Wearable performance power & interconnect harness

System design overview — a body-worn power distribution and camera/IMU signal harness built for a live performance-arts installation

Payload: 1× NVIDIA Jetson Orin NX, 2× FRAMOS GMSL3 camera links, 1× Teradek Ranger wireless video transmitter, 1× LP-Research CAN IMU. Powered by hot-swappable Li-ion packs.

Purpose

A performer wears a compute-and-camera rig through a live show and needs the batteries changed mid-performance without the system ever losing power. That single requirement — never dark, even for the second it takes to swap a battery — turned out to be the constraint that shaped almost every other decision in this system: which battery chemistry, how many cells in parallel, which power-management board, how the fuses are sized, and even which connector families were usable at all. This page documents the reasoning trail, not just the final bill of materials — the goal is to show how a body-worn power system earns its reliability margin at every layer, from cell chemistry up through the physical connector shell.

System overview

Power flows through one hot-swappable path; telemetry and IMU data run on two independent, lower-current buses

Power architecture: why hot-swap drives the cell topology

The requirement that the system never lose power during a battery change rules out a single-battery design outright — there is no way to physically remove and replace one battery without a gap unless something else is carrying the load in the meantime. The solution is a dual (or triple) battery bay with a battery management board that combines them electrically, so that pulling one pack still leaves the other(s) carrying full system current with zero interruption.

That single requirement cascades into a chain of smaller decisions:

Nominal voltage, 10.8V (3S Li-ion): the camera SerDes deserializer's confirmed operating window is 9.6–13.3V. A 3S pack's native range (~8.25–12.6V) sits almost entirely inside that window, which means the system can run with no active buck/boost converter in the main power path — just wiring and fusing. A 4S (14.4V) pack would overshoot the 13.3V ceiling at full charge and require a converter anyway, reintroducing conversion heat on a device worn against skin.
Cell configuration, more parallel strings over fewer: for a fixed system current draw, more parallel strings (higher "P" in an "S×P" pack spec) means a lower discharge rate per individual cell — less heat, better cycle life, and — critically for hot-swap — more margin during the moment a single pack has to carry the entire system alone.
Battery-only power path: the chosen power-management board's output is sourced exclusively from the batteries, never from its charging input — so pre-show charging on mains and live performance on batteries can never accidentally cross paths and put an unregulated charging voltage onto the load bus.

Margin philosophy, applied consistently Every current-carrying decision in this system was sized the same way: identify the real worst case, then size the component so normal operation sits comfortably below its rated limit — not at the edge of it. That discipline shows up at every scale in this build, from cell-level C-rate down to a single fuse value.

Fusing and distribution

The power-management board's output is unfused by design — protection is added downstream, once the single supply rail splits into individual branches. Each branch gets its own fuse, sized independently:

Branch	Real / worst-case draw	Fuse	Why
Jetson Orin NX	~2.2–3A (up to max power mode)	5A	~65–100% headroom above worst case — a fuse sized right at the operating edge nuisance-trips during normal use
Teradek Ranger TX	~2A nominal	5A	Same margin logic; wireless transmitters can burst above their quoted average
GMSL camera A	~0.2A real, 1A budgeted	2A	Fuse respects the deliberate budget margin rather than the tighter real-world number
GMSL camera B	~0.2A real, 1A budgeted	2A	Same

The shared ground return gets a wire gauge sized differently from any single branch: because it carries the combined return current of every branch simultaneously, it needs to be rated for the sum of all branches, not any one branch alone — a detail that's easy to miss if "ground" is treated as an afterthought rather than a conductor doing real work.

Protective earth vs. functional ground A 9-position connector considered for this harness offered a PE (protective earth) pin alongside the power pins. In a mains-referenced system, PE bonds an exposed enclosure to true earth so a fault has somewhere harmless to go. This system is a fully isolated, floating battery supply with no mains reference anywhere — so PE has nothing to bond to, and was deliberately left unterminated rather than tied to the functional ground return, keeping the two paths permanently separate.

Connector engineering: where the real depth lives

This is the part of a wearable hardware build that rarely gets documented, and it's where most of the actual engineering time went. A body-worn harness has to route power and several distinct signal types (I²C/SMBus telemetry, CAN bus, MIPI camera data, USB) through a small number of sealed panel connectors — and getting a connector "roughly right" is not the same as getting it right. Several real mismatches were caught in this build before they reached hardware:

What looked right at first	What was actually wrong	How it was caught
A connector's "female"/"male" label	Shell gender and contact gender are independent axes — a "female connector" can carry male pins	Cross-checked the full spec string, not just the summary label
Two similar-looking connector families, same position count	One was a genuine wire-splice product; a near-identical part number was actually a cable-shield-to-ground terminator — wrong physical function entirely	Verified the manufacturer's own product-family name, not the distributor's generic category tag
A small industrial connector rated for the job	Its per-pin current rating (common on compact sensor connectors) sat below the system's real worst-case draw	Checked pin-level current rating against the fused branch rating, not just overall connector rating
A crimp tool recommended for a similar-size connector pin	Tool frame compatibility and contact-geometry compatibility are separate specs — a correct frame with the wrong die still produces an unreliable crimp	Verified against the connector manufacturer's own tool cross-reference before ordering

None of these were exotic failures — they're the ordinary, easy-to-miss traps in any harness build that mixes several connector families. The discipline that avoided them was simple and repeated dozens of times over the build: read the actual manufacturer datasheet for both halves of a mating pair before ordering either one, rather than trusting a part number's resemblance to something already confirmed.

Signal integrity: choosing the right bus for the run length

The harness carries two distinct low-current signal types alongside the power distribution, and each was matched to a bus protocol suited to its actual physical run:

Two signal buses in one harness, each selected for the length it actually has to cover

Keeping the IMU on CAN rather than a more common short-range bus turned out to be validating in hindsight: CAN's differential signaling is specifically designed to reject noise over real cable distances, which matters in a harness that routes this signal alongside battery power lines and camera data. The two things that actually determine whether 5 meters of CAN works reliably have nothing to do with distance capability itself — they're implementation details:

Termination: exactly two 120Ω resistors, one at each physical end of the bus — missing or duplicate termination is the most common real-world cause of CAN bus problems, far more than raw cable length.
Twisted pair for the differential signal: a generic 4-conductor cordset with parallel, untwisted conductors gives up most of CAN's noise immunity even though the protocol itself would tolerate the distance — the cable construction has to match the protocol's assumptions, not just the connector.

Tooling: the crimp is only as good as the die that made it

A connector's electrical spec is only half the story — the crimp joining wire to contact has its own tolerance stack, and using the wrong tool on the right connector produces a joint that looks fine and fails later. This build used six distinct small-pitch connector families (JST VH, Molex Micro-Lock Plus, Molex KK254, Molex Mini-Fit Jr, Molex SL, and an aerospace-grade Amphenol M23 series), each requiring its own verified crimp tool or die:

Connector family	Manufacturer's own tool	Approach taken
JST VH (battery interconnects)	~$900+ dedicated tool	Verified generic multi-pitch ratchet crimper against the manufacturer's own AWG range
Amphenol M23 aerospace-grade (enclosure connectors)	$2,300+ full production tool	Sourced the same tool frame generically (it's a rebadged MIL-spec Daniels tool), then confirmed the correct positioner against the connector maker's own contact part numbers
Molex Mini-Fit Jr (power distribution)	~$660 premium-grade tool	Found the manufacturer's own lower "service grade" tool at a fraction of the price — same brand, same guarantee, correctly scoped for prototype volume
JST VH, smaller pin variant	~$500–1,200 dedicated tool	Found a real, tested community report confirming a specific affordable interchangeable-die tool for this exact contact — and equally, confirmed via the same source that it does not work on the connector's larger sibling contact

The recurring lesson: a "compatible tool" claim is only as good as whether it names the exact contact part number, not just the connector family or pitch. Two contacts can share a pitch and still need different dies — this build hit that boundary more than once, and every time, checking before ordering was cheaper than finding out after a bad crimp failed in the field.

Design principles, in summary

Let the hardest constraint set the architecture. Uninterrupted hot-swap wasn't a nice-to-have bolted on later — it determined the battery topology, the power-management board's feature set, and the fusing plan from the start.
Size for the real worst case, with margin, not the average case. Every fuse, wire gauge, and connector current rating in this system was checked against the system's actual worst-case draw, not its typical draw — and re-checked when a "typical" number turned out to have more headroom than first assumed.
Match the signal path to the physical run, not the other way around. A 5-meter cable run picked its own bus protocol; the protocol wasn't chosen first and the distance squeezed to fit it.
Verify both halves of every mating pair against the manufacturer's own documentation before ordering either one. Nearly every real mistake avoided in this build came from this single habit, applied without exception.