UWB for Follow-Me Devices: Algorithms, Power, and Practical Tips

Ultra-Wideband (UWB) has moved from niche industrial systems into consumer gadgets and mobility devices. If you build “follow-me” products — smart luggage, shopping trolleys, assistive mobility, or last-meter delivery bots — UWB is one of the most practical ways to get reliable, low-latency, centimeter-class ranging. But getting a commercial product to behave in messy real environments requires a reality check: careful algorithm choices, robust NLOS handling, and smart power management.

This article collects the practical algorithms (TOF / TDoA / PDoA), fusion and filtering patterns, low-power tactics, and real-world pitfalls — with pointers to the recent standards, chip docs and papers that matter today.

1. Quick snapshot — where UWB stands in 2025

UWB (impulse radio) is standardized under IEEE 802.15.4x families; the 802.15.4z work introduced HRP improvements and security extensions that matter for commercial ranging._{IEEE Standards Association}
Modern transceivers (e.g., Qorvo/Decawave DW3000 family) provide practical building blocks: sub-nanosecond timing, configurable preambles and sleep modes; their datasheets also show the typical power envelopes and timing knobs you’ll need to use._{Mouser Electronics}
UWB is entering consumer ecosystems (phones, trackers, smart locks); platform support (Apple, Samsung, Google/Find Hub updates in 2024-25) is accelerating adoption. _verge

2. The core ranging methods — formulas you’ll actually implement

2.1 TWR / TOF — Two-way ranging and Time-of-Flight

The simplest conceptual model is ToF: measure the radio propagation time, multiply by c (light speed). In practice, you use two-way ranging (TWR) to remove clock offsets between devices.

A canonical TWR exchange (A ↔ B):

A transmits at t₁, B receives at t₂ and replies at t₃, A receives reply at t₄.
Round-trip minus device processing time gives the two-way flight time:

T_{round} = (t_4 - t_1) - (t_3 - t_2)

So one-way ToF ≈ T_round/2 and distance

d = c \cdot \frac{T_{round}}{2}

Implementation notes:

Use hardware timestamps (chip timestamps) not host timestamps to avoid OS jitter.
Account for asymmetric processing delays if the responder has variable processing latency.

2.2 TDoA — Time Difference of Arrival (anchor-based)

TDoA is anchor-centric: a tag emits, multiple synchronized anchors receive; the differences of arrival times define hyperbolas. For anchors at (x_i,y_i) and measured arrival times t_i:

\Delta d_{ij} = c(t_i - t_j)

Solve the resulting nonlinear system (often as a linearized least-squares problem after algebraic manipulation).

TDoA pros/cons:

Pros: tag is simpler / lower active duty cycle; good for multi-tag scenarios.
Cons: strict anchor synchronization (PPS/PTP or over-the-air sync) and anchor geometry (GDOP) matter.

2.3 PDoA / AOA — Phase / Angle

PDoA (phase difference) or direct AOA from antenna arrays yields angle estimates. For two antennas separated by distance d_a , phase difference Δϕ relates to angle θ :

\Delta \phi = \frac{2\pi}{\lambda} d_a \sin \theta \Rightarrow \theta = \arcsin \frac{\lambda \Delta \phi}{2\pi d_a}

PDoA is useful to supplement range with bearing, but practical deployment must manage phase wrapping, precise antenna calibration, and greater sensitivity to multi-path.

3. Mitigating multipath and NLOS — the real engineering work

Real deployments are rarely clean line-of-sight. In warehouses, parking garages, and airports reflections dominate. Recent studies continue to show that NLOS is still the primary error source for UWB ranging; modern systems combine CIR analysis, machine learning classifiers and geometry checks to detect and mitigate NLOS. _MDPI

Practical recipe:

CIR / early-path detection: inspect the channel impulse response (CIR) to find the first arriving path and reject late strong echoes. Many chips provide CIR or preamble correlation metrics.
SNR and power checks: low SNR or unusual power profiles often correlate with NLOS.
Geometric consistency: run a lightweight consistency check — if a TDoA solution jumps far from the last trusted state, down-weight or discard it.
Hybrid sensor fallback: if UWB is NLOS, use IMU dead-reckoning + vision/LiDAR to bridge the gap; when UWB returns, use EKF or graph optimization to correct drift.

A recent practical approach is two-stage NLOS detection (decision tree + shallow NN) on CIR features — this gave robust rejection in harsh indoor tests. _MDPI

4. Sensor fusion patterns that work in the field

For a follow-me system the minimal useful sensor stack is UWB + IMU; the most resilient stacks add vision or LiDAR.

4.1 A practical EKF skeleton

State vector (example):

x = [p_x, p_y, p_z, v_x, v_y, v_z, q_w, q_x, q_y, q_z, b_a, b_x]^T

Predict: integrate IMU (pre-integration when necessary).
Update: UWB gives position or range constraints; vision gives relative pose or re-ID observations; LiDAR gives local scan matching constraints.
Gate: use Mahalanobis distance to reject outlier measurements.

4.2 Sliding-window graph optimization (backend)

For longer runs, run a pose-graph or bundle-adjustment backend every few seconds (1–5 Hz) to absorb drift. Include UWB distance/anchor constraints as edges for global anchoring.

4.3 Practical fusion tips

Time-synchronize: hardware timestamps or PTP; software interpolation when needed.
Dynamically weight sensors: when UWB CIR shows NLOS, reduce its weight; when vision loses track, raise UWB weight.
Keep a “safe” local controller that can stop the vehicle if fused uncertainty grows beyond a threshold.

5. Power management — make it survive a day

Power is what decides whether your follow-me feature is a novelty or a product. DW3000-class parts offer sleep modes and configurable preamble rates; but continuous high-rate ranging is power-hungry. Datasheets and application notes show typical currents and emphasize the need for system-level duty cycling. _{Mouser Electronics}

Practical power strategy:

Adaptive ranging rate: base ranging rate on relative speed and required responsiveness. Use high rate when user is moving fast; low rate when stationary.
Event wake: combine a low-power proximity trigger (BLE RSSI, PIR, or accelerometer) to wake the UWB engine.
Selective anchor polling: only poll a subset of anchors to reduce radio airtime and MCU work.
Hardware power gating: turn off peripherals (cameras, radios) when not needed.

Example budget calculation (illustrative):

If active UWB average current ≈ 40–80 mA during ranging bursts and sleep current ≈ µA, a 300 mAh battery can last a few hours if the duty-cycle is high; but with smart wake and adaptive rates you can extend to 8–12+ hours for many consumer designs.

6. Standards, security and ecosystem updates (2024–2025)

IEEE 802.15.4z contains PHY/MAC improvements and security features relevant to commercial ranging. Use libraries and stacks that support the secure ranging extensions where available. _{IEEE Standards Association}
Ecosystem momentum: major platforms continue adding UWB support (smartphones and tracking ecosystems), making interoperability and use-cases like ‘find my’ and device handoff more practical. Google, Apple and Samsung moves in 2024–25 widened the consumer reach for UWB-based location services. _verge

7. Field lessons — real pitfalls and fixes

Below are concise field notes from live projects.

Pitfall: Anchors too close to metal pillars → systematic bias and frequent NLOS.
Fix: re-place anchors for geometric diversity, use early-path CIR detection, add an extra anchor to reduce GDOP.
Pitfall: Tag embedded in a metal enclosure (phone case, luggage frame) → signal attenuation.
Fix: move antenna to non-metal location, use PCB antenna with clearance, or add RF window/slot.
Pitfall: Multiple tags/robots in same frequency/time slot → collisions.
Fix: MAC scheduling (TDMA), short randomized backoff, or assign separate channels.
Pitfall: Thinking theoretical ±2 cm is typical in all conditions.
Fix: design for 10–30 cm worst-case, show graceful degradation, and add fallback navigation.

8. Where to read next (short reading list)

Qorvo / Decawave DW3000 datasheet and application notes (power, CIR metrics). _{Mouser Electronics}
IEEE 802.15.4z standard overview (security and HRP features). _{IEEE Standards Association}
Recent papers on NLOS detection & mitigation (MDPI / Sensors 2024–2025). _MDPI
Industry perspectives (Pozyx state-of-UWB 2025 briefing). _pozyx.io

9. Final takeaway — product checklist

If you’re shipping a follow-me product, make sure you have:

Anchor placement plan and GDOP analysis.
CIR/NLOS detection and dynamic measurement weighting.
UWB + IMU fusion (EKF) and a low-latency local controller that can stop safely.
Adaptive ranging duty cycle and wake triggers for power economy.
Security: mutual authentication, anti-replay and encrypted ranging where privacy or safety matter.
Field test plan: include metal-dense environments, people flows, and adjacent device interference.

UWB makes follow-me features feasible in the real world. The rest — robustness, safety, battery life — is still engineering: thoughtful architecture, good timing, and lots of field testing.