LODESTONETechnical report TR-2026-01 · revision 3Հայերեն
Contents
  1. Cover and contents
  2. 1Introduction
  3. 2State of the art
  4. 3Physics of detection and identification
  5. 4System architecture
  6. 5Induction subsystem
  7. 6Radar subsystem
  8. 7Positioning and timing
  9. 8Signal processing and machine learning
  10. 9Work plan
  11. 10Team and responsibilities
  12. 11Budget and funding
  13. 12Risks, safety and quality
  14. AAppendices
Research project · Technical report TR-2026-01

LODESTONE

A low-cost instrument that finds metal objects buried in the ground and tells what they are, how large they are and how deep they lie.

Design complete · seeking grant funding · full report as PDF (74 pages)

What it is

An ordinary metal detector makes a sound when there is metal under its coil. It cannot say whether the metal is a coin, a nail or a piece of foil, how big it is, or how deep it is buried. To find out, someone has to dig.

LODESTONE is a research instrument designed to answer those questions before anyone digs. As the operator walks across the ground, it measures every buried metal object from many positions and reports on a map, for each object, the most likely type of metal, its approximate size and shape, its depth and its exact position.

Why it is needed

Most of what metal detectors find is scrap. Each signal has to be checked by hand, because the detector cannot tell a harmless object from one that matters. This makes surveys slow and expensive wherever buried metal has to be found:

Instruments that can classify buried metal exist, but they cost more than 100,000 USD and their designs and data are closed. LODESTONE aims to reach a useful part of that capability with components costing about 2,354 USD, and to publish the design, the software and a labelled dataset openly so that others can build on it.

How it works

soilnavigation satellitesbase stationsatellite antennamotion sensorscameradatasensor head: coils and radarwalking directiondepth from radarand signal shapeBCDArecorded track, one measurement per cmABCDresult: a map of objectsAaluminium can, 35 cm deepBcopper coin, 12 cm deepCsteel nail, 20 cm deepDstone, not metal1234
Figure I.How LODESTONE works. The numbered markers correspond to the four steps below. Lettered objects A to D appear on the result map with their estimated type and depth.
  1. The sensor sends a magnetic field into the ground. Coils in the sensor head carry an alternating current at twelve frequencies at the same time, from 188 Hz to 18.75 kHz. The resulting magnetic field passes through soil and reaches buried objects.
  2. Buried metal answers. The field makes small electric currents flow in any metal object, and these currents produce a weak field of their own, which receive coils in the same head pick up. How strongly the object answers at each of the twelve frequencies depends on its metal, size and shape. This pattern works like a fingerprint; a stone gives no answer at all.
  3. Depth and position are measured. A small radar in the sensor head measures how deep objects lie. Satellite positioning, motion sensors and a camera record where the sensor head is, to about 2 cm, at every moment. The same object is therefore measured from many known positions as the operator walks past it.
  4. A computer builds a map of identified objects. For each object it fits a physical model to all measurements, removes the effect of depth and orientation, and compares the result with a library of known objects. The output is a map showing each object with its likely type, size and depth.
about 40 cmdetection range for a coin-sized object, calculated
2 cmposition accuracy of every measurement
12frequencies measured at once, 188 Hz to 18.75 kHz
3 × 3transmit × receive coils: the full target tensor
2,354 USDequipment, including import and contingency
102,936 USDgrant request for 12 months, team of four

Technical summary

The induction part uses a six-input, eight-output audio codec (CS42448) to drive three orthogonal transmit coils and read three receive coils, so that the full magnetic polarizability tensor of a target can be estimated. The radar is a stepped-frequency ground-penetrating radar from 0.5 to 3 GHz on a software-defined radio (ADALM-Pluto). RTK satellite positioning, inertial sensors and a ground-facing camera are fused in a Kalman filter, and all data share one GPS time base through the receiver's pulse-per-second signal. Every design value in the report was derived from physical models before construction; the noise budget predicts a detection range of about 40 cm for coin-sized objects, in line with commercial detectors.

Induction head 3 TX coils (x, y, z), 3 RX coils class-D drivers, current sense ×33 low-noise preamplifiers Radar ADALM-Pluto (AD9363) PA, LNA, coupler 2 Vivaldi antennas, 0.5–3 GHz Positioning RTK GNSS base and rover IMU, magnetometer camera, time-of-flight ranger CS42448 codec 6 in, 8 out, 48 kHz, one clock DAC ADC Host Teensy 4.1 or Raspberry Pi TDM Processing computer 1 demodulation of 12 tones 2 SFCW sweep to range profile 3 pose estimation (EKF) 4 time alignment by PPS 5 dipole inversion, β(ω) 6 classification and fusion 7 map, dataset, reports Python; data stored with metadata and checksums on the team server USB USB, IQ samples UART / USB, 10 Hz fixes, 200 Hz IMU, 60 fps PPS Common time base: GNSS PPS is recorded on codec input 5 and blinks an LED in the camera view
Figure II.System overview: the three sensing subsystems, the codec and the processing computer. Details in Chapters 4 to 7.
measurement band0.00.20.40.610 Hz100 Hz1 kHz10 kHz100 kHz1 MHzfrequency (log scale)response (quadrature)Copper: strongest response at 989 HzCopper: strongest response at 989 HzCopper, 989 HzAluminium: strongest response at 1.6 kHzAluminium: strongest response at 1.6 kHzAluminium, 1.6 kHzStainless steel: strongest response at 40.5 kHzStainless steel: strongest response at 40.5 kHzStainless steel, 40.5 kHzMild steel (ferrous): strongest response at 84.6 kHzMild steel (ferrous): strongest response at 84.6 kHzMild steel (ferrous), 84.6 kHzCopperAluminiumStainless steelMild steel (ferrous)
Figure III.Why the frequency pattern identifies the metal: computed response of four metal spheres of 1 cm diameter (exact solution, Chapter 3). Each metal answers most strongly at a different frequency. The shaded band and the ticks show the twelve measurement frequencies.

Plan

The project runs for twelve months with a team of four in seven work packages. Each technical package ends with a measurable milestone, followed by a decision gate with a prepared fallback (Chapter 9).

Q1Q2Q3Q4WP0 Project managementWP0 Project management: weeks 1 to 52, led by PIWP1 Bring-up and receive chainWP1 Bring-up and receive chain: weeks 1 to 8, led by HWWP2 Static induction rig and calibrationWP2 Static induction rig and calibration: weeks 6 to 18, led by PIM1, week 18: Calibration agrees with theoryM1WP3 Positioning and timing platformWP3 Positioning and timing platform: weeks 10 to 26, led by SWM2, week 26: Positioning validatedM2WP4 Full tensor, dataset and classificationWP4 Full tensor, dataset and classification: weeks 18 to 36, led by DSM3, week 36: Classifier validated on unseen objectsM3WP5 Radar subsystemWP5 Radar subsystem: weeks 22 to 40, led by HWM4, week 40: Radar imagingM4WP6 Fusion, field trials and releaseWP6 Fusion, field trials and release: weeks 36 to 52, led by PIM5, week 52: Fused demonstration and releaseM5113263952weekled by PIled by HWled by SWled by DSmilestone
Figure IV.Work packages, coloured by the role that leads them, with milestones. M1 measured spectra agree with theory; M2 sensor position known to 2 cm; M3 classifier validated on objects not seen in training; M4 radar locates a can at 35 cm; M5 fused field demonstration and public release.

Budget and funding

The request of 102,936 USD covers four full-time positions for twelve months, equipment, running costs, dissemination and contingency. The main target is a NATO Science for Peace and Security Multi-Year Project with a partner group in a NATO member country, complemented by national research funding in Armenia (Chapter 11).

Personnel, 4 full-time positionsPersonnel, 4 full-time positions: 76,800 USD (74.6% of the request)76,800 75%OverheadOverhead: 11,520 USD (11.2% of the request)11,520 11%ContingencyContingency: 8,832 USD (8.6% of the request)8,832 9%DisseminationDissemination: 2,500 USD (2.4% of the request)2,500 2%EquipmentEquipment: 2,354 USD (2.3% of the request)2,354 2%Instruments and running costsInstruments and running costs: 930 USD (0.9% of the request)930 1%Total request: 102,936 USD
Figure V.Budget request by cost category, USD. Equipment is a small share because the platform is built from inexpensive components; most of the cost is the work of the team.

About this document

ReportTR-2026-01, revision 3
Date8 October 2026
StatusDesign complete; construction starts when funding is in place
Project12 months, team of four, 7 work packages, 5 milestones
Funding sought102,936 USD
DownloadFull report as PDF (74 pages)

Contents

  1. 1Introductionpage 2
  2. 2State of the artpage 3
  3. 3Physics of detection and identificationpage 4
  4. 4System architecturepage 5
  5. 5Induction subsystempage 6
  6. 6Radar subsystempage 7
  7. 7Positioning and timingpage 8
  8. 8Signal processing and machine learningpage 9
  9. 9Work planpage 10
  10. 10Team and responsibilitiespage 11
  11. 11Budget and fundingpage 12
  12. 12Risks, safety and qualitypage 13
  13. Appendicespage 14

Use the page numbers at the bottom of each page, or the arrow keys, to move through the report.