← Back to Index

Dennis Burke - Lightning & Thunder (TDOA/ToF) Project Status

Contact: Dennis Burke dennisfburke@verizon.net
Location: 1465 Cobblecreek St, Manteca, CA 95336 (510-828-3036)
Last Updated: February 10, 2026

🎯 Project Overview

Lightning & Thunder (L&T) — Time Difference of Arrival (TDOA) / Time of Flight (ToF) acoustic ranging system for ZTAG proximity detection.

Lightning = RF packet (ESP-NOW broadcast)
Thunder = Acoustic burst (16-18kHz ultrasonic tone)
Goal: 10-15cm spatial resolution within 2-3 meter range
Target jitter: 200-300 microseconds (>1ms = 30+ cm variance, unusable)

📅 Recent Timeline

Feb 3, 2026 — Doppler Analysis + Espressif Follow-Up

Dennis calculated Doppler guard-band formula
19-20 frequency slots between 10-10kHz with ~10% guard-band
Assumption: ΔV ~20 MPH (arm-waving motion)
Asked: "Any feedback from Espressif regarding ESP-NOW enhancements?"

Jan 29, 2026 — ESP-NOW Enhancement Request to Espressif

Critical request: Dennis needs Espressif cooperation
Specific ask: Add 4x 16-bit timestamp fields to ESP-NOW packet (managed at stack level)
Quan forwarded request to Espressif (via email Jan 29)
Timestamps needed:
1. T1 (Build-Send): Packet placed on stack (sender local time)
2. T2 (Transmission): Packet positioned to transmit (updated every retry)
3. T3 (Reception): Packet received on stack (receiver local time)
4. T4 (Consume): Packet arrives at application (receiver local time)
WiFi/RF delta = (T2 - T1) + (T4 - T3) — self-referential, clock-rate error negligible
Resolution: 16-bit field gives ~65ms range with 1μs resolution
Benefit to Espressif: Provides tight time-sync useful beyond ZTAG (tri-corner hat, etc.)

Jan 22, 2026 — L&T Architecture Discussion

Dennis proposed 4-timestamp solution to eliminate RF RX jitter
Quan demonstrated acoustic pulse detection works with good accuracy
Remaining uncertainty: interval between lightning packet generation and reception
Consensus: TDM scheduling must come into play for meaningful L&T function

Jan 13, 2026 — Hardware Limitations Discovered

M5Stack Fire hardware constraint: PDM mic (SPM1423) and Internal DAC (GPIO 25) both require I2S0
Result: True simultaneous TX+RX impossible on this hardware
Workarounds:
- I2S1 supports PDM but requires software PCM→PDM conversion
- DAC can be bit-banged (performance hit)
Audio startup issue: DAC tear-down for half-duplex creates broadband clicks
Dennis suggested AC coupling issue: pumping low-frequency signal during idle reduces clicks

Jan 11, 2026 — Global Clock Approach

Quan's new strategy: Stop using RF RX timestamps (too much jitter)
New approach: Phase-lock all devices to shared timebase
- Encode TX timestamp in RF payload
- Schedule acoustic emission at known offset
- Only timestamp acoustic arrival against global clock
Rationale: RF becomes coordination channel, not timing sensor
Early results: Far more stable, avoids pathological outliers
Aligns with TDM: Global time infrastructure needed anyway

Jan 2, 2026 — Milestone 5 Achieved

Quan: Single burst at 16k-18kHz, 2ms-10ms length, detectable up to 3 meters
No broadband noise in signals (big physical layer milestone)
Working on multiple bits back-to-back for preamble detection
Original LT: Used relative frequency intensities to encode 0/1 (like differential signaling RS485)

Dec 29, 2025 — Marco Polo vs. Lightning & Thunder

Quan explored Marco-Polo (RF trigger → acoustic response)
Abandoned: Jitter in Marco→Polo response exceeded tolerance
Returned to L&T: Device always sends ESP-NOW + audio burst together
Once refined, can layer call-and-response on top

Dec 8, 2025 — Dennis' Deep Dive Summary

Timing consistency critical: 200-300μs jitter budget
Clean phase-stable tones required
ESP32 hardware cosine generator can't change frequencies mid-output (discontinuities)
Moving to raw DAC control + custom audio driver
Multi-frequency signatures needed (pure tones fail with reflections/multipath)
Adjacency gating approach:
1. RSSI threshold (optional, coarse)
2. Acoustic preamble validation
3. Slot-timing identity bursts
4. SNR + echo-rejection filters (direct-path arrivals only)

Nov 24, 2025 — UWB vs. L&T Discussion

Dennis explored UWB positioning (DW1000 modules)
UWB pros: Better accuracy, anchors define plane
UWB cons: Anchors not viable for ZTAG, $15-20/unit cost, hardware changes required
Quan's constraints:
- 60,000+ ZTAGGERs deployed across 500-600 systems
- Enhancements must be software-only
- 2-minute setup time non-negotiable
- UWB impractical across application, hardware, business dimensions
Dennis' ad-hoc localization dream: Anchorless UWB (experimental)

Sep 25, 2025 — Hardware Request

Dennis unable to run standard ZTAG code on M5Core/M5Core2
Requested ZTAG kit from Quan

Aug 6-20, 2025 — Onboarding

Dennis signed NDA
Initial Zoom meetings with Quan + Malachi
Set up M5Stack modules for testing

🔧 Technical Architecture

Current L&T Design (Quan, Jan 2026)

TX Device:
  1. Generate RF packet (ESP-NOW) with TX timestamp
  2. Schedule acoustic burst at T_offset from TX timestamp
  3. Broadcast both over air

RX Device:
  1. Receive RF packet → extract TX timestamp
  2. Detect acoustic arrival → timestamp against global clock
  3. Calculate distance: (T_acoustic_rx - T_tx - T_offset) × speed_of_sound
  4. Phase-lock to shared timebase (global clock sync)

Dennis' Proposed ESP-NOW Enhancement

ESP-NOW Packet Header (4x 16-bit fields):
  T1: Build-send (sender local time, packet queued)
  T2: Transmission (sender local time, updated on retry)
  T3: Reception (receiver local time, packet on stack)
  T4: Consume (receiver local time, packet at app)

WiFi/RF latency = (T2 - T1) + (T4 - T3)
Thunder pulse age = T_acoustic_rx - (T_tx + WiFi_latency)
Distance = Thunder_pulse_age × speed_of_sound

Multi-Tone Encoding (Dennis' Research)

Frequency range: 10-10kHz (audible range limit)
Doppler compensation: Non-uniform spacing (~10% guard-band)
Capacity: 19-20 identity slots
Detection: Goertzel algorithm (falls down at some point) → FFT required
Challenge: Continuous frequency detection consumes heavy ESP compute

Adjacency Gating (4 layers)

RSSI threshold (ESP-NOW exposes RSSI, coarse filter)
Acoustic preamble validation (strong, properly-timed signatures only)
Slot-timing identity bursts (avoids ambiguity with multiple nearby devices)
SNR + echo-rejection (direct-path arrivals, suppress multipath)

🚧 Current Blockers

CRITICAL: Espressif ESP-NOW Enhancement

Status: Request sent Jan 29, 2026 (Quan → Espressif)
Blocker: Need Espressif to add 4x 16-bit timestamp fields to ESP-NOW vendor header
Rationale: User-land RF RX timestamps have unacceptable non-deterministic jitter
Impact: Without this, must rely on global clock sync (more complex, less robust)
Espressif incentive: Feature useful for any ESP-NOW time-sync application

Hardware Limitations (M5Stack Fire)

I2S conflict: PDM mic + DAC both require I2S0 (can't do full-duplex)
Audio clicks: DAC startup/teardown creates broadband noise
Workaround options:
- Software PCM→PDM conversion on I2S1 (CPU hit)
- Bit-bang DAC (performance hit)
- Keep DAC on continuously (power/thermal?)
- Pump low-frequency signal during idle (reduces clicks)

RF RX Timestamp Jitter

Problem: ESP-NOW RX timestamps exposed to userland show non-deterministic jitter
Cause: WiFi/ESP-NOW reception scheduled above MAC/PHY boundary (not true RF arrival time)
Impact: Intermittent outliers imply unrealistically short arrival times
Solution: Either Espressif stack-level timestamps OR abandon RF-based timing entirely

✅ Action Items

For Quan

Follow up with Espressif on ESP-NOW enhancement request (sent Jan 29, status unknown)
- Contact: Likely via Malachi Burke's relationship with Espressif team
- Provide Dennis' technical spec (4x 16-bit timestamp fields)
- Emphasize mutual benefit (time-sync useful beyond ZTAG)
Request from M5Stack/Espressif: Clarify if lower-level ESP-NOW hooks accessible
- Dennis mentioned: "I don't think we have access to that level of the code"
- Alternative: If source available, changes are "pretty trivial"
Test global clock sync approach (current path if Espressif declines)
- Quan's Jan 11 approach: phase-lock devices, encode TX timestamp, avoid RF RX timing
Resolve M5Stack audio hardware limitations
- Evaluate software PCM→PDM conversion overhead
- OR: Accept half-duplex with DAC click mitigation (idle pumping)
- OR: Future hardware revision with separate I2S channels

For Dennis

Doppler analysis complete (Feb 3)
ESP-NOW enhancement spec delivered (Jan 29)
Continue UWB research (personal curiosity, not blocking ZTAG)
Test ZTAG hardware (requested Sep 25, unclear if received)
Validate multi-tone detection (Goertzel vs FFT performance on ESP32)

For Malachi Burke

Facilitate Espressif communication (Dennis + Malachi have relationship with team)
Provide ZTAG hardware to Dennis (if not already sent)
Review Dennis' timestamp solution (validate technical feasibility)

📊 Technical Metrics & Constraints

Parameter	Target	Current	Notes
Spatial resolution	10-15 cm	~10 cm achieved (early test)	Quan's first L&T proof-of-concept
Jitter budget	200-300 μs	Variable (RF RX unreliable)	>1ms = 30+ cm variance (unusable)
Detection range	2-3 meters	3 meters (Milestone 5)	Acoustic burst at 16-18kHz
Frequency range	10-10 kHz	16-18 kHz tested	Dennis researching wider band
Identity channels	19-20 slots	TBD	Doppler-compensated spacing
Update rate	12 Hz	TBD	Quan's Marco-Polo target
Player capacity	30-50	TBD	Limited by tone count × TDM slots

🔗 Key Documents & Repos

🤝 In-Person Meetings

Jan 6, 2026 — Oakdale Training

Location: OJUSD Technology Center, 331 Hinkley Ave, Oakdale, CA 95361
Time: 8:00 AM
Attendees: Quan, Steven Hanna, Dennis Burke
Purpose: ZTAG training for teachers + face-to-face L&T discussion
Outcome: (meeting occurred, no notes captured in emails)

Dec 2, 2025 — Twin Rivers (Cancelled)

Original: Dec 5 (Thursday) → Rescheduled to Dec 2 (Tuesday)
Location: Twin Rivers School District, 5115 Dudley Blvd, McClellan Park 95652
Time: 10:00 AM
Dennis' response: "HVAC issue, can't make it"
Outcome: Did not occur

Nov 25, 2025 — Sacramento Training (Proposed)

Location: TBD (Northern California, early Dec)
Purpose: Group training for teachers, invite Dennis to observe + meet in person
Outcome: Evolved into Dec 2 Twin Rivers meeting (which was cancelled)

💡 Dennis' Technical Insights

On Doppler Effect (Feb 3, 2026)

"The doppler effect may be uniform but the result is frequency dependent so it makes it possible to use non-uniform spacing to squeeze in more useable frequencies. A delta V of just over 20 MPH [assumes waving arms]. 19-20 slots between 10 and 10 KHz inclusive with ~10% guard-band."

On ESP-NOW Limitations (Sep 20-23, 2025)

"ESP-NOW by itself is a no-go for the ZTAG use case. [It] detected more than one master (isolated devices) and didn't self-heal when master A became unavailable."

Resolution: Malachi clarified master device concept is software-layer, not ESP-NOW behavior. Dennis discovered example code hard-coded 2 peers (recoded to 32).

On Localization vs. Adjacency (Nov 24, 2025)

"Is an ad-hoc localization scheme something I should pursue or is L&T to be a confirmation of adjacency?"

Quan's answer: "For now, I'm okay with foregoing a full spatial map unless we have a permanent, controlled environment. L&T is for proximity/tagging within 2m cluster."

On Audio Ranging (Dec 8, 2025)

"When I look at the range of the audio I think 'If I hear you, we're in a cluster, otherwise not.' Marco sends RF saying there will be a subsequent audio pulse. Then all responding Polo RF packets will only come from adjacent units. Resolving RF latency is much less time-constrained than flipping audio pulse onto Polo units."

On Multi-Tone Detection (Jan 11, 2026)

"Goertzel algorithm falls down at some point making an FFT the best/only option. Continuous frequency detection consumes a great deal of ESP compute power and can limit L&T update rate. It may well impact 'system' behavior."

🎯 Strategic Decisions

Why Not UWB?

Anchors not feasible — ZTAG customers prioritize 2-minute setup, no infrastructure
60,000+ devices deployed — Software-only enhancements required
Cost prohibitive — $15-20/unit adds significant margin impact
Use case mismatch — Full spatial map unnecessary; proximity/tagging within 2m sufficient

Why Global Clock Sync Over RF RX Timestamps?

RF RX jitter unacceptable — Non-deterministic outliers above MAC/PHY boundary
Espressif enhancement uncertain — Can't block on vendor cooperation
Aligns with TDM — Global time infrastructure needed anyway for scheduled slots
Proven stability — Quan's early results show far fewer pathological outliers

Why ESP-NOW Over Other Mesh Protocols?

Dennis tested painlessMesh — Required down-rev libraries, "not optimal"
ESP-NOW advantages: Low latency, vendor-supported, already integrated in ZTAG
Limitations addressed: TDM scheduling + global clock sync overcome ad-hoc issues

📧 Communication Preferences

Dennis Burke:

Preferred: Email (paper trail, checks frequently)
Secondary: Text (510-828-3036, ringer often off due to spam)
Availability: Flexible, based in Manteca (close to Oakdale, Sacramento)

Best contact method: Email first, text for time-sensitive

🔮 Future Considerations

If Espressif Declines ESP-NOW Enhancement

Path A: Fully commit to global clock sync approach (Quan's Jan 11 design)
- Pro: Software-only, no vendor dependency
- Con: More complex, requires tight phase-lock discipline
Path B: Hybrid approach — use RF as coordination channel only
- Encode TX timestamp in payload
- Accept RF RX timestamp unreliability, only trust acoustic timing
Path C: Explore ESP32-S3 sub-microsecond time sync (Dennis' Jan 12 link)
- Requires hardware migration (current: ESP32, not S3)

Hardware Evolution

Current: M5Stack Fire (ESP32, I2S limitations)
Next-gen considerations:
- Separate I2S channels for PDM mic + DAC (full-duplex)
- Or: ESP32-S3 with sub-μs time sync primitives
- Or: External audio codec with independent I2S buses

Scaling to 30-50 Players

TDM slots required: 19-20 identity channels × time multiplexing
Update rate tradeoff: 12 Hz target may drop with player count
Collision mitigation: 160 channels (10 time slots × 4 frequencies × 4 preamble phases)

📝 Notes from Recent Conversations

Breakfast Discussion (referenced Jan 11, 2026)

"TDM scheduling must come into play, as I mentioned when we spoke over breakfast last week."

Context: Quan and Dennis met in person (likely early Jan 2026, exact date unclear). Discussed need for Time Division Multiplexing to make L&T work at scale.

Fathom Transcripts

No Fathom meeting notes found with Dennis Burke in Google Drive search
Meetings likely occurred via Zoom (Malachi's account) without Fathom recording

⚠️ Open Questions

Espressif Response Status?
- Quan forwarded request Jan 29, 2026
- No follow-up email found through Feb 10
- Need to check: Did Espressif respond? What was their answer?
M5Stack vs. ZTAG Hardware
- Dennis requested ZTAG kit Sep 25, 2025
- Unclear if hardware was sent
- Last mention: Malachi provided "small collection of ZTAG units" (Sep 23)
Legacy LT Code Status
- Dennis: "Old Lightning and Thunder [code] doesn't run on [new ZTAGGERs]. Haptic motor comes on and nary a peep."
- Quan (Dec 8): "Entire audio path and DAC configuration changed on new ZTAG hardware, legacy implementation obsolete."
- Question: Did Quan ever provide updated LT code compatible with new hardware?
Global Clock Sync Implementation
- Quan described approach Jan 11
- No updates found through Feb 10
- Question: What's the current status of phase-lock implementation?
Oakdale Meeting Outcome (Jan 6, 2026)
- Meeting confirmed, Dennis attended
- No follow-up email discussion found
- Question: What was discussed? Any decisions made?

🔬 Experimental Insights

From Quan's Testing (LT2/LT3 branches)

Milestone 5 complete: Clean single bursts at 16-18kHz, 2-10ms length, 3m range
Current challenge: Extending to multiple bits back-to-back (preamble detection)
Previous implementation: 1.6ms bit width
Current: 4-8ms bit width for reliable separation
Goal: Optimize down toward original 1.6ms (tighter jitter, faster update rate)

From Dennis' UWB Experiments

Testing DW1000 eval kits (not integrated with ZTAG)
Exploring anchorless, ad-hoc localization (no success reported)
"Pretty cool stuff but architecturally reliant on stationary reference points"

🎖️ Dennis' Background

Expertise: RF/audio systems, signal processing, embedded systems (ESP32/Arduino/ESP-IDF)
Tools: VS Code, PlatformIO, Arduino, painlessMesh (experimented), M5Stack hardware
Projects: FPV drones (Mobula 7 HDZero), UWB positioning research
Relationship: Met through Malachi Burke, signed NDA Aug 2025
Collaboration style: Email-first, detailed technical write-ups, references academic papers

📚 Acronyms & Terms

L&T / LT: Lightning & Thunder
TDOA: Time Difference of Arrival
ToF: Time of Flight
TDM: Time Division Multiplexing
FFT: Fast Fourier Transform
Goertzel: Algorithm for detecting specific frequencies (faster than full FFT)
ESP-NOW: Espressif's connectionless WiFi protocol (low latency, no AP required)
I2S: Inter-IC Sound (digital audio bus)
PDM: Pulse Density Modulation (mic format)
DAC: Digital-to-Analog Converter (speaker output)
UWB: Ultra-Wideband (alternative positioning technology)
RSSI: Received Signal Strength Indicator
SNR: Signal-to-Noise Ratio

End of Report
Last email in thread: Feb 3, 2026 (Dennis on Doppler + Espressif feedback question)
Status as of Feb 10, 2026: Awaiting Espressif response on ESP-NOW enhancement request