132 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 1, JANUARY 2005 Cole Electrical Impedance Model—A Critique and an Alternative Sverre Grimnes and Ørjan G. Martinsen* Abstract—The Cole single-dispersion impedance model is based upon a constant phase element (CPE), a conductance parameter as a dependent parameter and a characteristic time constant as an independent parameter. Usually however, the time constant of tissue or cell suspensions is conductance dependent, and so the Cole model is incompatible with general relaxation theory and not a model of ﬁrst choice. An alternative model with conductance as a free parameter inﬂuencing the characteristic time constant of the biomaterial has been analyzed. With this free-conductance model it is possible to separately follow CPE and conductive processes, and the nominal time constant no longer corresponds to the apex of the circular arc in the complex plane. Index Terms—Bioimpedance theory, biomaterial, cell suspension model, Cole model, constant phase element (CPE), dielectric, electrical network theory, electrical relaxation theory, tissue model. Fig. 1. Single-dispersion equivalent circuits. (a) Debye circuit with ideal components, (1). (b) Cole circuit with the ideal capacitor replaced by a CPE with frequency-dependent components, (2) and (4). [ ] 1 =1 Z = R1 + R0 0 R1 1 + (j!Z ) I. INTRODUCTION TheColeimpedancemodel was introduced in itsﬁnalform byKenneth Cole in 1940 . It is based upon the replacement of the ideal capacitor in the Debye model  with a moregeneral constantphaseelement(CPE)as shown in the equivalent circuit diagrams, Fig. 1. The idea of introducing a CPE came after a number of ﬁndings, both in electrochemistry and with tissue and cell suspensions, that measured impedance loci in the complex plane were in the form of circular arcs with depressed circle centers, and that such arcs were best modeled with a CPE. The CPE together with a circular arc complex impedance analysis were introduced by Cole as early as 1928 . A Cole impedance model consists however of three parts: an equation, an equivalent circuit, and a complex impedance circular arc locus. The impedance model developed gradually in the 1930s until completed with the mathematical equation . In 1941 the Cole brothers introduced a similar model for dielectric permittivity, the Cole-Cole model . A Cole model is, therefore, an impedance model, a Cole-Cole model a permittivity model. The popularity of the models is due to the condensed complex form and the elegant and simple mathematical appearance, but as we shall see this is unfortunately at the expense of general applicability. It is the purpose of this paper to present a critique of the Cole model, and to analyze an alternative model with the time constant as a conductance dependent parameter. Such a model is in general better suited for tissue and cell suspensions, and actually this model often implicitly has been used. Only the basic Cole impedance model will be treated describing a single ideal dispersion. Often a biomaterial shows multiple dispersions, but it is not the purpose of this paper to analyze such systems, a more detailed description also of multiple-dispersion systems can be found in . II. DEBYE AND COLE MODELS The Debye circuit  is shown on Fig. 1(a). The impedance of the circuit is Z = R1 + 1 Gvar + Gvar j!Z Z = C=Gvar (1) Manuscript received June 6, 2003; revised May 9, 2004. Asterisk inidcates corresponding author. S. Grimnes is with the Department of Clinical and Biomedical Engineering, N-0027 Oslo, Norway and also with the Department of Physics, University of Oslo, N-0316 Oslo, Norway. *Ø. G. Martinsen is with the Department of Physics, University of Oslo, POB 1048, Blindern, N-0316 Oslo, Norway (e-mail: firstname.lastname@example.org). Digital Object Identiﬁer 10.1109/TBME.2004.836499 [ ] where Z is complex impedance ; R is resistance at very high frequencies, j is the imaginary unit, ! is the angular frequency [1/s], Z is the characteristic relaxation time constant [s] of the circuit cor=Z ; C is the responding to a characteristic angular frequency !Z parallel capacitance [farad, F], and Gvar is an independent parameter conductance [siemens, S]. The Cole empirical equation for the frequency dependence of tissue or cell suspension complex impedance is  (2) where R0 is the resistance at very low frequencies, Z is the characteristic time constant of the system corresponding to a characteristic angular frequency !Z =Z , and is an exponent [dimensionless]. The product !Z is dimensionless, and j!Z represents a CPE as long as is constant because =1 ( ) j = cos(=2) + j sin(=2) where = 'CPE =90 . (3) The equivalent circuit of a CPE consists of a resistor and a capacitor, both frequency dependent (nonideal) so that the phase 'CPE becomes frequency independent. Equations (1) and (2) describe and deﬁne one ideal dispersion. A more detailed description can be found in . = G the Cole equation becomes By introducing R0 0 R1 =1 1 1 Z = R1 + 1G + 1G(j!Z ) : (4) From (4), the capacitance CCPE of the CPE capacitor is found to be CCPE = 1G (!Z ) sin(=2): (5) ! The dimension of G=! and, therefore, CCPE , is [S s] or [F]. An equivalent circuit is shown in Fig. 1(b). As described in , (4) reveals that the Cole model presupposes a CPE element in parallel with an ideal conductance G. However, the CPE admittance G j!z is proportional to that same conductance G. Thus, a parallel conductance is not an independent parameter in the Cole model. Equation (5) shows that CCPE is proportional to G with the consequence that z is independent of G, just as Z in (1) is constant if C and Gvar vary by the same factor. is related to the constant phase of the CPE according to (3), as well as to the frequency exponent in the term ! . This double inﬂuence of reveals that the Cole model is based upon a Fricke compatible system. According to Fricke’s law , , the phase angle ' and the frequency exponent m of the impedance in many electrolytic systems are related so that 1 1 ( 1 ) 1 ' = m 1 90 Fricke's law: 1 (6) = In such cases it is common practice to set m . An admittance version of the Cole equation  is shown in Fig. 2 0018-9294/04$20.00 © 2005EEE Y 0 G0 = G0 + 1 +G1(j! Y )0 (7) IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 1, JANUARY 2005 Fig. 2. Admittance Cole model. (a) Equivalent circuit according to (7) with G G = R. (b) Complex plane Y -plot. ! is the characteristic frequency, from which is found: =! . 0 =11 =1 ( =1 ) III. NON-COLE MODELS A general equation without any constraints of Fricke compatibility is 1 Gvar + Kj !m Z (8) where K is a real proportionality factor [S sm0 ] for the CPE admittance. This equation is compatible neither with the Cole model, nor the Fricke law, nor Kramers–Kronig transforms. It is not a very attractive alternative, but shown here for the sake of completeness. It is possible to stick to the condition that the system shall be Fricke compatible and, therefore, also Kramers–Kronig compatible by introducing an independent ideal conductance Gvar in parallel with the CPE (in the case of admittance: ideal resistance Rvar in series). These earlier proposed  equations are 1 Gvar + G1 (j!Z ) 1 Y = G0 + : Rvar + R1 (j!Y )0 (9) Fig. 3 shows the most adequate equivalent circuits. To analyze (9) we use the Z -version as an example in the following. The complex admittance of the CPE element is: CPE G1 j!Z . When (9) is plotted in the Wessel plane (Fig. 4), the (angular) frequency corresponding to the apex of an arc is the characteristic frequency. From Fig. 4, we see that then the real impedance value is equal to Z 0 R1 = R0 0 R1 . With that expression put into (9) the measured characteristic frequency at the apex of the arc !Zm and the measured relaxation time Zm is Y + 1 2( = ( ) ) Zm = 1 = Z !Zm G1 Gvar 1= 1 : (10) The characteristic frequency corresponding to =Z is !Z . The Z parameter in (9) corresponds to the apex of the arc only when G1 Gvar : =Z and that is the Cole case. However, with !Z then !Zm G1 6 Gvar Z no longer corresponds to the apex of the arc, !Z has become just a nominal characteristic frequency. The new time constant = = =1 Fig. 4. Free conductance model, a Z -plot example. Three parameters R ;R ;' deﬁne a circular arc locus. Two additional parameters together with ! deﬁnes the frequency scale along the arc; for the Z-version of (9) it is the product G . ( ) Zm given in (10) shows that a change in Gvar now indeed inﬂuences the time constant in accordance with relaxation theory: a higher value of Gvar results in a shorter time constant Zm . The true time constant Zm =!Zm does not appear in (9) but is given by (10). With (10) it is found that !Zm is the crossover frequency when the CPE admittance YCPE G1 !Z becomes equal to Gvar . =1 = ( ) IV. COMPLEX PLANE Z-PLOT The Cole (2) has four parameters: R1 ; R0 ; Z , and ; and ! as the variable. Equation (9) has ﬁve parameters, for the Z-version the parameters are: R1 ; Gvar ; G1 ; Z , and . The ﬁve parameters are independent parameters, Zm in (10) is a derived, dependent but measured parameter. A. Forward Analysis Z = R1 + = Fig. 3. General equivalent circuits for (9) with a free-conductance/resistance parameter and a dependent parameter. (a) Z-version, (b) Y-version. 1 where Y is the complex admittance Y =Z Z [S], G is the conductance at very high frequencies, and G0 is the conductance at very low frequencies. Immittance values can be plotted in the complex (Argand or Wessel ) plane, Fig. 2(b) shows an admittance and Fig. 4 an impedance example. The immittance loci for all the equations given in this paper have the form of a circular arc. The Cole equations are compatible with the Kramers–Kronig transforms . A control of both Fricke  and Kramers–Kronig compatibility can be used for data quality evaluation. Z = R1 + 133 = Setting up a Cole model with the four parameters given is straightforward, so is setting up a model according to (9) with the ﬁve parameters given. In the latter case the geometry of the Z-arc is determined by the three parameters R1 ; Gvar , and . This is illustrated on Fig. 4 with R0 =Gvar R1 and 'CPE 1 . Two additional parameters together with ! deﬁne the frequency scale along a given arc: G1 and Z . The true relaxation time Zm is found from (10). =1 + = 90 B. Inverse Analysis The inverse problem of determining the four parameters of the Cole equation (2) from measurement results is simple; the parameters are read directly from the Wessel plot. For the Z case: R0 ; R1 , and 'CPE determine the arc geometry, and the characteristic time constant Z =!Z is found from the measured frequency at the apex of the arc. However, it may be impossible to interpret the results correctly if the experiment continues and new sets of data are sampled. Let us take the example that only !Z changes during an experiment. According to (4) that cannot be due to a change in the ideal conductance G, a conclusion which may be seriously wrong. Determining all ﬁve parameters in (9) from measured data is not possible. A circular arc is ﬁrst ﬁtted to the data in a Wessel plot. The = 1 1 134 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 1, JANUARY 2005 1 three geometrical parameters R ; Gvar , and are found from the arc geometry. For the frequency scale on the arc, (10) is rearranged G1 Z = Gvar : !Zm (11) Here the three parameters of the right hand side are found from the Wessel plot: Gvar ; !Zm , and . Accordingly, the product G1 Z is known, but not the individual values. As long as the product G1 Z and all other parameters are constant, arcs with different G1 0 Z combinations are indistinguishable from each other. A ﬁxed nominal value for Z together with G1 as a free parameter is then a preferred alternative. A ﬁxed nominal value for G1 together with Z as a free parameter is not preferred since Zm should be the dependent variable, not Z , (10). G1 can then be selectively followed during an experiment with a ﬁxed nominal value for Z . From (10) we have G1 = Gvar 1 (Z !Zm) : (12) The nominal time constant Z is not the true characteristic time constant Zm as found by measurement. In practice (10) is used in forward analysis to ﬁnd Zm with a given Z value. Equation (12) is used to follow G1 values with a ﬁxed nominal Z value. In conclusion, the inverse analysis has shown that the four parameters of the Cole model can be determined directly from the measurement results, but the interpretation of the results may be seriously misleading. In the new model four parameters can also be determined; in addition a ﬁfth parameter G1 can be calculated and selectively followed during an experiment. C. An Example Four skin impedance spectrographic complex data sets were obtained with three pregelled electrocardiogram-electrodes positioned on the skin of a human underarm. The effective skin-wetted area was 3 cm2 and electrode center distances 4 cm. The ﬁrst data set was taken 3 min after electrode onset to dry skin (Fig. 5 and Tables I and II), and the subsequent three sets after 6, 9, and 16 min (Tables I and II). Each spectrographic data set was measured in the frequency range 1 Hz–1 MHz with a Solartron model 1260/1294 setup and a measuring voltage amplitude of 10 mV(rms). The four circular arcs were determined with a ZView software package (Scribner Associates), Fig. 5 shows as an example the arc found with the ﬁrst data set (continuous line). With the non-Cole model the G1 values were calculated with a nominal Z [s]. With the non-Cole model (Table II) the samples showed a clear correlation with time between the and G1 (or G1 Z ) parameters. This is in agreement with the commonly accepted electrical model of human skin: the stratum corneum can be modeled by a CPE, and the sweat duct ﬁlling by a dc conductance in parallel , . The postulated process is, therefore: First, the electrolyte penetrates the most superﬁcial and electrically dominating layers of the stratum corneum resulting in a rapid increase of the admittance G1 of the CPE. Then, gradually, the sweat ducts are ﬁlled with contact electrolyte increasing the conductance path Gvar in parallel with the CPE. With the Cole model (Table I) the four data sets showed no clear correlations between the Z ; , and G parameters with time, making it impossible to postulate any particular process in the skin. =1 ( ) 1 V. DISCUSSION The Cole equation (2) or (4) is with the time constant as a conductance-independent parameter. In relaxation theory  the time constant is the time necessary to discharge a resistance–capacitance network, so that a large capacitance and resistance result in long time constants. The Cole equation is, therefore, incompatible with general relaxation theory. The alternative free-conductance model of (9) and (10) is in Fig. 5. Skin impedance plotted in the complex plane (broken line), human underarm. First spectrographic dataset as measured 3 min after electrode positioning on dry skin. Best circular arc found by regression analysis (continuous line) by the ZView program (Scribner Associates), calculated arc data given in the right lower corner. TABLE I COLE MODEL APPLIED TO THE MEASURED DATASETS TABLE II FREE CONDUCTANCE MODEL APPLIED TO THE SAME MEASURED DATASETS AS IN TABLE I. G WAS CALCULATED WITH (11) AND G WITH (12) agreement because Gvar inﬂuences the discharge process of the CPE capacitor so that higher conductance leads to higher characteristic frequency. This is true also if Z is regarded as a parameter describing a distribution of relaxation time constants . In biomaterials the capacitance values are usually much more stable than the resistance/conductance values. When that is the case, the Cole model with the time constant independent of conductance is not a logical starting point for model selection. When data show that the time constant or characteristic frequency indeed varies , this may be due to a variation in an independent conductance according to the alternative model of (9), or a change in the nominal time constant independently of conductance changes according to the Cole (2). Some authors have explicitly used a model similar to (9). Yamamoto et al. , for instance, called the independent parallel resistor R2 and had it connected directly in parallel with the CPE as in Fig. 3(a). They measured the skin impedance of about 60 persons on four different occasions, and found low correlation between R2 and the CPE capacitance, but a high correlation between R2 and , in agreement with our argumentation. As long as many published varying time constant results are due to changing conductance, and the authors do not discuss the use of Cole models based upon a time constant as an independent parameter, many of them implicitly actually may have used the free-conductance model of (9). IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 52, NO. 1, JANUARY 2005 An immittance equation should be in accordance with the theory of physics and electrical networks . The problem of dimensions is already apparent in Frickes classical paper . There, he presented 1 the formula for his frequency-dependent capacitance as Cp !0m , which implies that either the constant cannot have the dimension of capacitance or that the formula is dimensionally heterogeneous. Many authors have used such formulas, for instance –. However, network theory requires an angle in immittance equations, and the problem of dimensions is solved if ! (angular frequency) is replaced by ! (angle). may be seen as a simple frequency scaling factor, and thereby looses some of its value as a biomaterial characteristic constant. An alternative to Frickes formula is then Cp C1 ! 0m where C1 , compare is a capacitance with the value corresponding to ! with (5). Several authors have also used the ! term alone but with hidden into a “constant” , –, so that their equations actually are homogeneous. This is in accordance with the frequency scale being determined by the product G1 Z . Commercial software packages do it the same way, e.g., in ZView (Scribner Associates) with the parameter CPE1-T being equal to G1 [S], and the value of Z chosen to be Z [s]. Equations (1), (2), (4), and (7)–(9), in both impedance and admittance versions, have perfect circular arc loci in the immittance plane and may, therefore, be equally preferable candidates for a ﬁt to experimental data. Having the same descriptive power, a choice has to be made on the basis of knowledge about the biomaterial examined and the desired properties to be modeled. For instance, if the biomaterial is the human skin in vivo, the impedance model of Fig. 3(a) and (9) is a good starting point with the stratum corneum modeled by the CPE, and the independent ﬁlling of the sweat ducts by Gvar , cf. the example given in Section V. The Cole model of Fig. 1(b) is not a preferred model because it does not have such an independent conductance. If the biomaterial is a cell suspension, the Y equivalent circuit of Fig. 3(b) and (9) may be a good starting point. Then, Rvar models the cell interior, and the CPE models the cell membranes. R as a dependent parameter according to the Cole Y-model (Fig. 2) is less preferable. In this paper, we have limited our treatment to the CPE as an adequate model element for biomaterial dispersion analysis, without discussing its physical meaning. Cole himself was very well aware of the poor explanatory power of his models, , . Cole apparently with time became less comfortable with his own model, and did not focus on it in his summing-up book . Fricke did not use the Cole equations. = const = ( ) =1 =1 1 VI. CONCLUSION The concept of dispersion is very fundamental to the ﬁeld of bioimpedance, and it is important to know the properties and limitations of the models describing the immittance spectra usually found. The existence of a dependent conductance but an independent time constant parameter in the Cole model makes it incompatible with general relaxation theory and a more specialized model than usually realized. The analysis of the alternative free-conductance model with the relaxation time as a dependent parameter has shown that the model is preferable both because it is in accordance with general relaxation theory and actually measured data. Our results show that the time constant in the non-Cole equations no longer corresponds to the apex of the circular arc, and that the G1 and Z parameters cannot be determined separately from measured data, only the product G1 Z . However, it is possible to follow the relative changes in G1 selectively during an experiment. We ﬁnd that dispersion analysis by circular arc plots in the complex plane together with equivalent circuit modeling is still an important part of the analysis of bioimmittance data. 135 REFERENCES  K. S. Cole, “Permeability and impermeability of cell membranes for ions,” in Proc. Cold Spring Harbor Symp. Quant. Biol., vol. 8, 1940, pp. 110–122.  P. Debye, Polar Molecules. New York: Dover, 1929/45.  K. S. Cole, “Electric impedance of suspensions of spheres,” J. Gen. Physiol., vol. 12, pp. 29–36, 1928.  K. S. Cole and R. H. Cole, “Dispersion and absorption in dielectrics. I. Alternating current characteristics,” J. Chem. Pys., vol. 9, pp. 341–351, 1941.  S. Grimnes and Ø. G. Martinsen, Bioimpedance and Bioelectricity Basics. New York: Academic, 2000.  H. Fricke, “Theory of electrolytic polarization,” Phil. Mag., vol. 14, pp. 310–18, 1932.  H. P. Schwan, “Electrical properties of tissue and cell suspensions,” in Advances in Biological and Medical Physics, J. H. Lawrence and C. A. Tobias, Eds. New York: Academic, 1957, vol. V, pp. 147–209.  P. Fatt, “An analysis of the transverse electrical impedance of striated muscle,” Proc. Roy. Soc., B, vol. 159, pp. 606–651, 1964.  T. Yamamoto and Y. Yamamoto, “Dispersion and correlation of the parameters for skin impedance,” Med. Biol. Eng., vol. 16, pp. 592–594, 1978.  F. F. Kuo, Network Analysis and Synthesis. New York: Wiley, 1966.  L. Gougerot, “Des schemas electriques d’impedance equivalents a la peau humaine,” J. Radiol. Electrol., vol. 29, pp. 479–486, 1948.  G. Falk and P. Fatt, “Linear electrical properties of striated muscle ﬁbers observed with intracellular electrodes,” Proc. Roy. Soc., B, vol. 160, pp. 69–123, 1964.  I. Epelboin and M. Keddam, “Faradaic impedances: Diffusion impedance and reaction impedance,” J. Electrochem. Soc., vol. 117, pp. 1052–1056, 1970.  E. T. McAdams and J. Jossinet, “The detection of the onset of electrodeelectrolyte interface impedance nonlinearity: A theoretical study,” IEEE Trans. Biomed. Eng., vol. 41, pp. 498–500, May 1994.  B. Rigaud, J.-P. Morucci, and N. Chauveau, “Bioelectrical impedance techniques in medicine, bioimpedance measurements, impedance spectroscopy,” Crit. Rev. Biomed. Eng., vol. 24, no. 4–6, pp. 257–290, 1996.  J. R. Macdonald, Impedance Spectroscopy, Emphasizing Solid Materials and Systems. New York: Wiley, 1987.  Y. Yamamoto and T. Yamamoto, “Analysis for the change of skin impedance,” Med. Biol. Eng., vol. 15, pp. 219–227, 1977.  T. Yamamoto and Y. Yamamoto, “Non-linear electrical properties of skin in the low frequency range,” Med. Biol. Eng., vol. 19, pp. 302–310, 1981.  K. S. Cole, Membranes, Ions, and Impulses. Berkeley, CA: Univ. California Press, 1972.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project