GB2434443A

GB2434443A - Spectrophotometer

Info

Publication number: GB2434443A
Application number: GB0702411A
Authority: GB
Inventors: Ian Goodyer; Tracy Ann Nuttall; Rudi Labarbe; Nicholas Salvatore Arini; Andrew Seddon; Nick Stock; Gary Ross; Ralf Patz
Original assignee: Zinir Ltd
Current assignee: Zinir Ltd
Priority date: 2005-12-14
Filing date: 2005-12-14
Publication date: 2007-07-25
Anticipated expiration: 2025-12-14
Also published as: GB0702411D0; GB2434443B

Abstract

A spectrophotometer has two photodetectors which are displaced spatially in the direction of increasing wavelength in the spectrum. The diffraction grating of the monochromator of the spectrophotometer rotates such that at any given time the second photodetector receives light at a wavelength which is substantially greater than that being received simultaneously by the first photodetector at that time. The two photodetectors have different but overlapping spectral response characteristics. Thus the range of operation is extended, and data in two different ranges is processed simultaneously. The optical components are mounted to a chassis and the chassis is connected to a housing by shock absorbing mounts. The spectrophotometer is suitable for analyzing a sample in an environment where the spectrophotometer could be prone to detrimental effects. The position of the light source may be adjustable. The spectrophotometer comprises a portable unit containing the light source, detection optics and data processing means. The unit is connected to a probe to guide the light to and from the sample.

Description

SPECTROP1-IOTOMETER This invention relates to a device to identify and

quantify a substance, and more particularly the present invention relates to a spectrophotometer.

Spectrophotometry is the study of electromagnetic spectra. Spectrophometry involves the use of a spectrophotometer. A spectrophotometer is a photometer; a device for measuring light intensity, that can measure the intensity of light as a function of color, or more specifically, the wavelength of light. The most common application of spectrophotometers is the measure of light absorption, but can be designed to measure, for example, diffuse reflectance or the emission spectrum of a fluorescent or phosphorescent molecule.

The increasing demand for product quality improvement in different industries has led to the gradual replacement of time consuming analytical techniques such as gas chromatography, high performance liquid chromatography, nuclear magnetic resonance, mass spectroscopy and non-specific control procedures such as the measurement of temperature, pressure and pH by more specific and environmentally compatible techniques such as spectroscopy.

Spectroscopy has the advantage over many other analytical techniques in that it does not destroy the sample being analysed.

However, in general spectrophotometers are not robust and are prone to damage or malfunction if moved. In particular the spectrophotometer optics and monochromators are prone to damage due to movement/vibration of the instrument.

Furthermore spectrophotometers are sensitive to the environment in which they operate such as for example a dusty, corrosive or hazardous environment. As a result most spectrophotometers are to be found in the laboratory and if found outside the laboratory are located in an environment where their movement/vibration is minimal or controlled, such as part of a production line. Their use outside of the laboratory such as for example, in agriculture or in the analysis of biological or environmental material, or in an environment where the spectrophotometer may be subjected to movement such as vibration has been limited.

Spectrophotometers that are portable and/or can operate in a non laboratory environment have found Limited application. For example US patent no 6,483, 583 describes apparatus for Near Infra-Red (NIR) spectrometry for real time analysis of substances for agricultural use, wherein the apparatus forms an integral part of agricultural machinery such as a combine harvester or chopper.

International patent application no PCT/US99/l 8963 describes a spectrophotometer system mounted on a memory card of unitary construction which can be inserted into the side of a compact computer. The spectrophotometer of this system is a solid state spectrophotometer.

However such spectrophotometers to overcome the problems of the detrimental effect of movement/vibration on the instrument use opticslmonochromators which offer poor sensitivity and resolution and the spectrophotometers are limited in their range of capabilities. Further these spectrophotometers are often slow.

It is therefore an object of the present invention to provide a spectrophotometer which is capable of analysing a sample where the sample is in an environment where the spectrophotometer could be prone to detrimental effects.

Accordingly, a first embodiment of the present invention provides a spectrophotometer, the spectrophotometer comprising excitation optics, detection optics and a microprocessor in which the excitation optics comprises a light source, light focussing means and means for guiding the light to a sample to be analysed, and in which the detection optics comprises means for receiving modified light from the analysed sample, and means to convert the received modified light from the sample into an electrical signal for interpretation by a microprocessor, the microprocessor comprising a reference library and a sample determination algorithm for comparing the received electrical signal to a library of reference electrical signals to identify a compound or class of compounds in the sampLe whereby, in use, the light emitted by the light source emits light which is focussed by the light focussing means onto the means for guiding the light to the sample, the light guiding means guides the light to the sample, the sample modifies the received light and transfers a modified light signal indicative of the nature of the sample which signal is received by the light receiving means which conveys the signal to the light converter means, characterised in that the light converter means comprises a plurality of photodetectors, and in that the photodetectors are capable of detecting different wavelengths of light such that the information received can be formed to make a contiguous spectrum which is usable for interpretation of the modified light to determine the nature of the sample.

Advantageously, a spectrophotometer of this type provides a means for guiding the light to a sample to be analysed and a means for receiving modified light from the analysed sample hence there is spatial separation of the components of the spectrophotometer which are sensitive to movement/vibration from the sample which reduces the likelihood of these components being subjected to movement/vibration.

This separation also allows for the sensitive components not to be in the environment of the sample which may be detrimental to the spectrophotometer.

Furthermore this separation allows information to be obtained from environments where it would be impractical to operate a conventional spectrophotometer. Such environments may include but are not limited to for example, difficult to reach or confined spaces.

A further advantage of this spectrophotometer is that it comprises a plurality of photodetectors, which allow for fast processing of the spectral data.

Other desirable features of the present invention are described in the dependent claims appended hereto.

A second embodiment of this spectrophotometer comprises a moving monochromator which offer the possibility of a high resolution spectra.

A third embodiment of the present invention provides a spectrophotometer comprising an internal scaffolding upon which sensitive components are mounted and said scaffolding is connected to an external cover by means for removing or reducing shock.

Advantageously, a spectrophotometer of this type provides a means to directly shield the movement/vibration sensitive components of the spectrophotometer from an environment in which it may be subject to movement or vibration.

A fourth embodiment of the present invention provides a spectrophotometer which is a hand held or vehicle-, vessel-or aircraft-mounted instrument.

Embodiments of the present invention will now be described in more detail, by way of example only, with reference to and as illustrated by figures 1 to 54 of the accompanying drawings of which: Figure 1: Representation of the instrument. It is composed of three parts: 1) a control module containing the electronics and the optics, 2) a probe connected by a fibre optic, 3) an optional computer connected by a USB link. The instrument can be connected to several interfaces to increase its functionality including a memory card, a GPS receiver and a barcode reader.

Figure 2: Schematic representation of the principles of the NIR Sensor instrument.

Figure 3: Schematic representation of the sensor (top view). The case is divided in two compartments: the excitation optics (A) and the detection optics (B) Figure 4: A) Ray traces diagram of the light source (ES), spherical mirror (EM), lens (EL) and fibre input (EF). (B) Ray trace diagram for the light directly coupled from -the light bulb into the fibre.

Figure 5: Power emitted by the lamp over the whole spectrum (blue) and in the NIR (red: 900nm to 2300nm) as a function of the voltage applied to the liglit bulb Figure 6: Energy distribution at the position of the fibre optic A) along the X axis; B) along the Y axis.

Figure 7: EYE JC 6V/20W 20H/G4 light bulb Figure 8: Irradiance of a QTH light source as a function of the wavelength at 2850K Figure 9: Lampholder, bulb and mirror in the illumination optics. All units are in mm Figure 10: The Detection optics. Optical diagram of the spectrometer with the grating at the blaze angle. Red lines: I 300nm. Purple lines: 1 800nm. A) Top view B) Sagital view. The section plane in B) orthogonal to the plane in A) Figure 11: Grating angle versus wavelength of light reaching detector 1 and detector 2 Figure 12: Point Spread function at different image fields (top) and image of the fibre (bottom) on the detector I (left) at 1 300nm and on detector 2 (right) at I 800nm Figure 13: ff# and bandpass as a function of the wavelength of light Figure 14: Diffraction grating efficiency curve Figure 15: Orientation of the local reference frames with respect to the slit surface Figure 16: Position of the baffles around the opening from the fibre optic Figure 17: Position of the baffles around the detectors Figure 18: Spectral response of the two detectors Figure 19: Detector size and image size at the detector Figure 20: Fibre optic with SMA connector Figure 21: Attenuation of the fibre optic (dB/km) as a function of the wavelength (nm) for the LOH type fibre Figure 22: Optical diagram of the diffuse reflectance probe. A) Ray tracing for the illumination of an object situated at 10mm of the front lens of the probe. B) Ray diagram for the detection of the light from an object situated at 10mm from the front lens Figure 23: Encircled energy. Image of the 10mm object formed by the lens on the end of the fibre optic Figure 23: Rotation of the Diffraction grating by the stepper motor and associated mechanics Figure 24: Specifications of the right angle grear box PF2O-120 from OnDrive Figure 25. Mirror holder component Figure 26. Diffraction Grating holding component Figure 27. The component to fix and adjust the detectors. The cooling backplates are shown in green Figure 28. The completed optical design and the positioning of the other components.

A) The excitation optics B) Mirrors and mirror holder, C) Diffraction grating and holder, D) the slits, the filters, the detectors and their holder, E) the gearbox F) the stepper motor Figure 29. The internal scaffold and other internal components. A Fibre optic cable to excitation optics, B) Fibre optic cable from probe to detection optiocs, C) Gearbox, D) Stepped Motor E) Main electronics board F) The two battery packs.

Figure 30. External design 1 Figure 31. External design 2 Figure 32. External design 3 Figure 33: Block diagram of the electronic system Figure 34: Photodetector frontend and ADC Figure 35: Photodetector frontend with two ADCs Figure 36: Stepper motor control Figure 37: Distributed NIR system Figure 38: Example of NIR spectrum (sample: polystyrene). The vertical axis is log(l/R), where R is the reflectance. The X axis is the wavelength (nm) Figure 39: N1R Reflectance spectrum (bottom) and second order derivative spectrum (top) of a white ceramic tile Figure 40: Organisation of the battery packs Figure 41: NiMH battery discharge curve. [4] Figure 42. Main architectural components Figure 43. Use of the visualisation mapper Figure 44. Data processing pipeline Figure 45. Inheritance diagram for Process Objects Figure 46. A state diagram showing a dynamic model of the system describing the macro-states of the system (ovals), the transitions between states (arrows) and the events which trigger the transitions (captions on the arrows).

Figure 47. The core processes. Each process is implemented as a standalone executable (represented by boxes in the diagram) with conmuinication between processes (solid lines in the diagram).

Figure 48. Example of probability density function of distribution of the values of the cross-correlation functions between two spectra randomly selected in the database.

Figure 49. P.d.f. of the cross correlation function (red) and the auto correlation functions. The Xcor and Autocor functions were computed for mixture of two components combined in different proportions: a=l (dashed line and yellow line), a = 0.5 (solid line and green line), a = 0.1 (dotted line and blue line).

Figure 50: Model of substrate spectrum. Modelled as a sum of Gaussian peaks. The thickness of the line represents the variation of the background spectrum. The variation was modelled by using a 0.2% SD in the position, height and width of the peaks.

Figure 51: p.d.f. of the intensity of the background spectrum at 275nm. The red curve is the best fitted Normal distribution to the p.d.f.

Figure 52: Instrument noise. The noise is modelled by a normal random variable with a mean =0. The amplitude of the noise (SD) is lO (i.e. l0 times the maximum peak intensity in this example).

Figure 53: Computed proportion of target substance (alpha) versus the actual concentration of the target substance for a set of model systems. Log-log plot Figure 54: Model of spectrum. Yellow: Background spectrum. Red: Target compound spectrum (concentration = 1.) Blue: instrument noise. Green: "Measured" spectrum, which is the sum of all the other spectra.

It should also be noted that certain aspects of the drawings are not to scale and that certain aspects are exemplified or omitted to aid clarity.

As used herein the term "spectrophotometer" and "sensor" refer interchangeably to the spectrophotometer of the invention.

The term "monochromator" refers to an optical device that transmits a mechanically selectable narrow band of wavelengths of light chosen from a wider range of wavelengths.

The term "lens" refers to a device for either concentrating or diverging light The term "photodetector" refers to a sensor of light or other electromagnetic energy

EXAMPLES

Section 1. Overview.

The spectrophotometer is a near infra-red (NIR) based universal sensor that can accurately detect the presence of chemical compounds in a number of diverse environments. The sensor can also be used in order to monitor the biological state of products in the food industry (e.g. ripeness).

The sensor is composed of two main parts (Figure 1): 1. The sensor module, a completely independent and stand alone unit. It operates on batteries that can be recharged using a battery charger siniilar to those of mobile phones or laptop computers. It contains the light source and the detection optics of the sensor. It also contains the control electronic module and an on-board micro-controller chip that carries out the data analysis. It stores the recorded data in memory until the sensor is connected to a PC to download the data. The communication between the user and the sensor takes place via the numeric keypad and a monochrome LCD display.

2. An optical probe connected to the sensor module by fibres optic. It delivers the light to the sample, collects the reflected light and sends it back to the sensor module. Depending on the type of sample being investigated, the geometry of the probe will be different. Swapping the optical probes can be done very easily by the customer by disconnecting the fibre optic from the connector on the sensor module and connecting the new optical probe.

The sensor module is designed to have an open architecture so that it can be connected to many different devices for the collection of metadata and for communicating with the external world. The connectivity of the sensor is achieved by the presence of three connection ports: 1. PCMCIA port. The PCMCIA port transforms the sensor in a very versatile device as many commercially available standard PCMCIA cards can be connected to the sensor. The main purpose of these devices is to automatically record the metadata for every spectral measurement. For example, a GPS card can be used to record automatically the geographical position alongside the spectral information of the sample. A Bluetooth card allows the sensor to communicate with many different external devices, such as, for example, a bar code reader. Another application of the PCMCIA port is to connect the sensor to the internet via a mobile phone card. The instrument can then automatically upload/download data from a remote database or perform remote sensing.

2. USB port. The USB port is used to connect the sensor to a docking computer.

The computer is used to download the data recorded in the sensor, to upload new spectral fingerprints in the sensor or to upgrade the device drivers loaded in the sensor.

3. Memory Card port. A memory card port allows the insertion of memory chip.

These memory chips provide a larger capacity for recording data.

Furthermore, they can also be used to load new spectral fingerprints in the sensor.

Figure 2 shows a schematic representation of the sensor module. It is composed of three main components: I. The excitation optics. The light is focused on the core of the fibre optics. The intensity of the light bulb is under the control of the power supply. The light is delivered to the sample by the fibre optics.

2. The detection optics. The light reflected by the sample is collected by the fibre optics and is delivered to the detection optics. The detection optics contains a diffraction grating that separates the spectral components of the light. The different wavelengths are focused at different points in space. By placing a slit in front of the detector, it is possible to only record the intensity of a small fraction of the light spectrum. By rotating the grating, light of different wavelengths is scanned in front of the detector. Two detectors, with different spectral sensitivity, are used to optimise the detection limit and increase the data acquisition speed.

3. The electronic module. The electronics module contains the electronic circuits necessary to drive the system: stepper motor controller and pre-amplifler or the detector. It also contains the microprocessor that carries out the data analysis.

The description of the spectrophotometer is divided into four sections: * Section 1 The optical design. The design and optimisation of the optical set-up of the instrument. This section also indicates the tolerance in the design, therefore indicating how critical the mechanical design is. A proper optical design is essential to reduce the risk on the spectral resolution and sensitivity of the instrument. It will also reduce the risk associated with mechanical design by increasing the tolerance margin.

* Section 2. The mechanical design. Details the design of the internal scaffolding that will hold the optical components in place and allow them to be adjusted during manufacture. This section also describes how the internal casing and optical components are shielded protected from environmental shocks and how the components are pieced together. The design allows the moving diffraction grating in the spectrometer to rotate with reproducible and accurate positioning and for accurate positioning of the slits and detectors in the spectrometer. Finally various optional designs for the outer casing of the instrument are provided. Furthennore, based on the tolerance defined by the optical design, the mechanical design of the supporting structure of the optics allows a reduction in the risk associated with the optical system.

* Section 3. The electronic design. The design of the different electronic boards, the power supply system and other electronic components are described * Section 4. The software. Description of the associated software used with the sensor.

Section 1 Description of optical set-up of the portable sensor This section describes the optical layout of the instrument which is based on a modified version of the Czemy-Turner mount, using a planar line grating to achieve the diffraction of the light. Contrarily to the standard design, the system has two detectors in order to increase the acquisition speed and achieve a high sensitivity across the wavelength range.

Section 1.1 Optical Set up of the sensor Figure 3 shows a schematic representation of the sensor. The sensor module is divided in two compartments. The excitation optics (A) is composed of a light source and the conditioning optics to couple the light into a fibre optic, the detection optics (B) consist of two mirrors, a diffraction grating and two detectors with associated slits and filters.

The fibre optics delivers the light to the optical probe. The probe sends the light onto the sample and collects the reflected or transmitted light and feeds it back into another fibre optic that delivers the light to the entrance of the monochromator.

Section 1.1 a. Excitation optics Figure 4 shows a schematic of the set-up of the excitation optical train. The filament of the bulb is placed near the focus of a spherical mirror that collimates the light. The body of the bulb is placed parallel to the surface of the mirror. The rays close to the optical axis are reflected by the mirror and collected by the lens. The position of the mirror and of the bulb can be set independently. The lens focuses the collimated light onto the fibre input. The fibre is connected to the casing of the instrument by a releasable connector such as a SMA connector. The Lens not only focuses the light reflected by the mirror, it also focuses the light travelling directly from the filament.

The distance between the filament and the fibre is selected so that the image of the filament, for the direct light, is focused onto the fibre input. Furthermore, the distance between the filament and the lens is as short as possible in order to increase the solid angle of the light collected by the lens.

The Numeric Aperture (NA) of the fibre is 0.22, corresponding to aperture half angle of 12.7 . The core has a diameter of 600Mm.

Element Description Semi-Absolute Angle Toleran-Toleran-Toleran-Toleran-Toleran-Code Diameter co-ordinate (deg.) ce ce ce ce Tilt X ce Tilt Y (mm) (mm) Shift X Shift Y Shift Z (deg.) (deg.) (mm) (mm) (mm) _____ ES Filament X =0 0 -0.50 -0.50 N.A. -1.00 -1.00 Y=0 +0.50 +0.50 +1.00 +1.00 z=o EM Spherical Edmund Optics 5 0 0 -0.40 -0.40 -1.00 -0.28 -0.28 Mirror Code: 46-231 0 +0.40 +0.40 +1.00 +0.28 +0.28 -6.00 Diarn.: 10mm EL Lens Edmund optics 3.0 0 0 -0.40 -0.40 -1.00 -1.00 -1.00 Code: 45-077 0 +0.40 +0.40 +1.00 +1.00 +1.00 Curv. Rad. 4,71mm 10.00 Piano convex lens EF Fibre 0.3 0 0 -0.10 -1.00 -1.00 -1.00 -1.00 0 +0.10 +1.00 +1.00 +1.00 +1.00 20.63 Table 1. Co-ordinate of the element of the excitation optical train in the global reference frame.

Specifications of the excitation optics. Table 1 gives the co-ordinate of the centre of the first surface of the different optical elements in the global reference frame (defined in Figure 4. The Z-axis is along the optical axis. The X and Y axis are orthogonal to the optical axis. The Y axis is parallel to the long axis of the filament).

The origin of the reference system is the object (filament). All distances are in millimetres. Table 1 also reports the angle of the elements with respect to the main optical axis and the tolerance on the position and the angles of the elements.

The collection efficiency is determined by: the diameter of the fibre and the size of the image of the light bulb and the acceptance angle of the fibre (NA. 0.22). The coupling efficiency between the light bulb and the fibre in this design is around 4%.

As shown in FigureS, the total power emitted in the NIR is around 10 W and consequently approximately 160mW are coupled into the fibre.

The tolerances on the lateral position of the fibre are determined from the plot of the energy distribution as a function of position at plane of the fibre. The energy distributions along the X and Y axis are shown in Figure 6. These distributions were used to compute the tolerance on the X and Y position of the fibre. The tolerance on the other optical components was computed by setting the condition that the Root Means Square (RMS) radius of the Point Spread Function should not be larger than 0.5 mm for the rays directly travelling from the filament to the lens.

The position of the fibre along the Y axis is not critical. However, because the lens forms a very narrow image of the filament along the X axis, it is important that the fibre is properly aligned in factory with the optical train. The small size of the filament image is necessary in order to obtain a good coupling of the light into the fibre.

The SMA connectors are designed to give a very reproducible positioning of the fibre within the connector. Therefore, if the SMA connector is properly aligned with the optical train in factory and does not move subsequently when inserting a new fibre in the connector, the fibre should automatically be aligned. As the user will have the opportunity to plug fibres optic in and out of the sensor, it is critical that the relative position of the connector and the excitation train does not change during operation.

It should therefore be noted that: * A displacement of the filament along the Y direction when replacing the light bulb is not critical, as the image of the filament is significantly longer than the diameter of the fibre.

* A displacement of the filament in the Z direction (i.e. along the optical axis) is not critical, provided it is within the specification (i.e. +1-1.00mm).

* A displacement in the X direction is very critical. A small displacement of the filament will shift the image. As the paraxial magnification of the system is around 0.7, this means that a displacement of 500.tm of the filament will colTespond to a displacement of around 35Otm of the image. The mechanical design (described in section 2.1) provides a mean to adjust the position of the light bulb in the X-direction allowing the maximum coupling to be obtained The light source is a QTH (Quartz tungsten Halogen) light bulb (Figure 7).

Preferably, the light source is an "Eye" Light bulb from Iwasaki Electrics (catalogue number JC6V/20W) or equivalent. QTH lamps are popular visible and near infrared sources because of their smooth spectral curve and stable output and they do not exhibit the sharp spectral peaks that are characteristic of arc lamps exhibit. QTH bulbs also emit little UV radiation.

The Eye light source has the characteristics outlined in Table 2.

Manufacturer EYE Iwasaki Electrics Model JC6VI2OW 20W04 Product code 83677 Lamp wattage 20 W Voltage 6V Rated current 3.3 Amp Luminous output 280 lumen Colour Temp 2850 C Filament C-6 Base G4 Diameter 9 mm Length 33mm Average life 2000 hours Table 2. Characteristics of the Eye Light Source QTH lamps use a doped tungsten filament inside a quartz envelope. They are filled with a rare gas and a small amount of halogen. The QTH light source has a high irradiance in the wavelength range 900nm to 2.5im and can be considered as a black body radiator. This allows the theoretical prediction of the radiated power at different wavelengths as a function of the temperature of the filament. It is also possible to predict theoretically the temperature of the filament as a function of the voltage applied to the lamp. Because of the steep dependence of the emitted light power on the applied voltage, it is important to have a well stabilized power supply for the light bulb (see section 3.1).

The power emitted by a black body at a given wavelength can be predicted at different temperatures using the Plank formula: 2,r.h.c2 lO I = A.e. -s:.

e' ..1 Equation 1 where I is the irradiance (W/nxn) of the black body, A is the surface of the black body, e is the emissivity, h is Plank'sconstant, c is the speed of light, ) is the wavelength of the light, k is Boltzmann constant and T is the temperature of the black body (in Kelvin). The power radiated in different wavelength bands for different temperatures is shown in Figure 8.

It is possible to relate the luminous flux to the irradiance of the lamp by using the following equation: CD = 683. J!(1%,TK(4dA. Vible

Equation 2 where K() is the luminous efficacy at wavelength, I(,T) is the irradiance of the black body at wavelength and temperature T. The number 683 lumens/watt is based on the sensitivity of the human eye at 555 urn.

The luminous flux of the lamp is 280 lumen and the temperature of the filament is 285 0 C. Therefore by inserting Equation i in Equation 2, we predict a theoretical luminous flux in the visible equal to 290 lumen, which is close to the specified luminous flux. Therefore, the light bulb can adequately be described as a black body radiator.

The total electrical power P dissipated by the light bulb is equal to: elec = Equation 3 where U is the voltage applied to the lamp and R is the resistance of the tungsten filament. The electrical resistance of tungsten varies with the temperature according to the equation:

S

Equation 4 where p(T) is the resistivity of the metal (rim), T is the temperature (K), 1 is the length of the filament and S is the cross section of the filament.

The total power PBD emitted by a black body at temperature T is given by the Boltzmann equation: F81) = 6.t7.T Equation 5 where c is the emissivity of the body (0.4 for tungsten), a is the Boltzinann constant (5.67x104 W m2K). The electrical power defined in Equation 3 is dissipated in several ways: I. Black body emission from the filament 2. Black body emission from the enclosure 3. Conduction of heat from the filament to the enclosure This is summarised in the following equation that is set equal to Equation 3: 1 =Afi,.Efi,u.T4 +A e.cr.T k.(T-1j= where Afil is the surface of the filament, A is the surface of the enclosure, k is a constant related to the conduction between the filament and the enclosure, T is the temperature of the filament, T is the temperature of the enclosure. We can therefore derive the relation between the voltage applied to the light bulb and the filament temperature: = P(1 JAfi,.gfi,cr.T4 +A.a2' k.(T-T) Equation 6 From electron micrographs of a typical bulb filament it is possible to estimate the size of the filament and the enclosure (Table 3) Filament Radius 65 pm Length 29 cm Surface 1.18 10m2 Section 1.33 1 O m2 Emissivity 0.4 Enclosure Radius 4.65 in2 Surface 2.72 i03 in2 Emissivity 0.5 (estimated) Table 3. Sizes of elements of a typical bulb The temperature of the filament is around 2850K and under normal operating conditions, the voltage of the bulb is around 6.0 V and a current of 3.3Amp according to the manufacturer specifications. Therefore, k should be chosen so that T-2850K when V6.O V and l-3.3Amp. A value k = 7x10 WTK is therefore adequate. Figure shows the total emitted power versus the applied voltage of the light bulb.

It is also possible to predict the power emitted by the filament in the NIR as a function of the voltage applied to the filament. By solving Equation 6, it is possible to determine the temperature of the filament as a function of the applied voltage.

Furthermore, by using Equation 1 and Equation 2, it is possible to compute the power PNIR emitted in the NW at a given temperature: 2jrJLc2 l0 NIR (r) = j A.e. * dA.

900 e-l The PNIR is plotted in Figure 5 (the lower line). Because of the steep dependence of the emitted light power on the applied voltage, it is important to have a well-stabilised power supply for the light bulb.

The light bulb is installed on a socket providing easy removal by the customer when the bulb has reached its lifetime. The software monitors the lamp intensity, via the reference photodiode in order to determine when the light bulb is too old and requires replacement.

The reflector (EM) is a spherical mirror that is used to collimate the light emitted from the back of the filament and redirect it towards the lens. Its characteristics are

shown in Table 4.

Supplier Edmund Optics Model C46-23 I Outside diameter 10 mm Edge thickness 3.63 mm Material BK7 Coating Aluminium Table 4. Specification of the spherical mirror focusing the light of the bulb into the fibre optic The lens (EL) focuses the collimated light from the parabolic mirror onto the fibre input. It also focuses the light directly emitted by the filament onto the end of the fibre. The characteristics of the lens are shown in Table 5.

Supplier Edmund Optics Model 45-077 Shape Piano convex Diameter 6 mm Effective focal length 6 mm Centre thickness 2.5 mm Edge thickness 1.42 mm Radius of curvature 4.75 mm Material SF11 Coating None TableS. Characteristics of the lens (EL).

The excitation optics, the light source, the mirror (EM) and the lens (EL) are assembled as shown in Figure 4. The characteristics of the light bulb and the lens are shown in Figure 9.

Section 1.1 b. Detection optics The light collected by the probe is fed into a spectrometer contained in the main body of the sensor. The optical design is a modified version of the Czemy-Turner mount. A plane grating separates the different wavelengths (Figure 10).

The Czemy-Turner (CZ) monocbromator consists of two concave spherical mirrors and one plane diffraction grating. Although the two mirrors function in the same separate capacities as the single spherical mirror of the Fastie-Ebert configuration, i.e., first collimating the light source, and second, focusing the dispersed light from the grating, the geometry of the mirrors in the Czerny-Turner configuration is flexible.

By using an asymmetrical geometry, a Czemy-Tumer configuration has been designed to produce a flattened spectral field and good coma correction at one wavelength. Spherical aberration and astigmatism will remain at all wavelengths. It is also possible to design a system that may accommodate very large optics.

The sensor contains two detectors. This increases the sensitivity of the system (each detector has a different spectral response) and the acquisition speed (the two detectors record half of the spectrum in parallel).

In the Near Infra Red spectrum of compounds there are several interesting spectral regions that can be used to determine composition and concentration of individual chemical components: Short wavelength (700-1,500 nm). This is the spectral region of detector 1.

This region tolerates longer sample path-length (5-30mm) but yield poorer spectral resolution. The blaze wavelength of the diffraction grating (l300nm) is in this spectral region. The monochromator design has been optimised so that the light at 1 300nm hits the monochromator at the blaze angle.

Long wavelength (1,500 -2,500nm). This is the spectral region of detector 2.

This region requires shorter path length but provide better spectral discrimination. This region is more adequate to identify a component whose bands are not well resolved from the sample matrix or is present in a lower concentration because the absorption bands will be stronger. The monoebromator design has been optimised so that the coma aberration is reduced to a minimum in this spectral range. This will give the highest spectral resolution on detector 2. This is useful, as the spectra usually contain some narrow bands in this region.

Description of the optical set up

The schematic of the optical set-up is shown in Figure 10. The main components of the optical set-up are: 1. Fibre optic (DS1). The end of the fibre optic delivers the light in the monochromator.

2. A spherical mirror (DM2) used to collimate the light from the fibre optic onto the diffraction grating 3. A plane ruled grating (DG). The different wavelengths are reflected by the grating at different angles.

4. A spherical mirror (DM2), re-focussing the diffracted light onto two different detectors 5. Two long pass filters (DF1 and DF2) to remove the higher diffraction orders.

6. A slit (DSLI and DSL2) in front of each detector to only allow a limited range of wavelengths to reach the detector.

7. Two detectors with different spectral responses to detect simultaneously two different wavelengths.

The RMS spot Y for the optimum design at the detector I and detector 2 are shown in Table 6. The maximum acceptable value for the RMS spot Y is also reported in the Table 3. The critical parameter for defining the spectral resolution of the instrument is the spot size along the Y axis (where the wavelength separation takes place). The tolerances on the lens position and angles are determined by varying each lens parameter in turn, till the maximum RMS Spot Y is reached.

Optimum RMS Spot Y Maximum acceptable (Inn) RMS Spot Y (i.tm) Detector 1 33 45 Detector 2 26 35 Table 6: Detection optics. Optimum RMS Spot Y size and maximum RMS Spot Y size The tolerances assume that the Z-position (in the local reference frame) of the slit can be used as a compensator. This means that the slit must be installed on a translation stage that is adjustable in the Z direction (in the local reference frame) for alignment during assembly.

Once the tolerance on the lens elements are defined, a Monte Carlo simulation is run in order to study the combined effect of the variability on each element. The specifications on the RMS Spot Y that are reported in Table 7 represent the average RMS Spot Y size following the Monte Carlo simulation (20 runs). These numbers represent the typical resolution that can be expected from the monochromator, taking into account the combined effect of the tolerances on each optical element.

Detector 1 (I300nm) Detector 2 (lSOOnm) RMS Spot X size (tim) 455 154 RMS Spot Y size (rim) 45 34 Image X size (.un) 1720 720 Image Y size (l.Lm) 300 300 Table 7: Size of the diffraction spot and size of the image of the input fibre in the plane of the output slit. RMS Spot Y size calculated by taking the average of the RMS Spot Y size for 20 Monte Carlo simulations taking into account the combined variations introduced by the tolerance on all the optical elements of the lens Table 8 shows the specifications of the monochromator chosen for the sensor design.

By rotating the grating from 15.050 to 2.650 degree (with respect to the bisector of the deviation angle between the incoming and diffracted beam), the spectrum will be scanned in front of the two detectors. The detectors will record the intensity as a function of the grating angle. With appropriate calibration, the grating angle can be converted into wavelength (see Figure 11).

Bandpass (theoretical) 6.8 nm Deviation angle D 22.8 Diffraction Order (k) I Grating Groove density (n) 293.53 groove Imm Effective exit focal length L8 100 mm Linear Dispersion d)Jdx 31 nm / mm Grating Blaze wavelength 1300 nm Grating Blaze angle 11 40' F/# 3.0 Table 8: Monochromator parameters Plane Grating systems (PGS) spectrometers exhibit certain aberrations that degrade spectral resolution, spatial resolution, or signal-to-noise ratio. The most significant are: Astigmatism: It is characteristic of an off-axis geometry. A spherical mirror illuminated by a plane wave incident at an angle to the normal will present two foci: the tangential focus, F, and the sagittal focus, F5. Astigmatism has the effect of taking a point at the entrance slit and imaging it as a line perpendicular to the dispersion plane at the exit, thereby preventing spatial resolution and increasing slit height with subsequent degradation of optical signal-to-noise ratio. The mirror can have a cylindrical deformation (toroidal mirrors) to correct for astigmatism. However, cylindrical mirrors are expensive. As the astigmatism is in a direction perpendicular to the dispersion plane, this does not reduce the spectral resolution of the spectrometer. It only prevents imaging, which is not important in this application.

* Coma: The coma manifests itself as an elongation along the Y axis of the spot diagram. It cannot be corrected without employing asphenc optics. It is the result of the off-axis geometry of a PGS and is seen as a skewing of rays in the dispersion plane enlarging the base on one side of a spectral line. Coma may be responsible for both degraded bandpass and optical signal-to-noise ratio.

Coma may be corrected at one wavelength in a CZ by adjusting the angle between the mirror and the incident ray and by carefully choosing the curvature radius of the mirror (see below for more details).

* Spherical aberration: It is the result of rays emanating away from the centre of an optical surface failing to find the same focal point as those from the centre.

This camiot be corrected without the use of aspheric optics.

* Defocusing. Defocusing should not be a problem in a PGS monochromator used with a single exit slit and a PMT detector.

PGS systems are used off-axis, so the aberrations Will be different in each plane. In addition stray light can degrade the signal to noise ratio. Stray light can have several origins: * Defects on the surface of the optical elements: such as surface degradation, dust and particulate material.

* Imperfections in the position of the lines of the grating * Incorrect setting of the optical elements before the entrance slit of the monochromator. The cone if incoming light must match the NA of the monochromator, otherwise the light that will miss the grating will become stray light.

All the aberrations cannot be corrected in the Czerny-Tumer configuration. Therefore, the image of the input slit will become an arc of circle at the exit slit. This effect will increase when the size of the device is reduced because the mirrors will have to operate at larger incidence angle.

In the Czerny-Turner monocbromator, the difference between the incidence angle x (angle between the normal to the grating surface and the incident beam) and the diffraction angle (3 is constant. This is defined as the deviation DV: D =fl-a Equation 7 The incidence angle can be computed for a specific monochromator configuration and a wavelength using: lO.k.G.1t D a = asm 2.cos(.) Equation 8 Where k is the diffraction order, G is the groove density (groove 1mm) and ( is the angle of diffraction (angle between the grating normal and the diffracted light beam), X the wavelength of light (nm).

According the manufacturer specifications, this diffraction grating has a blazing angle of 110 40' at 1 300nm. The maximum efficiency of the grating is obtained when the blaze angle 0 is such that, for the blazing wavelength, the specular reflection angle for the angle of incidence is equal in magnitude and opposite in sign to the angle of diffraction. In equation, this translates in: 2.0 = fl-a Equation 9 Furthermore, the incidence angle a and the diffraction angle f3 are related by the grating equation: G.k.t = sin(a)+sin(fl) Equation 10 Where G is the groove density (groovelmm), k is the diffraction order, ?. is the wavelength of the light. Solving the system of equations: Equation 8, Equation 9 and Equation 10, it is possible to calculate that the deviation angle D that will result in the maximum grating efficiency for the blazing wavelength is 22.8 .

In order to minimise the coma in the system, it is necessary that the optical system respects the sine condition [1], [2]. It can be shown that it is equivalent to respecting the following conditions: 1. The angle between the chief ray and the optical axis of the first mirror should be as small as possible, i.e. try to be as paraxial as possible 2. The chief ray hits the centre of both mirrors 3. The angle between the chief ray and the optical axis of the mirror should be equal for both mirrors Figure 10 shows an optical diagram of a Czemy Turner monochromator satisfying these requirements.

* The minimum distance between the detector is 16mm. As the monochromator linear dispersion (Table 8) is 31 nn/mm, this means that the minimum separation between the wavelengths of the two detector is: 31 x 16= 496nm. Therefore, if the first detector receives light at l300nm, the second detector will have to receive light at I800nm.

* At 1800 nm, the angle between the chief ray and the optical axis of the first and second mirrors are equal (7.5 ). Therefore, the coma in the image formed on the first detector should be minimal. The spectral resolution of the second detector is optimised. Indeed, this is the wavelength range that gives spectra with the better resolved bands. It is therefore important that the spectral resolution is as high as possible for this detector.

* On the other hand, the angle between the chief ray at l300nm and the optical axis of the second mirror is 12.5deg. There will be some coma left in the image formed onto the first detector, which will reduce the spectral resolution at the second detector. This is less important as the spectra in this spectral region have usually less defined bands.

Figure 12 shows the spot diagram (point spread function) and the shape of the image of the fibre optic at 1 300nm (detector 1) and 1 800nm (detector 2) when the monochromator is at the blaze angle (11.4 ). Table 7 shows a summary of the root mean square (RMS) spot Y size (i.e. along the axis where the spectral resolution takes place) and the size of the image of the fibre.

The image of the circular fibre optic is a line. This is due to the astigmatism of the system. However, the only critical parameter for the determination of the spectral resolution of the system is the dimension of the image in the Y direction. The RMS Spot Y size is around 40jim. As the theoretical Airy disk for the system is l4pm, the optical design is not diffraction limited.

Table 9 shows the characteristics of the two detectors used in the sensor.

Spectral range Spatial Spectral (nm) Resolution Resolution lmage size (nm) _____ _____ (tm) _____ Detector 1 900 to 1600 300 9.3 I Detector 2 I 1400 to 2100 I 300 I 9.3 1 Table 9: Spectral resolution and detection range of the two detectors Table 10 shows the specifications of the monochromator used in the system.

Bandpass (theoretical) 6.8 nrn Deviation angle D 22.8 Diffraction Order (k) 1 Grating Groove density (n) 293.53 groove / mm Effective exit focal length LB 100 mm Linear Dispersion dA/dx 31 nm / mm Grating Blaze wavelength 1300 nm GratingBlazeangle 1l 40' F/# 3.0 Table 10: Monochromator parameters The linear dispersion of the monochromator is a measure of the ability of the spectrometer to separate the different wavelengths. A higher linear dispersion means a better separation of the wavelengths. The linear dispersion is usually expressed in nm/mm. The linear dispersion (d)Jdx) perpendicular to the diffracted beam at a central wavelength is given by: dA 106.cos(fi) dx k.n.L8 Equation 11 where L8 is the effective exit focal length in mm and dx is the unit interval in mm, In a monochromator, LB is the arm length from the focusing mirror to the exit slit.

In the case of a diffraction grating-based sensor such as the one described, the image of the entrance slit is not imaged 1:1 in the exit plane. The geometric horizontal magnification depends on the ratio of the cosines of the angle of incidence, alpha, and the angle of diffraction, beta, and the LB/LA ratio. Magnification may change substantially with wavelength.

w,_cs(a) L8 -cos(fi)LA Equation 12 where W is the entrance slit width and W' is the width of the image of the slit.

The bandpass B of the instrument can be computed from the entrance slit size (W = 0.2mm, i.e. the diameter of the central fibre optic In the bundle), the magnification (see Equation 12) and the linear dispersion (see Equation 11): B=W'.

Equation 13 Figure 11 shows the correlation between the grating angle and the wavelength of light reaching detector 1 and detector 2. There is a good linear relationship between the monochromator angle and the wavelength. Figure 13 shows the shows the f7# and the bandpass of the monochromator as a function of the wavelength in the working range of the monocbromator.

The above equations are based on geometrical optics, they ignore the wave nature of light. They assume that the image of a point object is a point. However, because of the diffraction effect of the pupil, it is not the case. The minimal image size of a point, called the diffraction limit, is defmed as the diameter of the Airy disk for the mirror in the same geometry: a = 2.44.

W.cos(fi) Equation 14 The result for geometrical optics and ray tracing should not be considered valid if the dimension of the image is found to be near or below the diffraction limit calculated with Equation 14. For the current optical set-up, the diameter of the Airy disk is approximately equal to 14tm.

Table 11 shows the characteristics of the diffraction grating (DG) used in the sensor.

Supplier Spectra-Physics Groove Frequency: 293.53 g/mm Grating Type: Flat Ruled Coating Aluminium Substrate TBD Catalogue Nb 53*64lR Nominal Blaze Angle: 11.40 Blaze Wavelength: 1.3.tm MaximumRuledArea: 3Ox3Omm Table 11. Characteristics of the diffraction grating A diffraction grating is an optical component that diffracts polychromatic light into its component wavelengths. The diffraction grating has been selected to optimally reflect the light in the wavelength range 900nm to 2.3im. When working in NW above 1.2im, and with very low groove density (less than 600 g/mm) ruled gratings are preferred to holographic gratings (see Figure 14).

The blaze wavelength is l300nm, for the first order diffraction. It should be noted that there is therefore also a blaze wavelength of 1300/2=650 nm for the second order diffraction. The long pass filters in front of the detectors will block the second (and higher) diffraction orders.

Two optical filters are placed front of the detectors in order to remove the higher diffraction order from reaching the detectors. The characteristics of the long wavelength filter (DF2) are shown in Table 12 and the characteristics of the short wavelength filter (DF 1) are shown in Table 13.

Supplier Edmund Optics Product reference number C54-665 Dimensional Tolerances 0.015" (0.38mm) Material Thickness 3mm P(d) Reflection Factor P(d) 0.91 Diameter 12.5 mm Cut off wavelength 770 nm 0.00 1% internal transmittance Wavelength 50% internal transmittance l000nm Wavelength internal transmittance >= 99% 1 300nm Density (gIcm) 2.75 Table 12. Long wavelength detector (DF2) Supplier Edmund Optics Product reference number C54-662 Dimensional Tolerances 0.015" (0.3 8mm) Material Thickness 3mm P(d) Reflection Factor P(d) = 0.91 Diameter 12.5 mm Cut off wavelength 660 0.001% internal transmittance Wavelength 50% internal transmittance 830 Wavelength internal transmittance >= 90% 900 >=99% 1100 Density (g/cm3) 2.94 Table 13. Short wavelength detector (DF1) In the following section, when referring to the slits (DSLI and DSL2) the axis refer to the local reference frame. The local Z axis is defined as the local optical axis at the position of the slit. The normal to the surface of the slit is along the local Z axis. The local Y axis is perpendicular to the local Z axis and in the plane of the Figure IOA.

The short dimension of the slit is along the Y axis. The local X axis is perpendicular to the local Z axis and perpendicular to the plane of Figure IOA. The long dimension of the slit is along the local X axis (perpendicular to the sheet of paper in Figure 15).

The orientation of the local reference frames is shown in Figure 15.

The magnification of the system is close to I. The core of the fibre input is 200pm, the size of the image is close to 300pm along the local Y axis. The optical system is astigmatic, the size of the image along the local X axis is 1720pm.

The image formed on the slit is not symmetrical. The elongation in the X direction is 6 times larger than the elongation in the Y direction for detector 1. The spectral resolution is limited by the spread in the Y direction. Therefore, the spatial resolution of 300pm will be the limiting factor for the spectral resolution. The slit should have a width of 300pm. The maximum extend of the image in the X direction is 1.7mm.

Therefore, the slit should have a length of 2mm. At detector 2, the slit should have a width of 300pm. The slit should have a length of around 1mm.

This is summarised in Table 14.

Width (pm) 300 300 Length (mm) 2 1 Orientation Length along the local Length along the local X axis in Figure 15 X axis in Figure 15

Table 14: Specification of the detector slits

The Z position (in the local reference frame) of the slit is used as a compensator. The Z-position (in the local reference frame) of the slits is adjusted during initial assembly for the precise alignment of the system. The slit should therefore be installed on two separate translation stages that can be adjusted separately (in factory) in the direction of the normal to the slit surface. Once the position adjustment is done, the slits are locked in place and should no longer move during the lifetime of the instrument. The specifications of the adjustment stage are shown in Table 15.

Travel distance Step precision (mm) (_ SIitDSL1 -2to+2 15 SIitDSL2 -1.5to+1.5 10 Table 15: Travel distance and precision of the adjustable stage for the position of the detector slits.

The position of the slits along the local Y axis does not require a factory adjustment.

Indeed, if a slit is shifted in the Y direction, this only results in a shift of the calibration curves of Figure 11. As each instrument will be calibrated individually, the only requirement for the tolerance on the Y position is that the light going through the slit hits the photosensitive surface of the detector behind the slit.

The NA of the fibre is larger than the NA of the monochromator. Therefore, the cone of light emerging from the fibre will be larger than the surface of the mirror DM1.

Some of this light could be reflected by DM2 directly into the detectors. In order to reduce the amount of stray light, some baffles are installed near the fibre input The baffles are shown in Figure 16 and their specification outlined in Table 16.

Shape Empty Cone Length 10mm Opening angle 16 deg Hole Diameter 5 mm Table 16. Characteristics of the baffles surrounding the fibre optic Further baffles are installed in front of the detectors in order to collect on the detectors only the light reflected by mirror DM2. The baffles are composed of a single triangular block 35 x 35 mm (i.e. parallelepiped cut along its diagonal). Two holes are drilled in the block. The holes have the shape of a truncated cone. The apex of the cone is 27mm behind the edge of the block. The opening angle of the cones is 15g.

The chief rays reaching each detector define the axis of the cones. The axis of the two cones are parallel and are 13.7mm apart. The axis of the cones are orthogonal to the surface of the triangular block. The position of these baffles are shown in Table 17 and depicted graphically in Figure 17.

Detector 1 Detector 2 Hole shape Open cone Open cone Opening angle (deg) 15 15 Position of cone apex 27 27 (w.r.t. block) (mm) Distancefromapexof 10 23.7 block Table 17. Specification of the baffles surrounding the detectors.

The system uses two InGaAs photodiodes type detectors. Each detector scans one half of the spectrum. Using two detectors has several advantages: * The sensitivity of each detector is opthnised for the partof the spectrum that it scans * The time taken to record the spectrum is reduced by a factor 2 The specifications of the detectors in the sensor are outlined in Table 18 and the sensor is designed to change-over between the two detectors at a wavelength of 1 600nm.

Detector 1 Detector 2 Manufacturer Hamamatsu Hamamatsu Model G8605-22 Detector G5852-21 Active area diameter (mm) 2 1 Number of TE cooling stages 2 2 Typical Dark current (nA) 0.15 25 Max. dark current 0.75 250 Photosensitivity (A/W) 0.9 1.2 Table 18. The specification of the two detectors used in the sensor The spectral responses of the two detectors are shown in Figure 18.

The sensor also has a reference photodiode placed in front of the QTH light source, near the fibre optic coiuiector. This photodiode is used to provide a reference measurement of the intensity of the light source. It also detects a fault in the light source. An example of photodiode is an RS instruments component number TSL252 or BPW34.

As mentioned earlier (Table 2) the sensor uses a JC6V/20W 20H/G4 light bulb from EYE (Iwasaki Electrics) which has a luminous output of 280 lumens. This corresponds approximately to a power of 2.9W in the visible. We can assume that all that power is radiated isoiropically. Assuming that the distance between the photodetector and the bulb is R = 6mm away from the light bulb. Let also assume that the photodector has a sensitive area with a radius of r = 1 mm. The fraction F of light intercepted by the photodiode is the ratio of the area of the spherical cap to the area of the sphere and will be: =0.007 2R Therefore, the amount reaching the photodetector will be: 2.9 x 0.007 = 20mW. This is the total power in the visible (all wavelength) reaching the detector. An optical density (OD) 4 filter will be required so that the reference photodiode is not saturated.

The slit defines the spectral resolution of the monochromator. However, the light leaving the slit still needs to reach the detectors. As the beams are divergent after going through the slit, the size of the spot on the detector will be larger than the size of the slit. One therefore must determine: * The size of the spot of light on the detector.

* The relative distance between the two photodetectors The distance between the slit and the photosensitive area of the detector is calculated by adding: * The distance between the photosensitive surface and the top of the detector cover * The maximum travel distance in Z of the slit (see Table 15) * Add one extra millimetre The relative distance between the two slits or the two detectors can be computed using the following equation: Ii 2 I \2 I d=px1 -x2j +ty -y2; +z1 -z2j Equation 15 The (x,y,z) co-ordinates are given in the absolute reference frame in Figure 19 and the results are shown in Table 19.

Distance between slits 16.42 mm Distance between centre of detectors 16.64 mm Table 19. Distance between the slits and the centre of the detectors The diameter of the detectors is 15.3mm and they are held firmly in a block. The minimum separation between the detectors should be 16mm. If the centre of the sensitive surface of the photodetector is placed on the optical axis, the distance between the centre of the detectors is 16.64mm.

The detector G5852-21 has a sensitive area of 1mm diameter. In Figure 19, the size of the image is in the Y direction (vertical axis) is 1.8mm. Therefore, the image is larger then the size of the photodetector.

In conclusion:

* The two photodetectors are assembled on the electronic board so that the distance between the centre of the detectors is 16.64mm.

* The slits are aligned with the centre of detector G8605-22 (short wavelength detector) * The distance between the slit is 16.42mm.

The position of the centre of the photodetectors in the global reference frame is summarised in the Figure 19.

Alternatively, it is possible to use a 3mm detector (e.g. the 05852-23). Using this detector will not change the position of the detector.

Section 1.2 Optical Set up of the probing optics Different geometries will be used for the opticaL probes, depending on the type of sample to be analysed and depending on the physical state of the sample (solid, liquid or gas). This example uses a probe that is suitable for diffuse reflectance measurements.

The light is guided from the light source to the measurement probe and back to the detection optic using two fibres optic. There is a fibre to deliver the light to the optical probe and a fibre to send the reflected light back from the probe to the detection optics. The two fibres are inserted in protective tubing, in order to make their handling easier and to protect them. The specification of the fibre optic cables are detailed in Table 20, an example is shown in Figure 20 and the attenuation of the fibre optic (dB/km) as a function of the wavelength (nm) for the LOH type fibre is shown in Figure 21.

Quartz glass fibres are used in optical spectroscopy (UV/VIS and NIR). The so-called LOH fibre (Low OH < 2ppm) has very few OH groups. This type of fibre is required for NIR applications, since otherwise the attenuation will be too high even on short distances. The fibres are inserted in SMA 905 connectors and are locked in place by screwing the protective cap, making the procedure to change the fibre straightforward.

The liglit led into the probe over the 6 x 200pm optical fibres. The light reflected (diffuse reflectance) by the sample is directed back to the sensor over the central 200pm fibre. The fibre bundles are protected by a flexible chrome plated brass outer protection tube for extreme conditions.

Supplier GetSPEC.com Model GS-R-200x7-IR-V2A Wavelength range 500-2100 nm Core diameter 20Otm Number of fibres in 7 bundle Numerical aperture 0.22 Length im Maximum temperature -100 to 400 C Connector SMA 905/ SMA 905

Table 20. Specification of the fibre optic cables

The optical probe is composed of a single lens. The same optical train is used for both illumination and collection of the reflected light. The optical parameters of the optical probe are summarised in Table 21. The optical system has a magnification of 1.66 for the illumination path and 0.6 for the detection path.

Illumination Detection Source size (nun) 0.6 1 Image size (mm) 1 0.5 (fibre diameter: 0.2) Transmission efficiency (%) 99.9 20 Table 21: Object and image size for the illumination and detection with the probe.

The transmission efficiency represents the fraction of rays entering the probe that will reach the image plane of the probe.

The probe has been designed for the observation of an object situated at 1 cm in front of the front lens. The probe forms a light spot of around 1mm diameter on the sample.

The same probe is used to observe the diffuse reflectance from the sample. The lens re-collimates the reflected light into the central fibre optic of the fibre bundle. Figure 22 shows the optical diagrams for the probe.

Optionally the optical assembly could be inserted in a cylindrical tube that extended 10mm beyond the front lens. This would allow the user to reproducibly position the front lens at the correct distance from the sample surface by placing the tube in contact with the sample. Furthermore, the tube would prevent external light from reaching the monochromator.

The tolerances of the optical probe were computed by determining the value of the optical element position and angle that will lead to a RMS Spot of 300pm for the detection optics. The encircled energy of the image on the fibre optic of the illuminated spot on the sample is shown in Figure 23. From this graph, as the fibre diameter is 200pm, it can be seen that the lateral displacement of the fibre from the centre position should be less than 100 jim.

The position of the fibre with respect to the lens should be adjustable during manufacture in order to obtain the maximum coupling of light into the fibre. It will then be locked in place. The adjustments needs to be made along the X,Y and Z axis.

The fibre should then move by less than 0.1mm during the lifetime of the probe.

The global co-ordinates of the optical elements of the probe are given in Table 22.

Element Description Semi-Absolute Angle Toleran-Toleran-Toleran-Toleran-Toleran-Code Diameter co-ordinate (deg.) ce ce ce ce Tilt X ce Tilt Y (mm) (mm) Shift X Shift Y Shift Z (deg.) (deg.) ___ _____ ___ (mm) (mm) (mm) PL Lens Edmund optics 3 0 0 -1.00 -1.00 Code: C45-077 0 +1.00 +1.00 Piano-convex lens 10 Material: SF! 1 Curv. Rad. 4.71mm Thickness 2.5mm PF Fibre optic 0 0 -0.100 -0.100 -1.00 -1.00 -1.00 input 0 +0.100 +0.100 +1.00 +1.00 +1.00 22.5 Table 22: Global co-ordinate of the optical elements of the probe. The origin of the co-ordinate system is the centre of the parabolic mirror.

Table 23 sununarises the transmission efficiency of the different parts of the optical system and estimates the amount of energy that will reach the photo-detector for each wavelength band.

Wavelength (nm) 900 1300 1750 2200 Excitation optics Excitation optics 4.02 l0 4.02 l0 3.96 10 3.72 10.2 Coupling into fibre Fibre transmission 0.99954 0.9977 0.99977 0.988553 Optical Probe: 9.99 10' 9.99 10' 9.95 101 9.50 10' Illumination Detection optics Optical Probe: 4.53 10 2.37 i0 1.15 i0' 9.32 10 Detection Fibre transmission 0.99954 0.9977 0.99977 0.988553 Monochromator 0.150356 0.231941 0.160682 0.139724 efficiency Total transfer 2.73 i0 2.20 l0 7.29 10 4.55 10 Efficiency Power P8 1.10 10 1.00 102 6.38 10 3.85 iO3 Light bulb Power (W/nm) Bandpassof 9 9 9 9 monoebromator (nm) Power in spectral 0.099 0.09 0.05742 0.03465 band (W) P = Pa.

Power reaching 2.71 10 11.98 10 J 4.19 l0 11.58 i04 detector in a spectral band (W) Table 23: Energy budget through the optical system. The table summarises the coupling and transmission efficiencies of the different parts of the optical train. The last section of the table gives an estimate of the light power reaching the detector.

The RS440-420 stepper motor has a step angle of 1.8' and the grating has to rotate by an angle of 12.4' in 500 steps. This corresponds to 0.0248' per step. An ideal direct drive gearbox would be 72:1. The right-angle gearbox PF2O-120 from Ondrive (Ondrives Ltd., Foxwood Road, Foxwood Industrial Park, Chesterfield, Derbyshire, England. 541 9RN) is a good solution. It has a gear ratio of 120:1 and reduces the height and has a bearing for the grating mount. Figure 24 summarises the features of the mechanical coupling system between the stepper motor and the diffraction grating.

Section 2. Mechanical Design Section 2.1. Internal scaffolding design The design of the chassis had to account for the mechanical clearances and adjustments that are needed during both manufacture and use. The design is this example was also constructed within the constraints of finding the most economical and practical shape to house and protect the critical functional elements, whilst allowing the best ergonomic layout options for the creation of the external appearance concepts.

As there are a number of components that needed to be accurately fixed a number of parts had to be accurately designed. These included: * Mirrors. A cavity machined block holds the two mirrors in place (see Figure 25).

The mirrors are brought in from the back and hit a lip at the front of the part that holds the mirror in position. A ring is then attached to the back of the mirror to hold it securely in place. This design has a number of advantages 1) the mirrors are well supported and protected ii) the mirrors are held firmly and accurately in position. This design does not require any adjustment of the mirrors while assembling the sensor but instead relies on precision machining of the metallic block holding the mirrors iii) it is easy to fit the mirrors without touching the sensitive front of the mirror. A nylon washer may be used between the ring and the back of the mirror to act as a shock absorber if necessary and depending on the final environment in which the instrument will be used.

* Diffraction Grating. This part has to be able to rotate during the operation of the sensor. Figure 26 shows a machined holder for the diffraction grating. The grating is placed in the holder in a similar way to the mirrors: the grating is brought in from the back in the holder and hits a lip at the front of the holder, so that is placed at the correct position. Two holes are drilled in opposite corners of the holder to attach a strap holding the grating into place from the back. A nylon push screw is placed at the top of the holder to fix the grating tightly in the block. The machined piece has a long shaft on the bottom that connects to the gearbox on the stepper motor allowing it to rotate (see Figure 26). Apart from the rotation in one plane this part does not move and no adjustment is necessary.

* Detectors, filters and slits. There are three pieces of the optical design that fit on this machined piece, i) the filters, ii) the slits and iii) the detectors and there are two of each of these components. The piece is shown in Figure 27. Monochromatic light bouncing off the second mirror first hits the filter and is then passed through the slit before hitting the detector. To adjust the optical layout the distance between the slits and the detector should be adjusted during manufacture. This part therefore contains a turn screw system whereby the slits can be moved (through a distance of -4.5mm) towards or away from the detectors. The detectors are equipped with thermoelectric coolers that require a heat sink. The photodetectors holder will also act as a cooling block. 6.7. Detectors and heat sink. The detector packages are 15.3mm diameter and the Peltier cooling system on the detector release around 2.25W of heat. This heat is dissipated by a heat sink which is clamped over the top of the detectors to hold them in close contact with the heatsink. This design consists in a block 4-5mm thick, with two holes (diameter 14.2mm), bored through to act as a fixing/clamping plate. The detectors fit into the block until they reach the flange on the detector. The block is then fixed to the heatsink, clamping the detector against it. In the optical design configuration the centres of combination DFI, DSLI & Detectorl lie on angle A = 62.9 , and DF2, DSL2 & Detector2 lie on angle B=61.41 . For practical reasons, it is more convenient to mount the filters and detectors on a single block with the slits and their adjustment stages placed in the space between the filter and the detector. For ease of manufacturing the assembly, the pairs of optical elements will therefore be placed in the same plane rather than at a small angle to each other. The line normal to the surface of the optical elements would then be at an angle of(A+B)/2, and the two assembles of filter, slit and detector would be parallel to each other.

The internal scaffold holds the three machines pieces shown in Figures 25-27 in place are all positioned relative to themselves on a fourth piece -the machined scaffold. The other components -the four circuit boards, keyboard, battery holders were positioned outside this casing. The design maximised the space available for electronics and is shown in Figure 28. As there is only 1.5mm distance between the fibre input area and the grating this piece had to be precision engineered the fibre holder is an unusual shape in order to accommodate the rotation of the grating and avoid a clash. The overall internal design with the positioning of the battery packs and electronic boards are showing in Figure 29.

Vibration is often a big problem in instruments like these. In this instrument the entire internal scaffolding (Figure 28) is suspended so that it is free to move relative to the outer casing by encasing areas of the scaffold in rubber.

Section 2.2. External Designs Using the internal scaffolding described above and in Figures 28 and 29 as a template three designs for the outer box are presented. These are shown as examples of what are possible and the final choice of design will depend upon the application to which the instrument will finally be used. In all three designs it is intended that the instrument is carried by a shoulder strap.

Design 1 (Figure 30) and design 2 (Figure 31) use the same layout for the internal components, only the shape of the external box is different. In design 3 (Figure 32), the position of the electronic bards is slightly different than in the other two designs. In all three examples the PCMCIA and other outlets were hidden behind a plastic covering to help with water resistance and all had a water resistant touchpad. It is possible to affix the probe to the side of the case in each example so that a hand is not required to hold the probe.

The keypad is made of a membrane, ruggedised and water ingress resistant.

Section 3. The Electronic Design The instrument consists of a combination of electronic hardware and embedded software, which is required to control the measurement procedure and analyse the measured data. The hardware has to interface to a range of communication interfaces, including USB, RS232, PCMCIA and Memory Cards. The data are measured via an optical interface using photodetectors and A/D converters. A stepper motor control circuit is required to rotate the diffraction grating. Figure 33 shows the block diagram of the electronic system.

Section 3.1. Electronic components in the instrument The sensor interface of the system consists of four main blocks: the light source, the photodetectors, the analog/digital conversion and the stepper motor control. Each of these will be discussed in detail.

The light source emits a wavelength range of 900 nm to 2500 nm (see Table 2) for details. The emitted power of the light source depends on the applied voltage. A stabilised voltage supply, capable of delivering 0.85 A over 2 s, is therefore required. Although the light source is powered by its own battery, it might be advisable to use a voltage regulator. The advantage would be a constant voltage (< 2%) to the light source. However, a small amount of energy will be lost during the voltage level regulation. A switch-mode power supply could be used to improve efficiency, but it could add in turn high frequency noise (f5j> 300 kHz typically) onto the light source power supply. This noise might not have an effect, as it is usually low in amplitude and might be too fast for the light source to react on. A selection of suitable voltage regulators is shown in Table 23. An alternative way is to power the light source directly from the battery.

This would remove the requirement for the voltage regulator, but emitted light power would depend on the state of the battery. Although this can be measured using the reference photodiode, it would increase the dynamic range of the measurement. It would also apply a larger voltage onto the light source as specified by the manufacturer reducing the life of the light source.

Manufacturer Name VoId L Shut-Tolerance Price in (A) h down (%) quantities (kHz ($) ) Linear Technology LTC1879 Adj. 1.2 550 Yes 2 4.26 (<100) Micrel M1C4690 Adj. 1 500 Yes 2.88 (l 00) Table 23: Voltage regulators.

Table 18 specified two photo detectors that are suitable for the detection of the required light spectrum. The expected power received by the photodiodes is estimated to be in the range of 0.01 1.tW to I mW. Both photodiodes have a sensitivity of around 1 A/W, resulting in a current in the range of 10 nA to I mA. A current-to-voltage amplifier performs the conversion of the photodiode current to a voltage level. This amplifier is located close to the photodiodes reducing interference from other sources. The photodetector and current-to-voltage amplifier require a very clean power supply. Assuming a supply voltage of 5V, the photodiode current can be translated into a voltage range of 50 lLV to 5V.

The dynamic range of the analogue signal is 100,000. By providing a 0.1% resolution for the signal the required resolution becomes lOOx 106, which is equivalent to a 27-bit resolution B=log2(EL+l] Equation 16 This resolution is difficult to achieve even with a low sampling frequency of I kllz. The effective resolution of an Analog Devices AD7732 is reduced from 24 bit to 21 bit at 500 Hz sampling frequency. Two solutions for this problem can be suggested. The first approach is to use a programmable gain amplifier (PGA) in combination with a 16-bit ADC. With a resolution requirement of 0.1% within each range, the quantisation step in the lowest range is 50 nV (1/1,000 of 50 ?V). An example of possible ranges is presented in Table 24. The resolution within range is detennined at the switching points between ranges. The programmable gain amplifier needs to provide a max mum gain of 1667 in order to amplify a 3 mV signal to a full-scale voltage of 5 V. Range Full-scale voltage Quantisation Resolution within range 1 3 mV 45.8 nV 0.09 % @50.iV 2 lOOmV 1.53 0.05%@3mV 3 5 V 76.3 j.tV 0.076% @ 100 mV Table 24. ADC input ranges.

The measurement of each sample is perfonned in 2 seconds and taking into account the time for the adjustment of the stepper motor and the programmable gain amplifier, the sampling frequency of the ADC needs to be above 3 kSps (see.Figure 34) The second approach is to provide more than one signal path, each with a different gain setting in the amplifier section. Each path, however, contains the same ADC for digitisation. Each amplifier in the amplifier -ADC signal path has got a different amplification in order to provide a properly scaled signal to its ADC. The ADCs run simultaneously and sample with a frequency of lkSps. This reduces the constraints on the ADCs, while increasing the demands on the microcontroller. The microcontroller in this solution has not only to control the multiple ADCs but also to decide which ADC output is the best scaled one. Figure 35 shows the principle of this approach using two ADCs.

The noise within the circuitry needs to be minimised. Good layout practise and close location to the photodetector is required. A possible multi-layer PCB could improve the performance.

Furthennore, it is possible to over-sample the analogue signal and to average the sampled data.

This would reduce the influence of the noise floor.

A third photodiode is used as reference signal and needs to be sampled at least once measurement. The same approach as for the two other photodiodes can be used.

A stepper motor is used to rotate the diffraction grating for each measurement sample. A gearbox is used to achieve a step size of 0.02 degree and the stepper motor controller needs to issue a maximum number of 500 steps to provide a total of a 10-degree rotation. A set of three position sensors are used to keep track of the diffraction grating. Two sensors are used to indicate that the stepper motor has reached either end of the allowed movement, while the third sensor provides the actual rotational data of the stepper motor (see Figure 36).

The power supply of the instrument consists of two independent batteries packs. The first battery is solely used for the light source and is discussed below. The second battery powers the remaining system. The voltage level used for most processors is 3.3V, which is sufficiently below the voltage of the drained battery (assumed to approx. 5V). A standard switch-mode voltage regulator can be chosen to efficiently supply power to the processor and its peripherals.

The analogue and the ADC section must be supplied separately to avoid interferences from the digital electronics. A linear regulator is advisable for this section combined with line filtering.

The voltage supplied to this section should be as high as possible to maximise the input swing for the ADC. Assuming a drained battery voltage of 5V, and linear low-dropout voltage regulator would confidently achieve a supply voltage of 4.5 V. It is advisable to locate the voltage regulator close to the analogue circuitry to avoid pick-up' on possibly long power supply lines.

Supplying a voltage of I 2V externally charges both batteries. Each battery has got its own build-in battery charger. This is to avoid early charge termination or damage to one of the batteries.

The charge of the batteries will be a normal charge, with trickle-charge when filly charged.

Description of the batteries are in section 3.2

A battery voltage monitor will be implemented into the design in order to alert the user about a low battery condition.

The NIR sensor system needs to be controlled by a microprocessor. The required high-level operations that will be performed by the processor are: 1. Record spectra 2. Spectrum analysis 3. User interface 4. Download data 5. Upload fmgerprint spectra Operations 1 and 2 define the spectrum acquisition time and are completed in 5 seconds.

However, operations 3-5 require interaction with the interfaces. The required microprocessor controls the acquisition, calculates the spectra and communicates with the peripheral interfaces.

There are a number of options available -in this embodiment we chose the distribtited' system but a discussion of the options follow.

A standard DSP processor is usually specialised to perform very fast calculations. A small amount of general-purpose input/outputs can be found on the new hybrid DSP processors. A single DSP processor can be used to carry out all operations required. The block diagram of the system is shown in Figure 33. The performance advantage of a DSP processor comes from its optimised arithmetic units, including hardware multipliers, and its ability to perform more than one operation at the same time. In order to achieve the best performance, a floating-point DSP processor should be used. These processors cost usually more than the fixed-point version.

Fixed-point DSP processors are only available up to 24-bit, therefore falling short on the 27-bit resolution requirement.

One big disadvantage of DSP processors is their lack of general 110 pins (GPIO). Table 25 shows a few DSP processors and their I/O provisions. A number of DSP processors have an interface for ADC's. However these interfaces are made to support audio codecs, which are unsuitable for this system.

Although the compiler for the DSP processor is frying to optimise the code during the compilation process, it can be assumed that the best performance is achieved by keeping the DSP architecture in mind when programming the processor. This limits the possibilities for later modifications of the algorithm, as specialist programming knowledge is required. A specialist compiler is also needed in order to compile the source code.

Manufacturer DSP Performance Power Comments consumption Texas TMS320C67 1200MFLOP 635mA @ Floating-point, 2 Instruments 11 S 200MHz 200MHz serial ports, HP!, 8

GPIO

Analog Blackfin 531 800MMACs l6OmA @ Fixed-point, 16-bit Devices 400MHz 400MHz multiplier, RTC, 15 GPIO, SPI Analog SHARC 198MFLOPS 670mA @ Floating-point, 2 -Devices @ 66MHz 50MHz serial ports Table 25: Selection of floating-point DSP processors.

An alternative approach is the use of a general-purpose 32-bit microprocessor. The main advantage of a general microprocessor over a DSP is the increased number of I/O pins, its support for peripheral interfaces and the wide availability of operating systems. Although an operating system (OS) is not required for the acquisition of the spectra, it could provide support for the interface peripherals with ready-made functions. The use of a general-purpose 32-bit processor has a number of advantages,including being better equipped for control tasks. The modification of the algorithm at a later date does not require architectural knowledge, as only one operation can be performed at one time.

Furthermore, the availability of a number of embedded OS supports the programming of the interface peripherals as well as future extensions. Many embedded OS have interface drivers already build-in. The block diagram of such a system is shown in Figure 33 and a selection of 32-bit processors is shown in Table 26.

Manufacturer Processor Performance Power Comments consumption Intel PXA255 273mA MMC, PCMCIA, UART, 400MHzj GPIO, RTC, USB, 16-bit multiplier, 40-bit accumulator Motorola 68k 8MIPS @ 324mA @ GPIO, Serial /Coldfire 33MHz 33MHz Table 26: A selection of 32-bit general-purpose processors.

The programming of the 32-bit processor can be done with and without an operating system. The advantage of a non-operating system design is that the real-time control of the peripherals can be achieved easily. However, an embedded OS offers the advantage of build-in routines for some of the peripherals, which can be adopted for NIR sensor system. The OS is also better in hiding the details from the programmer, which is useful for later modifications. A possible disadvantage is that unless a real-time OS is used, the control of the ADC and stepper motor might cause a problem. Furthermore, an embedded OS requires a boot time on power-up, therefore, introducing a delay on switch on. This delay depends on the OS and can take up to a few seconds. Embedded operating systems include Embedded Linux, VxWorks, LynxOS, eCos and Windows CE.

The use of an operating system would ease future development by hiding the low-level complexity to the developer. Especially an operating system like embedded Linux could be very beneficial in this respect. It provides a number of interfaces as functions within the OS itself Furthermore, it provides a very familiar programming interface to the high-level programmer.

An alteration of the algorithm is therefore much easier. There are other OS available, but they do not offer such a familiarity in the programming interface. An exception might be the Windows family of embedded OS, Windows CE and Embedded Windows XP. However, as many other OS, they incur a purchasing and/or a royalty cost. This includes, with respect to the purchasing costs, also some of the commercial available embedded Linux distributions. These are usually covered by purchasing a development kit.

The disadvantage of embedded Linux is that it is not a real-time OS. This disadvantage only affects the recording of the spectra, not the spectrum analysis and the interface functionality.

Given a sample throughput of 1 kHz, with a possible sample frequency of greater than 3 kHz, the timing of the embedded Linux OS is likely to be sufficiently real-time.

A third alternative to using a DSP or a 32-bit general processor is to avoid the dependency of the recording on the timing of the OS and use a distributed system, where a small secondary processor performs the recording of the spectra. This is beneficial for the physical layout of the unit, offering a smaller opto-electronic interface board.

A distributed processing system, where the ADC and stepper motor control is performed in a second processor, can have some advantages over a single processor system. It allows to physically locating the ADC closer to the photodiodes, thus improving the system's performance. It furthermore allows the main processor to use an OS that is not performing in real-time.

The physical separation of the system into a processor board and a set of daughter boards, i.e. analogue board, acquisition board and power supply board, offers the ability to upgrade individual system components without affecting others. Furthermore, each daughter board can be optimised in performance, i.e. noise, interference, etc. This approach also allows the purchase of a ready-made processor board that is bundled with an embedded OS such as Linux, thus saving development time. A later upgrade with a purpose build board is then still possible.

Figure 37 shows such a system and a selection of suitable processor boards is shown in Table 27.

Manufacturer Board OS Interface Strategic Test Trito-Linux PXA25S, upto 32MB SDRAM, upto 16MB LP Flash, 12C, 4 serial ports, LCD controller, PCMC!A/CF, RTC, lOMbit/s Ethernet, Silicon serial number, USB, credit card size, upto 30 1/0 Arcom Viper Linux / PXA255, upto 64MB SDRAM, upto 32MB Windows Flash, 12C, 4 serial ports, LCD controller, CE CF, RTC, 10/100 Mbit/s Ethernet, Dual USB, Table 27: Processor boards.

An additional advantage of using a separate analogue and acquisition board is that cost and development time can be keep lower, as possible optimisation iterations of the analogue layout can be limited to a relatively small board. The split into an analogue and an acquisition board offers the advantage of a smaller analogue board which can be located close to the photo detectors, thus reducing the effect of interferences. The analogue board will therefore contain the signal conditioning circuitry to amplify the signal. The ADC(s) can be located on either the analogue or the acquisition board.

Furthermore, such an approach would allow interfacing the acquisition board via a RS232 interface to a PC. This way the sampled data can be transferred to the PC and an early verification of the measured data can be performed on the PC Of the three possible options, DSP processor, 32-bit board or a distributed system the preferred solution is the distributed system. This has the most advantages. It allows early testing of the analogue front-end as well easy upgrade of the analogue, processor and power supply section.

The system also contains an embedded operating system. The requirements for interfaces and future expandability of the hardware (PCMCIA) and Software (PCMCIA driver and algorithm) make the use of an OS an efficient solution in the long term.

The electronic module is composed of 4 boards labelled: 1. Main board with the power supplies and the connectors.

2. Processor board. Its size is 68 x 37 mm with a maximum installed size of 7.3mm 3. Acquisition board with the data acquisition electronics.

4. Analogue board containing the photo detectors.

The analogueboard is the optical front end of the electronic circuit. It carries the photodetectors and the pre-amplifier. The system uses two detectors. The detectors are InGaAs photodiodes.

Each detector scans one half of the spectrum. A pre-amplifier is placed close to the photo-diode in order to amplify this small signal as soon as possible in order to minimise the effect of the electrical noise. The position of the two photodiodes is dictated by the optical design (see previous section). The two photodetectors should be placed as close as possible to each other on the analogue board, in accordance to the optical design. The analogue board should be connected to the main electronic board using flexible wires so that the position of the analogue board can be adjusted. The sensor also has a reference photodiode placed in front of the QTH light source, near the fibre optic connector. This photodiode is used to provide a reference measurement of the intensity of the light source. It also detects a fault in the light source. The amount of light hitting the Hamamatsu photodiodes is expected to be in the range of 1mW to O.OIuW. As the photodiodes have a photosensitivity close to 1 A/W according to the specification sheets, this gives an electrical current in the range of ImA to 0.OluA. A pre-amplifier is placed close to the photo-diode in order to amplify this small signal as soon as possible in order to minimise the effect of the electrical noise. The 2 photodetectors will be placed on a heat sink in order to dissipate the heat generated by the Peltier cooling. Each photodetector generates around 2.25W of heat. It is possible connect the heat sink to the external chassis to prevent the heat from being dissipated inside the instrument casing.

The acquisition board contains the different module necessary to control the different components composing the detection module (e.g. stepper motor). The ADC converter digitises the output of the two photodiodes into a digital signal that is processed by the microcontroller chip. This section describes the dynamic range of the signal, the resolution required to digitise the signal, the precision and accuracy of the measurement. To illustrate the purpose of this board Figure 38 shows an example NIR spectrum. In this example the Y-axis is Y = logio(j) where where I is the intensity of light measured by the detector for the sample and Jo is the intensity of light measured by the detector for the refererice material (ideally, the reference material will give a intensity close to the maximum intensity). These units are used because there is an almost linear relationship between the concentration [C] of an absorbing compound and the logio(lfR).

A low value on the Y corresponds to a large intensity of light reaching the detector. A high value on the Y axis corresponds to a low intensity of light reaching the detector. In Figure 38, the maximum is 2 units (at 2460nm), i.e. there is a 100-fold decrease in the intensity of light at the peak compared to the baseline light intensity (la) at that wavelength.

The absorption of light by the sample is proportional to the concentration of the absorbing substance and on the thickness of the sample (optical path length). For example, if the sample concentration was doubled, scale on the Y axis would be doubled, i.e. the peak at 2460nm would become 4 units.

The variation of light intensity can in some cases have a dynamic range of 5 orders of magnitude as while the spectrum is being scanned, the light intensity can vary from full transmission (ImA at the detector) to nearly complete extinction (lOnA). This defmes the full dynamic range of the signal to be detected.

The resolution defines the minimum variation of intensity that is detectable. This defines the number of different levels that exists within the dynamic range. For this application, the instrument does not require the same absolute resolution across the full dynamic range. For example, while the instrument needs to measure the current in the range 10 to 11 tA with a resolution of 23nA, it does not need to measure the current in the range 0.9 to ImA with the same resolution, it would only require a resolution of 2.3pA.

In other terms, the resolution of the instrument is not defined in absolute values, i.e. not defined as a resolution with steps of I OnA in the range 1 OnA to 1 mA. Rather, it is defined in terms of a percentage of the current to be measured. A resolution of 0.1% across the range is acceptable.

This means that in the range 0.9 to imA, the instrument can digitise the signal in steps of imA x I0 = ijiA. In the range 10 to 1 IpA, the instrument will digitise the current with steps of l0p.A x 10 = lOnA (see Table 28).

Intensity Step size for I digitalisation (Resolution) imA lpA lOOpA lOOnA lOpA lOnA IpA lOnA lOOnA lOnA lOnA lOnA Table 28: Specification of the required step size for the digitalisation of the signal as a function of the signal intensity.

The digitalisation steps (resolution) have been chosen so that the steps are slightly smaller than the amplitude of the instrument noise.

For the lowest signal intensity, it would not make sense to digitise the signal in smaller steps than lOnA. Indeed, the noise from the detector is lOnA, therefore, the precision on the measurement is at least I OnA. There is no point digitismg the signal with a step size significantly smaller than the amplitude of the noise.

The instrument is used to obtain data on the concentration of a target substance in a sample. One would like to have a good precision on the measurement of the concentration. If the experiment is repeated several times, one would expect to obtain a reading of the concentration [C] with a precision 0.1%.

This is the precision on the measurement, i.e. how close different readings should be when the experiment is repeated several times. The precision describes how close the readings are. It does not specify whether the readings should be close to the correct value or not. The variability of the readings can come form many different sources: * Instrument variability (thermal noise, photon noise etc) * Sample variability * Operator variability Ignoring variation from sample or operator we can therefore define the following test for the precision of the instrument: The same sample is read five times by the same operator at 1 minute interval. The error bar on the determination of [C] should be less than 0.1%.

Let's now convert the specification on the instrument precision into the precision for the measurement of the light intensity. Firstly, the concentration is proportional to Y = log,o(IoII).

Furthermore, in a practical experiment, one usually tries to measure the concentration with a Y around I unit. Therefore, to simplify things, we can in fact specify the precision on the determination of [C] by saying that: If the same sample is read five times by the same operator at 1 minute interval, the error bar on the determination of V should be less than 0.001.

This means that Y must be measured with a precision Y 0.001 i'his is a short term precision as it only takes into account variation of the intensity during a short time scale, of the order of a few minutes. There is also the long-term precision, i.e. if the instrument read the same sample at a few days interval will it give the same readings? The long-term precision will depend on the calibration protocol and the periodicity of the calibration protocol.

The relation between the standard deviation AY on Y and the standard deviation A! on the measured intensity I is: = with I(L) ala! In(IO) =ii))in(1O), Therefore: t:Y = iio)fTJ Therefore: = in(iO)y = In(1O). iO = 2.3. iO In conclusion, the relative precision on the measurement of the light intensity should be: -i=2.3.lO

I

Equation 17 If the same sample is read five times by the same operator at 1 minute interval, the standard deviation on the determination of intensity I of the light should be less than 2.3 i -I. The Table 29 summarises the short-term precision that is required for the measurement of the current at the photodetector for different light intensity (and therefore different current at the light detector).

Intensity Standard deviation Precision on Y I of intensity imA 2.3pA 0.00 1 100iA 230nA 0.001 lOp.A 23nA 0.001 lp.A lOnA 0.004 lOOnA lOnA 0.04 lOnA lOnA 0.4 Table 29: Precision on the measure of the intensity at the photodetector The accuracy of an instrument specifies whether the instrument measures the true value of the parameter that is determined. The accuracy is determined by measuring samples of known composition and checking whether the reading form the instrument is biased or not.

The accuracy is a different concept from the precision. For example, if the true value to be measured is 5.0, an instrument that returns a result 4.5 0.01 will be precise but inaccurate (there is a 0.5 unit bias). The accuracy of the instrument depends on the calibration protocol of the instrument, as well as on the availability of good calibration samples.

Another important parameter is the long-term stability of the instrument. If the instrument baseline drifts with time, there will be a bias introduced in the measurement with time.

Therefore, while the measurement can still be precise, they will be inaccurate after some time.

The instrument accuracy will be determined by the calibration protocol. The calibration of the instrument is an important process, as it will determine the accuracy of the instrument and the long-term stability of the readings. The calibration process can be divided in two steps: 1. Instrument calibration. This calibration is used to relate the current intensity measurement done by the instrument to the intensity of light that is reflected/transmitted by the sample.

2. Experiment calibration. This calibration relates the intensity of light reflected/transmitted by the sample to the parameters of the physical or chemical process that are to be measured.

Only the first type of calibration concerns us here. The instrument calibration is used to correct for different artefacts that are introduced by the instrument: * Fluctuation of the intensity of light bulb. As the light bulb and/or the batteries age, there will be a change in the intensity of the light. This variation must be accounted for.

* Variation of the transmission of the optics. As the different optical components age in the instrwnent, there will be a variation in the transmission efficiency of the instrument. This must be accounted for.

* Wavelength dependence. The different optical components do not transmit the light with the same efficiency for every wavelength. Therefore, the transmission efficiency of the system will show some wavelength dependence. l'his must be accounted for by the calibration process.

* Variation of the sensitivity of the photodetectors. As the photodetectors age, they can show some variation in their efficiency (which could even be wavelength dependent). This must be accounted for in the calibration process.

* Variation of the gain and offset of the amplifier. There can be a variation of gain and offset of the amplifier from day to day. This must be corrected by the calibration process.

There are therefore two different types of artefacts to correct: 1. Wavelength dependence 2. Variation of absolute intensity with time The two artefacts are corrected by the calibration protocol. The calibration will be carried out at least once a day, before starting a new series of experiments. The instrument is calibrated by measuring the reflectance of a white tile. The white tile has a nearly flat reflectance spectrum (see Figure 39). Therefore, the gain and offset of the instrument should be stable within 0.1% between two calibrations, i.e. for a period of 24 hours. This is the long term stability of the instrument.

Because of the wavelength dependence of the instrument sensitivity, the instrument will not record a flat spectrum for the white tile. The spectrum of the white tile will be used to correct all the other spectra recorded by the instrument during that day: i, (a) i_(a) tile Equation 18 Where 1(A.) is the corrected intensity of the sample at wavelength A., I(A.) is the intensity measured by the instrument at wavelength A. and l(A.) is the intensity measured for the tile at wavelength A..

If there is a variation of intensity of the light bulb between two calibrations with the white tile, this can also be accounted for by using the measure of the intensity of the light bulb from the reference photodiode: I 1 nomJ -Equation 19 where I is the sample intensity that has been normalised to accountfor the fluctuation of light intensity, Icor is the intensity corrected according to Equation 18, ito is the intensity of the light bulb as measured by the reference photodiode at the time of the calibration with the white tile and It is the intensity of the light bulb as measured by the reference photodiode at the time of the measurement.

The instrument will read the intensity of the reference photodiode before and after recording the spectrum. If the intensity has fluctuated more than a given threshold, the spectrum will be rejected. The instrument will monitor the intensity of the light bulb during a few seconds. If a significant fluctuation is observed, the instrument will issue a warning message to the user to replace the light bulb or the batteries.

The reference photodiode should measure the light intensity with a precision of 0.1% (short-term precision).

The sampling frequency should be at least 1kHz (see Table 30) However, if the ADC is sampling at a higher frequency, the data points can be averaged to reduce the noise.

Sampling frequency At least 1 kHz per channel Dynamic range 16 bits Amplification Variable gain Number of channels * measurement photodiodes: 2 channels with at least 1kHz sampLing reference photodiode: I channel at 1Hz Table 30. Sampling frequency of the spectrophotometer The wavelength selection is done by rotating the grating through an angle of 10 degrees. The rotation of the grating is operated by a stepper motor. The rotation of the stepper motor is under the control of the Stepper motor control board.

Each detector will record 500 data points during the rotation of the monochromator. The number of data points is defined by the spectral resolution of the monochromator.

The angle of the diffraction grating determines which wavelength of light is sent onto the photodetectors. It is therefore essential that the main micro-controller can keep track of the current position of the diffraction grating. Furthermore, to prevent mechanical damages, the diffiaction grating must' not be rotated beyond the limit angle in the clockwise and the anticlockwise direction. Therefore, three switches are used to give a feed back on the grating position to the main micro-controller: 1) Two limit switches, to detect when the diffraction grating is fully rotated clockwise and anticlockwise. These switches should be directly connected to the stepper motor control electronic so that the stepper motor is stopped automatically, without the main micro-controller intervention. The stepper motor control electronic would stop the stepper motor and send a "limit error" signal to the main micro-controller when a limit switch is closed.

2) A slot sensor, so that the microcontroller can keep track of the number of rotations that the stepper motor has actually done. An example of inductive slot sensor could be the component RS285-403. The associated electronic should send a signal to the microcontroller each time a piece of metal go through the slot sensor (i.e. at each complete revolution of the stepper motor shaft). The program in the microcontroller will keep track of the number of signal it has received.

The main board contains the power supply and the connector. It also contains the connector to dock the processor board.

The NIR sensor system processor has two main internal peripherals; a real-time clock, program and data memory. In addition there are a number of I/O ports required. Table 31 shows the estimated [JO count without taking into account resource or bus sharing.

Peripheral I/O Bus RTC 2-3 SPI,12C Memory Up to 45 Memory Bus Memory card 5 SPI PCMCIA 24 (addr) + 16 (data) Memory Bus + Control Keypad 7 None LCD 7(4-bit); 11(8-bit) (Memory Bus) RS232 2 none USB 2 none Loudspeaker 1 none ADC Mm. 3 SPI Stepper Motor 5 none Light source 1 none Table 31: 1/0 resources without resource or bus sharing.

The number of required general I/O pins are 21 if the PCMCIA interface is mapped to the memory bus and excluding the serial interfaces (USB, RS232, SPI).

A main ON/OFF flick switch is used to power the instrument. A flick switch is suggested to make it easy for those wearing gloves. Following powering, the instrument carries out an internal check and calibration procedure.

The device contains a clock that keeps track of the time and date even when the device is switched off. The real-time clock (RTC) is required to provide a time and date stamp for the recorded spectra. The main system battery and a backup battery, which is dedicated to the RTC, supply the RTC. A suitable device is shown in Table 32.

Manufacturer RTC Interface MaximlDallas DS1339 12C Table 32: RTC selection.

The memory is split into two categories, program memory and data memory. Flash-based program memory is used for easy development and the ability to upgrade in the field. The data memory can be split into the processing and the storing part. The first is required to calculate the spectra and should be RAM-based. A size of approx. 60 kB if sufficient. However, this is dependent on the processor and its OS, if any. The second part of the data memory is used for storing the recorded spectra and should, therefore, be non-volatile. As each measurement consists of: * 3000 bytes for each spectrum (1000 data points 3 bytes/data point) o Two recorded spectra per measurement o Two fingerprint spectrum (1000 data points @3 bytes/data point) o User data (100 bytes/spectrum) o Intermediate analysis results (assumed 3 kB) * Several background spectra (1000 data points @3 bytes/data point) * A minimum of 200 stored measurements each measurement is composed in the worst case of 2 * 3 kB + 2 * 3 kB (fingerprint) +3 kB (intermediate) +0.1 kB (user data) = 15.1 kB. This assumes a new fingerprint spectrum for each measurement and a 1000 data point intermediate result. This would require a minimum of 3020 kB, or 2.95 MB. However, the size of the storage memory should be at least 4 MB as the previous calculation does not include the background spectra. This area of the data memory could be incorporated into the programming memory.

The NIR sensor system contains an interface for removable memory card. A number of memory cards are available. A selection of these cards and their interfaces are: * Multi-Media-Card (MMC), serial interface * Compact Flash (CF), 16-bit data interface * Secure Digital (SD), serial interface A docking station is included in the sensor to insert a memory card. The memory card is used to: * store new fingerprint spectra * offer additional memory to collect spectra The inclusion of a PCMCIA, or Cardbus, allows the extension of the system with new standard hardware, such as GPS, network, modem, etc. However, use of such devices requires the availability of suitable drivers. Although these could be written for each device, this is not practical. This could therefore pose a problem with the system development. However, a number of OS contain already many of these PCMCIA devices. The use of a PCMCIA interface for standard hardware is a strong argument for the use of an OS. The PCMCIA interface requires 68 pins, of which 26 are address pins, 16 data pins and the rest are control and power pins.

The sensor has a docking station and the necessary interface to accept PCIMCA card. This offers the possibility for future extension of the system. Indeed, the standard PCMCIA developed for laptop PC could be inserted in the sensor to provide functionality such as GPS, network connection, modem, bar code scanner or other peripherals as may be required.

A numeric keypad is included in the NIR sensor system in order to enter numeric information for each measurement. The keypad has 14 keys. The user can encode some numeric and textual information with each spectrum recorded by the sensor.

Ten keys representing a numeric keypad similar to those of mobile phones (i.e. with only digits) is used to introduce both number and letter. Entering text is done using the same procedure as for mobile phone, i.e. by repetitively pressing the same key to scroll through the sequence of letters.

* Two scroll buttons are used to scroll the line of text displayed on screen if more than 4 lines of text axe available for display. At a sub menu level they can be used when writing text to scroll across the text.

* One Start/Stop' (reading) key * One'Menu'key The interface of such a keypad consists usually of 71/0 lines.

The menu key will take the user to a range of options contained within the program, for other textual symbols or commands such as Save, Send, Look Up (saved records) and other basic symboissuchas:.,' ?I\&*Q+= # Thekeys should be sufficiently large so that the user can key in the next while wearing gloves i.e. an actual mobile phone keypad would not be suitable.

User information is displayed via an alphanumeric LCD display. Alphanumeric LCD displays are split into LCD's with less or equal than 80 characters and in LCD's with more than 80 characters. A LCD with less than 80 characters should suflice for the NTR sensor system. These LCD's can have I to 4 lines of usually 20 characters and have their own build-in driver. The LCD driver can be driven via a 4-bit or 8-bit bus interface. The LCD display requires 7 control lines (4-bit interface) or 11 control lines (8-bit interface). The LCD could be mapped into the memory space of the processor.

readiness bar which is used to show the user the progress of detection (for example like the horizontal bars that go from empty to full to show the progress of software downloads). The progress bar could be displayed by using special text characters: black rectangles of various thickness. Therefore, there is no need for a graphic display. Only a text display will be required.

The display will not need to be backlit. The specifications of the display are shown in Table 33 Type LCD display (monochrome) Mode Text and limited graphics (only for progress bar) Number of lines 4 Number of characters 20 per lines

Table 33. Specifications for the display

The NW sensor system includes two serial communication interfaces. The first interface is internal only and is a simple RS232 interface. It serves mainly for diagnostic and debugging. It contains only the transmit and receive line.

The second interface is an external USB 1.1 compliant peripheral interface. This is the main up-and download port for the NIR sensor system. The interface allows the upload of the measurement data, setting of the time and date as well as an upgrade of the software. The implementation of the USB interface can be achieved by either using a processor with a USB block or an external USB chip. Both are available depending on the choice of the processor.

The system contains a small loudspeaker for user notification. A simple bleep' is used to notify the user about the end of the measurement. This is accompanied by a small red LED light.

It is well known that the operating temperature of NIR spectrophotometers affects the spectra that are collected. This instrument is designed to operate in environments between -5 C and +35 C and the spectra are corrected within this temperature range as described by Hernandez et al [3]. The temperature in the detection compartment will be monitored by a temperature sensor.

As the shape of the NIR spectrum can change with the temperature, it is important to record the temperature in the detection compartment to control that the temperature has not deviated too much from the calibration temperature. If a large deviation is detected, either a new calibration will be required or a compensation will have to be applied to the spectrum during the data analysis. The operation temperature range of the temperature sensor is -SOC to + 35 C and it has an accuracy of 1 C.

The processor board has dimensions of 68 x 37 mm with a maximum installed size of 7.3mm and is installed on the main board. The processor is a 32-bit micro-controller with floating point arithmetic will be used for the analysis algorithm. This allows more flexibility with complex algorithms.

The processor carries out the following operations I. Record spectra. This involves the following operations * Control rotation of the stepper motor and record monochromator angle * Recording of the readings from the 2 channels of the ADC card * Store angle and 2 readings in memory * Store user input with corresponding spectrum 2. Spectrum analysis * Pre-processing, involve deconvolution of instrument response from the measured spectrum * Analyse the spectrum to detect the presence of the target substance (involves Fourier transforms (IO24points) and principal component analysis) * Store the analysis result with the associated metadata 3. User interface * Input using the keypad * Output using a LCD text display * Red LED * Loud speaker * Control PCMCIA devices (including GPS, barcode reader) 4. Download data * Transfer the data stored in the memory to an external PC using the USB interface * Download new devices drivers from PC 5. Upload fingerprint spectra * The fingerprint spectrum can be downloaded in the sensor via the USB interface.

Alternatively, a removable memory card containing the fmgerprint spectrum can be inserted in the sensor.

* Alternatively new fingerprint can be downloaded from one of the PCMCIA devices (for example a modem).

Table 34 lists the external connectors and interfIces on the outside casing of the sensor: Position Type ON/OFF button Casing Push button Numeric keyboard Casing Numeric Red LED Casing Power supply Casing Mobile phone charger socket USB Casing Memory card slot Casing Serial port Internal Fibre optic IN Casing SMA Fibre optic OUT Casing SMA Table 34. External connectors and interfaces of the instrument Section 3.2. Power considerations and battery layout To explain the power requirements of the system it is important to understand the sequence of operations that are performed during one acquisition. From cold, the sequence of operations is: Warm up to stabilise the Peltier temperature: 3Oseconds. If required, the stabilisation of the light bulb could be done at the same time. However, if the light bulb is switched on for longer, this will reduce the number of spectra that can be recorded between battery recharge. Furthermore, there is a 30s boot time to load the operating system in the CPU.

This can be done at the same time as the Pettier cooling.

* Calibration: around 1 second * Data acquisition: I to 2 seconds * Processing and user interaction (e.g. UPS position, WiFi,...): maximum 4 minutes (probably less, but let's assume a maximum of 4 minutes to compute the electrical power requirement).

The maximum voltage required for the supply is 6V. During one acquisition the batteries will need to deliver a peak current of 4.6A during is during the data acquisition. The remaining of the time, the peak current will be below IA. Furthermore, the estimated capacity of the batteries will need to be 51.39As or 14.275 mAh per acquisition (this is summarised in Table 35. The power profile of the system is as such that the unit will have a peak current around 4.6A for the acquisition time (is). The remaining time the current peak should be below IA, probably even below O.5A.

Required voltage 6V Peak current 4.6A Duration of peak current Is Operating current <1 A Capacity per acquisition 14.275 mA.h Table 35: Summary of the minimum specification requirement of the power supply The estimate battery life based on the unit being turned on for 4 minutes during each reading.

This is the best estimate based on the time for the operator to set a measurement, do the scan and turn the equipment off. The major warm-up will be the time taken by the Peltier to cool the photodetectors, in about 30 seconds Any remaining warm-up settling time will be dependent upon relatively large power dissipation in the vicinity of and warming the analogue board. This could be, for example, the heat output from the Peltier heatsinks or radiated emission from the bulb.

The instrument contains two packs of batteries. Each pack consisting of 6 Nickel Metal Hydride (NiMH) batteries. Pack 1 provides a peak current of -4.3A for 1 second and -0.8A for 30 seconds (Peltier cooling time). Pack 2 only needs to provide a -0.3 A peak for I S with -O.2A for -4 minutes (full on-time of unit).

NiMH batteries are available as consumer products. They are readily available and an example is

shown in Table 36.

Manufacturer Uniross Model R6 Mignon Rechargeable Part No. RB 102746 Voltage per battery 1.2 V Capacity per battery 2300 mA.h Battery Type NiMh Number of recharge 1000 Price per battery 2.75 Weight 30g Table 36. Batteries for the instrument Alternative batteries can be selected if they are more adapted to the instrument.

The batteries are arranged in a series parallel arrangement, i.e. 2 packs of 6 batteries and place the 2 packs in parallel (Figure 40). This gives, for example, 6V nominal at 2x 2.3 A.h = 4.6A.h.

By placing the 2 battery packs in parallel, the load is shared evenly between the two packs.

Around 320 spectra can be acquired by the machine using this power arrangement and unplugged from the mains supply before the batteries have to be recharged.

The peak current (4.6A) represents twice the capacity of a battery (but only one time the capacity of the 2 packs in parallel). By referring to the typical discharge curve shown in Figure 41, one would not expect a discharge voltage of less than 1V when the lamp is switched on. This means that the battery pack for the light bulb would need at least 6 x 1.2V batteries. Indeed, the voltage needs to be stabilised for the 6V lamp and thus one needs a little overhead to be able to do this.

Also, the cell's voltage reduces as they discharge. The cell voltage, when fully discharged, is generally regarded to be 1.OV.

Both power packs are charged by a single battery charger by plugging a 12V DC power supply in the sensor (e.g. using the car cigarette lighter socket) on a socket on the sensor casing, using a sandard phone charger socket. The two battery packs are charged by using a single charger and a single socket. A battery voltage monitor will be implemented into the design in order to alert the user about a low battery condition.

Using the battery arrangement described the peak current that can be delivered is 7.6A (-2C), which means that the actually required current is only-i.2C.

Section 4. Software Maximum flexibility in the software is be achieved by distributing' the system into a set of communicating and collaborating processes. l'his allows for changes to the software should the user's requirements change (e.g. the instrument is used for a different purpose). The objects within the system are also designed with extensibility in mind by utilising object oriented and generic programming techniques.

A description of the main architectural components that will meet both the design goals and the software requirements for this system is included.

A preferred option in developing the software is to use the Qt toolkit (a mature and well established toolkit in the Linux/Open Source community). The Qt toolkit has a number of advantages over other systems that are available. Firstly Qt is an entirely C+ toolkit and is an open source product, which means that the full source code for all the components is available to the user. Several years ago Qt was selected to form the basis of the KDE project (K Desktop Environment). KDE has since become the dominant windowing system along with Gnome) on the Linux operating system and is supplied with all the major distributions. Literally hundreds of applications have been created and continue to be created for the KDE project using Qt (which is free for software developed under the open source GPL license). Qt has been used for some major commercial projects as well. It is particularly well suited for scientific applications due to its strong support for visualisation. One of the main reasons for Qt's increased use in the commercial world is the fact that it is a cross platform tool. This means that the same source code will run natively on UNIX/Linux, Windows and MacOS X with only a recompile required.

This is a major advantage for organisations that want applications which are cross platform but only want to maintain one code base. There are some other advantages to this even for projects like this one where cross platform code is not an immediate requirement. Code which can compile across platforms tend to contain far fewer bugs as differences in architecture and compilers throw up different problems with the code. Code that will run on multiple platforms will therefore be more robust and usually better designed than single platform code. Qt also has a very elegant object oriented design.

To help with usability the software presents the user with a single point of entry to all the flmctionality of the software. Given the inherently distributed nature of the system a monolithic application cannot be created. It is best, therefore, to develop the system by providing a number of self contained but communicating processes. Each process can be designed and implemented separately (by different people and even in different languages if necessary) as long as the communication mechanism is well designed.

Before defining the processes that need to be developed it is important to consider the uses that the instrument will be put to. Given that the instrument is designed to be a field based portable instrument we will assume that in each case the instrument is remote from the base station.

Figure 42 shows the main architectural components in the system. The system is split into two parts, the field based component that comprises the instrument itself with the embedded computer. For the purposes of this software design this component can be considered as a remote PC or process. The second component is the base station that comprises a server, a storage and data management system and optionally a user interface. Each of these sub components can be geographically remote so long as they are connected by a network of some description. The requirements of the network connection will be discussed in the context of the messaging system.

The system is split into twop, the field based component that comprises the instrument itself with the embedded computer. For the purposes of this software design this component can be considered as a remote PC or process. The second component is the base station that comprises a server, a storage and data management system and optionally a user interface. Each of these sub components can be geographically remote so long as they are connected by a network of some description. The requirements of the network connection will be discussed in the context of the messaging system.

Several typical use cases for the instrument are: Case 1 -standalone mode. In this scenario the instrument is being used as a standalone instrument in the field. An active network connection to the base station is not required or present. This is conceived as the usual way to use the instrument and will be how the instrument will be configured when it is actually being used in the field to collect sample data. The software is designed to be usabLe by a non-technical worker and the instructions and results are displayed on the small text based display integral to the unit.

The user will position the probe and capture a measurement (spectrum of the sample).

This data will be processed with the data analysis algorithm on the embedded PC in the instrument. The data analysis is performed with respect to the expected target and background spectra which are relevant to the current experiment. This data is preloaded onto the instrument prior to taking the readings. In such a configuration the instrument is specific to a set of experimental conditions. Since the target and the background are known, the processing requirements are minimal. The presence or absence of the target compound is calculated and the result fed back to the user. The user has the option of storing the collected spectrum to a local medium such as a flash card, alternatively only the processed results (compound x is present at y concentration) can be stored.

Case 2-Load new instrument protocol -in this case the supporting scientist wants to change the settings of the instrument for different experimental conditions (e.g. look for a different target compound). The collections of setting and data which comprise the instrument configuration for a particular experimental condition is called a protocol.

Given the fact that a number of different options will need to be considered in the definition of a protocol a more complex user interface than the simple one on the instrument is required and is provided on a PC. The instrument will need to be connected to the PC or base station via a wireless or network cable connection. This software will be designed with the trained chemist in mind and the user will require training to use the software. The user will select background and target compound datasets from their spectral database. The user will also select a data analysis algorithm, algorithm settings (such as confidence limits) and operating parameters such as whether the individual measurements are to be stored, transmitted or deleted in the field. Once a protocol is defined it can be saved for future reference and uploaded via the network link to the instrument computer. The instrument is now set up for use in the field.

Case 3-creating spectrum database entries. In this use case a spectral database is populated with known compound spectra which can be used for protocol definition. For background and target spectra several measurements are required to eliminate statistical variations. In this operating mode the instrument in positioned in the field or lab and collects multiple statistically independent samples of the compound spectra. These are uploaded in real time to the base station that combines them in a statistically rigorous way and stores them in the data management system along with the appropriate metadata.

Once stored these measurements can be used as background or target compound data in an instrument protocol. Although the data can be collected by an unskilled worker, a scientifically trained user will be required at the base station to process and input the data.

* Case 4-online analysis of unknown targets. In this scenario the user does not have a priori knowledge of the target compound and therefore a complete instrument protocol cannot be created. In order to identify unknown compounds it will be necessary to match the collected spectrum with a compound in the database. If the database contains the unknown target compound an identification will be possible. This analysis cannot be done in a standalone manner on the instrument for two reasons. Firstly this kind of analysis will be computationally complex and the processing power required will be more than is available with the embedded processor. Secondly, potentially vast spectral database will be required. The storage capacity of the instrument will be extremely limited and could not possible contain the number of data points required. In this case the instrument is connected to the base station via a mobile network link (GPRS modem).

The instrument will collect a sample and transmit it to the base station where the superior processing power available will allow for computationally complex analysis and essentially unlimited data storage capacity will mean databases of any size can be utilised. Once a match with a compound is made the results are returned to the instrument

in the field via the network.

The following design overview will help to explain the processes involved in the software. The object model identifies each object in the system, its properties and its relationships to other objects in the system.

The Data Processing Pipeline consists of two fundamental types of object; data objects and process objects. Data objects are used to represent data within the system and process objects are classes which operate on data objects and which may produce new data objects. There are three types of process object within the system; sources, filters and mappers. Sources produce data, filters take in data and process it to produce new data and mappers accept data for output to some other system.

The data processing pipeline ties together data objects and process objects. Given the fact that there are several different ways of using the software depending on whether there is an instrument connected and whether there is a database connection a simple way of handling the permutations is required. By decomposing the processes involved into process objects and linking these process objects together in a processing pipeline we can present the user with a consistent interface while managing the complexity. This pipeline approach also offers future proofing. By using derivation and inheritance from existing process objects we can specialise and extend the pipeline.

To see how we can use the pipeline to decompose user operating modes we will present a

number of examples:

* The user loads a spectrum from a file and displays the results. This user mode can be decomposed into two process objects, a source and a mapper. The data is already in existence and resides on the file system so we can initiate the pipeline with a FileReader source object. We don't need to modify the data in any way so no filters are required. We do however need to map the data to a form in which it can be displayed to the user via a GUI widget. This mapping may be simple or complicated and is provided by a Visualisation Mapper object (Figure 43).

* The user creates a new spectral database entry by acquiring new raw data. The data is saved to the database and no initial analysis is performed. This user mode requires three process objects, a source, a filter and a mapper. No data exists prior to the pipeline being executed. In fact no protocol information is in existence either. Since we cannot execute this experiment without any protocol information we need to initiate the pipeline with a Protocol Source process object. This object creates a new set of protocols with default values. Since the user will most likely want to modify the protocol the protocol wizard will be used if we are in GUI mode. We then need an Acquisition Filter to generate the raw data. This requires an instrument of course and so it is the responsibility of the Acquisition Filter to communicate with the instrument. If there is no instrument available then the AcquisitionvFilter cannot be used. Since we are not doing any data analysis we can simply use a Database Writer mapper object to map the data to a database. The defined protocol will make sure the correct information is written to the database according to the appropriate ruleset (see Figure 44).

Figure 45 shows the inheritance diagram for Process Objects. As can be seen in Figure 45 there are currently three types of Process Object source filter within this system. The output of a source filter is a data object. The three types are i) a File Reader which reads data or a protocol file from a disk, ii) a Database reader which reads data or a protocol from a database and iii) Protocol Source which creates a new protocol data object populated with default values that can be modified to create new protocols.

Figure 45 also shows that there are two types of Process Object Filter the Acquisition Filter which manages communication with the instrument (connections and messaging) to acquire new raw data according to the loaded protocol and the Analysis Filter which manages the interface with the analysis engine. Using this design different analysis algorithms can be used by switching Analysis Filters.

There are currently three types of Process Object mapper filter within this system. As Figure 45 shows these are i) the Visualisation Mapper which transforms the data into a form which can be displayed using the systems visualisation capabilities, ii) the File Writer which data and protocol files to disk and iii) the Database Writer which writes data and protocol information to a database and manages the database connections.

There are different types of data used in our system these include: * Experiment Data including all the metadata, raw data and derived data (analysis results) associated with an experiment (background and target compound data etc.).

* Experiment Meta Data containing all other experiment metadata like the date, operator etc. * Instrument Parameters * Analysis Parameters containing the analysis configuration for the experiment.

The RunProcessor is an object which executes the data processing pipeline according to the current experimental protocol. A protocol is a set of rules or configurations for how the experiment is run. This can include instrument settings, analysis settings and pipeline settings.

While the object model in Figure 45 describes the static portion of the system, the dynamic model describes the sequence of events and time dependent aspects of the system (see Figure 46).

Figure 47 shows a representation of the core processes. Each process is implemented as a standalone executable with communication between processes. Processes may run several threads of execution.

The Inter Process Communication (IPC) allows the different processes to be able to pass information between themselves at various stages. There are several methods of 1PC but our system uses the XML-RPC messaging protocol. XML-RPC is a very simple protocol, which uses XML messages traveling on HTTP to represent client-server remote procedure calls (RPC). The XML messages identify methods, parameters and the results of calling the methods. The XML document uses a simple but effective set of data types to pass information between processes.

Since the amount of data that will be passing between processes is relatively small, and given the speed of modem computers (and networks) speed will not be a major issue. Alternatively shared memory and semaphores could be used in place of XML-RPC.

The IPC is also responsible for operating the instrument. All instrument control functions and firmware calls will be made from this process. It accepts requests from the control software (run control process) and returns the results of any RPC's. The IPC will be implemented as an XML-RPC server and in the context of communication between this process and the run control process the rcp will be an XML-RPC client.

The run control process (RCP) is the central process. You can consider it as the Grand Central Station for data, information and commands passing through the system. The RCP can initiate a run' in a couple of different ways. Either the user has initiated a run from the Graphical User Interface (GUI) which will pass a message to the RCP to start a run, or the RPC receives a message to start from an external scheduler. The GUI itself is part of the RCP. However it exists on a separate thread from the main RPC process. This is because it is undesirable for a GUI to become unresponsive during processor intensive operations. In fact in some cases (where a set of protocols have already been defmed) it may not be necessary for a GUI to be presented to the user at all. If a batch job of experiments is to be run overnight and controlled from a script then a command line user interface (CU!) is more appropriate. The system is designed so that it could also easily be controlled remotely (using a secure login shell such as ssh) from the CUI.

In most cases the system will be operated from a graphical user interface (GUI). In order to give the user the illusion of a mono-app there will be data display facilities which can feedback information from any of the running processes or threads.

The Data Analysis Process is responsible solely for receiving raw data from the main RPC and transforming it into analysed results which it then returns to the main RPC. As mentioned previously the actual analysis code is supplied in a dynamically loaded library. It is important to run the analysis as a separate process as this is the most processor intensive aspect of the system. If this were run in the same thread or process as the GUI then the GUI

would lock up and become unresponsive until the analysis process has completed. The analysis process will be created by the RCP if and only if it is required in the processing pipeline. It is implemented as an XML-RPC server which, once created, waits for remote procedure calls from the RCP. When it receives data from the RCP it will load the appropriate dynamic library and execute the data analysis algorithm. It will then return the results to the RCP.

The Input Output Process (lOP) is designed to be an abstraction layer between the RCP and what ever data management system is required by the user. This data management system is most likely a simple file system or a relational database management system (RDBMS). This process will receive data from the RPC and direct it to the appropriate place. This process will also be responsible for retrieving data from the storage medium.

Responsibility for communication between the lop and the RPC is provided by the appropriate Mapper object. For the purposes of the analysis of an unknown sample the data analysis process will communicate directly with the lOP.

The Data Analysis Process requires a method for analysing the raw spectra and converting it into a meaningful qualitative and in some instances quantitative measure. There are a number of candidate algorithms known to perform these tasks and we will examine a few in the following section.

The Cross Correlation Algorithm is one such example. The principle of this analysis algorithm relies on the use of a cross-correlation function. This technique could be used in order to asses the presence of the target substance in a mixture. It is not expected that it would be able to determine the concentration of the substance in the mixture. The spectra can be thought of as a function of the wavelength A.. The spectrum from the sample is defined as SQ.), representing the light intensity reflected by the sample at a given wavelength. The spectrum of the target substance is defined as a function R(A.) representing the light intensity reflected by the target substance at a given wavelength.

The cross-correlation X(S,R) function between the sample spectrum and the reference spectrum is defined by: J' s(R(4d x(s,R)= J' s(,JdLf' R()dL Equation 20 where A and A.,, are the lowest and highest wavelengths in the spectra.

If one takes two spectra S and S2 that are from unrelated compounds, it is also possible to compute the cross-correlation function X(S1, S2) between S1 and S2. It is possible to think of the cross-correlation function as a random variable. Indeed, if one has a database of spectra S1,... SN from unrelated compounds, it is possible to take any two spectra S1 and S and compute their cross-correlation function X(SI, Si). We can then compute the cross correlation function X(S1, S) for every pair of spectra in the database and plot an histogram of the number of time that we observe a value X(SI, S) in a given range (see Figure 48).

This histogram therefore defines the probability density function of the distribution (p.d.f.) of the X(S1,S2). In Figure 48 the filled area represents 95% of the area under the p.d.f. curve. Therefore, when taking two spectra at random, there is a 95% chance that the cross correlation function will have a value less than XLjmjt. This is defined as the 5% confidence level.

Let's now compute the cross-correlation function between a measured spectrum and the reference spectrum. If the result is less than XLimjt, there is no indication that there is a correlation between the two spectra. On the other hand, if we find a X(S,R) that is larger than Xmit, there is less than 5% chance that this correlation is due to chance. This is therefore a good indication of the presence of the reference compound in the measured spectrum.

Therefore, the identification of the presence of the compound is done in several steps steps: I. Compute the cross-correlation X(S,R) between the measured spectrum and the reference spectrum 2. Fix a confidence level, for example 5% 3. From the p.d.f distribution, determine the value XLImII for the given confidence level.

4. If X(S,R) > Xi..jmjt, then the compound has been detected, at the specified confidence level.

The key problem is of course to determine the p.d.f. There are two ways of determining the p.d.f.: empirically and theoretically.

Empirically using a database of spectra experimentally measured from unrelated compounds, one can compute the cross correlation function X(S1,S2) for any pair of spectra. Then the histogram of the frequencies of the X(S1,S2) is the experimentally measured p.d.f. Instead of using the spectrum of pure mixture, one can also use the spectrum of a mixture of compounds.

Theoretically we can assume that a spectrum can be modelled by a sum of K Gaussian functions: K (z_A, j 2c Equation 21 As the spectrum is chosen at random, it can be composed of multiple absorption bands of random position Aj, width o and height L. Therefore Xj, aj and I, can be defined as random variables with a uniform distribution.

It is possible to show that the cross correlation between two spectra defined by Equation is given by: x(s1,s2)=E Ri.R 2 j k 242r.Aoft where jk1J LcTk =a + NK=Ik K,with k &.Jk, &Tjk, R, and Rj are function of the random variables A., a and I. Therefore, they are random variables. It is possible to determine the p.d.f. of &jk, LCTjk, R1 and Rj from the p.d.f.

of A., a and I (which are uniform variables). Knowing the p.d.fOf&jk, , Ri. and Rj, it is then possible to determine the p.d.f. of X(S,,S2) as it is a funciton of these random variables.

To test the cross correlation function simulated NLR spectra were used. The spectra ( I(A.) ) were modelled as a sum of Gaussian components:

K__-

1(A.)= 4-.e -j2ir,o Equation 22 This equation contains several parameters: * K: the number of Gaussian bands composing the spectrum * I: intensity of the band * ?: the position of the centre of the Gaussian band number i.

* a: Full width at half maximum of the Gaussian band number i.

The value of these parameter is assumed to be described by a random variable (see Table 37): Parameter Random Range Range Variable minimum Maximum Distribution K Uniform 20 20 A.j Uniform 0 1000 a, Uniform 10 50 II Uniform 1 10 Table 37. Parameters used during simulation of N data To compute the probability density function of the cross correlation and auto-correlation function a random sample of spectra (I(?.)) were generated as described above. To simulate the spectrum of a mixture of component, a new series a mixture spectra IkjQ') was generated from all the possible sums of spectra: J,(A)= 1k(') aJ1(A.) Equation 23 Where a is a parameter related to the proportion of component j in the mixture.

In total a set of 20 random spectra were generated. From these spectra a set of mixture spectra was generated. This process was repeated for three different value of a (a 1, a = 0.5 and a 0.1). This represents 20 x 20 I 2=200 different mixture spectra.

All the cross correlation functions between the mixture spectra and the individual spectra were computed: X[Ikj(A.), 11Q)] with k!= i andj!= i. This represents 200 x 19=3800 correlation function. The probability density function (p.d.f.) of the cross correlation functions is computed by representing an histogram of the distribution of these 3800 values.

All the auto correlation functions between the mixture spectra and the individual spectra were computed: X[Ikj(?.) , IkQL)]. This represents 200 x I = 200 auto correlation functions. The p.d.f. of the autocoreelation function is computed by plotting the histogram of the distribution of these values.

The results are shown in Figure 49. It can be concluded that: * For all the mixture (whatever the proportion of the two components), the cross correlation function between the mixture spectrum and a spectrum that is unrelated to the mixture components will give a number comprised between 0.0005 and 0.0015. The most likely result is 0.001. The same distribution is obtained for all the mixture composition (see Figure 49).

* The correlation function between a mixture spectrum and the spectrum of one of its component will give a value that is on average larger than the cross correlation function with unrelated spectra (see blue, yellow and green curves in Figure 49).

* If the proportion of the component in the mixture decreases, the correlation function between the mixture spectrum and the spectrum of that component will become more and more similar to the cross correlation function with unrelated spectra.

* Therefore, for spectra containing a very small proportion of the target compound, the autocorrelation function will not be significantly different from the cross correlation function. That component will therefore not be detected by this procedure.

There are several reasons why this techniques does not deal very well with mixture containing a small proportion of target compounds: * It assigns the same weight to all the wavelengths in the spectrum, even those that do not carry any information about the target compound.

* No information is provided on the spectrum of the background. It is assumed that the background spectrum is totally random. This is the main problem in these simulations.

Indeed, a small concentration of target compound is translated in a very small variation in intensity. As the shape of the background spectrum is unknown, it is virtually impossible to detect a very small intensity of target compound.

It is therefore expected that by using weighted wavelength in the computation of the cross correlation function, this method should improve. This would suggest that the method of the Principal Component Analysis could be used in order to obtain new co-ordinate with the proper weighting.

It is also expected that very low level of target substance cannot be detected if no information on the background spectrum is available. When the background spectrum is known and the variability of the background spectrum is known, the method of Maximum Likelihood estimator is expected to be able to predict if an observed deviation from the background is due to background variability or the presence of a target substance.

Another example of an algorithm that may be suitable for analysing the raw spectra in the Data Analysis Process is the Maximum Likelihood Estimator algorithm. This technique is based on the following assumptions: 1. The measured spectrum is modelled as the sum of the substrate spectrum, the target compound spectrum and the instrument noise (see Equation 24).

2. The substrate spectrum is known.

3. The variability on the substrate spectrum is known. In the simplest case, the variability of the intensity of the substrate spectrum at each wavelength is described by a Gaussian distribution with a known standard deviation.

4. The standard deviation of the instrument noise is known.

5. The spectrum of the target substance is known.

The practical implications of these assumptions are: * The nature of the background substrate must be known before the experiment, so as to known the background spectrum. This means, for example, that we need to know whether the target substance will be measured on apples or pears, before the analysis is done.

* The variability of the background spectrum must be estimated. This implies, for example, that the spectra of a collection of apples must be recorded and analysed to define a

reference background spectrum and its variability.

* The instrument must be calibrated so as to have an estimation of the instrument noise.

If these assumptions are valid, the method of the MLE allows: 1. The definition of a simple test to reject the mill hypothesis (target substance absent from sample) at a given confidence Level 2. The estimation of the most likely concentration of the target substance 3. The estimation of the confidence interval on the most likely concentration Once the spectrum has been recorded, there are two problems that must be solved: 1. Determine whether the target substance is present or absent in the spectrum 2. If the target substance is present, determine its concentration and the confidence interval on the concentration.

The recorded spectrum I(A.) is constituted of three major contributions: I(%)= I8(A.)+a.I(.t)+ !N(#) Equation 24 * IaQ) is the spectrum of the substrate on which the target substance is located. It is assumed that the nature of the substrate is known before the experiment is carried out. As in most cases, the substrate will be biological, we can expect some variation in the shape of the background spectrum. Therefore, B() is a random variable. The exact shape of the p.d.f. of that variable must be determined by experience.

* Ir(X): the spectrum of the target substance. This spectrum is known with some precision, as it has been determined in the laboratory before the measurement. The spectrum is weighted by a coefficient a that is proportional to the concentration of the target substance.

* lwR) is the measurement noise. This can be estimated during the calibration of the instrument. This is a random variable with a p.d.f. that is determined experimentally during the calibration procedure.

It is important to note that this method assumes that the shape of the background spectrum is known. Indeed, if the nature of the substrate is unknown, the shape of the background spectrum is totally random. This introduces a great uncertainty in the model, that will be translated by a poor detection limit for the target substance.

The problem consists therefore in fitting Equation 24 to the experimentally measured spectrum.

The parameter a is varied and the likelihood of observing the experimental spectrum (knowing the theoretical model) is computed for each value of a. The value of a giving the largest likelihood is the most probable a. The likelihood expresses how likely is the observed mismatch between the theoretical model and the experimentally observed spectrum. In order to be able to compute this probability, it is necessary to have enough information on the variability of the substrate spectrum and on the instrument noise. The more information is available, the more accurate is the computation of this probability and therefore the more accurate is the estimation of the concentration of the target substance.

In theory the analysis method relies on the following steps: 1. Determination of the variability of the background spectrum 2. Determination of the instrument noise 3. Computation of the likelihood function 4. Computation of the maximum likelihood estimator 5. Statistical test for the presence of the target substance 6. Compute the confidence interval on the concentration As the target substance is deposited on a substrate and the spectrum of that substrate will show some variability, it is necessary to have an estimation of the variability on this substrate. Indeed, if one measure a larger intensity than usual at a given wavelength, it is necessary to know whether this increased intensity is more likely to be due to the presence of the target substance or to the intrinsic variability of the substrate spectrum.

In a real experiment, the substrate spectrum would be measured experimentally. In this example, the background spectrum is modelled by a sum of Gaussian peaks. In order to model the possibility of variation in the substrate spectrum, it was assumed that each parameter describing the individual peaks (centre wavelength, peak height, peak width) can show some variation, described by a normal variable with a standard deviation equal to 0.2% of the amplitude of the parameter (e.g. = 0.002 x Peak height). In Figure 50, several background spectra have been overlaid. The variation in the background spectrum manifests itself by the broadening of the spectrum at different wavelength.

Let assume that the spectrum of the background is measured for K different reference samples (e.g. K different apples). Each reference sample will give a background spectrum Bk(), with K = k. From this set of reference samples, one can determine the probability density function (pd.f.) of the background intensity at each wavelength by plotting the histogram distribution of the intensity measured at that wavelength. The p.d.f. determines the probability of measuring a given intensity. There is one p.d.f. for each wavelength.

P8(18,A.) is the probability of observing an intensity I at wavelength ? for the background.

There are two ways of describing the p.d.f.: 1. Using a numerical description of the function based on the experimentally measured spectra.

2. Model the p.d.f. by a normal distribution centred at the mean intensity of the p.d.f. and with the same SD as the p.d.f. In this case: P (1 21-_________ r j2jr.o8 Equation 25 The instrument will add some noise to the measured spectrum. This noise is usually assumed to be normally distributed. The magnitude of the noise has to be determined experimentally. Figure 52 shows an example of modelled instrument noise. In this model, the probability of observing some noise with an amplitude IN at wavelength is: ( \2

J N

D (7 21-_________ 1NoueYN") r-- sJ2.ff7 Equation 26 Where a is the standard deviation of the instrument noise The likelihood (L) of an observation (0) is defined as the probability of making that observation, knowing the theoretical model (M) describing the system: L = P(OIM) As the measurement at each wavelength are independent events, the probability of observing a given spectrum is equal to the product of the probabilities of observing the intensities at each wavelength: L = P(OjM) = HP(O(M(J) where 3 is to total number of data points.

The model M(A.) is described by Equation 26 with MQ.)4(X). The model is the sum of two normal random variables (B() and NQ)) plus a constant term (a.liQ.)). Therefore, the probability P is described by normal random variable with mean 1(A) and a standard deviation a equal to: ci(it)= .J) o'W) Equation 27 Therefore, the likelihood is equal to:

J L=fl

Introducing Equation 26 ()[I u(.) ) Equation 28 For practical reasons, instead of using the likelihood, one usually uses the logarithm of the likelihood: J,,(a).z.i(a,,(2)-O(4)' pL(a) = -2.In(L(a))= _2.ln[fl Je Using the well known properties of the logarithm function, this can be rewritten as: pL(a) = 241n( (AJ)+[" aJT(+ )?1(*7) y2] Equation 29 The spectrum of the target substance can also show some variability depending on the nature of the environment in which the substance is embedded. The relative amplitude of the peaks can be modified and the position of the peaks can shift.

This variability can be taken into account in the computation of the likelihood. In Figure 29, the standard deviation aQ.) is now computed using: o(A.) = 4Jor (,) + o (A.) + cy,' (A.) Equation 30 a(A.1) is the standard deviation representing the variability of the target spectrum at wavelength . Furthermore, in Equation 29, Ii() is now defined as the average intensity at wavelength ?i.

Maximum likelihood: most likely concentration of target Equation 29 contains one unknown variable, a, which is the concentration of the target substance. In order to determine the most likely concentration of the target compound, one must identify the value of a that minimises Equation 29. This corresponds the value of a that maximises the likelihood function (Equation 28). The minimum of Equation 29 is determined by solving the following equation: pLaO J12A.

2. (.%,) + N O2,) + a.2.E 21) = 1=1 a Therefore: ,.(0(,) I,(,)-1(A,)) a = J j21 Equation 31 Equation 31 gives the most likely value of the concentration a of the target substance. This is the answer to the second question of section Figure 53 shows the proportion of target substance predicted by Equation 31 as a function of the actual concentration of target substance in the model system. The systems were modelled by using Equation 24. The instrument noise is modelled as described in Figure 52. The spectrum of the target substance was simulated. The spectrum ( IrQ.) ) was modelled as a sum of Gaussian components: K (2-k) I(2)= _& Equation 32 This equation contains several parameters: * K: the number of Gaussian bands composing the spectrum * I: intensity of the band * ?j: the position of the centre of the Gaussian band number i.

* a: Full width at half maximum of the Gaussian band number i.

The value of these parameters is drawn at random from the distribution of a random variable (Table 38) and produced data as shown in Figure 54.

Parameter Random Range Range Variable minimum Maximum Distribution K Uniform 20 20 Uniform 0 1000 a1 Uniform 10 50 Uniform 1 10 Table 38. Random variable used during simulation The computed alpha is linearly dependent on the concentration down to 1 (y4* It should be noted that the concentration units axe arbitrary. Indeed, they represent the relative proportion of the intensity of the target compound spectrum to the background spectrum in the model. They do not represent actual chemical concentrations. The relation between 1K intensity and concentration is described by the extinction coefficient of the substance and must be determined during the calibration of the system. Alpha = means that the target substance can be detected is its intensity is larger than 10 times the background intensity. This value of results from the specific parameters that were chosen for this simulation. A real experiment may present different values for the parameters.

The only question that remains unanswered is question I: after having observed the spectrum we need to asses whether or not we have evidence against the null hypothesis: a -0 (i.e. the target substance is absent).

The method of generalised likelihood ratio test [4] has been developed to deal with this case.

Let's define the null hypothesis as Ho: cx=O. The alternative hypothesis is HQ: aE (where is the set of possible values for a). It can be shown that under some general conditions, if the observed dataset 0 contains n data points, then: 2' (L(Ho) L(H00) tends to a f(dim H -dim Ho) distribution as n-9oo.

Here, dim H0 and dim H0 are the dimensions of the sets containing the parameter a in both models. This equation can be rewritten as: 2. ln( ::?) -2. in(L(Ho 10))-(-2. ln(L(H jo))) = pL(HoI0)-pL(H(0) This leads us to reject the null hypothesis whenever: pL(HofO)-pL(Hnl0)>

(dizn HQ -dim!!0) Equation 33 where quantile is a predefined quantile of the distribution. Usually, one chooses the 95% confidence level. With one degree of freedom, X295% = 3.84 [7]. In Figure 53, the data points for which the null hypothesis (a=O) has been rejected using Equation 33 are plotted in blue. The data points for which there is not enough evidence for rejecting the null hypothesis (a=0) are plotted in red. In this example, one can confidently reject the null hypothesis (target substance absent) when the "concentration" of the substance is above 10'...DTD: The last step is to define the confidence interval on the estimate of the concentration a. The likelihood ratio test allows us to find the confidence interval using the 2(dim H0 -dim Ho) as a general chi-square statistics [5]. The procedure to follow is: * Define a limit value for PLL(MIO).

* Any parameter concentration a that gives a value pL(MIO) <pLL(MIO) is defined as acceptable and within the joint confidence interval * Any concentration a that gives a value pL(MO)> PLL(MIO) is defined as unacceptable and is outside of the joint confidence interval.

The limit value of pLj(MO) can be defined in several ways: 1. Assume that pL(MtO) is adequately described by a chi-square variable. An F statistics with 1 parameters, i-i degrees of freedom and a given confidence level (usually CL=95%) can be used to define what is an acceptable value of pL(MO) [6] Therefore pLL(M(O) is defined as: pLL(MIO) = _PL(Ma_ii(O)(l + .11.F(l,J -lCL)) where pL(MjO) is the minimum value of pL(MO), that is obtained using the optimal value of the concentration a.

2. Compute the value of pL(MjO) for a range of value of the concentration (centred on the optimal value of a). Then the confidence interval is defined by the 2 values of a that enclose 95% of the are under the pL(MO) curve. This method does not assume any a-priori shape for the statistical distribution of pL(MIO), it computes the confidence interval directly from the probability distribution.

Using method I and Equation 29, it is clear that the lower and higher limit of the confidence interval are given by the values of UL defined by the equation: PLL = 2[In(.5cr(A)) ( B(1)+ aJT(A,) IN (j....

After some simple arithmetic transformations, we can rewrite this equation as: cr2[ ii)] + a[2. IT(AJ ()+ 1N("?) o(Av))] a, (i,)+ JNQ)_O(A,))2...L.pL + In( x.a(A))] = which gives the following two solutions for al: -2.jkL.(I1(A,)+ !N(AT)-O(2)) S., aL 2.Z 5.1 0' ) 4[ ITd1][E(t1 1N(Al)-0(AS)Y _!.PLL 147))] 0 I_I C 0' S_a

I

2.----a Equation 34 The MLE method relies on a proper description of the model. The more information is provided to the analysis algorithm, the more precise will the determination of the concentration be.

Ideally, the nature of the substrate should be known. However, in some cases, it may not be practical to have to specify the nature of the substrate for each measurement. The MLE can also deal with the situation when the nature of the substrate is not defined.

Let suppose that the algorithm has access to a database of substrate spectra (IBQ'43) and aBQ,P)).

The nature of the substrate is specified by the parameter f3. For example, l for apples, 2 for pears,...

The likelihood is now a function of two variable parameters: a and I. pl(a./J) = The algorithm must now find the values of a AND that minimise the function pL(aj). The algorithm will therefore identify the most likely substrate from the database. At the same time it will determine the most likely concentration of the target substance.

In conclusion the Maximum Likelihood Estimator (MLE) is based on the following assumptions: I. The measured spectrum is modelled by the sum of the substrate spectrum, the target compound spectrum and the instrument noise.

2. The background spectrum is known.

3. The variability on the background spectrum is known. In the simplest case, the variability of the background spectrum at each wavelength is described by a Gaussian distribution with known standard deviation.

4. The standard deviation on the instrument noise is known.

5. The target spectrum is known.

The practical implications of these assumption are: * The nature of the background substrate must be known before the experiment, so as to known the background spectrum. For example, this means that we need to know whether the target substance will be measured on apples or pears, before the analysis is done.

* The variability of the background spectrum must be estimated. This implies that the spectra of a collection of apples (for example) must be recorded and analysed to define a

reference spectrum for the background,

If these assumptions are valid, the method of the MLE allows: 1. The definition of a simple test to reject the null hypothesis (target substance absent) at a given confidence level. See Equation 23. 2. The estimation of the most likely concentration of the target

substance. See Equation 21.

3. The estimation of the confidence interval on the most likely concentration. See Equation 24.

As the MLE depends only on a single variable parameters (a) and it is linearly dependent on this parameters, the optimisation function can be solved analytically. This means that the MLE method only requires straightforward calculations. There is no iterative algorithm that is necessary to compute the optimal concentration. Therefore, the calculations can be done very quickly and only require a small amount of memory.

The MLE method uses as "fingerprint" the following data:

* The background spectrum IB(A)

* The variance of the background spectrum cr2Q.)

* The target spectrum l1(X) The method of MLE relies on the fit of a model to the experimentally measured data. The estimation of the goodness of the fit relies on the estimation of how probable the observed mismatch between the observed spectrum and the theoretical spectrum is. In order to obtain a good limit of detection for the target substance, it is necessary to: * Define the model correctly, i.e. to specify correctly the shape of the background spectrum * Defme correctly the variability of the background spectrum Furthermore, if the target substance is on a substrate that shows large variations of spectrum, the limit of detection of the target substance will become worse. Indeed, the MLE algorithm will not be able to estimate whether the large deviation between the model and the data is due to the variability of the background or the presence of the target substance.

It is important to note that the detection limit is mainly determined by the precision with which the substrate spectrum is known and on the magnitude of the variability of the substrate spectrum.

References [1] Hecht. Optics, 3 edition. Addison Wesley press, 1998.

[2] M. Born, E. Wolf. Principles of optics, Cambridge University press, 1999.

[3] Hemandez et al. Journal NIR spectroscopy 11,97-107(2003) [4] Probability and statistics by M. J. Evans & J. S. Rosenthal (2003) W. H. Freeman and company [5] S. Baker et a!. Nuclear Instruments and Methods in Physics Research (1984) 221, 437-442 [6] J. R. Lakowicz M. (1999) Principles of fluorescence spectroscopy, Plenum Publisher, New York [7] E. Keyszig (1993) Advanced engineering mathematics (7th edition) John Wiley & sons

Claims

Claims 1) A Spectrophotometer comprising excitation optics, detection

optics and a microprocessor in which the excitation optics comprises a light source, light focussing means and means for guiding the light to a sample to be analysed, and in which the detection optics comprises means for receiving modified light from the analysed sample, and means to convert the received modified light from the sample into an electrical signal for interpretation by a microprocessor, the microprocessor comprising a reference library and a sample determination algorithm for comparing the received electrical signal to a library of reference electrical signals to identifS' a compound or class of compounds in the sample whereby, in use, the light source emits light which is focussed by the light focussing means onto the means for guiding the light to the sample, the light guiding means guides the light to the sample, the sample modifies the received light and transfers a modified light signal indicative of the nature of the sample which signal is received by the light receiving means which conveys the signal to the light converter means, characterised in that the light converter means comprises a plurality of photodetectors, and in that the photodetectors are capable of detecting different wavelengths of light such that the information received can be formed to make a contiguous spectrum which is usable for interpretation of the modified light to determine the nature of the sample.

2) The spectrophotometer according to Claim 1, characterised in that the light source is polychromatic.

3) The spectrophotometer according to Claim 1 or Claim 2, characterised in that the means for guiding the light to the sample comprises an optical probe.

4) The spectrophotometer according to any one of Claims 1 to 3, characterised in that the optical probe optimises both the light received by the sample and the modified light reflected, scattered or transmitted by the sample.

5) The spectrophotometer according to any one of claims 1 to 4, in which the modified light received by the optical probe is transmission light, reflected light, transfiected light and interacted light.

6) The spectrophotometer according to any one of Claims I to 5, characterised in that the detection means comprises means to select a narrow band of wavelengths of light chosen from a wider band of wavelengths of light.

7) The spectrophotometer according to Claim 6, characterised in that the means to select a narrow band of wavelengths is mechanical.

8) The spectrophotometer according to Claim 6 or Claim 7, characterised in that the means to select a narrow band of wavelengths is a monchromator.

9) The spectrophotometer according to Claim 8, characterised in that the monochromator is a moving monochromator.

10) The spectrophotometer according to Claim 9, characterised in that the moving monochromator is a Czerny-Tumer type monochromator.

11) The spectrophotometer according to Claim 9 or claim 10, charactensed in that the moving monochromator comprises a light diffraction grating selected from the group comprising one or more of; volume holographic transmission grating, ruled surface relief diffraction grating, and holographic surface relief reflection grating.

) The spectrophotometer according to any one of the preceding claims, characterised in that the light source emits light in the wavelength range of about 900nm to about
2.5 Iun.

13) The speetrophotometer according to any one of the preceding claims, characterised in that the light source is selected from the following group; bulb, arc lamp, light emitting diode and laser.

14) The spectrophotometer according to any one of the preceding claims, characterised in that the light source is a quartz tungsten halogen bulb.

15) The spectrophotometer according to any one of the preceding claims, characterised in that the position of the light source is adjustable.

16) The spectrophotometer according to any one of the preceding claims, characterised in that the excitation optics light focussing means is selected from any one of the group comprising a lens, a prism or a reflective surface.

17) The spectrophotometer according to Claim 16 characterised in that the light focussing means is a piano convex lens.

18) The spectrophotometer according to any one of the preceding claims, characterised in that the excitation optics comprise a collimator optimally positioned to direct the light emitted by the light source to the light focussing means.

19) The spectrophotometer according to Claim 18, characterised in that the collimator is a light reflective surface.

20) The spectrophotometer according to claim 19, characterised in that the collimator light reflective surface is parabolic.

21) The spectrophotometer according to any one of the preceding claims, characterised in that the means for guiding the light to a sample is a fibre optic cable.

23) The spectrophotometer according to Claim 21, characterised in that one end of the fibre optic cable is associated with the body of the spectrophotometer and the other end of the fibre optic cable is associated with the body of the optical probe by a connector means.

23) The spectrophotometer according to Claim 22, characterised in that the connector is a releasable connector.

24) The spectrophotometer according to any one of the preceding claims, characterised in that the light intensity of the light source is monitored with a reference photo-detector optimally placed within the excitation optics.

25) The spectrophotometer according to any one of the preceding claims, characterised in that the photo-detector is selected from the group comprising; photo-resistor or light dependent resistor, photo-voltaic cell, solar cell, photo-diode, photo-multiplier tube, photo-tube containing a photo-cathode, photo-transistor, or a sensor that detects light via a change in temperature due to illumination.

26) The spectrophotometer according to Claim 25, characterised in that the photo-detectors collectively detect light in the wavelength range of about 900nm to about 2100mm 27) The spectrophotometer according to Claim 26, characterised in that the photo-detectors comprise a first photo-detector which detects light in the wavelength range of about 900nm to about l600nm and a second photo-detector which detects light in the wavelength range of about l400nm to about 2lOOnm.

28) The spectrophotometer according to any one of claims 25,26 and 27, characterised in that the photo-detector is selected from one or more of the following group comprising; photo-resistor or light dependent resistor, photo-voltaic cell, solar cell, photo-diode, photo-multiplier tube, photo-tube containing a photo-cathode, photo-transistor, and sensor that detects light via a change in temperature due to illumination.

29) The spectrophotometer according to any one of claims 25 to 28, characterised in that the position of each detector within the spectrophotometer is adjustable.

30) The spectrophotometer according to any one of the preceding claims, in which the detection optics comprise a slit optimally arranged in-front of the light sensitive surface of the photo-detector between the photo-detector and the means to select a narrow band of wavelengths of light, to limit the range of wavelengths of light reaching the photo-detector.

31) The spectrophotometer according to Claim 30, characterised in that the position of each slit within the spectrophotometer is adjustable.

32) The spectrophotometer according to any one of the preceding claims characterised in that the detection optics comprise a plurality of filters optimally arranged in-front of the slit, between the and the means to seLect a narrow band of wavelengths of Light, to prevent unwanted light impinging upon the photo-detector.

33) The spectrophotometer according to Claim 32, characterised in that the position of each filter

is adjustable.

34) The spectrophotometer according to any one of the preceding claims, which comprise one or more baffles optimally positioned to prevent unwanted light impinging on the photo-detector.

35) The spectrophotometer according to any one of the proceeding claims, comprising a power supply in which the power supply for the spectrophotometer is selected from one or more of the following group; mains power supply, battery power supply and power supply from a personal computer.

36) The spectrophotometer according to Claim 35, characterised in that the power source is a rechargeable battery.

37) The spectrophotometer according to any one of the preceding claims, in which the photo-detector cooperates with the microprocessor by a wireless connection.

38) The spectrophotometer according to any one of the preceding claims characterised in that the spectrophotometer further comprises one or more communication ports.

39) The spectrophotometer according to any one of the preceding claims, characterised in that the spectrophotometer comprises a internal scaffolding upon which movement sensitive .omponents are mounted and said scaffolding is connected to an external cover by one or more shock absorption means.

40) The spectrophotometer according to claim 39, characterised in that the external cover protects the spectrophotometer from its environment.

41) The spectrophotometer according to any one of the preceding claims characterised in that the spectrophotometer can operated in an environmental temperature of about minus 5 degrees celcius to about plus 35 degrees celcius.

42) The spectrophotometer according to any of the preceding claims characterised in that the algorithm account for the modifications of electrical signal induced by sample temperature before comparing it to a library of reference electrical signals to identify a compound.

43) The spectrophotometer of any of the preceding claims characterised in that the spectrophotometer is a hand held instrument or is mounted on a vehicle, vessel or aircraft or forms part of a production line.

44) The spectrophotometer according to any of the preceding claims characterised in that the spectrophotometer further comprises a means to input and receive information and a means to view said information.

45) The spectrophotometer as herein described with reference to and as illustrated by figures 1 to 54 of the accompanying drawings. * -

Amendments to the claims

CLAIMS

1. A spectrophotometer comprising means for directing light having a plurality of wavelengths to means arranged to produce a spectninl of different wavelengths distributed spatially, a first photodetector for receiving at any given time light in said spectrum within only a relatively narrow range of wavelengths, and means for effecting relative movement between the first photodetector and the spectrum so that the range of wavelengths received by the photodetector varies as a function of tune; wherein there is provided a second photodetector which is displaced spatially from the first photodetector in the direction of increasing wavelength in the spectrum, the means for effecting relative movement between the first photodetector and the spectrum also eflcting relative movement between the second photodetector and the spectrum, whereby at any given time the second photodetector receives light in said spectrum within a relatively narrow range of wavelengths which are substantially greater than those of the relatively narrow range of wavelengths being received simultaneously by the first photodetector at tlmt time; and wherein the first photodetector has a first range of wavelengths over which it is operable and a first upper operating limit, and the second photodetector has a second range of wavelengths over which it is operable and a second upper operating limit, the second : * 20 range overlapping the first range and the second upper operating limit being greater :..::: limit. **.*

* * 2. A rophotometer as claimed in claim 1, wherein the means arranged to * produce a spectrum of different wavelengths distributed spatially is a monochroinator.

* :;
3. A spectrophotometer as claimed in claim 1 or 2, wherein the means for * u effecting relative movement between the first photodetector and the spectrum comprises means fOr moving an optical component of the monochromator.

4. A spectrophotometer as claimed in claim 3, wherein the optical component of the monochromator is a diffraction grating.

5. A spectrophotometer as claimed in claim 4, wherein the morochromator is a Czerny-Tumer mcmochromator. l:3Q

6. A spectrophotometer as claimed hi any preceding claim, wherein the first and second photodetectors provide a combine operable wavelength range of at least about 1000 nm to about 2000 nm.

7. A spectrophotometer as claimed in claim 6, wherein the operable wavelength ranges of the first and second photodetectors overlap in a wavelength range of about 1400 nmto about l600nm.

8. A spectrophotometer as claimed in claim 7, wherein the first photo-detector has an operable wavelength range of about 900 nm to about 1600 rim, and the second photo-detector has an operable wavelength range of about 1400 am to about 2000 am.

9. A spectrophotometer as claimed in any preceding claim, wherein the spatial positions of the photodetectors and the operation of the means for effecting relative movement between the photodetectors and the spectrum are such that the first and second photodetectors are exposed to different but overlapping parts of the spectrum.

10. A spectrophotometer as claimed in any preceding claim, wherein there is provided a light source, a probe for directing light to a sample to be analysed and ibr receiving modified light from the sample, and means fbr directing the nxxhfled light to the means arranged to produce a spectrinn of different wavelengths. I. * * * * * **

* :.: ii. Aspectrophotometer as claimed in claim 10, wherein data processing means are provided for processing signals from the first and second photodetectors so as to : : produce data fbr use in identifying identify a compound or class of compounds in the *. .: sample.

12. A spectrophotometer as claimed in claim 11, wherein in at least part of the region where the second range of wavelengths overlaps the first range of wavelengths, signals from both the first and second photodetectors are used.

13. A spectrophotometer as claimed in claim 10,11 or 12, comprising a portable bodyunit to be carried by auser, th odyunit comprising apower source the light source, a fibre optic connection to the probe to transmit light to the sample, a fibre optic connection to the probe to receive the modified light from the sample, the means arranged to produce a spectrum of different wavelengths, the photodetectors and the data processmg mean& 14. A spectrophotometer as claimed in claim 13, wherein the data processing means has access to a library of data associated with compounds or classes of compounds, for comparison with data derived from signals from the photodetectors 15. A spectrophotometer as claimed in claim 13 or 14, wherein components of the means arranged to produce a spectrum of different wavelengths and the photodetectors are mounted within the body unit by shock absorbing means.

16. A spectrophotometer as claimed in any preceding claim, wherein the positions of the first photodetector and second photodetector are adjustable.

17. A spectrophotometer as claimed in any preceding claim, wherein between each of the photodetectors and the means arranged to produce a spectrum of different wavelengths there is provided a respective slit to limit the range of wavelengths of light reaching the photodetector at any given time.

18. A spectrophotometer as claimed in claim 17, wherein the position of each slit

is adjustable. * ..* * * 19.

each slit and the means arranged to produce a spectrum of different wavelengths is provided a respective filter to select a narrow band of wavelengths of light.

20. A spectrophotometer as claimed in claim 19, wherein the position of each filter is adjustable.

21. A spectrophotometer as claimed in claim 1, comprising a housing containing a light source, the means arranged to produce a spectrum of different wavelengths distributed spatially, and the first and second photodetectors; a first light transmitting connection between the housing and a probe for transmitting light from the light source to a sample to be analysed, a second light transmitting connection between the housing and the probe for receiving light from the sample to be analysed, and means for directing the received light to the means arranged to produce a spectrum of different wavelengths distributed spatially; wherein optical components of the means arranged to produce a spectrum of different wavelengths distributed spatially and the photodetectors are mounted rigidly to a chassis which is in turn mounted in the housing by means for at least partly absorbing shock and I or vibration.

22. A spectrophotometer as claimed in claim I wherein a light source is mounted by means of an adjuster providing for adjustment laterally with respect to the optical axis of the light source.

23. A spectrophotometer comprising a housing containing a light source, a first light transmitting connection between the housing and a probe for transmitting light from the light source to a sample to be analysed, a second light transmitting connection between the housing and the probe for receiving light from the sample to be analysed, and, within the housing, optical components comprising means for directing the received light to means arranged to produce a spectrum of different :. 20 wavelengths distributed spatially, and at least one photodetector for detecting light in :..::: the spectrum, wherein the optical components are mounted rigidly to a chassis which is mounted within the housing by means for at least partially absorbing shock and / or vibrations. * ***

S S

24. A spectrophotometer comprising a housing containing a light source, a first :: light transmitting connection between the housing and a probe for transmiUing light from the light source to a sample to be analysed, a second light transmitting connection between the housing and the probe for receiving light from the sample to be analysed, and, within the housing, means for directing the received light to means arranged to produce a spectrum of different wavelengths disthl,uted spatially, and at least one photodetector for detecting light in the spectrum, wherein the light source is mounted by means of an adjuster providing for adjustment laterally with respect to the optical axis of the light source. ks3

25. A spectrophotometer comprising means for producing a spectrum of different wavelengths or light distributed spatially and for directing the spectrum to a plurality of photodetectors spaced from each other in a direction such that at any given time each of the photodetectors rccewes light in a pail of the spectrum which is different to that pail of the spectrum in respect of any other of the photodetectors, the photodectors having different response characteristics so that different photodetectors provide optimal outputs for different parts of the spectrum. * . * *** *.. * * **** ** . * * * * *. * . * *** * . ** * * * . * * *. . * S I * *.