Introduction

1.1Motivation

A metal detector tells its user that something conductive is under the coil. It does not say reliably what the object is, how large it is, or how deep it lies. Commercial detectors reduce the received signal to a single "target identity" number, and experienced users still dig up large numbers of bottle caps, foil and nails for every object of interest. In survey and clearance work the same limitation appears as a high false-alarm rate: each excavation of harmless scrap costs time.

The physics does not impose this limitation. A buried metal object responds to an alternating magnetic field in a way that depends on its size, shape, orientation, conductivity and magnetic permeability. With enough measurements, taken at several frequencies, from several directions and at known positions, most of these properties can be separated. Military programmes for unexploded ordnance have shown this with vehicle-mounted instruments costing well above one hundred thousand dollars (Chapter 2).

LODESTONE asks whether the same class of measurement can be made with inexpensive, openly available components: a multichannel audio codec, a software-defined radio, and consumer positioning sensors. The project name refers to magnetite, the naturally magnetised stone used in the first compasses. The project combines magnetic sensing with precise navigation.

1.2Research questions

The work started from a series of practical questions. They are stated here in the form in which the report answers them.

1. RQ1. Can a software-defined radio be used to detect buried metal, and in which frequency range does it help? (Chapters 3 and 6.)
2. RQ2. Is a measurement across many frequencies more informative than a measurement at one frequency, and what exactly does it add? (Chapter 3.)
3. RQ3. Can a trained model predict the type, size and location of a buried object, and how many transmit and receive coils does that require? (Chapters 5 and 8.)
4. RQ4. How much does precise knowledge of the sensor's motion (inertial sensors, camera, RTK satellite positioning) improve the result? (Chapter 7.)

1.3Approach

The platform has three measurement subsystems that share one time base:

• an induction subsystem operating between 188 Hz and 18.75 kHz, with three transmit and three receive coils driven and read by a CS42448 codec (six 24-bit inputs, eight outputs, one sample clock);
• a radar subsystem, a stepped-frequency continuous-wave ground-penetrating radar built on an ADALM-Pluto software-defined radio (AD9363), sweeping 0.5 to 3 GHz;
• a positioning subsystem combining RTK satellite positioning, an inertial measurement unit, a ground-facing camera and a height sensor.

The pulse-per-second output of the satellite receiver is recorded on a spare codec input and shown to the camera through an LED. Every sample of every sensor can therefore be placed on the same GPS time line.

Every design value in this report was calculated before any hardware is built. Where an exact solution exists (for example the response of a conducting, permeable sphere), the figures are computed from it directly in the page. The last check of the design was a noise budget. It predicts a detection range for coin-sized targets of about 40 cm, which agrees with the performance of commercial detectors.

1.4Expected contributions

1. An open hardware and software design that recovers the full magnetic polarizability tensor of buried targets for a hardware cost of about two thousand dollars.
2. A method for recording induction, radar and positioning data on a common GPS time base with single-sample precision.
3. A labelled, openly published dataset of induction and radar measurements of known objects at known positions and depths.
4. A physics-based classification pipeline, evaluated on objects that were never seen during training.

1.5Scope and limitations

LODESTONE is a research instrument. It is not a demining device, and it will not be certified or used as one. Field work is restricted to prepared test beds and to land that is known to be free of ordnance (Chapter 12). The project does not address detection of non-metallic objects by induction, and it treats the radar only as a supporting sensor for depth and shape.

1.6Status of this document

This is revision 3 of the report. The design work is complete; construction will start when funding is in place (Chapter 11). The one piece of hardware already characterised is the CS42448 codec board, and even there the noise and full-scale values are derived from the datasheet and must be confirmed on the bench (task T1.3). Measured results will replace calculated ones as the work packages in Chapter 9 are completed.

1.7Structure of the report

Chapter 2 reviews related work. Chapter 3 sets out the physics that the design depends on. Chapters 4 to 8 describe the system: architecture, the induction, radar and positioning subsystems, and data processing. Chapters 9 to 12 form the research plan: work packages and schedule, team and responsibilities, budget, and risks. The appendices collect key numbers, code, references and photographs.