UWB vs BLE Channel Sounding: How to Run an Accuracy Bake-Off
Datasheets report best-case medians. Here is a field test protocol that tells you what UWB and Bluetooth 6.0 Channel Sounding will do in your building.
Run a bake-off before you commit. In line-of-sight tests, 802.15.4z UWB with its 499.2 MHz channel ranges to roughly 10 to 30 cm, while Bluetooth 6.0 Channel Sounding, working with about 80 MHz of 2.4 GHz spectrum, lands closer to 30 to 100 cm and degrades faster near metal. Vendor specs report best-case medians. Measure P95 error, non-line-of-sight behavior, and update rate in your own facility for a week, then price the gap against your workflow.
Key takeaways
- UWB's 499.2 MHz bandwidth lets receivers pick the first-arriving path; BLE Channel Sounding's roughly 80 MHz span cannot, which is the whole accuracy gap in one sentence.
- Judge radios on P95 error from your own building, not the median from a vendor's line-of-sight chamber test.
- A fair bake-off uses identical anchor geometry, surveyed ground truth, moving tags, and at least one week of logging.
- Match P95 to the task: bin-level picking needs about 30 cm, zone tracking tolerates 2 to 3 m, and that threshold decides the radio.
- The radio is the cheap decision. Turning noisy ranges into stable positions is where teams lose quarters, and where licensing granted IP shortens the schedule.
Why won't vendor accuracy specs match your warehouse?
Every ranging spec you have read was measured somewhere friendlier than your building. Silicon vendors characterize accuracy in open line-of-sight, often in an anechoic chamber or an empty lab, and they report the median error. Your facility has steel racking, forklifts, concrete columns, and 40 access points sharing the 2.4 GHz band. A radio that shows 30 cm median in the lab can show 2 m at the 95th percentile next to a loaded pallet rack, because reflections off metal arrive almost as strong as the direct path.
The median is also the wrong statistic. An asset tracking system fails on its worst readings, not its typical ones. If 5 percent of position fixes put a pallet in the wrong aisle, workers stop trusting the system within a week, and adoption dies regardless of how good the other 95 percent were.
So the question is not which spec sheet wins. The question is what each radio's error distribution looks like in your building, under your interference, at your tag speeds. Datasheets sell medians. Buy the tail.
What accuracy does each radio deliver, and why?
The gap comes from bandwidth, not cleverness. 802.15.4z UWB uses a 499.2 MHz channel, which means pulses about 2 ns wide. A receiver can separate the direct path from a reflection arriving a few nanoseconds later and time-stamp the first edge. That is why Qorvo and NXP Trimension parts deliver 10 to 30 cm in typical indoor deployments, and why the error stays bounded near metal.
Bluetooth 6.0 Channel Sounding, standardized by the Bluetooth SIG in September 2024, does phase-based ranging across up to 72 channels of the 2.4 GHz band, roughly 80 MHz of usable spectrum, with a round-trip-time mode as a cross-check. Early silicon like Nordic's nRF54 series shows 10 to 50 cm in clean line of sight. Indoors with multipath, expect 0.5 to 1 m, and worse behind racking, because a narrowband phase measurement sums the direct path and its reflections into one number it cannot untangle.
Neither figure is a lie. They are answers to different questions. Bandwidth buys accuracy. Nothing else does.
How do you run a fair bake-off?
Treat it like an experiment, not a demo. The protocol that produces decision-grade data:
- Survey ground truth first. Mark 15 to 25 test points with a laser distance meter or total station, good to 1 to 2 cm. Without surveyed truth you are comparing two unknowns.
- Identical anchor geometry. Mount both systems' anchors at the same positions and heights. Anchor placement moves accuracy more than radio choice does, so hold it constant.
- Static and moving runs. Log each test point for 5 minutes static, then repeat on a cart at walking speed, about 1.5 m/s, and on a forklift at 3 to 4 m/s if that is your use case.
- Deliberate non-line-of-sight cases. Put tags behind loaded racking and inside a metal cage. This is where the two technologies separate.
- Run a full week. Interference from Wi-Fi, other tags, and shift activity varies by hour. A Tuesday afternoon demo misses the Monday morning congestion that breaks systems.
Then plot the error CDF per scenario and read off P50 and P95. One week of data beats any datasheet.
Which numbers actually decide the choice?
Four metrics settle it, in this order.
P95 error against your workflow threshold. Directing a picker to a specific bin needs about 30 cm. Finding a pallet in an aisle needs about 1 m. Knowing which zone a returnable container sits in tolerates 2 to 3 m. If BLE Channel Sounding's P95 in your building clears your threshold, the cheaper radio wins and the UWB premium buys nothing you can use.
Update rate at your tag count. A Channel Sounding procedure sweeps dozens of channels per range, so each measurement takes longer than a UWB two-way exchange. At 50 tags that rarely matters. At 5,000 tags, airtime becomes the ceiling, and your bake-off should include a loaded-network test, not just one tag.
Position stability, not just range accuracy. A system whose output jitters 50 cm between fixes reads as broken even when the mean is right.
Cost per tracked asset, tags plus anchors plus install, amortized over the fleet.
Score all four, but weight the first. Pick the metric your workflow breaks on.
After the bake-off, where does licensed IP fit?
The bake-off picks a radio. It does not give you a product. Raw ranges from either technology are noisy, and the work that turns them into positions an operator trusts, outlier rejection, filtering across anchors, fusing RF with motion data, handling the hand-off when a tag moves from open floor into racking, is where tracking teams spend 12 to 24 months. That math is also where the patent thicket is densest, and it applies whether your ranges come from a 499.2 MHz UWB pulse or a 2.4 GHz phase sweep.
Position Imaging licenses hundreds of granted US patents in radio-frequency ranging, real-time positioning, and computer vision, including US 11,774,249, US 12,079,006, US 12,066,561, and US 12,000,947, a portfolio cited in filings by Apple, Bosch, and other major firms. Licensing proven methods does two things a bake-off cannot: it shortens the ranges-to-positions engineering from years to months, and it gives you freedom to operate on the methods you ship.
License the ranging math. Ship the product.
Frequently asked questions
Is Bluetooth 6.0 Channel Sounding accurate enough for asset tracking?
For zone-level and aisle-level tracking, usually yes. Expect 0.5 to 1 m typical indoors, which covers use cases with a 1 to 3 m threshold. For bin-level picking or anything needing 30 cm at the 95th percentile, especially near metal racking, current Channel Sounding silicon falls short and UWB is the safer choice.
How much more accurate is UWB than BLE channel sounding?
Roughly 2x to 5x in realistic indoor conditions. 802.15.4z UWB typically delivers 10 to 30 cm because its 499.2 MHz bandwidth lets receivers isolate the first-arriving path. Channel Sounding works with about 80 MHz in the 2.4 GHz band, so multipath folds into the phase measurement and errors grow to 0.5 to 1 m, with a longer tail near metal.
How long should an RTLS pilot run before I trust the accuracy numbers?
At least one full week of continuous logging, covering every shift. Interference from Wi-Fi and facility activity varies by hour, and the failures that kill adoption live in the 95th percentile, which a short demo never samples. A one-hour vendor walkthrough tells you almost nothing about tail error.
Can I mix BLE channel sounding and UWB in one deployment?
Yes, and it is often the right answer economically. Use low-cost Channel Sounding tags for zone-level visibility across the whole fleet, and reserve UWB for the zones or assets that need sub-30 cm precision. Your bake-off data tells you exactly where each radio's P95 clears the local requirement.
If I buy ranging chips from Qorvo, NXP, or Nordic, am I covered on patents?
Chip purchases generally cover the silicon vendor's own IP in the part, not the system-level methods you build on top, such as positioning algorithms, multi-anchor fusion, and tracking logic. Those system methods are heavily patented. A freedom-to-operate review, and licensing granted patents where they apply, belongs in your plan before launch, not after a demand letter.
Send us a one-paragraph description of what you track and to what accuracy, and we will map your product to the specific patents in the portfolio that cover it.
Tell us the product. We map the exact scope, what a license covers, and how fast you can ship, all in a 20-minute call.
Book a 20-minute call