Report to IAEA



Use of back-scattered monoenergetic fast neutrons for humanitarian de-mining.



2001 IAEA, A-1400 Vienna, Austria





Vienna, Dec.11th, 2001





IAEA has been active in Humanitarian De-mining for some time on several avenues, like the Coordinated Research Program (CRP) on "Application of Nuclear Techniques to Anti-Personnel Landmine Identification" [1], the Advisory Group Meeting (AGM) on "A Remotely Controlled Multi-sensor Platform for Humanitarian De-mining" [2] or the TC Regional Project RER/1/005: "Field Testing and Demonstration of a System Using a Pulsed Neutron Generator for Humanitarian De-mining", a system known under the acronym PELAN [3].

In addition, a variety of groups are contributing to this field, like the University of Cape Town [4,5] or the Italian EXPLODET project [6]. All kinds of probes have been proposed and have been tried, among them fast neutrons.

But up to now the energy-selective methods of fast monoenergetic neutron technique have not been applied. Experiments and simulations exploring this avenue which are based on two Austrian patent applications of the author [7,8] are described in this report.

1. De-mining background

1.1 General

Humanitarian de-mining requires a multifold of methods, depending on the local environment, climate etc. In many cases the verification process is the costly (and dangerous) part in the process of de-mining, when a detected anomaly in the ground must be investigated (often by digging with the hands).

Basically there a three situations to be considered:

a) The anomaly can be accessed by a heavy de-mining vehicle. In this case the device verifying a mine may weigh a few hundred kilograms.

b) Access with a smaller vehicle (like a fork lift) is possible. The device may still be rather heavy.

c) The device must be carried. In that case it is reasonable to combine the detecting process with the verification process.

In addition, the sophistication of the method will (and must) depend on the application: a need of some understanding of the device is acceptable in case a), but in the case c) the system must be foolproof.

At this stage it is not possible to predict whether the method described here may result in a portable device with detecting and verification features, even if there is potential in this direction. So the discussion will be based on verification alone.

Two basic characteristics in verification work are

- false alarm rate

- missing alarm rate.

Quite obviously, a false alarm rate can be tolerated to some extent, the other not really. Therefore, also deep lying explosives must be noticed, i.e. a good ground penetration is necessary.

1.2 Fingerprinting of buried mines

Explosives are largely composed from the elements hydrogen, carbon, nitrogen and oxygen. As a rule, at least 1 out of 4 atoms in an explosive is hydrogen. Therefore, if the presence of hydrogen is measured at the spot of an otherwise detected anomaly in the ground, a necessary condition for the presence of a mine is established. To further reduce the false alarm rate, information on e.g. monoenergetic neutron back-scattering from nitrogen or on the shape of the buried object would be beneficial.

The presence of hydrogen may be detected, regardless of its chemical form, by observing the strong energy-moderating effect on multiply scattered fast neutrons using a source of fast neutrons and a neutron detector that is insensitive to fast but very sensitive to low energy (thermal) neutrons. This method has been known for some time as "neutron back-scattering" method or as hydrogen-density anomaly detection (HYDAD) [5].

Elemental analyses using gamma rays from nuclei excited by neutrons, like done in PELAN [3], are not well suited to detect hydrogen, because the gamma intensity from the neutron capture in hydrogen is low. So, this method must concentrate on the other constituents of the explosives.

In the case of a measurement of back-scattered monoenergetic neutrons a unique nuclear property of hydrogen (protonium) is taken advantage of: the maximum scattering angle of singly elastically scattered neutrons is 90 degree.

Therefore, at places in the ground, where hydrogen containing objects are buried, fewer (monoenergetic) neutrons are back-scattered than in their neighborhood producing a "shadow" in the shape of the object. This method is described in my Austrian patent application [7].

2. Monoenergetic neutron back-scattering with resonance penetration (MNBRP)

2.1 Principle

The method is based on two effects

- no back-scattering from hydrogen (protonium) [7]

- penetration of the soil by resonant monoenergetic neutrons [8].

In both cases the use of monoenergetic neutrons both in the input as well as in the output channel is essential.

2.1.1 Elastic neutron scattering from hydrogen.

Fig. 1 shows the correlation between angle and scattered neutron energy for elastic scattering from 1H. Beyond 45 deg the energy is strongly reduced, diminishing at 90 deg. There are no neutrons beyond 90 degree. This scattering into a forward cone occurs because the neutron mass is bigger than the proton mass. There is no other nucleus for which this holds. So, this property is a unique fingerprint for hydrogen (protonium). As a consequence, no neutrons are back-scattered from hydrogen reducing the back-scattered neutron intensity, i.e. a shadow is obtained.

But hydrogen has two more beneficial properties for mine detection in a back-scattering geometry

- the cross section is sufficiently big, especially at not too high energies

- the intensity of neutrons back-scattered from UNDER the mine is modified twice by the presence of the mine: first by the

attenuation of the primary neutron beam, then by that of the back-scattered neutrons. To contribute to the second effect the back-scattering angle must be as big as feasible.

Fig. 1. Dependence of scattered energy on angle for fast neutron scattering from 1H.

2.1.2 Monoenergetic techniques.

There are several advantages in using "monoenergetic" neutrons as can be seen by comparison with visible light. Using monochromatic light does not only allow to take advantage of resonance absorption and penetration, but makes, by means of filters, discrimination against background light possible. So, the use of monoenergetic neutrons allows taking advantage of resonances in the cross sections and of energy selective processes, like the dependence of elastically scattered neutron energy from angle and mass. This effect has been taken advantage of for material classification by fast neutron scattering [9,10]. Table 1 gives the ratios of back-scattered neutron energies to the primary energy for elements of interest in this work.

Table 1: Energy ratios of outgoing to incoming neutron energy at 150 and 180 deg for the main isotope of some (light) elements.

Isotope: \ angle:





















In fast neutron physics "monoenergetic" means that the energy spread is about an order of magnitude smaller than the maximum energy. A smaller spread is impractical, because of the necessity of a sufficient neutron intensity from the source which, to a first order, is proportional to the energy spread. Considering the classical energy region of fast neutron physics between about 1 and 14 MeV energy spreads between 1.10-1 and 1.10+0 MeV can be considered to be typical.

Therefore, only cross section resonances with a corresponding width can be utilized.

2.2 Problem specific essentials

2.2.1 General

There are two contributors to an optimum counting rate ratio for the mine-out / mine-in cases:

- areal density of hydrogen nuclei in the mine

- maximum ratio of (singly) elastically back-scattered neutrons from the soil below and around the mine to those from the layer above the mine.

Therefore, for a given mine (a given areal density of hydrogen) only the property of the layer of soil above the mine must be compared to that of the bulk around and under the mine. Thus, to nobody's surprise, the method works best for mines with little soil cover.

Considering a homogeneous soil the primary neutron intensity decreases more or less exponentially with depth. The same is true for the intensity of the elastically back-scattered neutrons. So, for detecting mines that are not too close to the surface it is important that the primary neutrons have a good penetration, i.e. that the soil attenuates the primary neutron intensity as little as possible. Before we go deeper into that let us look for a condition that would make the method fail, even for mines with only little soil cover.

Such is the case, if the intensity of elastically back-scattered neutrons from the ground below and around the mine is low, as it would be with very wet soil (or water). At first glance it is surprising, that wet soil (or a layer of water) ABOVE the mine has not a disastrous effect on the ratio mine-out/mine-in, but mainly on the returned neutron intensity requiring a longer measuring time, as will be shown later.

2.2.2 Resonance penetration

More than half of the atoms in all kinds of soil (or even concrete) are oxygen; quartz sand would even be 2 thirds oxygen (1 third silicon). A more or less typical soil composition is given in Table 2.

So it is not surprising that the neutron attenuation in oxygen dominates that of soil. Fig. 2 shows the transmission of fast neutrons through soil with an areal density of 225 g/cm2 (about 125 cm thick). Obviously and not surprisingly a neutron energy of 2.35 MeV is optimum because at this energy there is a resonant dip in the neutron cross section of oxygen, as shown in Fig. 3. This resonance is at 2.3506 MeV and has a half-width of 0.10 MeV, so that it can easily be utilized in fast monoenergetic neutron work. Therefore, neutrons of energy between 2.3 and 2.4 are best suited for the penetration of soil. This fact I have described in greater detail in one of my Austrian patent applications [8]. At 6.500 MeV there is a second maximum in Fig. 2 based on an alternate resonance of 16O with a half width of about 0.23 MeV. Its effect is increased by the 6.583 MeV resonance of silicon, the second most abundant element in standard soil (see Fig. 4).

Table 2: Mass and atomic percents of elements used for "standard earth" in the predictive simulations Element mass atomic % % O49.664.9 Si25.819.2 Al7.55.8 Fe4.61.7 Ca4.42.3 Na2.62.4 K2.41.3 Mg1.91.6 S1.20.8

Fig. 2. Energy dependence of primary neutron transmission through 225 g/cm2 soil

Fig. 3. Total neutron cross section of 16O between 2 and 3 MeV.

Fig. 4. Total neutron cross section of 16O and Si between 6 and 7 MeV.

2.2.3 Verification procedure

The basic method is simple and straightforward: comparison of the back-scattered neutron spectrum above an anomaly in the ground with that back-scattered from some area nearby which is mine-free. The advantage is that just a difference-spectrum must be interpreted, i.e. correlated uncertainties in the measurements drop out. In addition, the answer is, to a first order, independent of the soil composition. There is no need to establish spectral catalogues of various soils in order to interpret the measured spectra, as is the case for PELAN [3].

The disadvantage is the necessity of two or more "measurements", either simultaneously by producing a "shadow" picture with position sensitive neutron detectors (or other means of picturing, like coded apertures) or by "scanning" the ground.

With a picture or/and measurement of the monoenergetic neutrons back-scattered from nitrogen [9,10] the false alarm rate could be further reduced.

As mentioned in 2.2.1 the method fails if the ground below and around the mine does not return sufficient neutrons, like in the case of a thoroughly soaked soil. However, such a condition, even if it is not noticed by the operator, will be recognized by the instrument because the elastically back-scattered neutron intensity is much reduced already in the no-mine situation, i.e. all the area is "dark".

3. Monte Carlo simulation of the method

3.1 Set-up

The Monte Carlo neutron transport code MCNPv4B of LANL was used to test the basic ideas of this method. In order to give enough room for "unknowns" rather stringent conditions were chosen: the mine (plus housing) was taken to contain only 3.67 g of hydrogen, and a soil cover of 33.8 g/cm2 corresponding to 15 cm solid rock (or 22 cm of fine grained quartz sand) was chosen as a standard condition.

3.1.1 Soil simulation

A cylindrical volume, 100 cm deep with a diameter of 150 cm filled with "standard earth" (see Table 2) was chosen to represent the "earth" well enough. To be on the save side an unlikely high density of 2.25 g/cm3 (solid rock) was chosen rather than that of sand which is around 1.56 g/cm3. The cross section libraries chosen are shown in Table 3.

Table 3. Cross section libraries used in the simulation of the soil and of the explosive.

H-1 from ENDF/B-VI.1

C-nat from ENDF/B-VI.1

N-14 from special LANL ENDF/B-VI evaluation

O-16 from ENDF/B-VI

Na-23 from ENDF/B-VI.1

Mg-nat from ENDF/B-VI

Al-27 from ENDF/B-VI

Si-nat from ENDF/B-VI

S-nat from ENDF/B-VI

K-nat from ENDF/B-VI

Ca-nat from ENDF/B-VI

Fe-54 from ENDF/B-VI.1

Fe-56 from ENDF/B-VI.1

Fe-57 from ENDF/B-VI.1

Fe-58 from ENDF/B-VI.1

3.1.2 Explosive simulation

Somewhat arbitrarily the anomaly in the ground was chosen to be in the shape of a cylinder with 6 cm diameter and 3.3 cm thick, filled with material with a density of 1.45 g/cm3 and a elemental composition like that of the explosive Hexogen (see Table 4). So the anomaly contains 3.67g hydrogen, corresponding to an areal density of hydrogen of 0.13 g/cm2, and 51 g nitrogen.

Table 4: Atomic and mass percents of elements of the explosive Hexogen































3.1.3 Simulation geometry

A simple cylindrical geometry was chosen with the source neutrons impinging along the axis onto the cylinder made of soil. The detector was situated 15 cm off the surface in the shape of a concentric ring with 3 cm inner and 8 cm outer diameter. This results in an average scattering angle of 165 deg if said anomaly is covered by 4 cm of soil, increasing with the thickness of the cover. The standard soil cover was 33.8 g/cm3, the standard size of the explosive a cylinder with 6 cm diameter and 3.3 cm height.

3.2 Simulations

3.2.1 Optimizing external conditions

Although it is not impossible to make a "complete" simulation (including shielding, detector response etc.) such a thing is not reasonable, not only because of the enormous effort, but also because it does not make sense to do it for an apparatus that has not been designed yet.

Therefore, one must restrict oneself to simulate the crucial parts, to optimize the principle. The most important parameter in this experiment is the energy of the source neutrons. Disregarding technical questions like neutron production at the moment the best neutron energy is not necessarily the one with the deepest penetration, even if it is likely to be for finding deeply buried mines.

Obviously, the ratio in the number of detector counts with (A) and without (B) mine in place is crucial. Instead of this ratio B/A the net effect, i.e. (A-B)/B is given in Fig. 5 in dependence on the incoming neutron energy for our standard soil cover of 33.8 g/cm2 (the full line is only meant to guide the eye).

Fig. 5. Depression of the back-scattered neutron intensity by 3.67 g hydrogen at a soil cover of 33.8 g/cm2. Energy dependence of the "net effect" (full line) and of a figure-of-merit (dashed line, see text).

For a practical application the counting rates and the background conditions are very important, too. Whereas prediction of background is rather limited without an exact knowledge of the apparatus, count rate information can be extracted from the return ratio, i.e. the ratio C between information bearing counts in the detector and the numbers of neutrons started in the simulation. As long as B is not too much different from A one can thus define a relative figure-of-merit by multiplying the net effect with sqrt(C). Such a figure-of-merit is shown in Fig. 5 by the dashed line. From this figure it is clear that one optimum energy for MNBRP is at 2.35 MeV and somewhat above of it. The second maximum at 6.50 MeV cannot be utilized in a light weight instrument. If weight is not an issue, there is the advantage of much more efficient neutron sources in this energy range (three orders of magnitude!) and the disadvantage of a more demanding shielding.

A detailed interpretation of Fig. 5 at higher energies is rather difficult because several effects overlap (penetration of primary neutrons, of back-scattered neutrons, "shadowing" by hydrogen, by carbon -at the resonances of the carbon cross section, e.g. at 6.295 MeV and several others-).

Figs. 6 to 8 show simulated back-scattered neutron energy spectra at 2.35, 7.00 and 7.745 MeV primary neutron energy. These figures have in common that the counts in the highest energy bins of the sample-in case (solid line) are lower than those of the sample-out case (dashed line), whereas at about 3 quarters of the maximum energy it is the other way around. The first is due to hydrogen caused decrease of the intensity of neutrons back-scattered from soil, and the latter by additional neutron back-scattering from the light elements (nitrogen, carbon) in the mine, resp.. Comparing the spectra at 7.00 and 7.745 MeV one can see at 7.745 MeV the very strong carbon contribution around 5.6 MeV scattered neutron energy (from the strong resonance in the carbon cross section).

Fig. 6. Comparison of the back-scattered neutron spectra at 2.35 MeV primary neutron energy without (dashed line) and with mine (3.67 g hydrogen, full line) for 33.8 g/cm2 soil cover.

Fig. 7. Same as Fig. 6, but for 7.00 MeV

Fig. 8. Same as Fig. 6, but for 7.75 MeV

Furthermore it is important to see that the scale of the relative yield at 2.35 MeV is about a factor of 4 to 5 higher than at the higher energies. This scale measures the ratio of neutrons entering the detector volume to the source neutrons entering the surface of the soil above the mine position.

3.2.2 Maximizing the effect of hydrogen

Comparing Fig. 9 with Fig. 6 shows the strong increase of the measuring effect with reduced soil cover (from 33.8 to 11.3 g/cm3, or about 22 to 7 cm soil).

Fig. 9. Same as Fig. 6, but for a 11.3 g/cm2 soil cover.

Fig. 10 compares the depth dependence of the effect at 2.35 MeV and 2.79 MeV primary neutron energy. For the more convenient energy of 2.79 MeV the effect is not much less than for 2.35 MeV over all the depth range shown. The reason for that is that neutrons with about 2.79 MeV will be back-scattered from silicon with an energy of about 2.35 MeV, so that the back-scattered radiation has the advantage of the resonance in the oxygen cross section.

Fig. 10. Dependence of the net effect on the thickness of the soil cover.

The following discussion is for primary neutrons of 2.35 MeV and a soil cover of 33.8 g/cm2 (=15 cm solid rock).

Fig. 11. Dependence of the net effect on the areal density of hydrogen in the mine.

Fig. 11 shows the dependence on the areal density of hydrogen in the mine. The saturation to be expected at very high densities is due to neutrons back-scattered from the soil layer above the mine.

Fig. 12. Dependence of the net effect on the diameter of the mine. (The outer diameter of the detector ring is 16 cm.)

Fig. 13. Dependence of the net effect on the thickness of a water layer on top of the soil surface, for soil covers of 11.3 and 33.8 g/cm2.


Fig. 12 shows the increase of the effect with increasing diameter of the mine with constant areal hydrogen density of 0.13 g/cm2 and otherwise constant geometry. Obviously, the effect is not very dependent on the total mass of hydrogen, because of the decreased neutron intensity off-axis. The outer diameter of the detector ring is indicated by the dashed line.

Fig. 13 shows that a layer of water between neutron source and mine does not destroy the mine detecting capability of MNBRP. Even a layer of 3 cm water reduces the effect by less than 40 %, both with a soil cover of 11.3mg/cm2 and of 33.8 g/cm2.

3.2.3 Simulating the back-scattering from nitrogen

As mentioned in 2.2.3, back-scattering from nitrogen may be used as an add-on to reduce the false alarm rate. Whereas in the presence of hydrogen the highest energy component of the back-scattered neutron intensity is decreased, single elastic back-scattering from nitrogen increases the neutron flux at an energy of about 3/4th of the primary neutron energy. This discriminative effect has been taken advantage of for material classification by fast neutron scattering [9,10], not only for nitrogen but for all light elements (except light hydrogen, see also Table 1).


Fig. 14. Energy dependence of the back-scattering effect from 51 g of nitrogen (and carbon) for a 33.8 g/cm2 soil cover.

Therefore, the energy dependence of the difference in the numbers of neutrons with the corresponding energy was simulated with and without mine (see Fig. 14). The nitrogen mass was 51 g (mine with Hexogen, see 3.1.2), the soil cover again 33.8g/cm2. In this case an increased back-scattered neutron intensity is measured, so the effect is defined by (B-A)/A.

Like above, A is the case without mine, B that with mine. From this figure it is clear that also in this case both 2.35 MeV and 2.79 MeV are a good choice. The maximum at 7.745 MeV is from a contamination of carbon back-scattering (see also Fig. 8).


4. Experimental verification of the MNBRP method

4.1 Background

There are not many installations left that are equipped for lower energy monoenergetic neutron work. For two practical reasons the UCT (University of Cape Town) installation at NAC (National Accelerator Centre), at Faure, South Africa, appeared to be the installation of choice. Firstly, the main investigator, Prof. Frank D. Brooks, has developed a de-mining device based on thermalization of neutrons [5] and besides, he and his collaborators have measured monoenergetic neutrons, elastically scattered from light elements [9,10]. So, both the know-how and the equipment is available at this installation.

Consequently, from July 23rd to July 27th, 2001 120 hrs of machine time at the single-ended Van-de-Graaff of NAC was scheduled for this cooperative effort with the University of Cape Town. This time seemed more than sufficient for a very extensive investigation of MNBRP and related subjects that were of specific interest to the collaborators.

Actually, only 60 hours of beam time was available at one quarter to one third of the rated beam intensity because of vacuum problems in the accelerator. Thus only measurements at three energies could be performed. Fortunately, expecting a strong effect high accuracy was not required.

4.2 Experimental arrangement

4.2.1 Source and scattering geometry

A bunched proton beam from the Van-de-Graaff hit a Li-metal target (of thickness about 115 keV for 4.4 MeV protons) producing monoenergetic neutrons via the 7Li(p,n)7Be reaction. Actually, at the proton energies used in this experiment, there is a second, lower energetic neutron line present from the 7Li(p,n1)7Be* reaction (excitation of the 0.429 MeV level of 7Be) with about 1/10th in intensity. Another complication arose from beam retraces which were attributed to malfunctioning of the klystron bunching system. Due to the shortage in machine time (see 4.1) this retracing which produced a second time image with a delay of about 130 ns and an intensity of about 2.4 to 12% of the main image could not be repaired. Using the time-of-flight technique these two additional components in the time spectrum could be discriminated against so that they do not affect the evaluated results. Four sets of measurements could be completed, two at 2.38 MeV (nominal) primary neutron energy (resonance of oxygen at 2.350 MeV), and one at each side of the resonance (2.16 MeV and 2.72 MeV).

Fig. 15 shows a schematic diagram of the general arrangement used for scattering experiments at NAC. For the present case - back-scattering - the 150 deg detector position applies. A shield consisting of 30 cm of iron, of 15 cm of borated paraffin wax and of 3 cm of hevimet (high density metal) shielded the detector from the primary neutrons and provided at the same time a collimation with an aperture of diameter 2.5 cm at a distance of about 60 cm from the target.

The neutron beam was directed through the center of a sand-filled, cardboard box, of depth 50 cm and area of cross section 35 x 35 cm2. The distance between neutron target to the upstream face of the cardboard box was 79.5 cm.

Fig. 15. Schematic diagram of the experimental arrangement

Measurements were made with and without a scattering sample (to simulate a plastic, antipersonnel landmine) placed in the sand at various depths along the beam axis. The amount of sand in front of the sample (i.e. the "soil cover" or better its areal density) determines the size of the measured effect (sample-out vs. sample-in). Its thickness was known to within 1 cm. The scattering sample LX2 used in these experiments consisted of 0.130 kg urea enclosed in a cylindrical Lexan (plastic) container of mass 0.080 kg and outer dimensions 7.5 cm diam x 5.0 cm (see Table 5 for chemical compositions).

Table 5: Atomic and mass percents of elements in Urea and Lexan





















































The neutron detector (NE213 liquid scintillator, 5 cm diam x 5 cm) was placed (see Fig. 15) at 150 deg for the back-scattering measurements, with its horizontal coordinate 21.0 cm upstream of the front of the box.

Fig. 15 shows some examples of paths taken by neutrons from the source to the detector. Neutrons that travel directly from the source to the detector (through shielding or sand) are shown by solid lines. Paths of neutrons that were scattered (by sand or sample) from the beam axis into the detector are shown by dashed lines. A Monte Carlo simulation showed that in the case of the 150 deg position less than 3.10-6 of the primary neutrons penetrated the shield between detector and source.

A second, identical neutron detector (not shown in Fig. 15) was used to monitor the neutron output from the 7Li(p,n) source viewing the Li-target from a distance of 2.23 m, at an angle of 90 deg to the incident proton beam.

4.2.2 Electronics and data acquisition

Each detector was equipped with Link Model 5010 Pulse Shape Discriminator units which were operated at a pulse height threshold of approximately 180 keVee (keV electron-equivalent). The Link units selected neutrons efficiently and discriminated strongly against gamma rays.

For the monitor detector only neutron time-of-flight was measured (for events arriving at its 90 degree position). For the main detector (at 150 deg) neutron time-of-flight and pulse height were measured. These three quantities were digitized to 12-bit accuracy (4096 channels) by analogue-to-digital converters (ADCs) and recorded in event mode two-dimensionally for later off-line analysis.

4.3 Experiment simulation

4.3.1 Simulation parameters Soil simulation

The Monte Carlo neutron transport code MCNPv4B of LANL was used to predict the effect under the actual conditions of this experiment, too. A cylindrical volume, 55 cm deep with a diameter of 19.75 cm filled with quartz sand of density 1.56 g/cm3 was taken to represent the cardboard box (see 4.2.1) well enough. The same cross section libraries were used like in the explorative simulations (see Table 3). Explosive simulation

The anomaly in the ground was chosen to be in the shape of the sample LX2 (outer dimensions 7.5 cm diam x 5.0 cm) but with a uniform composition averaged over the data of Table 5 ( 13.4 g hydrogen, 113.8 g carbon, 31.4 g nitrogen and 50.9 g oxygen). The sand cover towards the neutron source was between 10.5 and 14 cm according to the energy dependent values of the experiment (see Table 6). Its thickness was known to within 1 cm. Detector geometry

The face of the detector was 15 cm off the surface of the soil and its shape a concentric ring with 14.75 cm inner and 19.75 cm outer diameter providing a simple cylindrical geometry. This results in an average scattering angle around 150 deg, depending on the depth of the sample. To find out about the penetration at 2.38 MeV also a fictitious case with a sand cover of 28.5 cm was simulated.

4.3.2 Simulation results

The results for the three experimental neutron energies are given in graphical form in Figs. 17 to 19, and in numbers in Table 6 under s-effect. As can be seen, the upper portion of the energy spectrum is strongly reduced (around 20 and more percent) in the sample-in cases (full line).

Fig. 20 predicts for a deep lying sample (2.38 MeV neutron energy, 28.5 cm sand cover) a meager 4% as difference between sample-in and sample-out. Due to the shortage in beam time (see 4.1) no experiment was performed with these conditions.

Fig. 17. Simulated energy spectra for a neutron energy of 2.16 MeV and a sand cover of 16.4 g/cm2. The full line is the sample-in case, the dashed line the sample-out case.

Fig. 18. Simulated energy spectra for a neutron energy of 2.38 MeV and a sand cover of 17.2 g/cm2. The full line is the sample-in case, the dashed line the sample-out case.

Fig. 19. Simulated energy spectra for a neutron energy of 2.72 MeV and a sand cover of 21.8 g/cm2. The full line is the sample-in case, the dashed line the sample-out case.

Fig. 20. Simulated energy spectra for a neutron energy of 2.38 MeV and a sand cover of 44.5 g/cm2. The full line is the sample-in case, the dashed line the sample-out case.

However, the primary experimental information is not in the energy domain (like in these simulations) but in the pulse height domain. To a first order a conversion can be made by assigning each energy bin a rectangular pulse height distribution down to zero energy. This was done in Fig. 21 for the 2.38 MeV case. As expected, the depression in the sample-in case extends now to lower equivalent neutron energies, at the cost of the size of the depression. Of course, both a high count rate and a good effect is desired. By multiplying the effect size with the square root of the counts a (relative) figure-of-merit can be found for each bias. This is illustrated in Fig. 22 where the simulated effect in the pulse height domain and its figure-of-merit are shown in dependence of the bias.

Fig. 21. Simulated pulse height spectra for a neutron energy of 2.38 MeV and a sand cover of 17.2 g/cm2. The full line is the sample-in case, the dashed line the sample-out case. The pulse height is given in equivalent neutron energy.

Fig. 22. Bias dependence of the effect (higher curve) and its figure-of-merit (lower curve) for the case shown in Fig. 21.

Although the effect is highest for big pulses (nearly 40% in Fig. 22) its figure of merit is low because it is based on few counts only. A (flat) maximum in the figure-of-merit can be seen at about 1.3 MeV equivalent neutron energy. So an optimum fraction of the pulse height bias value (given as equivalent neutron energy) of the primary neutron energy can be found, as done in Table 6 under "s-effect/fract.". Thus, in Table 6 "s-effect (En)" gives the simulated effect of hydrogen in the energy domain and "s-effect/fract." that in the pulse-height domain at the optimum threshold.

Table 6: Data summary of sample-in runs (see text). Run #





Enav [MeV]





soil cover [cm]















fnorm (Mon.)





av. current [nA]





neutrons per s


9.4 .104



duration [hrs]





time window





x-effect/at bias





counts: A/B












4.4 Data evaluation

A summary of relevant data of the sample-in runs of this ex-periment is presented in Table 6 together with the final results.

4.4.1 Normalization

Time-of-flight spectra obtained from the monitor detector were used to normalize data taken with and without the sample LX2 in position to the same incident neutron fluence.

Fig. 23. Normalized monitor spectra for sample-in (full line, run #40) and sample-out run (dashed line, run #39)

Fig. 23 shows a pair of spectra obtained in runs 39/40 resp., so normalized that the counts in the primary peak are equal. The
prominent features in each spectrum are two peaks, the primary peak and the small retrace peak (at channel number 200) that is attributed to malfunctioning of the buncher (see 4.2.1).

The loss of signals due to missing stop pulses, due to beam retracing and due to dead time affects the area of interest in the monitor and the main time-of-flight spectrum the same way so that, at least to a first order, these corrections are taken care of by using the monitor counts for normalization. Therefore, for each pair of runs, taken with and without the sample LX2 in position, normalizing constants were determined from the ratio of the integral of counts over the primary peak in the respective monitor time spectra. These (relative) normalizing factors are listed in Table 6 (1./ means that the sample-out run got adjusted with this number rather than the sample-in run dealt with there).

Although the method of MNBRP relies just on such a relative comparison sample-out vs. sample-in, a rough absolute normalization was established as well by means of the current integrator. Thus, knowing the target thickness a quantitative comparison with the simulated results is made possible (see also Table 6).

4.4.2 Data reduction and results

The data reduction was made off-line. The two-dimensionally (time vs. pulse-height) recorded events were replayed to find the proper time slice to select the appropriate types of neutron events.

At first, it was tried to find a time slice that gave the highest difference in counts between the corresponding sample-out and sample-in runs. Those are shown in Table 7. This choice fulfills all but one conditions for the correct choice: the width of the time window, a smooth energy dependence of its position, the decrease of the effect at the highest primary neutron energy and the order of magnitude of the size of the effect. All these properties fit. The effect could even be seen in the time domain without any background reduction by using a pulse height bias (see effect-t in Table 7). In one case even the statistical significance was checked and found to be 3.6 sigmas.

However, these time slices lie in the earlier hump which is expected to stem straight from the source and not in the second hump at lower channel numbers attributed to the back-scattering from the sand. Thus these slices do not fulfill the necessary condition to be in the correct time interval. No explanation was found up to now why there could be a difference between sample-in and sample-out for events occurring already in the first hump.

Table 7: Data summary of sample-in runs of preliminary time slice selection. Run #





time window





Enav [MeV]










effect w/o bias





effect/at bias





net error [%]










Fig. 24. Overlay of the modified neutron source spectrum (dashed line) over the measured scattered sample-out spectrum of run #55 (full line, see text).

To make sure that this necessary condition is fulfilled the correct time window was constructed from a measurement of the 2.16 MeV source spectrum (with a flight path of 439 cm) and laid over the scattered sample-out spectrum of run #55 (flight path of 92.5 cm) as shown in Fig. 24. This flight path correcting procedure did not take the time resolution and the dispersion in the scattering process into account making the time window of interest too narrow. As can be seen in this figure the centroid of the overlaid spectrum is above channel # 165. As can be seen from Table 6 the finally selected time window of run # 55 is from 163 to 168, in very good agreement with this prediction.

Table 6 shows the final time channel limits chosen for all four runs. These channel numbers are based on a full scale of the time spectrum of 256 channels. The position of the time window at the other energies were obtained after correcting for the different time-zero-channels due to the difference in travel time of the charged particles from the beam pick-off to the target.

Taking time windows as shown in Table 6 to cut the two-dimensional data one gets pulse height spectra for the four cases as shown (in logarithmic form) in Figs. 25 to 28. In all four cases is the full curve (sample-in) noticeable lower at higher channel numbers than the dashed curve (sample-out). This difference is the result of the shadowing by hydrogen, reduced by back-scattering contributions of the other light components of the sample (for pulse heights below a threshold that corresponds to about three quarters of the primary neutron energy). In principle, this counteracting effect in the pulse-height domain could be avoided by unfolding these pulse height spectra to energy spectra and applying energy discrimination rather than pulse height discrimination.

Like for the simulated spectra the numerical experimental effect depends on the bias chosen. Therefore, the experimental numerical results (Table 6) are given with their bias channel (x-effect/at bias) together with the actual number of counts and their number of sigmas to show their statistical relevance. This relevance is ample in all cases but not as good as that of the (fake?) result of Table 7.

Fig. 25. Pulse height spectrum with (full line) and without (dashed line) sample at 2.16 MeV primary neutron energy.

Fig. 26. Pulse height spectrum with (full line) and without (dashed line) sample at 2.38 MeV primary neutron energy (run # 38).

Fig. 27. Pulse height spectrum with (full line) and without (dashed line) sample at 2.38 MeV primary neutron energy (run # 40).

Fig. 28. Pulse height spectrum with (full line) and without (dashed line) sample at 2.72 MeV primary neutron energy.

4.4.3 Discussion of results Experimental results

It can be seen from Table 6 that the actually measured effect is noticeable bigger than the simulated one, despite the fact that no pulse height to energy conversion was done reducing the background and thus increasing the effect. This increase of the effect is the result of a very effective background suppression by means of "time slicing" which was not applied in the simulation.

As was discussed before, the MNBRP method depends on the ratio of back-scattered (fast) neutrons from the location of the anomaly with and without hydrogen in the anomaly. The back-scattered neutrons from the soil above the mine contribute background, "only". This background can be considerably reduced by an anti-coincidence that removes back-scattered events that come too early (and too late), like done in this experiment. For lower energy neutrons this is facilitated by their lower velocity (about 2 cm/ns for neutrons with 2.35 MeV).

So it can be expected that applying this method for deeper lying mines will yield a very good effect, too, although the reduced intensity will require longer measuring times. Simulated results

Because the time domain was not taken advantage of in the simulations reported here all results in chapter 3 can be assumed to be pessimistic, i.e. it appears that by using time-discrimination for background suppression even better results can be expected than suggested by these simulations.


5. Conclusion

It was shown that the MNBRP-method is much more sensitive for the detection of buried hydrogen-containing objects than anticipated if the time-slicing method is applied as described

(see Table 6).

This improvement based on background reduction by time-slicing requires timing capability, i.e. bunched beam or application of the associated particle method. The latter is handicapped by its limited source strength. The requirement of a narrowly bunched beam rules out presently available hand-held neutron generators. As a by-product of time-slicing also information on the depth of the buried object can be derived (from the position of the optimum time window).

Thoughts on a technical realization of a possible verification device are given in the Appendix.





I am very grateful to Prof. Frank D. Brooks from the University of Cape Town (UCT), South Africa, for his hospitality and his cooperation, and for making the experiment possible. The help of Dr. Saalih Allie with the experiment and also of my son, Bernhard is rightfully acknowledged. The Physics Department of UCT and the National Accelerator Centre in Faure made my stay there a pleasant experience. The expertise of Mr. Karl Springhorn in handling the Van-de-Graaff was duly admired.

APPENDIX. Thoughts on a technical realization of an apparatus for humanitarian de-mining using MNBRP

A.1 Boundary conditions

The feasibility of an application in humanitarian de-mining shall be discussed for one well defined situation, taking all obvious boundary conditions into account. As was discussed above, the sensitivity of the method relies, among others, on the ratio of back-scattered neutrons from below and in the volume of the anomaly to that above it. So, quite obviously, an optimization should be done for deep-lying mines because the others are much easier to detect. For ease of discussion I take here 25 cm below the surface as the position of the center for the back-scattered neutrons from below the anomaly. There, in the case of no mine, the primary neutron intensity will have been attenuated to somewhere between one forth and one third of the original intensity if one takes advantage of the resonance at 2.35 MeV as was done in this experiment.

The alternate resonance at 6.50 MeV has the advantage that neutron sources with much higher efficiency are available. However, 6.5 MeV neutrons are more difficult to produce, need more shielding and are faster by a factor of about 1.7, reducing the background rejecting capability by means of time slicing. In addition, neutron and gamma ray background could be prohibitive. So all this and a reduced hydrogen cross section speak against the choice of the 6.5 MeV resonance. This energy is, however, a fall back if no low energy neutron source of sufficient intensity is available.

A.2 Type of neutron source

To make the system acceptable for as many countries as possible, radioactivity in the neutron source should be avoided, if possible. Table 8 compares the specific neutron output (neutrons per steradian and picoCoulomb) for the production of monoenergetic neutrons at 0 deg with an energy spread of 0.10 MeV for use in connection with the oxygen resonance of 2.35 MeV.

Table 8. Properties of the best sources of 2.35 MeV neutrons with a 0.10 MeV energy spread


specific yield

beam energy


























a) radioactivity involved

b) by using triply charged ions the machine energy is correspondingly smaller

Recently, hand held neutron generators based on the d-D reaction have been developed. Unfortunately these sources cannot produce 2.35 MeV neutrons at 0 degree, but only at back angles. However, neutrons of energy 2.79 MeV which is another good choice (see Fig. 5) and of 6.50 MeV can be produced (see Table 9) by the d-D reaction.

Table 9. Neutron production based on the 2H(d,n)3He reaction.

av. En

d beam energy

emission angle

specific yield





















Using the 6.50 MeV resonance has the advantage that the d-D neutron source is about 1000 times as efficient at this energy. The most convenient neutron source would be based on a fully stopped deuteron beam of 0.1 MeV, bunched with an intrinsic time resolution of about 2 ns. The requirement of nanosecond pulsing for an efficient background rejection (reducing also the amount of shielding needed) rules out a light-weight solution, so other sources become compatible (see Table 8).

A.3 Geometry and detectors

Many reasons demand a well shielded neutron source; besides collimation is required, too. To save (precious) space heavy (metal) shielding is advisable. Materials with good shielding properties above about 1 MeV to 3 MeV must be selected. By making the neutron flight path from target-to-detector noticeable smaller than the path from the target to the mine (and back) time discrimination can be used for background suppression so that shielding becomes less important.

Let us, rather arbitrarily, assume that such a shielding/collimator can be restricted to a volume of 3 to 4 liters (a conical shape with typical dimensions of 15 to 20 cm). Towards the ground the shielding may be less thick for obvious reasons so that I take a somewhat optimistic target-to-surface distance of 6 cm. To provide the necessary local discrimination an outer collimator diameter of 1 cm at a distance of 5 cm from the center of the target appears to be an upper limit resulting in a solid angle of 0.031 sr.

Let us further make the plane of the detector face 15 cm from the surface, i.e. it cuts the axis 9 cm back from the target position. Then a circular detector ring with outside radius of 25.5 cm and inner radius of 20.5 cm is at about 150 degree with regard to an assumed back-scattering center at a depth of 25 cm. For radiation from this depth it has an accepting solid angle of 0.293 sr (getting bigger with smaller soil cover).

Making the detector ring from several (independent) detectors some kind of areal resolution could be introduced which should speed up the location of a mine. The use of deuterated scintillators could further improve the sensitivity of the instrument as the pulse height distribution of such instruments peaks at highest pulse heights. Thus the signal-to-background ratio is enhanced because the information is contained in the higher energy portion of the spectrum (typical pulse height bias bigger than one half of maximum pulse height.

A.4 Choice of electronics

Although it is premature to go into details it might be worth mentioning that the requirements are minimal. For the final instrument resolutions of 5 bits in the pulse-height domain and 6 bits in the time domain are sufficient, differential linearity of the ADCs is no issue so that even ultra-fast flash converters may be used. It might be worth checking if the associated storage of time and pulse height could be done in cache memories of standard CPUs. Using (low energy) d-D instead of p-7Li will reduce the gamma ray intensity possibly rendering neutron-gamma ray discrimination unnecessary (if it is necessary at all?). This would allow the use of cheaper detectors which, in addition, are more sturdy.

It is foreseeable that a fully automatic data reduction could be done online in parallel to the measurement because the amount of data to be handled is small. Thus also the required measuring time would be determined by the instrument itself.

Depending on the measuring procedure there might be the need to measure aside from the time and pulse-height spectrum (or spectra if individual detectors are used, e.g. to get areal resolution) additional parameters like source-to-ground distance and source intensity (e.g. via the beam current).

A.5 Strength of neutron source

In order to be useful the time horizon for a verification process is of the order of few minutes. Short measuring times cannot necessarily be reached by increasing the neutron source strength correspondingly, both for technical and biological reasons. With the help of the numerical results of Tables 6, 8, 9 and 10 it is rather straightforward to estimate the required source strength.

Table 10. Comparison of geometric parameters of this experiment and those of a dedicated device (as sketched in this paragraph) measuring the same spectra.



new device

Solid angle of source [msr]



Area of detector [cm2]



According to Table 6 it took a total time of 146 minutes to get a result to better than 2 sigmas. An increase of the source solid angle by a factor of 22 and the detector solid angle by 29 (see Table 10) results in a measuring time of 0.23 minutes under the same source strength conditions as in the experiment (250nA). It is unlikely that well bunched beam with an intensity of more than 5 microamps will be available so that a lower limit of 0.012 minutes is obtained for an ideal p-7Li source. According to Tables 9 and 10 d-D for 2.79 MeV neutron production is less efficient by a factor of 46 (using a gas target!) giving 0.6 minutes. Solid targets are less efficient by a factor of 3 to 10, resulting in a measuring time between 2 to 6 minutes for a LX2 sample as described with a 11 cm sand cover when using a low energy d-D neutron source with such a high current.

Using one of the sources of Table 8 allows shorter measuring times by one to two orders of magnitude at the expense of a much heavier generator (accelerator). Possibly, such a device could not only be used for verification but also for detection (of mines without metal).



[1] U. Rosengard, ed., "Application of Nuclear Techniques to Anti-Personnel Landmine Identification", report IAEA/PS/RC-799, Dec. 1999.

[2] AGM on "A Remotely Controlled Multi-sensor Platform for Humanitarian De-mining", report IAEA/PS/AG-1093, 2000.

[3] P.C. Womble, F. J. Schultz, and G. Vourvopoulos, Nucl. Instr. Meth. B 99, 757 (1995)

[4] F.D. Brooks, A. Buffler, and M.S. Allie, "Detection of plastic landmines by neutron back-scattering", IAEA Research Contract No. 10987, report IAEA/PS/RC-799, Dec. 1999.

[5] F.D. Brooks, A. Buffler, and M.S. Allie, "Landmine detection by neutron back-scattering", contribution to Crete-2001 Conference, G. Vourvopoulos, chairman, 18-22/06/2001, Crete, Greece

[6] M. Cinausero, "The EXPLODET Project. Progress Report 2000", report DFPD 01/NP/10, March 2001.

[7] M. Drosg, "Vorrichtung und Verfahren zum Nachweis verborgener wasserstoffhaltiger Objekte mittels Neutronen."('Procedure for the detection of hidden hydrogen-containing objects by means of fast neutrons'), Austrian Patent Application A1038/2000,G01N, June 15, 2000.

[8] M. Drosg, "Vorrichtung zur Strukturanalyse von sauerstoffhaltigen Medien bzw. Auffinden leichter Elemente in sauerstoffhaltigen Medien mittels schneller Neutronen."('Procedure for the analysis of oxygen-containing materials or the detection of light elements in such materials, resp., by means of fast neutrons'), Austrian Patent Application A1039/2000,G01N, Vienna, June 15, 2000.

[9] F.D. Brooks, A. Buffler, M.S. Allie, K. Bharuth-Ram, M.R. Nchodu, and B.R.S. Simpson, Nucl. Instr. Meth. A 410,319 (1998)

[10] A. Buffler, F.D. Brooks, M.S, Allie, K. Bharuth-Ram, and M.R. Nchodu, Nucl. Instr. Meth. B 173, 483 (2001)