Order 0: measurement procedure¶
The boards are measured with a LibreVNA (100 kHz-6 GHz), so every Order 0 band lies inside 1-6 GHz. The LibreVNA measures no switch terms and its own TRL has no switch-term correction, so the calibration runs offline in two tiers: a coaxial SOLT at the cable ends, then multiline TRL on the boards’ own coupons, as two branches (two-tier on SOLT-corrected data, and raw data with switch terms estimated from the asymmetric coupon A14) whose agreement is checked.
Before ordering (owner, about 30 minutes)¶
Record
DEV:INF:HWREV?andDEV:INF:FWREV?and update to firmware 1.6.4 or later (1.6.3 has a flash-timing bug; 1.6.1 and older invert some points).Measure the isolation floor, 6 MHz-6 GHz at 1 kHz IF bandwidth, both cables terminated: without and with an isolation term, then again after moving the cables.
Ten sweeps of a through adapter and five disconnect/reconnect cycles: the instrument’s own repeatability.
Optionally, the same with 3 dB attenuators at the VNA ports; keep whichever gives the better residual on the verification line later.
Measure the thickness of any OSH Park 4-layer board at hand before buying the connectors. The published layers add to 1.511 mm against the nominal 1.6 mm, and Cinch’s slot is 0.068 in (1.73 mm): a board at the top of a ±10 % tolerance (1.76 mm) would not fit.
Equipment¶
The LibreVNA powered from its 5 V jack or a powered hub; two phase-stable SMA cables, taped down. Move the board, not the cables.
Tier 1: a LibreCAL if available, else SMA open/short/load standards and an F-F thru adapter.
Tier 2: the on-board multiline TRL kits, A01-A07 (region M) and B01-B05 (region W); O0-D carries its own thru and two lines (A01, A20, A04) for its lot’s εeff, loss and relative Z0.
Two SMA-male 50 Ω loads for the third port of the 3-port demos, each measured as a 1-port after tier 1 and referred to the demo’s reference plane through that port’s error box.
An SMA torque wrench at one fixed setting (8 in-lb is inside Cinch’s 7-10 in-lb), a 4-wire ohmmeter or a current source and millivoltmeter, a thermometer and a hygrometer.
Assembly¶
One operator, one alloy, a 3D-printed jig that holds each connector flush against the milled edge and presses the board up against the top legs (the 1.73 mm slot is 0.19 mm wider than the board), flux removed with isopropyl alcohol (residue changes εeff on mask-open lines). The tabs sit only on edges without a connector; file their nubs flush anyway. Photograph every joint.
The solder mask covers the 0.155-0.21 mm gaps between each pin pad and its coplanar ground (the launch was designed with mask there). OSH Park’s minimum mask web is 4 mil (0.10 mm), so the slivers are buildable; a registration shift (3 mil) opens one gap as it closes the other, which changes the pad’s impedance only to second order.
On arrival¶
Photograph the boards and note the core colour: FR408HR is milky white, EM528 red.
Measure the board thickness at three points with calipers (this checks the core, not the 0.2 mm prepreg).
Break out the sticks of copies 1 and 2; keep copy 3 as a spare and for the microsection.
Solder the connectors (O0-M: 33 per copy; O0-W: 16; O0-D: 12), then condition for 24 h at lab ambient. The back silkscreen of every stick names its upload (
O0-D A01): A01 and R1 exist on two uploads, which are two lots.
Set-up and calibration¶
30 min warm-up. Sweep 6 MHz-6 GHz in 1000 points (a uniform grid starting at its own step, as IEEE 370 needs), IF bandwidth 1 kHz, 4 averages, the highest stimulus level. Check for compression once (the thru at 0 dBm and −10 dBm). Never change the span after calibrating.
Tier-1 SOLT at the cable ends. Save the calibration file, then re-measure a load and a short as devices. Leave the isolation term out of the published data.
The floor: both cables terminated, the same settings, saved with the data; also with both cables open.
Sequence per copy¶
The thru, then the lines in increasing length, the reflect and the verification line.
The width set, A10, the ring, the stub, the C-pads (tier 1; their analysis is in the pre-registration).
A14 in both orientations (swap the cables).
R1 (and on O0-D, D1): each pair of ports (1-2, 1-3, 2-3) with a load on the third.
The thru again (drift check).
Three independent connections of A01, A03, A04 and each demo.
Re-measure the tier-1 load; if it moved by more than 0.1 dB or 1°, repeat the copy.
O0-W: the same with B01-B05, R1t and D2.
Then the DC meanders of A15 (4-wire, 100 mA, with the board temperature: copper changes by 0.39 % per kelvin), and optionally a microsection of A15 from copy 3 at the cut mark.
Files¶
A session follows the coupon session layout (examples/rf-coupons/measurements/), extended for Order 0 with the raw (uncorrected) sweeps, the loads and the floor:
<session>/
session.yaml # stackup and received substrate, board serials, generator commit, VNA,
# firmware, hardware revision, calibration file sha256, grid, IF bandwidth,
# power, temperature and humidity at start and end, cables, torque, notes
cal/tier1.cal # the LibreVNA calibration file (with its raw standards)
corr/O-1_A03-M-L13.0_r1.s2p # <upload>-<copy>_<stick>-<family>-<what>_r<connection>
raw/O-1_A03-M-L13.0_r1.s2p # uncorrected, from the raw data stream
corr/O-1_D1-M-P12_r1.s2p # 3-port demos: one file per port pair
loads/L1.s1p, loads/L2.s1p
floor/iso_noterm.s2p, floor/iso_term.s2p
dc.csv
The capture script (SCPI on the headless LibreVNA GUI, the raw stream on TCP 19000), the switch-term step, the two-branch comparison and the demo de-embedding are still to be written (Order 0 design, WP7); they are needed when the boards arrive, not for the upload.