Bluetooth 6.0 Channel Sounding vs UWB: What Accuracy Really Ships

What accuracy does Bluetooth 6.0 channel sounding actually deliver?

Channel sounding, ratified in Bluetooth 6.0 in September 2024, measures distance two ways at once: phase-based ranging across up to 72 channels in the 2.4 GHz band, and round-trip time as a check against spoofed responses. The physics problem is bandwidth. Those 72 channels span roughly 80 MHz end to end, while a single 802.15.4z UWB channel is about 500 MHz wide. Range resolution scales with bandwidth, so a UWB receiver can separate a direct path from a shelf reflection arriving 2 nanoseconds later, while channel sounding has to estimate distance through the combined smear.

What that means in numbers:

• Line of sight, open room: vendor demos on Nordic nRF54L15 and NXP KW47 silicon show 10 to 30 cm error. Believe those numbers in those conditions.
• Real facilities: around steel racking, forklifts, and people, expect a median closer to 0.5 to 1 m, with occasional multi-meter outliers when the direct path is blocked and the strongest reflection wins.

Buyers get burned when the lobby demo becomes the spec they promise customers. Plan for the tail, not the demo.

Where does UWB still beat channel sounding outright?

UWB earns its accuracy in the time domain. A 500 MHz channel produces pulses around 2 nanoseconds long, so reflected paths that differ by 60 cm arrive as separate, distinguishable pulses. That is why systems built on Qorvo DW3000 or NXP Trimension parts hold 10 to 30 cm through the same multipath that pushes channel sounding past a meter, and why UWB error grows gradually instead of jumping when line of sight breaks.

Three requirements keep deals in the UWB column:

• Sub-30 cm accuracy. Forklift anti-collision, robot docking, and bin-level pick verification sit below what 80 MHz of spectrum can deliver.
• Dense, fast tag populations. TDoA infrastructure with UWB anchors tracks thousands of tags per zone at multiple updates per second. Channel sounding runs as a connected procedure per link, and nobody has published results at that tag density yet.
• Secure ranging with a track record. The scrambled timestamp sequence in 802.15.4z already protects car access under the CCC Digital Key specification. Channel sounding's protections are newer and less field-tested.

Below 30 cm, UWB is still the answer.

Why channel sounding will still win plenty of 2026 deals

Cost structure, not accuracy, is channel sounding's weapon. The ranging capability lives in the same single-chip 2.4 GHz radio that already runs the BLE beacon: one SoC, one antenna, one coin cell that lasts years at typical asset-tracking update rates. A UWB tag adds a second radio, a second antenna, and a power budget that forces harder duty-cycling. That difference compounds across 20,000 pallets or 5,000 infusion pumps.

Infrastructure follows the same logic. Hospitals and warehouses that spent the last decade installing BLE gateways for RSSI-based zone tracking can move to real ranging by swapping tags and gateways to channel sounding silicon, staying inside one radio ecosystem instead of adding a parallel UWB anchor network with its own cabling and site survey.

So for requirements like "which room holds the pump" or "which aisle holds the pallet," where 1 to 2 m is genuinely sufficient, channel sounding will underbid UWB on price per tracked asset. Cheaper and good enough wins the RFP.

How should you pilot channel sounding against UWB?

Run the head-to-head in your worst aisle, not a conference room. A useful protocol:

• Survey 20 reference points with a laser rangefinder, including positions behind racking, near metal walls, and at floor level where pallets actually sit.
• Log raw per-link range estimates, not just the filtered position output, so you see what the radio delivers before software cleans it up.
• Report median, p95, and p99 error separately. A 40 cm median with a 4 m p99 fails a search-and-find use case even though the average looks fine.
• Add motion. Run forklifts through the test zone; both technologies degrade when large metal reflectors move.
• Mount tags realistically. A tag against a metal surface or a liquid container, IV bags, beverage pallets, behaves nothing like one taped to a tripod.
• Measure battery at the real ranging rate. A 1 Hz channel sounding schedule and a once-per-two-minutes schedule give very different coin-cell lifetimes.

Your worst aisle is the honest benchmark.

Why fusion patents decide who actually wins the deal

RFPs do not ask for a radio; they ask for guarantees. A typical clause reads like a percentile: locate any tagged asset within 2 m, 99% of the time, around the clock. Neither channel sounding nor UWB meets that alone through metal, liquid, and crowds. The systems that win layer techniques on top of the raw ranges:

• Motion models and particle filters constrained by the facility map
• Detection of non-line-of-sight links so bad ranges get down-weighted instead of averaged in
• Fusion with inertial data on the tag, and with camera identification at chokepoints like dock doors
• Handoff logic that keeps identity when an asset moves between radio zones

This processing layer, not the radio protocol, is where granted patents concentrate, because it is where a decade of hard-won methods lives. Two vendors can buy the same NXP silicon; the one with the better fusion stack posts the better p99 and takes the contract. The filter wins the deal, not the radio.

Should you build the fusion layer or license it?

Building it means hiring estimation and RF engineers, collecting months of labeled multipath data from real facilities, and then running freedom-to-operate analysis across a field that has been patented steadily since the early 2010s. Teams that go this route typically spend 12 to 24 months before their p99 numbers stabilize, and they still carry infringement risk on whatever methods they converge on, because good engineers rediscover the same solutions the patents already claim.

Licensing inverts that. Position Imaging licenses hundreds of granted patents across radio-frequency ranging, real-time positioning, computer vision, and machine learning, IP that Apple, Bosch, and other major firms cite in their own filings. Granted patents in the portfolio, including US 11,774,249, US 12,079,006, US 12,066,561, and US 12,000,947, cover positioning and tracking techniques that apply whether your radio is channel sounding, UWB, or both. You get proven methods plus freedom to operate in one agreement, and your team spends its time on product instead of re-deriving filters.

License the hard part, ship the product.