Instrumentation & Measurement Magazine 23-3 - 37

Fig. 1. Typical coil designs used in search heads of conventional metal detectors: receiver gradiometer design, double-D overlapping coils, coaxial design with a
bucking receiver coil, coaxial design with a bucking transmit coil, and circular mono-coil design. Adapted from [5] (© 2020 IEEE).

towards spectroscopic detector operation would however require abandoning the traditional tuned-coil approach and
adopting more complex signal generation methods, such as
the ones relying on DDS techniques, or binary excitation with
adequate spectrum-shaping properties.
In MDs employing pulsed excitation, a time decay of the
secondary field induced by eddy currents is monitored after
the transmitter has been shut off. Looking at different portions
of the decay curve and observing its shape may reveal important information on a target's size and possibly metal type [8].
In general, EMI responses of smaller and less conductive targets are observable only in "very early" time portions of the
response, so the ability to capture those signals could potentially open up new application areas such as detection of very
thin wires or non-metallic parts typically found in improvised
explosive devices. The engineering problem is to overcome the
speed limitations related to inductive load switching, coil ringing, deep saturation of receive amplifiers, etc.
When it comes to the receiver side of MDs, microvoltlevel signals are typically induced in the coil(s). Incorporating
higher-resolution analog-to-digital (ADC) converters (≥ 16
bit) with sampling rates on the order of a few MS/s, as well as
more powerful microcontrollers, could enable more efficient
digital signal processing and thus more reliable formation of
alarms. Latest generations of MDs have now incorporated
such designs to overcome some of the difficulties related to
analog-domain processing of the induced signals.
Finally, coil geometry of the detector's search head is another important factor influencing the overall performance
of a particular MD. Specific geometries of conventional MDs,
shown in Fig. 1, normally stem from various design trade-offs
related to sensitivity, depth coverage, pinpointing ability, immunity to soil effects, etc. [2], [5]. Possible modifications to
well-established coil geometries can be observed from two aspects. The first one is related to the problem of background
interferences, e.g., the influence of magnetic (so-called noncooperative) soils, where the goal is to maximize the ratio of
signals corresponding to a metallic target of interest and soil.
Designing proper metrics for the characterization and comparison of different coil designs is therefore crucial [4].
Another aspect for the possible optimization of the search
head geometry comes into action if a detector is to be upgraded
with target characterization of classification capabilities. In
that context, it is important to come up with coil designs which
May 2020

would be able to induce different directional responses from
the target, so that acquired EMI data can be used to recover
information on a target's size, shape, orientation and metal
type. Unfortunately, conventional search heads (Fig. 1) are
by no means optimized for such purpose. Intuitively, coil arrays might provide data of higher quality / spatial diversity,
but these are much more difficult to implement into practical
handheld devices [5].

Feature Extraction Level
Extraction of a target's intrinsic features (i.e., the electromagnetic signature), independent of the target's relative location,
orientation, or a particular type of search head, is a crucial step
towards the implementation of a discrimination-enabled MD.
For that purpose, approximate analytical models describing
EMI scattering phenomena in metallic objects can be of vital
help. In the widely-used induced dipole model, an object's intrinsic features are contained in six independent elements of
the magnetic polarizability tensor [6], [7]. The task of dipole inversion is to estimate these tensor elements and the unknown
target location from induced voltages obtained over a large
number of known sensor positions (Fig. 2). This results in a
well over-determined nonlinear inverse problem. The part introducing nonlinearity, and hence most difficulties with the
inversion process, comes from dipole localization [5].
For a dipole-like metallic target, the signal of an EMI sensor, acquired over N different sensor positions is given by (1),
where uind is N x 1 vector of induced voltages, S is the N x 6
sensitivity matrix, v is a vector containing six independent
elements of the target's magnetic polarizabiliy tensor (also referred to as directional magnetic polarizabilities), and k is the
scaling term which depends on sensor's excitation parameters
[5]. Column vectors of the sensitivity matrix S are called directional sensitivities, since they directly reflect the ability of a
search head to induce the corresponding directional responses
of a metallic target.

u ind = Sxx


Sxy

Sxz

Syy

Syz

 M xx 


 M xy 
M 
xz 
Szz  ⋅ 
⋅ k = S ⋅ v ⋅ k (1)
  M yy 


 M yz 


 M zz 

IEEE Instrumentation & Measurement Magazine 37

Instrumentation & Measurement Magazine 23-3

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 23-3