Research Paper

Hand-held instrument for fat/water determination in whole fish

M. Kent

The fat content is related to its water content, and measurement of one serves to determine the other. A compact microwave instrument is described which utilizes this fact to measure the fat content of pelagic and other fatty species of fish. The instrument discussed relies on a microstrip sensor. Its use with herring is demonstrated and its further development is considered.

Keywords: Fish; fat content; water content; hand-held instrument


Some years ago a novel type of sensor was described by the author for use in microwave moisture measurements (Kent, 1972; Kent, 1973). The principle of this sensor was the modification of the propagation characteristics of a form of transmission line know as microstrip or stripline by placing it in contact with the material of interest. The electrical behaviour of many materials in electromagnetic fields, especially those with polar constituents like water can be described by use of a complex dielectric permittivity in which the real part, , is related to the ability of the material to store energy and the imaginary part , is an energy-dissipating component. This is normally written as follows.

e* = e’ – je”

It was demonstrated that an important factor in this application of a stripline was the complex dielectric permittivity of the material, particularly the real part. Calibration of the measured power loss in the sensor could be made as function of the permittivity but since for most food materials and many others, this depends on the water content (see Mudgett et al., 1974, for example) then the calibration could be in more practical terms. Steele and Kent (1978) described the use of the sensor with different food systems but for the purposes of this paper the most relevant results were those obtained by placing the sensor on the surface of whole herring.
Figure 1. Correlation between fat content and water content of the herring used in this work. Contents are expressed as a total weight percentage fraction.

It is widely know that the water content of pelagic species is complementary to the oil content. The reason for this is basic and arises from the need of the fish to maintain a density slightly greater than that of water (Iles and Wood, 1965). Any short term adjustment of net buoyancy is made with the swim bladder. The fat content can vary slightly in these species depending on the season and so the water content also varies. An example of the relationship between the two can be seen in Figure 1. The observed scatter is predominantly due to biological variation. From this regression it can be taken that fat content could be determined with an uncertainty of (1.3% from a precise knowledge of the water content. Iles and Wood quote a figure of (0.6% for a very much larger number of fish. This paper describes an instrument which uses this relationship for rapid determination of the fat/water content (Patent: Kent, 1989). The instrument’s performance for herring is then examined.

Materials and Methods

The fish used in this study were all herring (Harengus harengus) but taken from widely different geographical locations on both sides of the North Atlantic. Both fresh and thawed frozen fish were used since it was ascertained that there was no significant difference in the meter response between these treatments. The fish were all measured whole at 0(C. Fat and water reference determinations were made using only fillets. Fat determinations were made using the Foss-Let method. Water contents were determined by oven drying in air at 105(C for 48 h and recording the weight loss. Fat and moisture measurements are expressed as a percentage of total sample weight.

In order to demonstrate the dependence of the instrument readings on water content, a number of solutions of AR grade iso-propanol and distilled water were made ranging from 100% alcohol to 80% water.

Stripline sensor


Figure 2. The strip line sensor a, construction of strip line sensor used in the instrument, not to scale (A) Panel jack, (B) PTFE insulation. (C) jack contact (D) metal base plate and ground plate (E) dielectric substrate, (F) epoxy. (G) soldered strip line to jack contact. (H) copper strip line, (I) PTFF cover b, Photograph of actual sensor with an adhesive strip placed to indicate the position of the strip line which is not visible beneath the PTFE cover.

The stripline sensor is shown in Figure 2. Its structure may be considered as that of a coaxial line, which has been split longitudinally and opened out to expose the centre conductor. The outer cylindrical conductor of the coaxial line becomes a flat conductor known as the ‘ground plane’, the dielectric filling the coaxial line becomes a flat ‘substrate’ and the centre conductor is flattened into a thin planar strip on its surface. Any dielectric material now placed in direct contact with this stripline will modify the transmission properties in a way which will be dependent on the dielectric properties of this material. As has already been stated, the dielectric properties of foodstuffs depend very much on their water content, so these changes in transmission properties can be calibrated against water content. In general, any particular constituent of a material could be measured if the microwave dielectric properties are affected by that constituent.

Referring again to Figure 2, a protective layer of PTFE 0.1mm thick is fixed directly over the strip which both protects it from erosion or damage and at the same time reduces the sensitivity to a level which can be easily handled. Without this layer nearly all of the microwave energy can be coupled out of the strip by the superimposed material to be measured. This can result in high sensitivity for low water contents but distinct lack of sensitivity to change at the high moisture end of the range. The substrate material is a copper clad PTFE and ceramic mixture (Epsilam 10, 3Ms Co. Ltd) with a permittivity of 10 and a very low loss factor. The stripline is formed by photo-etching one copper-clad surface.

The substrate is 1.27mm in thickness and the strip itself, as used in this work, is 0.7mm in width. The sensitivity of the sensor also depends very much on the dimensions and properties of the structure and in general the stripline both loaded with sample or unloaded does not have the same impedance as the 50( coaxial line feeding it. This results in some impedance mismatching at the coaxial-to-stripline transition as well as at the boundaries of the loaded portion of the line. It has been shown, however, that reflections at such mismatches account for only a small proportion of the observed power losses in the line which are predominantly due to the interaction with the material of interest (Kent, 1973).

To enable the transition from coaxial line to stripline to be made, the coaxial line is fed through the substrate and soldered to the surface of the strip at the points marked G. In order that a flat surface can be presented to the fish these transition points are recessed slightly as shown. This makes negligible differences to the impedance of the line at those points. The thin PTFE layer covering the whole structure is uniformly flat. To achieve this flatness, a small quantity of expoxy fills the gap between the recessed transition and the cover, again with slight but negligible effect on impedance. The presence of the PTFE layer which is in contact with the sample also complies with food compatibility requirements.

Microwave Circuit

Figure 3. Basic microwave circuit for measurement of attenuation. (a) Microwave source, (b) isolator, (C) sensor, (d) detector, (C) signal amplifier, (f) display.

The transmission property of the sensor that is measured it its insertion or loss or the attenuation of power in the stripline and sample. Microwave circuits for the measurement of this can be fairly simple. The most basic is shown in Figure 3. Here a microwave source (a) feeds power through an isolator (b) then via a transmission line to the system under measurement (c). In this case this is the stripline sensor, but it could be any system or component in which losses occurred. In applications such as those which relate to moisture measurement, the system would most commonly be transmitting and receiving with the sample between them. The isolator prevents any reflected power from reaching the source and interfering with its operation.

After passing through the system under measurement, the unabsorbed power is detected by a crystal detector (d). The signal in this detector clearly diminishes as more power is absorbed in the system and can be calibrated against water content for example. However such a simple device has a major disadvantage, namely that fluctuations in the output power level of the source can be wrongly interpreted as changes in attenuation or ultimately in water content. In addition the response is not linear to attenuation (normally expressed in decibels) and for large values of attenuation, changes in the water content have considerably less effect than the same absolute change would have at a lower attenuation.

In this respect a better solution is adopted for the instrument described here. Although more complicated, it allows for transient operation (a necessary condition for conservation of battery energy as will be seen) since the detected signal power is always referred to the measured input power. The complete system is shown in Figure 4.
Figure 4. Actual rnicrowave circuit and associated electronics. (a) Gunn diode. (b) isolator, (c) l0 dB directional coupler, (d) 10dB attenuator, (C) strip line sensor, (V) detectors, (g) pre-amplifiers, (h) trimming amplifiers, (i) log-ratio amplifier, (j) output display and/or microprocessor. Dashed line indicates components housed in hand- held portion of instrument.

The boxed area is that part of the instrument housed in the hand-held probe. The other parts plus batteries and switches are in a separate unit connected to the probe by multicore cable (Figure 6). Power from a Gunn-diode microwave source (a) is fed via an isolator (b) to a 10 dB directional coupler (c). The purpose of the isolator is to prevent any reflected power from interfering with the operation of the Gunn diode. The directional coupler feeds 10 dB of the power to a reference detector (f). The major part of the power is fed through a 10dB attenuator (d) to the stripline sensor. The attenuator reduces the power to a level comparable with that in the reference arm. This level is also low enough to ensure square law operation of the detectors given the input power of a few milliwatts. It also reduces the effect on both the Gunn diode and the directional coupler of reflected power from the various mismatches in the sensor. Apart from the obvious mismatch of the sensor impedance there are also reflections from the coaxial-to-stripline transitions which are difficult to eliminate entirely. After passing through the sensor (e) the remaining power is detected by another crystal detector (f).

The Gunn diode oscillates at 10 GHz with a supply voltage of 7.5V and delivers 10 mW of power. The current drawn from the power supply for this is 200 mA, so in order to conserve battery energy and life the Gunn diode is switched on for the duration of the measurement. Since it never achieves thermal equilibrium it is, therfore, most probable that both the frequency and the power output are to some extent time-dependent. This is where the value of the reference system is demonstrated since the ratio of the power in the signal detector to that in the reference detector is always measured. Thus variations in power level during this transient operation are not important. The frequency shift that takes place is small enough to render negligible any frequency dependent changes that might occur in the various components. All components have fairly broad band characteristics (8-12 GHz). Although the system described operates in this particular frequency range (X-band) it can be constructed to function at any microwave frequency. For some applications it might in fact be better to make the measurements at 2GHz.

Electronic Circuit

Figure 5. Electronic circuit in detail IC1 = op amp, 0P200GP; 1C2 = op amp, AD647; 1C3 = l.2V reference, AD589; 1C4 = analog comp unit, AD647; Dl = diode, 1N4148.

The circuit developed for the processing of the d.c. signals from the two detectors is also shown schematically in Figure 4. In order to minimize component costs it operates on d.c. signals. This incurs some penalties with very small signals due to d.c. offsets at the amplifiers, but the added complexity of modulation and detection were considered to be undesirable.

The output from the reference detector is 100 mV while that from the signal detector ranges from 100 ( Vup to 100 mV depending on the loss in the sensor. These two d.c. signals are each fed directly to low gain (x8) pre-amplifiers (g).

The pre-amplifiers are simple inverting amplifiers with a gain of 8 based on low offset voltage, low power, operational amplifiers OP200GP (Figure 5). The outputs from them are connected through multicore screened cable to a second stage of amplification (h) mounted in the main part of the instrument. These second stage amplifiers (Figure 5) are constructed around an Analog Devices ultra low drift BIFET operational amplifier type AD647, operated with an adjustable gain up to 5 for optimization of the signal levels.
Figure 6. General view of instrument in use.

The output from these amplifiers is then fed to an Analog Devices real time analog computational unit (ACU) type AD538, which functions as a log-ratio device (I). The output of this, proportional to the logarithm of the ratio of the two signals, is fed to a digital panel meter in the hand-held part of the instrument. A range of only 30 dB can be measured due to the limitations imposed by: (a) the requirements that the microwave power be low enough at the detectors for them to operate in a square law region, i.e. output voltage proportional to the incident microwave power, (b) the offset voltages in the pre-amplifier stages which at high values of attenuation expected with this sensor on the leanest fish (80% water content) would be 28 dB (see Figure 8). Normally such offset voltages would be compensated but in this case they exhibit some variability. The variation is an unfortunate consequence of battery operation. Stabilization of the power supply whilst feasible would impose further undesirable and clearly unnecessary current drain on the batteries (PP9s).

For reasons of frequency stability and reproducibility, the power supply to the Gunndiode is stabilized however, using a 7.5V Zener diode in a stabilizing network.

In a final version, the instrument will contain a microprocessor to perform a small number of operations. Firstly, it could store the calibration equations found from experiment to be necessary for each species. In fact, the relationship between fat and water content for other species have been shown to be so similar as to be represented by a single regression line (Wallace and Hulme, 1977). If this is the case then calibration for one species should be sufficient. This does not, however, take into account possible differences of spatial distribution of the fat in different species, which may be relevant for the calibration of this instrument. Secondly, it will provide a facility for averaging a predetermined number of readings. This latter function will allow the mean fat content of a batch of fish to be estimated.


Sensitivity Check with Alcohol

By using solutions of iso-propanol and water the dependence of the sensor on water content can be well demonstrated. Iso-propanol has a dielectric relaxation spectrum with a relaxation frequency at 25(C of 600 MHz. This means that at 10 GHz, (the frequency of operation of this instrument) the permittivity is almost entirely real, i.e. no loss, and has a frequency-independent value of 4.0. Water at this frequency is very close to its relaxation peak (maximum loss) which occirs at 18 GHz. At 10GHz the real permittivity of water is 62 and the loss factor is 30. Despite the synergistic effects of mixing these two polar liquids, at 10 GHz the dielectric properties vary monotonically between the two concentration limits.
At lower frequencies, 3 GHz the loss factor can peak at some intermediate concentration (see Mudgett et al., 1974).
Figure 7. Results of instrument reading versus water content for iso – propanol / water solutions.

The results of immersing the sensor in these solutions maintained at 25(C(0.5(C are shown in Figure 7 from which the sensitivity of the sensor to variation in water content is seen. The reproducibility of repeated measurement is at least (0.1dB which is too small to show in this figure. The maximum concentration measured is not a limit of the sensor which could be used up to 100% water content. The level of 80% was chosen simply as typical of the highest water content encountered in herring (see Figure 1). Also, since >30 dB attenuation is found above this level of water content with the sensor, higher losses would not be accurately processed by the circuit due to the chosen range being restricted to 30 dB.

Measurements on Herring

After wiping the surface of the fish dry the sensor was placed on the middle portion of the fish parallel to an above the lateral line (see Figure 6). Both sides were measured in this way but there was no significant difference between the two overall calibrations.

As can be seen from Figure 8, the sensitivity decreases with increasing fat content. As the water content decreases simultaneously, this is probably due to a greater portion of the water molecules in the flesh being those which are rotationally hindered due to binding forces. This water is that which is more tightly bound to other molecular species in the tissue such as proteins, sugars, ionic salts, etc.

The calibration curve can be closely approximated by a logarthimic function relating attenuation A and fat content F.
Where a = 4.85 and b = 0.159.
Fish Fat Meter
Figure 8. Results of instrument reading versus fat content for whole herring The dashed lines are the 95% confidence interval.

The consequences of drying the surface was investigated by making measurements on 50 fish and simply observing the effect of wiping dry the surface. There was a significant decrease in the measured value of attenuation corresponding on the calibration to about 1-2% increase in fat content for the region of 12 dB ( 12 fat content). At the same time there was an increase in variation between left and right sides but as this was only just significant at the 5% level it can be safely concluded that only a cursory wipe is needed to free the fish of excess surface moisture. This may not even be necessary if the sensor is pressed firmly into the surface thus extruding all free moisture.

Variability of readings for herring

The likely sources of variation in the measurements are as follows: (a) Variation in skin and/or scale thickness between the sensor and the flesh. The skin is known to have a relatively high moisture content compared with the underlying tissue. (b) Small variations in residual surface moisture. (c) Variations is distribution of fat below the skin and in light dark muscle. The dark muscle is known to have a greater fat content but is largely avoided by placing the sensor above the lateral line. (d) With decreasing water content, the penetration depth of microwave energy into the flesh increases and reflections of power could occur at boundaries both between flesh and bone or even flesh and whatever surface the fish is lying on.

However, the dielectric properties of herring flesh are such that for a fish of say 18% fat content this penetration would only be 7mm at the frequency of measurement. The penetration depth is defined here for the propagation of a plane wave and is the distance over which its power would be reduced by 1/e. Thus reflected waves would be of very small amplitude when returning from distances greater than the penetration depth. This may prove to be a problem with smaller fish but has not been for herring. (e) Variation in the relationship between water content and fat content (see Figure 1).

In this work the regression equation for fat and water was found to differ from that of Iles and Wood (1965) (see Table 1). This is due to major differences in the fat determination, which in this work was on fillets only, but in Iles and Wood’s work had been on the whole fish minus gonads. In addition the sampling in the present study was from a wider geographical area compared with the earlier work which was on North Sea Herring alone.

Despite these perturbing influences, the calibration data show a quite narrow 95% confidence interval as can be seen in Figure 8 where the dashed curves represent these limits. For a fat content around 28% the 95% confidence interval is (2.4% fat. At 10% fat this is reduced to (0.5% fat, and at even lower fat contents around 5% it becomes (0.2%.

The confidence interval refers of course to a number of independent measurements made on a representative sample of fish. For single fish the predication interval must be invoked and the figures for that are much worse ranging from -10.4 +16.6% at the high fat content to -1.8 +2.9% at the low end of the range. Due to the logarithmic nature of the fitted curve, these prediction limits are asymmetrical and in fact are unduly pessimistic at high fat contents. They show, however, that the instrument must be used essentially to measure the mean fat content of a batch. In its final form, it will contain the necessary electronics to perform this task in direct analogy with the Torry fish quality meter based on dielectric measurements of fish flesh at 2kHz (Jason and Richards, 1975).

It is often the case that herring is caught as a by-catch, that it is found as a minority species in a haul of predominantly some other species. It is therefore appropriate as a final point to note that it has been observed that fish from a single batch, accumulated from such by-catches vary in fat content from as little as 2% up to 20%. One must then question for each application how useful or meaningful a batch measurement would be. Clearly the logarthimic nature of the calibration requires that such averaging be geometric rather than arithmetic.


From the scatter of the data about the calibration line, it is concluded that such measurements could yield a value for average fat content of a batch within 95% confidence limits of (2.4% at worst (high fat content) and typically much less. Thus for the estimation of the fat content of the batches the instrument seems useful especially when it is noted that the technique is non-destructive. It must be remembered that part of the error could arise from the method of calibration. Better fat analysis in the calibration procedure could result in a better performance of the instrument. It must be stressed, however, that a more accurate calibration would be against water content since the correlation with the fat content is inferred from the known relationship between fat and water.


Iles, T.D. and Wood, R.J. (1965) The fat /water relationship in North Sea herring (Clupea harengus) and its possible significance. J. Mar. Biol. Assoc. UK 45, 353-366

Jason, A.C. and Richards, J.C.S. (1975) The development of an electronic fish freshness meter. J. Phys. E Sci. Instrum. 8, 826-830

Kent, M. (1972) The use of strip-line configuration in microwave moisture measurement. J. Microwave Power 7, 185-193

Kent, M. (1973) The use of strip-line configuration in microwave moisture measurement II. J. Microwave Power 8, 189-194

Mudgett, R.E., Wang D.I.C. and Goldblith, S.A. (1974) Prediction of dielectric properties in oil-water and alcohol-water mixtures at 3000 MHz, 25(C based on pure component properties. J. Fodd Sci. 39, 632-635

Steele, D.J. and Kent, M. (1978) Microwave stripline techniques applied to moisture measurement in food materials. In: Proc. 1978 IMPI Symp. On Microwave Power pp 31-36, Ottawa, Ontario, Canada

Wallace, P.D. and Hulme, T.J. (1977) The fat/water relationship in the mackerel Scomber scrombus L., pilchard Sardinia pilchardus (Walbaum) and sprat, Sprattus sprattus L., and the seasonal variation in fat content by size and maturity. Fisheries Res. Tech. Rep. 35, MAFF Fisheries Research Laboratories, Lowestoft, UK

Received 27 November 1992
Revised 23 February 1993
Accepted 23 February 1993

Table 1

Regression parameters for fat (F) and water (W) in herring

n, number of samples ; Sf standard deviation for F about the regression line ii); r, correlation coefficient.
This work n = 24 Iles and Wood n=895
i) W 79.66 – 0.755 F 79.67 – 0.862 F
ii) F 100.4 – 1.254 W 90.45 – 1.139 W
r 0.973 0.991
r 0.973 0.991
Sf 1.353 0.614