Bluetooth Channel Sounding vs UWB: Run a 2-Week Accuracy Bakeoff

What accuracy does each radio actually deliver in 2026?

Bluetooth 6.0, published in September 2024, added channel sounding: the tag and anchor exchange phase measurements across dozens of 2.4 GHz channels plus a round-trip-time check, and the chipset fits a distance to the phase slope. Chip vendors quote sub-meter results, typically 0.3 to 0.5 m, and those numbers are real in open line of sight. UWB under IEEE 802.15.4z uses 2 ns pulses in a 499.2 MHz channel and times the first arriving pulse directly, which is why Qorvo and NXP silicon holds 10 to 30 cm in the same conditions.

The gap widens indoors. Around steel racking and moving forklifts, channel sounding deployments commonly degrade to 1 to 3 m, while well-anchored UWB usually stays under 0.5 m. Neither number appears on a datasheet.

• Open aisle, line of sight: CS 0.3 to 0.5 m, UWB 0.1 to 0.3 m
• Dense racking, partial blockage: CS 1 to 3 m, UWB 0.3 to 0.5 m

Field conditions, not datasheets, decide the winner.

Why does bandwidth decide the accuracy ceiling?

Range resolution scales with bandwidth: roughly the speed of light divided by twice the bandwidth. A 499.2 MHz UWB channel resolves reflected paths about 30 cm apart. Bluetooth has about 80 MHz of usable 2.4 GHz spectrum, which resolves paths only about 1.9 m apart. When a reflection off a steel beam arrives 1 m behind the direct path, a UWB receiver separates the two pulses and times the first one. A channel sounding receiver sees both paths folded into one corrupted phase measurement and must estimate its way out.

Good CS chipsets fight back with multi-channel phase fitting and machine-learned multipath classifiers, and they recover much of the loss in mild environments. But 2.4 GHz also carries your Wi-Fi, your other Bluetooth traffic, and microwave oven noise, so the estimator works on contested spectrum.

Signal processing narrows the gap. Physics keeps it from closing.

You cannot software-patch a 6x bandwidth deficit.

How do you run the two-week bakeoff?

Buy two evaluation kits, around $500 to $2,000 each, and test on your real floor rather than a conference room.

• Week 1, static: mark 20 test points with laser-measured ground truth. Make at least 6 of them non-line-of-sight, behind racking or inside a metal cage. Log 1,000 or more range samples per point per radio.
• Week 2, dynamic: mount a tag on a cart or forklift, drive a known route at working speed, and log continuously during a normal shift with Wi-Fi and other traffic live.

Then plot the error distribution per radio. Score the median and the 95th percentile, never the mean, because one radio can average 0.4 m while its worst 5 percent of reads land 4 m off and route your picker to the wrong aisle. Also log ranging failure rate at each NLOS point; a radio that returns nothing is worse than one that returns 1 m error.

Trust the 95th percentile, never the mean.

What does each system cost per tag and per anchor?

At 10,000-unit volume, a channel sounding tag built on a single BLE SoC lands around $3 to $8. A UWB tag needs a UWB radio plus a BLE companion for discovery and control, so $10 to $20 is typical. On tag price alone, BLE wins by 2 to 3x.

The tag is rarely the biggest line. Anchors and installation usually are:

• UWB anchors run $150 to $400 installed, and tight time-difference positioning wants dense placement, often one anchor per 300 to 800 square meters in racked space.
• CS anchors are cheaper BLE hardware, but the accuracy penalty means you may need more of them, or you accept zone-level answers.
• Cabling and lifts for ceiling mounts frequently cost more than the anchor hardware itself.

Price the whole system against the accuracy your workflow needs. If 2 m tells a nurse which bay holds the pump, CS wins on cost. If a robot needs 30 cm, UWB's tag premium is noise.

Cheap tags mean nothing if reads fail.

Which chipsets belong on a 2026 shortlist?

RTLS chipset selection in 2026 comes down to a handful of parts per radio.

• Channel sounding: Nordic's nRF54 series and NXP's KW47 family shipped with CS support in silicon, and Silicon Labs followed in its Series 3 parts. Check that the vendor's ranging library exposes raw phase data, not just a distance output, so you can run your own estimator later.
• UWB: Qorvo's DW3000 series and NXP's Trimension line (SR150 class for anchors, SR040 class for tags) dominate. Look for FiRa certification if you ever want interoperability with phones, since Apple's U2 chip and flagship Android devices speak FiRa profiles.

Two forward-looking checks: IEEE 802.15.4ab, the next UWB revision, is in progress, so ask vendors about firmware paths. And NXP sells both radios, which matters if your roadmap includes a dual-radio tag that uses CS for cheap presence and UWB for precision zones.

Shortlist two chips per radio, then test both.

Where do you actually spend the engineering year?

Teams budget for the radio decision and then discover the radio was the easy part. Turning raw ranges into positions a customer trusts means anchor auto-calibration, motion filtering, outlier rejection under multipath, and fusing RF with inertial or camera data when the RF goes dark behind steel. That layer typically takes a positioning team 12 to 24 months to build and harden, and it is exactly the layer where granted patents already exist, whichever radio you picked.

This is where licensing beats rebuilding. Position Imaging holds hundreds of granted patents in radio-frequency ranging and real-time positioning, including US 11,774,249 and US 12,000,947, and the portfolio is cited by Apple, Bosch, and other major firms. Licensing the ranging and tracking methods gives you two things at once: proven algorithms your team does not have to reinvent, and freedom to operate before your first customer ships.

License the ranging math, ship the product.