothing decoded; voice and speech lit in the SAME bin confirm it.
    Voice alone is noise. The BOT / MID / EOT strips beneath are the same data at 1-minute bins, last 6 h.

not used: ATCS 900 MHz (retired to encrypted PTC in 2025; our "carriers" were a dongle spur) · AIS (inland)
the record: every finding is numbered in OBSERVATIONS.md (O-1 … O-35) · protocol notes EOT-HOT-PROTOCOL.md ·
channel plan CHANNELS.md · the data DATA.md
────────────────────────────────────────────────────────────────────