The Trembler Line CS-435 is a chain of buried CS-430 nodes on a fibre trunk: detect, localize, cue. Nothing in the chain transmits outward and the fibre carries everything. This page is the system, and what one node becomes when it is one of fifty.
Every other program in the Underground set is a machine. CS-435 is four segments and the discipline that joins them, and the product is what the chain says together rather than what any one node hears.
CS-430 passive seismic and acoustic nodes, emplaced at survey-set intervals and left still. They are the only hardware in the system that senses.
Node-to-node fibre carrying three things at once: data up the line, a shared clock for the localization solve, and a candidate power path. No radio anywhere in the system.
Array processing, classification, the arrival-time solver and the operator console, at one monitored end of the trunk. This is where node triggers become a track.
An interpose point, a heading and a confidence, offered to a person at the console. Detect, localize and cue is the whole of the system's output.
The architecture puts the cost where it can be recovered. Compute lives at the head-end rather than in the ground, so the nodes stay cheap, simple and low-power, and the station that reads them is upgraded without digging anything up. Between the two sits the interface that decides how much of this is a product: fibre in and out, timing, data, power and bypass on a failed node. Define it once and every node revision, string length and layout becomes a configuration.
Nothing in CS-435 is a new sensor or a new machine. CS-430 supplies the buried ear and CS-425 supplies the legs that move on a confirmed cue. The system's own engineering is the fibre network, the localization logic that ties the two together, and the rule that keeps a person deciding.
Every byte the system moves rides the buried fibre trunk, node to node, to one monitored head-end. There is no spectrum in the loop to jam, because there is no spectrum in the loop at all.
No radio and no GPS appears anywhere in the network, and under ground that costs nothing to hold. A wired line also solves what a radio network would struggle with: the localization solve needs a shared clock at the millisecond level across the whole picket, and disciplined time distribution over a wire is ordinary engineering. The one emitter in the system is the calibration thumper, a shop-built known-source tool used to set the velocity model, and it is maintenance equipment rather than part of the watch.
A cut is loud. Link death is its own alarm, so tamper detection is intrinsic to the trunk rather than a sensor added to it. Survivability past a cut is a layout decision taken in front of the customer: a dual-headed line survives one cut with full coverage and pays for it with a second head-end, and a node that passes traffic through on failure degrades the picket instead of blinding it.
Human safety comes first, on every line. The most exposed minute of a standing post is the walk to relieve it, and the line stands that watch instead. Detect, localize and cue is the whole of the output: there is no warhead, fuze or energetic material anywhere in the system, and no strike authority anywhere in the loop.
Digging, tunnelling, footsteps and vehicles are genuinely different signals, and the array hears each of them differently. The quietest one sets the whole design.
| Target class | Signal character | Range class, per node |
|---|---|---|
| Footsteps and personnel | Impulsive, low-energy ground-contact transients at ~1–100 Hz; short range, strongly ground-dependent | ~5–50 m depending on ground and noise |
| Hand digging | Repetitive impulsive strikes, tool on soil, with a distinctive rhythm and more energy than footsteps | Tens of metres |
| Machine tunnelling | Continuous machinery tones over broadband ground working; the most energetic and persistent source | Order 100 m+ in favourable ground |
| Vehicles, on the surface | Continuous broadband with engine harmonics coupled through wheels or tracks; high energy | Order 100–500 m |
Ranges are class figures. Seismic amplitude falls off with spreading and with absorption, and absorption climbs with frequency, so every radius halves or doubles with site conditions.
Two of the four are easy to classify. Hand digging is periodic, and little else in the ground repeats on that rhythm. Machine tunnelling is the longest-range class and the most persistent, so the array can integrate against it for as long as it takes, which makes it the case the picket is soundest for. Footsteps are the hard one: low energy, short range, and impulsive enough that there is little to integrate. They also set the node count, and therefore the price.
Single-node detection needs a spacing of s ≤ 2R. Localization needs the same event on three or more nodes, which is roughly s ≤ R. Both are set by the weakest source class the line must catch.
| Sizing target class | Per-node radius | Detect-only | Localization-grade |
|---|---|---|---|
| Footsteps, the expensive fence | ~20 m | 40 m · ~26 nodes | 20 m · ~51 nodes |
| Hand digging | ~40 m | 80 m · ~13 nodes | 40 m · ~26 nodes |
| Machine tunnelling, the cheap fence | ~100 m | 200 m · ~6 nodes | 100 m · ~11 nodes |
| Vehicles, on the surface | ~150 m | 300 m · ~4 nodes | 150 m · ~8 nodes |
A straight picket a kilometre long, at a literature-class per-node detection radius, strongly ground- and noise-dependent. Vehicles and tunnelling are caught free at footstep spacing.
Node count per kilometre varies by roughly an order of magnitude across that table, from about six to about fifty-one, depending on which target class must be caught and whether localization-grade coincidence is required. The spread is what makes the mission statement the cost model. A line specified for tunnelling alone is several times sparser than the same kilometre specified to catch a person walking.
The expensive end is only credible if the node is cheap. A footstep-grade kilometre at ~51 nodes is a purchase decision about a moulded part, bought-in sensing and a node that stays simple. The compute sits at the head-end for the same reason: every dollar in this system is a node-count dollar.
A linear array localizes well along the line and poorly across it. Where cross-line accuracy matters, the answer is geometry rather than a better sensor.
The cheapest layout and the weakest off-axis: the arrival-time curves run nearly parallel for sources away from the line, so the solve says where along the fence and only coarsely how far out.
Cross-line observability bought outright, and paid for in nodes and in bore length. Where the cue has to name an interpose point ahead of a heading, this is what makes the point worth naming.
One bore path, partial cross-line gain. Short perpendicular stubs do the same work where the approach is already known.
Stringing nodes also buys signal. Coherently summing N nodes against incoherent noise raises signal-to-noise by up to 10·log₁₀N dB, which is 5 nodes ≈ 7 dB, 10 nodes ≈ 10 dB and 20 nodes ≈ 13 dB, and detection range against an absorbing medium responds steeply to signal-to-noise. Real gain is capped by how far a signal stays coherent between nodes tens of metres apart in heterogeneous soil, so the ceiling favours persistent sources that permit long integration.
Both effects point one way. The design centre of gravity of this system is the array, not the node: a mediocre sensor in a well-coupled, well-timed, well-laid array beats an excellent sensor strung badly. Geometry buys localization, density sets the detection floor, and array gain recovers what the soil takes.
Emplacement is where the system stops being software. Node coupling and node coordinates are first-order performance items, and both are bought at install or not at all.
Soil class, water table and a noise survey of traffic, machinery and livestock. This sets the detection radius the spacing table is read at, and the false-alarm floor.
Spacing from the quietest target class that must be caught, offset or stagger from what the localization requirement asks. The layout is settled here, before a node is placed.
A CS-410 lays conduit and fibre, and the nodes are emplaced at depth along the bore. Shallow hand or auger burial is the fallback.
Each node set and verified against a standard test source. Ground coupling is a first-order signal consequence, and a badly coupled node is re-set rather than compensated in software.
Known surface sources at surveyed points calibrate wave speed and per-node sensitivity. The in-situ velocity model the localization solve depends on is built here, and maintained.
Two emplacement modes are genuinely different products. Laid by bore, the conduit is the fibre path and can carry an optional tether tap for power and backhaul, the nodes sit at depth, and the surface is undisturbed. Buried shallow, at tens of centimetres with trenched or surface-run fibre, the line needs no other platform, and the trench is itself a signature. Deep hears tunnelling; shallow hears footsteps.
Both modes stand on the same in-house drilling discipline, and as-built node positions are surveyed to sub-metre class during install because every localization solve starts from the coordinates it is given. The bore is CS-410, and conduit emplacement is a mission it already carries.
Arrival-time differences across three or more nodes solve the source position. Two terms dominate the error, and the wire is not one of them.
| Error source | Value | Position contribution | Read |
|---|---|---|---|
| Time sync over fibre | σ_t ~1 ms | ~0.2–0.5 m | Negligible. A wired line makes timing a non-problem |
| Waveform onset picking | ~5–20 ms in soil | ~1–10 m | Real but manageable, and improves with signal-to-noise |
| Soil velocity, uncalibrated | δv/v ~50%+, v spanning ~150–1,500 m/s across soils | ~10–50 m at 20–100 m | Dominant. It takes the cueing mission with it |
| Soil velocity, thump-calibrated in situ | δv/v ~10% class | ~2–10 m at 20–100 m | The achievable class |
| Node as-built position | Surveyed to sub-metre class | ~1 m | Bought with installation discipline |
The budget for a calibrated line. Localization here is geology-limited rather than electronics-limited.
Three findings come off that table. In-situ calibration is a hard system requirement, not a maintenance task, because the third row cannot name an interpose point and the fourth can. A calibrated line's accuracy class is metres to ~10 m along-line. And cross-line accuracy is bought with geometry rather than with better nodes, which puts the layout section ahead of this one.
The fourth finding reframes the requirement. For cueing, heading beats position: a persistent source solved again and again over hours gives a direction of advance, and a tunnelling face tracked that way is exactly the class the array is best at. The interpose point comes from the heading, not from one instantaneous fix.
A picket that alarms on rain and frost is a fence nobody watches. The chain is therefore built around agreement between nodes rather than sensitivity at any one of them.
A low-power detector runs in the ground and sends triggers with buffered waveform windows up the fibre. The line does not stream raw waveforms, and that choice sets the power budget.
Triggers are correlated across nodes, classified against a library, and solved for position on confirmed events. A single-node trigger is a health event, not an alarm.
A confirmed, classified, localized event becomes a track with its confidence and its history, and a cue proposal beside it. A person takes it or leaves it.
Multi-node coincidence is the first and best false-alarm filter, because a real source moves plausibly and shows up on its neighbours while a sensor fault and a patch of local noise do not. The classification library carries reject classes alongside target ones, and the reject list decides whether anyone trusts the fence: rain, thunder, traffic, trains, animals and frost action. That list is won with data from the site itself.
A fixed line in fixed ground is well placed to collect it. The head-end learns this site's diurnal and seasonal rhythm node by node, so anomalous is measured against this line's history instead of a generic threshold; every labelled event grows the classification basis; and a per-node ledger of trigger-rate drift, sensitivity against calibration thumps and battery draw-down means nodes are revisited on data rather than on failure.
Dwell is a power argument rather than a battery argument. Triggers and short buffered windows go up the fibre; raw waveforms never do.
A duty-cycled node, sleepy acquisition with a local trigger and a quiet fibre interface, sits in the ~1–10 mW class. On a ~50 Wh lithium primary pack that is 0.6–5.7 years in the ground, and the span between those two figures is the whole of the architecture: the low end of the draw range is a node left alone for years, and a node that streams collapses it outright. Cold soil cuts it further. Where the string rides a bored conduit, a tether tap removes the constraint and the node runs off the line.
The same logic sets what the node is made of. Housings are moulded polymer, glass-filled nylon where the part is structural and recycled PET where it is not, cut on the fleet's own tooling; the sensing element is a bought-in geophone or accelerometer class part rather than a made one. That is what makes a fifty-node kilometre a purchase rather than a project. The node's own page is CS-430.
Node to head-end transport is microseconds. The timeline that matters is confirmation and the human, and against the target set that arithmetic writes the doctrine rather than following it.
The cue timeline is generous beyond argument, and the source is persistent enough to be tracked over hours into a heading. This is the interpose mission, and the class the array is best at.
Detection and tracking are comfortable; a physical response inside that window is only credible with a responder already staged nearby. Alarm and track is the product for this class.
Alarm and track. The line reports it, times it and places it, and nothing in the loop pretends a physical response can be interposed at that speed.
That split is doctrine: interpose for slow subterranean threats, alert and track for fast surface ones. It falls out of the arithmetic rather than out of a preference.
What crosses the response boundary is an interpose point, a heading estimate and a confidence. A person at the console takes it or leaves it, and a confirmed cue tasks a CS-425 already staged near the line to move ahead of the intrusion path. It blocks the way and annunciates. It does not strike, nothing upstream of it carries the authority to order one, and no autonomous physical response path exists in the system at all.
The line suits a buyer with ground to hold and a post they would rather not stand. It suits nobody looking for a weapon.
Bring the line you have to watch, the quietest thing you have to catch on it and what you want to happen when it is caught. The node count, the layout and the calibration follow from those three.
The way in is a screened enquiry, not a quotation. The first questions are a buyer's: which target class sets the fence, whether a responder is staged or a person is alerted, and whether the surface may be disturbed at install. Those answers move the node count across an order of magnitude, and the rest is layout and ground truth. Enquiries are screened, counsel comes first on any international transfer, and permits are taken per shipment. Every design in the family is patent pending.
The Underground family is the whole set side by side. CS-430 is the node this line is made of, CS-425 is what a confirmed cue moves, and CS-410 lays the conduit the string can ride. Who it is for is the buyer's view.
All Underground family designs are patent pending; international transfer is export-controlled and subject to Canadian government permits taken per shipment, and nothing on this page is an offer.
