Different V2X applications have materially different timing envelopes — SAE J2945/1 sets the BSM broadcast cadence at 10 Hz with sub-100ms end-to-end, but collision-avoidance inference at the RSU must run inside a much tighter window so the downstream vehicle control loop still has headroom. Bundling VRU protection and work-zone alerts under a single latency budget almost always misprices the critical path. The application mix also drives which message sets (J2735 BSM, PSM, SPaT, MAP, TIM) must be decoded at line rate.

SAE J2945/1 fixes the BSM cadence at 10 Hz, which sets a 100ms application budget — but safety-critical collision avoidance needs the RSU inference to complete well inside that window so the vehicle has useful time to brake. 3GPP TS 22.185 and Release 16 NR-V2X define sidelink latency targets in the 3–10ms range for advanced applications. Infrastructure choices made after the SLA is set have an order of magnitude less leverage than choices that set the SLA correctly the first time — and cloud routing is simply not an option.

ISO 26262 is the automotive functional-safety standard and ASIL determines the engineering rigor required — verification depth, redundancy, and failure-rate targets scale sharply from QM through ASIL D. Roadside infrastructure has historically been treated as advisory (QM or ASIL A), but as V2X moves from driver-warning to vehicle-control input under SAE J3016 Level 3+ automation, the ASIL assignment is being reconsidered. Getting this wrong locks you into a design that either over-engineers advisory functions or under-engineers the safety-critical ones.

ISO 21448 (SOTIF) specifically covers the failure class that ISO 26262 does not — perception models that are functioning correctly but produce unsafe outputs because the scenario is outside their training distribution. For V2X this is the dominant risk mode: a VRU detector that works perfectly on the validation set but misses a pedestrian in heavy rain at dusk. SOTIF requires an explicit argument that unknown-unsafe scenarios have been systematically reduced, not just that the model is accurate on the known set.

The automation level of the consuming vehicles changes the engineering rigor required of RSU output, because an L2 driver still has supervision responsibility while an L3+ vehicle may act directly on an RSU alert. Mixed-fleet deployment is the common real-world case and it forces you to design to the highest automation level — you cannot selectively degrade output quality for older vehicles on the same corridor.

UNECE R155 and R156 took effect for new vehicle type approvals in UN-1958 markets in July 2022 and for all vehicles in July 2024, and while they nominally apply to the vehicle, the RSU is increasingly treated as a functional extension when its output is consumed inside the ODD. The practical effect is that OEM customers will require a CSMS and SUMS attestation from the RSU operator as a condition of fleet integration — retrofitting this after a pilot is significantly more expensive than designing to it from the start.

The FCC 5.9 GHz R&O of November 2020 reallocated 45 MHz of the band to unlicensed Wi-Fi and retained 30 MHz (5.895–5.925 GHz) for C-V2X, effectively ending DSRC in the US and making C-V2X the operative US standard. Europe has kept ETSI ITS-G5 for near-term deployments while C-V2X and IEEE 802.11bd coexist as forward options. The spectrum choice determines which message set decoders must be on the hot path and which coexistence constraints (adjacent-channel, Wi-Fi 6E) apply.