Modern network analyzers have greatly enhanced the productivity of the microwave industry. Errorcorrection techniques have made it possible to use non-perfect hardware and still achieve very good performance [1][2]. A significant question is, "what really contributes to the accuracy and performance of a network analyzer?" Certainly, the error-correction concept of mathematically removing hardware errors has made a significant impact. New error-correction methods like TRL, LRL, TRM, and LRM have simplified the calibration process and also provided better accuracy. Through refined machining, the quality of the calibration standards has improved the error-correction accuracy. Also the modeling of standards has significantly improved. However, one overlooked fact is uncorrected "raw" hardware performance and its effect on the system accuracy. A common misconception is that error-correction "does it all," and that it can calibrate out, or quantify, any level of uncorrected performance. It is clear that the more stable the hardware, the better the calibration process can correct the errors. The calibration will then remain stable as a function of time and temperature, and calibrations will not need to be updated as often.

Page 1

The new error correction methods will be reviewed. Then the measurement accuracy will be defined and measurement assurance will be established. The design requirements for the hardware will then be described that is necessary for good error corrected performance. The last issue is selecting the best calibration method for various applications.

Page 2

There are two basic methods used widely in the industry today. The first method uses known calibration artifacts to determine the system error terms. There must be as many known characteristics of the calibration standards as there are error terms. The second method uses the redundancies that exist in the system equations to determine the system error terms as well as some of the characteristics of the calibrations standards.

Page 3

the fictitious error adapter must be a four-port. and 2 transmission tracking terms. There are 2 directivity terms. 2 port-match terms. Page 4 . The errors of the switch can be removed as long as we can measure all four waves at the coupler ports at the same time. Eight of the error terms are leakage between the various ports. In this case. This error adapter has 16 error terms. 2 reflection tracking terms.SLIDE 4 All the linear errors of the imperfect reflectometer can be combined into an error adapter yielding a system model with a perfect reflectometer combined with a fictitious error adapter. The number of error terms is the square of the number of ports.

The T-parameters describe the input wave as a function of the output waves. However it is better to describe the system error adapter in terms of the cascading T-parameters [3]. The T-parameter solution yields a much more compact result.SLIDE 5 The equations for the device under test and the measured results are described in terms of S-parameters. Page 5 .

This is easily solved for the actual S-parameters.SLIDE 6 With the T-parameter formulation. Page 6 . the solution for the measured S-parameters is the matrix form of the bilinear transformation.

Due to ratio measurements and symmetry at the test ports. Many practical measurement systems in coax and waveguide do not to use all 16 error terms. in wafer probing systems or fixtures. but the method of solution is the same. DETAILED CALIBRATION ANALYSIS The 12-Term Error Model The errors associated with a two-port system are in general characterized by an error adapter. see Figure 1 and 2. Currently in industry the 12term error model is used. The matrix bilinear transformations can be easily written in linear equation form. However. This is adequate for coaxial and waveguide measurements since leakage and coupling errors are usually negligible in these systems. 8 of the error terms are leakage terms and are not significant. Page 7 . This simplifies the math a little. The system of four linear matrix equations can then be solved for the 16 system error terms. The most common method neglects 6 of the leakage terms resulting in only 10 (9 for ratio measurements) unknown error terms. the other error terms are more significant and could be considered. Since there are four 2x2 matrix error terms.SLIDE 7 The real benefit of the T-parameters is in the solution for the error terms. there are only 14 error terms required. four known two-ports can be used as calibration standards. which does not include all the leakage and coupling terms.

e11S11a . In the forward direction from Figure 1.Figure 1 Figure 2 The branches of the flow graph are the error terms (e-terms) that describe the linear microwave errors of the system.e22S22a + e11e22∆Sa b ´ S22a . the actual S-parameters (Sa) can be determined by measuring the 4 S-parameters and if the 12 error terms are known.e11S11a . e00 = Forward directivity e11 = Port-1 match in the forward direction e22 = Port-2 match in the forward direction e10e01 = Forward reflection tracking e10e32 = Forward transmission tracking e30 = Forward isolation And in the reverse direction from Figure 2.e22´S22a + e11´e22´∆Sa (5) ∆Sa = S11aS22a . b S11a . e33´ = Reverse directivity e22´ = Port-2 match in the reverse direction e11´ = Port-1 match in the reverse direction e32´e23´ = Reverse reflection tracking e23´e01´ = Reverse transmission tracking e03´ = Reverse isolation From Figure 1 and 2 the measured S-parameters (Sm) can be calculated.e11´S11a .e22S22a + e11e22∆Sa b S21a (2) S21m = 3 = e30 + (e10e32) a0 1 .e11´S11a .e22∆Sa (1) S11m = 0 = e00 + (e10e01) a0 1 .e22´S22a + e11´e22´∆Sa b ´ S12a (4) S12m = 0 = e03´ + (e23´e01´) a3´ 1 . Page 8 .S21aS12a From these 4 equations above.e11´∆Sa (3) S22m = 3 = e33´ + (e23´e32´) a3´ 1 .

These probes utilize coplanar transmission lines which inherently have leakage and capacitive coupling. The 16-Term Model However.S11m3 = 0 (19) Equations (17) to (19) can be solved for e00. in recent years. There are also common mode coupling between the probes through inductive ground paths. The errors shown by the dotted lines are the coupling and leakage terms not included in the 12-term error model.e30 S12m .e03´ ∆ = ( 1 + ( 11m e e10e01 ) 11)  e10e32  e23´e32´    e23´e01´  Calibrating the 12-Term Model Using Known Standards The 12 error terms can be determined be using three known one-port standards and a thru connection. The 16 term error adapter is shown in Figure 3.S11m = 0 Where (14) a = e00e11 .e10e01 (15) b = e00 (16) c = -e11 With three known one-port standards.cS11a1S11m1 . Other systems involving Microstrip and similar transmission lines also have errors that are not accounted in the 12-term error model.e22( e10e32 )  e23´e01´  ∆ ( (7) S21a = S -e ´ S21m .e11S11a Which can be written as the bilinear transform.e03´ e10e01 e10e32  e23´e32´   e23´e01´  ∆ (9) (10) S12a = S12m .e11e22 Since e00.e03´( 1 + ( S11m . If one-port standards are connected to port-1 in the forward direction Equation (1) becomes.e33´  ( e10e01 )  1 +  e23´e32´ e22´ .e00 S21m .e00) e11) . There has also been an increased interest in improving Microwave Integrated Circuit testing.e33´( 1 + ( S11m .e00 1 + S22m . And with the same thru connection Equation (2) becomes b 1 (22) S21m = 3 = e30 + (e10e32) a0 1 . and e10e01 are known.e30) S12m .e11´)) e10e01  e23´e01´  ∆ S . Equation (18) can be solved for e22.S11m2 = 0 (18) aS11a3 + b .e33´e22´ . e11. (13) aS11a + b . With matched load on port-1 and port-2 the isolation term can be determined directly from Equation (2). which can be solved for directly.e11´( S21m . Equations (1) is reduced to b e22 (21) S11m = 0 = e00 + (e10e01) a0 1 .cS11aS11m .S11m1 = 0 (17) aS11a2 + b .cS11a2S11m2 .cS11a3S11m3 . aS11a1 + b . e11.(6) S11a = S11m .e30) S12m . b S11a (11) S11m = 0 = e00 + (e10e01) a0 1 .e00) (e11 . aS + b (12) S11m = 11a cS11a + 1 And can also be written in the linear form. (20) S21m = e30 With a thru connection. It should be noted that this model completely accounts for all coupling and leakage terms in the error system. One system used for on-wafer testing involves probes. This is especially true in the area of Microwave Integrated Circuit (MIC) fabrication and testing. other forms of transmission lines such as microstrip lines and coplanar waveguides have become increasingly popular.e03´ S22m . three independent equations can be written. The 6 forward error terms are known and the same procedure can be used to solve for the 6 error terms in the reverse direction.e11e22 All the terms are known except e10e32. Page 9 .e22) e10e32 )   e23´e32´   ∆ (8) S22a = S22m .e22e11´( S21m .e30  1 +  22m 33 (e22´ . and e10e01.

The error adapter system can be represented in terms of T-parameters as shown below.E1)-1E2 + E4]-1 Theoretically by using different two-port and one-port calibration standards. 21a 22a By using linear algebra.Figure 3 From this model a system of equations can be developed. enough independent equations can be obtained to solve for the error terms. However.E4Sa)-1E3 (28) Sa = [E3(Sm . the error terms are difficult to solve for numerically because they are in terms of products and quotients. 21m 22m a1 b1 S S (26) Sa = ( S11a S12a) ( a2) = Sa ( b2) . solutions for the measured S-parameters (Sm) and actual S-parameters (Sa) of the DUT in terms of the error adaptor (E) may be obtained as: (27) Sm = E1 + E2Sa(I . The error adapter is depicted in matrix form given by: 0 0 a b b3 a3  = E (23) b1 a1 b2 a2 e00 e03 | e01 e02 e30 e33 | e31 e32 1 | E2 E (24) E = ----|---- = ------------|-----------E3 | E4 e10 e13 | e11 e12 e20 e23 | e21 e22 The measured (Sm) and actual (Sa) S-parameters of the DUT (Device Under Test) are given by: b0 a0 S S (25) Sm = ( S11m S12m) ( b3) = Sm ( a3) .       By using T-parameters to represent the error terms it is possible to obtain an easier solution. 0 1 b a b3 a2    (29) =T a0 b1 a3 b2 T1 | T2 (30) T = ----|---- T3 | T4 Solving for Sm and Sa yields Page 10 .

SmT4 = 0 If 4 known two-port standards and their resultant measurements are available.(31) Sm = (T1Sa + T2)(T3Sa + T4)-1 (32) Sa = (T1 .Sm4T3Sa4 . However the equations are homogeneous equations.SmT3)-1(SmT4 . (33) T1Sa + T2 . Numerically the solution to the T-parameters would be the trivial one.Sm2T3Sa2 . (34) T1Sa1 + T2 . There are a total of 16 independent and unknown T-parameters. This normalization reduces the error terms to 15. This procedure is valid as long as we are calibrating for ratio measurements. enough independent linear equations can be obtained to solve for the T-parameters.Sm2T4 = 0 (36) T1Sa3 + T2 .SmT3Sa . that is T=0. Page 11 . This requires reducing the rank of the matrix by one and normalizing the values of all the coefficients with respect to one of the Tparameters.Sm3T4 = 0 (37) T1Sa4 + T2 . These equations are linear in 'T' and by using appropriate two-port and one-port standards for Sa. 4 matrix equations can be written.Sm1T3Sa1 .T2) The equations (31) or (32) can also be written in linear form which makes it convenient to solve for the error terms. Also if there is symmetry at the test ports (e21 = e12) the system of equations reduces to 14 unknown error terms.Sm4T4 = 0 Expanding Equations (34) to (37) gives 16 equations in 9 unknown T-parameters each.Sm1T4 = 0 (35) T1Sa2 + T2 .Sm3T3Sa3 .

Page 12 .SLIDE 8 There are four combinations of common two-port standards that will give the required 16 linear equations that are needed to solve the 16 unknowns. Any known standard can be used as long as each is used only once on each port.

This generalized method [5][6][7] also ties together the traditional techniques with the new techniques in a unified manner. Since X and Y are 2-ports it would appear there are 8 unknowns to find. only 3 additional characteristics of the standards are needed.SLIDE 9 Further development of the TRL calibration concepts [4][8][10] has led to a more generalized method. This error adapter represents all the errors in the system that can be corrected. It can be split into two 2-port error adapters. X (at port 1) and Y (at port 2). If a thru (4 known characteristics) is used as one of the standards. there are actually only 7 error terms to calculate. This means that only 7 characteristics of the calibration standards are required to be known. however since all measurements are made as ratios of the b's and a's. after removing the leakage (crosstalk) terms as a first step in the calibration. The first step involves separating the system into a perfect reflectometer followed be a 4-port error adapter. Page 13 .

M = XAY. In a similar manner. and M3. M1 = XC1Y M2 = XC2Y M3 = XC3Y DETAILED CALIBRATION ANALYSIS System Definition This development will use both S-parameters and T-parameters (cascading parameters). M2. Page 14 . A flow graph of the measurement system is shown in Figure 1. of the DUT. each measurement of three 2-port standards.C2. M. C1. It has been assumed that the leakage terms or crosstalk terms have been determined previously. as corrupted by the error adapters as a simple product of the matrixes.SLIDE 10 It is convenient to use T-parameters because it allows one to represent the overall measurement. A. and C3 can be represented as M1.

x11 x12 1 e10e01-e00e11 e00 (10) X= x21 x22 = e10 -e11 1 e e -e e e y y 1 32 23 22 33 22 11 12   (11) Y= 1 y21 y22 = e32 -e33 System Equations It is convenient to initially use T-parameters instead of S-parameters because it allows one to represent the overall measurement. (1) e00 = Forward directivity (2) e11 = Port-1 match (3) e10e01 = Forward reflection tracking (4) e10e32 = Forward transmission tracking (5) e33 = Reverse directivity (6) e22 = Port-2 match (7) e32e23 = Reverse reflection tracking (8) e23e01 = Reverse transmission tracking The relationship between the S-parameters and the T-parameters follows. (13) M1 = XC1Y (14) M2 = XC2Y (15) M3 = XC3Y Where the three two-port standards are defined as follows.Figure (1) The S-parameter flow graph is shown in Figure 1. e-γl1 0  (17) C1 =   0 eγl1 The second standard is a matched two-port with unknown transmission coefficients C2 11 and C2 22. C1. each measurement of three 2-port standards. Equation (9) defines the T-parameters for M and A in terms of S-parameters. The first standard is a matched known two-port with low transmission loss. M2. of the DUT. C1 11 0  (16) C1 =   0 C1 22 Or a matched transmission line of length l1. A. In a similar manner. M. and C3 can be represented as M1. (12) M = XAY. T11 T12 1 S21S12-S11S22 S11 (9) T= T21 T22 = S21 -S22 1 The following equations define the fictitious error adapters in terms of S-parameters (e-terms). The S-parameter error terms (e terms) that need to be determined are now defined. as corrupted by the error adapters as a simple product of the matrixes. C2 11 0  (18) C2 =   0 C2 22 Page 15 . and M3.C2.

Or a matched transmission line of length l2. x11= a = e00 . start from (13) and solve for Y (21) Y = C1-1X-1M1 Substituting (21) into (14) gives (22) M2 = XC2C1-1X-1M1 Rearanging the terms of (22) yields (23) (M2M1-1)X = X(C2C1-1) . (20) C3 = Unknown one-port high reflect (Γ3) connected to port-1 and then port-2.l1) 0  (27) CA = C2C1-1 =   0 eγ(l2 .m12=0 C Eliminating  2 22 from (31) and (32) yields C1 22 2 x12  x12 (34) m21 x22 + (m22 . Solve for X Error Adapter To solve for part of X. and (24) MAX = XCA Where MA and CA are defined as m11 m12 (25) MA = M2M1-1 =  m21 m22 C2 11 0  -1 (26) CA = C2C1 = C1 11 C2 22   0 C1 22  And if matched transmission lines are used for C1 and C2. e-γ(l2 .m12=0 At this point it is actually easier to convert to S-parameters for the remainder af the analysis.m11)x21 .l1) Rewriting (27) gives  C2 11 m11 m12x11 x12 x11 x12C1 11 0   (28) m21 m22x21 x22 = x21 x22 0 C2 22 C1 22  Expanding the above matrix equation yields C (29) m11x11 + m12x21 = x11 2 11 C1 11 C (30) m21x11 + m22x21 = x21 2 11 C1 11 C (31) m11x12+ m12x22 = x12 2 22 C  1 22 C (32) m21x12+ m22x22 = x22 2 22 C1 22 C Eliminating  2 11 from (29) and (30) yields C1 11 2 x11  x11 (33) m21 x21 + (m22 .e10e01 (35) e11 x21 (36) x12= b = e00 x22 Page 16 . e-γl2 0  (19) C2 =   0 eγl2 And as the 3rd two-port standard.m11)x22 .

The T-parameters coefficients (m terms) used in the quadratic equations (33) and (34) can also be defined as S-parameters.S22m2) (38) (m22 . This allows the calibration standard C2 to have S21 and S12 equal to zero! The quadratic equations (33) and x x (34) can now be solved for  11 and  12. but not enough information is known at this time. Note the coefficients are the same for both equations and x21 x22 e e therefore the roots are the same. the following result is obtained.∆m2) (39) m12 = K(∆m1S11m2 . Solve for Y Error Adapter Using a similar approach the error box Y can be partially solved.e33 (47) e22 y12 (48) (49) y21= d = -e33 y22 c-d= e23e32 e22 The term e22 needs to be solved for. The root choices are obvious since e00 is small and 10 01 is large for a e11 typical system.m12´=0 y y Equations (45) and (46) can be solved for  11 and  21. Solve for Port Matches To solve for e11 and e22 the standard one-port measurement procedure can be used.m11) = K(∆m1 + S11m1S22m2 -S11m2S22m1 .m11´)y22 .Γmx (51) Γ3 = e11 a .m11´) 11 . e e (41) b .e11Γ3 Solving for Γ3 and using equations (35) and (36) 1 b . (e e )Γ (50) Γmx = e00 + 10 01 3 1 . it can be arbitrarily set to K = 1. Since the root solution of the quadratic does not depend on K.m12´=0 y12 y12 2 y21  y21 (46) m21´ y22 + (m22´ .a = 10 01 e11 But not enough information is available to solve for e11 at this time. The next task is to separate e11 from e10e01 as defined in equation (41).2 are the measured S-parameters of the calibration standards C1 and C2. With a termination Γ3 on port-1 or error adapter x. (42) YMA´ = CAY Where MA´ and CA are defined as m11´ m12´ (43) MA´ = M1-1M2 =  m21´ m22´ C2 11 0  -1 (44) CA = C1 C2 = C1 11 C2 22   0 C1 22  Solving (42) yields the following quadratics 2 y y (45) m21´ 11 + (m22´ . (37) m21 = K(S22m1 . y12 y22 y11= c = e23e32 .Γmx Page 17 .∆m2S11m1) 1 (40) and K = S21m2S12m1 Where Sxxm1.

S11m Substituting the value of e22 from (56) into (54) yields the solution for e11.d)e22 Solve for Transmission Terms The two transmission tracking terms (e10e32) and (e23e01) still need to be determined.S11m (56) e22 = e11 a .With the same termination Γ3 on port-2 of error adapter y (e e )Γ (52) Γmy = e33 + 23 32 3 1 . therefore (62) (e10e32) = S21m(1 .Γmxd + Γmya .a)e11 And from (49) (59) (e23e32) = (c . (e e ) e e-2γl1 (55) S11m = e00 + 10 01 22 -2γl 1 .S11m a . Expanding (52) yields the individual measured S-parameters.e11e22e 1 Solving (55) for e22.Γmy Eliminating Γ3 from (51) and (53) yields b . e22 can be measured (S11m) with the port-1 reflectometer. The reflection tracking error term can be solved for from (41) (58) (e10e01) = (b .e22Γ3 Solving for Γ3 and using equations (47) and (48) 1 d . This can be obtained from the connection of the two test ports through calibration standard C1. as long as we can measure all four waves at the coupler ports at the same time.Γmxc + Γmyb .e e e-2γl1)eγl1 (61) S12m = (e23e01) 23 01 12m 11 22 Remove Switch Errors The switch errors can be removed when measuring the S-parameters by using a generalized method of measuring S-parameters.e11e22e-2γl1)eγl1 and (63) (e e ) = S (1 .e11e22e-2γl1 The terms e11 and e22 are known.Γmxd + Γmy During the connection when measuring the standard C1 (assumed to be a transmission line e-γ(l1)).Γmy (53) Γ3 = e22 c . -1 bo bo´ao ao´ (64) Sm =  b3 b3´a3 a3´ Where the unprimed terms are with the switch in the forward direction and the primed terms are with the switch in the reverse position. bo bo´ a3 ao a3´ ao (65) S11m = ∆ Page 18 . Typically Γ3 is an open or short that can have an accurate estimate for it's reflection coefficient.Γmxc + Γmy (54) e11 = e22  a .e e e-2γl1 11 22 and e-γl1 1 . e-γl1 (60) S21m = (e10e32) 1 . e2γl1 b .S11m The root choice for (57) can be determined from (51) if there is an approximate value for Γ3. (57) e11 = eγl1 b .

The normal measurement of the S-parameters requires 4 measurements. (70) e00 = Forward directivity (71) e11 = Forward Port-1 match (72) e22´ = Forward Port-2 match Page 19 .(66) (67) (68) Where (69) bo´ bo ao´ a3´ ao a3´ S12m = ∆ b3 b3´ a3 ao a3´ ao S21m = ∆ b3´ b3 ao´ a3´ ao a3´ S22m = ∆ ∆=1a3 ao´ ao a3´ Convert from 8-Term to 10-Term Model The measurement method above used to remove the switch errors requires the measurement of 6 ratio terms. Figure (2) Forward Direction Figure (3) Reverse Direction The 5 error terms can now be defined for the forward direction. This model has 10 error terms and can be used with the standard 12 term model used in the industry. For this reason the generalized method is used only during the calibration step but not during device measurements after calibration. where it is assumed that the 2 leakage or crosstalk terms have been previously determined. The 8 term calibration model is divided into a forward and reverse model for measurement as shown in Figures (2) and (3).

S22m´ (83) e23´e01´ = S12m´(1 . (75) e33 = Reverse directivity (76) e22 = Reverse Port-2 match (77) e11´ = Reverse Port-1 match (78) e32e23 = Reverse reflection tracking (79) e23´e01´ = Reverse transmission tracking The primed error terms will now be determined.S11m´ (80) e22´ = e11 a .S22m´ (82) e11´ = e22 c . The primed measured S-parameters must not use switch correction.e11´e22e-2γl1)eγl1 Page 20 .e11e22´e-2γl1)eγl1 e2γl1 d . e2γl1 b .S11m´ (81) e10´e32´ = S21m´(1 .(73) e10e01 = Forward reflection tracking (74) e10´e32´ = Forward transmission tracking And 5 error terms for the reverse direction.

SLIDE 11 While there are 7 unknowns. X and Y can be solved for directly plus 5 characteristics of the calibration standards.Γmx (87) Γ3 = e 11 a . and the termination Γ3 is calculated by Equation (51). measuring three 2-port standards yields a set of 12 equations. 1 b .Γmx Transmission Line used as the Thru Connection If C1 and C2 are both transmission lines with the same propogation factor then the term eγl1 used in many of the earlier equations can be determined as follows. Due to this redundancy. DETAILED CALIBRATION ANALYSIS CONTINUED Solving for Characteristics of the Standards Five of the coefficents of C2 and C3 were not needed in the solution but can now be calculated. Page 21 . (88) e-γ(l2-l1) = e-γl = e-γl1(l2/l1 -1) From which γl1 can be determined by knowing only the ratio of the lengths l2 to l1 and γl. it is not necessary to know all the parameters of all the standards. (84) C2 11 = C1 11(m22 + m21a) And from (32) (85) C2 22 = C1 22(m22 + m21b) If transmission lines are used Equations (27) and (84) yields (86) e-γ(l2-l1) = e-γl = (m22 + m21a) The calibration standard C3 had the two transmission terms are defined to be zero. The two unknown coefficients of C2 can be determined from Equation (30) .

(89) γl γl1 = l /l -1 2 1 Page 22 .

The other parameters of C2 and C3 can be solved from the data. For this calibration method there are several combinations of standards that fit the requirements. one standard needs to be Zo based. impedance renormalization can be used to shift to a different impedance base. one needs to present a high mismatch reflection. The simplest of all standards is a throughline.SLIDE 12 All 4 parameters of C1 must be known but only 2 parameters for C2 and none for C3 if S11 = S22. there are also choices that generate ill-conditioned solutions or singularities. Page 23 . In choosing appropriate standards. In addition. However. so let C1 be a thru and C2 a Zo matched device. all three standards need to be sufficiently different as to be three independent measurements. If needed.

The unknown device characteristics can be easily calculated by knowing the parameters of the X error adapter, the known standard C1, and the measured data of the test device and measured data for C1.

There are two methods that can be used to solve for the DUT A. First lets use the 8 term error model that requires the measurement of 6 ratios as described earlier. Converting the e-terms to T-parameters is the first step. 11 x12 x x 22 x22 (90) X = x22x  21 1  x22  Wherethe parameters of the X adapter are best defined by the ratios x11 (91) =e e -e e x22 10 01 11 22 x12 (92) =e x22 00 x21 (93) = - e11 x22 Now starting with (94) M = XAY and solving for A (95) A = X-1MY-1 Next using the measurement of the first calibration standards C1 (96) M1 = XC1Y and solving for Y-1

Page 24

(97) Y-1 = M1-1XC1 Substituting (97) into (96) yields (98) A = X-1MM1-1XC1 Note the x22 divides out in (98). The second method uses the 10 e-terms in Equation (70) to (79) plus the two leakage terms directly in the equations for the 12-term industry standard error model described earlier in this paper.

Page 25

There are several possible strategies in choosing standards. The use of a zero length thru is an obvious selection but a non-zero length thru is also acceptable if its characteristics are known or the desired reference plane is in the center of the non-zero length thru. A second standard needs to be a Zo reference. In this solution, only the match of this standard needs to be of concern. Its S21 and S12 can be any value and do not need to be known. In fact, they will be found during the calibration process. This opens up the choices to a wide range of 2-port components, such as a pair of matched loads or an attenuator. For the final standard only one piece of information is needed. This could be a known reflection value of symmetrical reflections. Since the other standards have been well matched, this standard must be a high mismatch. The table shows a partial list of possible calibration configurations with appropriate three letter acronyms. Many are familiar combinations but there are some new ones of significant usefulness, such as TRM and LRM. There are many other possibilities not listed.

Page 26

Port 1 Unknown Short X X Port 2 Unknown Short X X Match X Unknown Open X X Match X Unknown Open X X And there are many other combinations still possible! Page 27 . Note that the number of connections for two-port calibrations is reduced over the standard OSLT calibration method. The remaining 3 terms can be determined with 3 one-port standards. and a known open. the following table can be developed. a known short. Port 1 Short X X X X Port 2 Short Match X X X X X X X X X X X Open X Match X Open X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X Another Combination of one-port standards that works well in wafer calibration use only one known match with equal unknown shorts and equal unknown opens. Using as one-port standards a known match.DETAILED OTHER COMBINATIONS OF STANDARDS Another combination of standards uses a known two-port thru or line connection which defines 4 characteristics.

The same problem exists if the device has different connectors at both ends.SLIDE 15 There are many times when a device with female or male connectors on both ends needs to be measured The problem is that the two test ports can not be connected together for the transmission calibration. like type N and waveguide. Page 28 .

These two calibrations can then be combined to remove the adapter and provide a complete non-insertable calibration. DETAILED NON-INSERTABLE ANALYSIS This first two columns of the table defines the e-terms that result from the two calibration steps at test port 1 and test port 2. Two Port Cal #1 Two Port Cal #2 Combined Cal e00 E00 e00 e11 E11 e11 (e10e01) (E10E01) (e10e01) e30 E30 See Equation (1) Below e22 E22 E22 (e10e32) (E10E32) See Equation (2) Below e30 + E30 (1) ( ) 2 (e10e01) (E10E01) The same process is used for the reverse direction. The redundant data gathered by this method is used to improve the final calibration results. The third column is the combined result.SLIDE 16 There is a mathematical method to calibrate in these difficult environments. (2) (E10E32) Page 29 . The process is to first do a two-port calibration at test port 1 then do a two-port calibration at test port 2. The first step is to locate an adapter that mates the two test ports.

SLIDE 17 Error correction theory is of limited value if the final measurement accuracy is unknown. Page 30 . A method that accounts for the errors in a simple and complete manner is a key desired feature.

2 and µ1.2) of the converter determines the sensitivity of the system. Drift and stability errors characterize the system changes with time and temperature.t2 describes the cable transmission coefficient change and Cr1. Page 31 . and a high level noise of the LO and IF (Nh1.2) characterize the calibration standards that are not perfect.2). Rt1.2). The cable and connector interface will change after calibration. These errors can also be viewed from a different perspective. The cable will remain calibrated if it is not moved.r2 characterizes the connector reflection repeatability error between calibration and measurement.t2 characterizes the connector transmission repeatability error and Rr1. Systematic errors are those errors that don't change after calibration and are stationary. The random errors have a zero mean and random distribution and can be reduced by averaging or multiple measurements. The device under test will be degraded by the following residual errors and hardware imperfections.SLIDE 18 The error corrected measurement system can be nicely described using flow graphs. and even after calibration there are residual errors present. The front end and If hardware will drift with time and temperature as characterized by the and stability terms (S1. The low level noise (NL1.r2 describe the change in the cable reflection coefficient. The residual microwave errors (δ. but this is not the case and Ct1. τ1.2) contribute to the noise on the measurement data. The nonlinearities of the system with measurement level are described by the dynamic accuracy (A1.

These equations calculate the magnitude uncertainty (∆S11 and ∆S21) with each error term defined by its absolute magnitude. The random. where the error terms are defined by their absolute magnitude. The first part of the equation describes the systematic errors and these errors typically add up in a worst case manner. repeatability errors of connector Nr = (Nh1S11)2 + (NL1)2 . where the error terms are defined by their absolute magnitude. The phase error is determined by taking the arcsin of ∆S11 or ∆S21 and adding any additional phase terms that are necessary. where the error terms are defined by their absolute phase. cable errors Rr = (Rr1 + 2Rt1S11 + Rr1S112)2 + (Rr2S21S12)2 . noise errors (4) Drift & Stab = S1S11 . DETAILED UNCERTAINTY ANALYSIS Equation (1) defines the reflection magnitude uncertainty.SLIDE 19 The flow graph can be solved to show the total system uncertainty. drift and stability errors are typically characterized in an RSS fashion in the second part. (6) (7) (8) ∆S21(magnitude) = Systematic + (Random)2 + (Drift & Stab)2 Systematic = (τ2 + µ1S11 + µ2S22 + µ1µ2S21S12 +A2)S21 Random = (Ct)2 + (Rt)2 + (Nt)2 Ct = S21 (Ct1)2 + (Ct2)2 + (Cr1S11)2 + (Cr2S22)2 . (1) (2) (3) ∆S11(magnitude) = Systematic + (Random)2 + (Drift & Stab)2 Systematic = δ + τ1S11 + µ1S112 + µ2S21S12 + A1S11 Random = (Cr)2 + (Rr)2 + (Nr)2 Cr = (Cr1)2 + (2Ct1S11)2 + (Cr1S112)2 + (Cr2S21S12)2 . cable errors Page 32 . (5) ∆S (magnitude) ∆S11(phase) = arcsin S 11 + 2Ct1 + S1 (magnitude)  11  Equation (6) defines the transmission magnitude uncertainty. drift and stability errors Equation (5) defines the reflection phase uncertainty.

drift and stability errors Equation (10) defines the transmission phase uncertainty. (10) ∆S (magnitude) ∆S21(phase) = arcsin S 21 + Ct1 + Ct2 + S2 (magnitude)  21  The definition of the error terms follows δ = Residual directivity τ1 = Residual reflection tracking τ2 = Residual transmission tracking µ1 = Residual port-1 match µ2 = Residual port-2 match A1 = Reflection dynamic accuracy A2 = Transmission dynamic accuracy Cr1 = Cable drift and stability (reflection) on port-1 Cr2 = Cable drift and stability (reflection) on port-2 Ct1 = Cable drift and stability (transmission) on port-1 Ct2 = Cable drift and stability (transmission) on port-2 Rr1 = Repeatability of connector (reflection) at port-1 Rr2 = Repeatability of connector (reflection) at port-2 Rt1 = Repeatability of connector (transmission) at port-1 Rt2 = Repeatability of connector (transmission) at port-2 Nh1 = Reflection high level noise Nh2 = Transmission high level noise NL1 = Reflection low level noise NL2 = Transmission low level noise S1 = Reflection drift and stability error S2 = Transmission drift and stability error A detailed analysis of each of these error terms will now be developed. Page 33 . noise errors (9) Drift & Stab = S2S21 . where the error terms are defined by their absolute phase.Rt = S21 (Rt1 + Rr1S11)2 + (Rt2 + Rr2S22)2 . repeatability errors of connector Nt = (Nh2S21)2 + (NL22) .

∆2. After a page of so of math the final result can be put in the following form. DETAILED ANALYSIS OF THE SYSTEMATIC ERRORS CAUSED BY THE STANDARDS Assuming that three reflection standards are employed. then the residual errors can be calculated. Γ2.Γ3´) (S11 .Γ3) Noting that.Γ1´)(Γ2´ .Γ3´) (S11 .SLIDE 20 The reflection residual microwave errors are mainly determined by the quality of the calibration standards. This is done by using the invariance of the cross-ratio principle of the bilinear transformation [11]. Page 34 . If the assumed values of the standards and their measured value (assumed plus error) are known. Γ3 Γ1´. the perceived value of the device under test is in error (2) S11´ = S11 + ∆S11 If the true reflection coefficient is required it can be obtained by substituting Equation 2 in Equation 1. Γ2.Γ1)(Γ2 . in fact. the measurements S11m of an unknown S11 can be written as (S11m . ∆3 = errors in standards Γ1. Γ3 = assumed reflection coefficients of standards ∆1. then define Γ1. Γ3´ = measured values (S11m) with standards connected Using the invariance property of the cross-ratio of complex numbers. Γ2´.Γ2´)(Γ1´ .Γ3) (1) = (S11m .Γ2)(Γ1 .

(D1Γ2Γ3 + D2Γ1Γ3 + D3Γ1Γ2) τ1 = D1(Γ2 + Γ3) + D2(Γ1 + Γ3) + D3(Γ1 + Γ2) µ1 = .ΓL τ1 = 2 εs2 + εo2 µ1 = ΓL2 + 4εs2 + 4εo2 1 1 1 The transmission residual error term (τ2) is calculated by a different method. (20) M2c = M2 + µ2 . Also assuming that the amplitude errors are small.Γ2)] Total reflection error Residual directivity & port-2 match Residual reflection tracking Residual port-1 match The ∆1.2oS112     (9) (10) δ = .Γ3)] D2 = ∆2/[(Γ2 .µ2 = . µ1 = residual port-1 match.µ2 = .3 terms of the calibration standards are determined from primary electrical standards that are carefully modeled from precision mechanical measurements. The resultant measurement yields: 1+T2 (17) S21m = 1-M1M2 where M1 and M2 are the raw test port matches and (1+T2) is the raw transmission tracking term.Γ1)(Γ3 .∆L ∆ ∆ τ1 =  2s . The final result after substituting Equations 19 and 20 into 18 and simplifying yields: (21) 1+T2c = (1+T2) (1 .Γ1)] D3 = ∆3/[(Γ3 .1. This is true even if the port match is stable and never changes.2o   The errors can be expressed in a different form that relates closer to how they are measured. and load (Γ3 = 0. µ2 = residual port-2 match. 1-M1cM2c (18) 1+T2c = S21m (1-M1cM2c) = (1+T2) 1-M1M2 where the calculated terms are defined as (19) M1c = M1 + µ1 .. short (Γ2 = . ∆3 = ∆L) calibration standards.2s . (7) (8) ∆ ∆ ∆ ∆ ∆S11 = .(D1 + D2 + D3)/(1 +τ1) where D1 = ∆1/[(Γ1 . (14) (15) (16) δ = . The slide shows the simplified results when using an open (Γ1 = 1. The calculated (1+T2c) is solved for from Equation 17 where the calculated values for M1 and M2 (defined as M1c and M2c) are used.2oS11 + ∆L .M1µ2 . open and load are random in both magnitude and phase it is realistic to RSS the tracking and match terms.Γ2)(Γ1 . (11) ∆L = ΓL ΓL = the reflection coefficient of the fixed or sliding load calibration standard.µ2 = .σs . The raw port match causes an error in the error corrected transmission tracking.2s . To calculate the residual transmission tracking (τ2) the two test ports are connected together. (12) ∆s = .2o   ∆ ∆ µ1 = ∆L .(3) (4) (5) (6) ∆S11 = δ + τ1S11 + µ1S112 δ = . ∆1=∆o).jεs σs = the amplitude error of the short εs = the phase error of the short in radians (13) ∆o = σo + jεo σo = the amplitude error of the open εo = the phase error of the open in radians Since the errors of the short.M2µ1) And the residual transmission tracking error is (22) τ2 = M1µ2 + M2µ1 Page 35 .2.∆L +  2s .Γ3)(Γ2 . ∆2 = ∆s).

01158( ) |S11 . Re(1 + S112) (5) Re1 = PsLaLs1Lc1S11Lc1L1cLci And the IF and detector residuals at port-2 after error correction. At high measurement levels the main error is caused by the front end compression. A/D bit resolution. The dynamic accuracy at port-1 for reflection measurements is (1) A1 = Pc1 + Re12 + Lx2 + Mp2 + Ge2 The dynamic accuracy at port-2 for transmission measurements is (2) A2 = Pc2 + Re22 + Lx2 + Mp2 + Ge2 Where the compression after error correction at port-1 is 2 PsLaLs1Lc1Lc1L1c 2 (3) Pc1 = . The dynamic accuracy does not include any noise or linear microwave errors.01158( ) |S21 . crystal filter non-linearity and the accuracy of the detectors as the phase is changed. DETAILED DYNAMIC ACCURACY ANALYSIS The dynamic accuracy terms after error correction is defined by the following equations. and coherent IF leakage signals.1| Pc The IF and detector residuals at port-1 after error correction. Re(1 + S212) (6) Re2 = PsLaLs1Lc1S21Lc2L2cLci Page 36 . In the middle ranges the errors are caused by the autoranging attenuators. At the low levels the residuals are caused by DC drift.SLIDE 21 The dynamic accuracy after error correction (A1.2) describes the system errors as a function of level and phase shift.1| Pc And the compression after error correction at port-2 is 2 PsLaLs1Lc1Lc2L2c 2 (4) Pc2 = .

(8) Mp = 2 M Where the other system terms that impact the dynamic accuracy are defined as follows.The linearity of the crystal filter after error correction. Ps = Power output of the source La = Loss defined by the RF attenuator setting in the test set Ls1 = Loss from source to port-1 Lc1 = Loss of cable on port-1 Lc2 = Loss of cable on port-2 L1c = Loss from port-1 to converter L2c = Loss from port-2 to converter Lci = Loss from converter to the IF Pc = Power at the converter for . IF and detector measured at the IF. (7) Lx = 2 L The change in magnitude as a function of phase rotation. after error correction. L = Linearity of the crystal filter as a function of level M = Magnitude error as a function of a 360 degree phase rotation Ge = Gain error of the IF autoranging attenuators Page 37 .1 dB compression Re = Residual errors of the converter.

The system sensitivity is determined by the noise of the front end converter and LO noise leakage through the converter (NL1. There is also noise (Nh1. Noise on the data (Nh1. If they are not moved after calibration the error can be very small but this is not the typical use of the system. The higher the harmonic of the LO the more this noise will increase and dominate. The connector repeatability (Rt1.2 and Cr1.2) is characterized by making multiple connections and measuring the vector difference in the data. The port match and transmission characteristics of the cables will change with use (Ct1.2).2) added to the data by the detector circuits and the RF source. Hard line cables tend to be more stable if the measurement requires very little movement.SLIDE 22 There are a number of noise sources. Cables are a major source of error.2). But if the cables must be moved often then a high quality flexible cables are a must.2) is caused by the system sensitivity plus noise from the LO close to the carrier. Page 38 . This analysis is done with a large sample base and characterized statistically. Typically the transmission phase error will be larger that the magnitude error. Also the phase shift of cables with temperature is mainly a function of the dielectric and can be very different from cable to cable. but these are usually smaller.2 and Rr1.

lets look at a simple transmission calibration model. (1) S21m = C + (1 + T2)S21 S21m = Measured S-parameters S21 = Actual S-parameters C = Crosstalk leakage term (1 + T2) = Transmission tracking term To determine C and (1 + T2) connect loads (S21ml) and a thru (S21mt) to ports 1 and 2. Substituting Equations (7). Nf = Noise floor of the front end converter measured at the IF.N ) N + (S21 . the total worst case low level noise is (11) NL2 = 3Nf |S21 + 1|2 + |S21|2 + 1 PsLaLs1Lc1Lc2L2cLci Avg Also from Equation 10.C S21m . define NL2 = ∆S21 as the total effect of the low level noise (Nf) of the front end and the calibration process. Nf and Nv are assumed to be 1 sigma noise power data with no averaging. The noise floor (Nf) is measured at the IF. Nft = Noise floor with a thru connection measured at the IF.S21∆S21mt Now defining as the noise terms S N Nf (7) ∆S21m = 21 v + Avg PsLaLs1Lc1Lc2L2cLci Avg Nv = Noise of the local oscillator.S21ml Solving for S21 S . Nfl (8) ∆S21ml = PsLaLs1Lc1Lc2L2cLci Avg Nfl = Noise floor with loads connected to test ports measured at the IF.S21ml Taking the partial derivative of Equation 5 and assuming T2 is small yields (6) ∆S21 = ∆S21m + (S21 -1)∆S21ml . the total worst case high level noise is (12) 2 N Avg v A similar analysis for reflection measurement noise yields Nh2 = 3 (13) (14) NL1 = 3Nf Nh1 = 3 |S11 + 1|2 + |S11|2 + 1 PsLaLs1Lc1Lc1L1cLci Avg 2 N Avg v The factor of 3 in Equations 11 to 14 determines the peaks of the noise which is defined to be 3 sigma. N Nft (9) ∆S21mt = vt + Avg PsLaLs1Lc1Lc2L2cLci Avg Nvt = VTO noise with a thru connection. The other terms are defined as follows: Ps = Power output of the source La = Loss defined by the RF attenuator setting in the test set Ls1 = Loss from source to port-1 Lc1 = Loss of cable on port-1 Lc2 = Loss of cable on port-2 L1c = Loss from port-1 to converter L2c = Loss from port-2 to converter Lci = Loss from converter to the IF Avg = Averaging factor Page 39 . (8) and (9) into (6) yields S (N . (2) S21ml = C (3) S21mt = C + (1 + T2) Therefore (4) (1 + T2) = S21mt . Sinse Nv = Nvt and are random variables.S21Nft (10) ∆S21 = 21 v vt + f Avg PsLaLs1Lc1Lc2L2cLci Avg From Equation 10.1)Nfl .S21ml = (5) S21 = 21m 1 + T2 S21mt . IF. define Nh2 = ∆S21/S21 as the total effect of the high level noise (Nv) of the VTO/IF and the calibration process.DETAILED NOISE ANALYSIS To calculate the impact of noise on a transmission measurement. and detector that is proportional to signal level. Since Nf = Nfl = Nft and are random variables.

SLIDE 23 As calibration methods have improved and the frequency range has increased. the quality of the RF hardware has become the limit in measurement accuracy. A major step forward in stable high performance hardware is a must. a detailed analysis of the hardware is required. Because of the importance of system stability in an error corrected system. Page 40 .

The system directivity is mainly determined by the directional coupler. including the coupler. T1=system reflection tracking error term. where D=system directivity error term. The system tracking error is determined by how well the reference and test channels track each other as a function of frequency. The system port match consists of the match terms of all the components in the system. These error terms correspond directly to the uncorrected "raw" hardware performance. and M1=system port match error term. Page 41 . step attenuators. splitters and switches. bias network.Slide 24 The flow graph of an uncorrected raw one-port measurement system is shown.

T1c. they are not equal due to an imperfect system and imperfect calibration standards.SLIDE 25 This analysis shows that the drift and stability of the system after error correction depends on the stability of the directivity. 1-M1S11 From Equation 1. and M1). DETAILED DRIFT AND STABILITY ANALYSIS The reflection tracking drift and stability error (S1) is determined as follows. residual system directivity 1c T1-T1c (5) τ1 = 1+T . Equation 1 can then be solved for the calculated return loss S11c of the test device: S11m-Dc (2) S11c = (1+T1c)+M1c(S11m-Dc) Substituting for S11m from Equation 1. residual system reflection tracking 1c (6) µ1 = M1-M1c . T1c. the calculated error terms (Dc. Hopefully. and M1c) can be determined. and M1c) are equal to the actual error terms (D. and deleting second order terms yields: D-D T1-T1c 2 (3) S11c ≅ 1+T c + ( 1+ 1+T ) S11 + (M1-M1c) S11 1c 1c Residual error terms (after error correction) can now be defined in a different manner from the earlier development: D-D (4) δ = 1+T c . and from the use of a calibration procedure. Typical values after error correction for the residual directivity (δ) are in the 35 to 60 dB range. Note that the drift and stability will be different as a function of the device under test's reflection coefficient. T1. But. and typical values for the residual port match (µ1) are from 30 to (1) S11m = D + (1+T1) S11 Page 42 . residual system port-1 match. The measured S11m can be calculated in terms of the actual test device S11 and the error terms from the reflection flow graph. Also the absolute value of the tracking term will impact the stability. the calculated error terms (Dc. and will cancel. tracking and port match raw uncorrected terms.

Note that both the stability of the error terms (dD. S1 = Page 43 . and the quality of the calibration standards. the equation for reflection drift and stability (S1) can be defined. T1. and M1) is determined by taking the partial derivative of S11c that is defined in Equation 3. The sensitivity of S11c to the uncorrected error terms (D. T1c. (7) dS11c = ∂S ∂S ∂S11c dD + 11c dT1 + 11c dM1 ∂T1 ∂M1 ∂D Taking the partial derivative of Equation 3 and dropping second order terms yields: S 1 (8) dS11c ≅ dD + 11 dT1 + S112 dM1 1+T1c 1+T1c Equation 8 shows clearly the effect of changes in directivity.1 dB. and dM1) and the absolute value of (1+T1c) contribute to the stability. Using the safe assumption that T1c ≅ T1. calibration method. (9) S dS11c  1 = dD + 11 dT1 + S112 dM1/S11 S11 1+T1 1+T1  S1 = Raw reflection tracking drift and stability error T1= Raw uncorrected reflection tracking term (includes loss) dD = Raw directivity drift and stability error dT1 = Raw reflection tracking drift and stability error dM1 = Raw port-1 match drift and stability error. The partial derivative is defined in Equation 8. Also the stability will change as a function of the test device. tracking. and M1c) are stationary and don't change after error correction. frequency. The wide range of error-corrected performance is determined by the connector size. dT1. Note that the calculated error terms (Dc. The tracking term (1+τ1) normally varies from 0 to ±0. and match on the resultant error-corrected measurement.60 dB return loss.

As long as each of the error sources can be determined.SLIDE 26 The 8510 specifications are developed by combining the effects of the specified error sources. Page 44 . The resultant uncertainties are not only a function of the measurement system but must include the parameters of the device under test. the final result is assured.

The residual IF dynamic accuracy is determined by measuring the individual IF and A/D error terms.SLIDE 27 Each one of the specified error sources has a traceable path back to NIST. The airline electrical characteristics are developed from mechanical measurements. These errors include the front end compression and the other IF and detector errors described earlier. The mechanical measurements and material properties are carefully modeled to give a very accurate electrical representation. The mechanical measurements are then traced back to NIST through various plug and ring gages and other mechanical measurements. The resistance is than directly traced back to NIST. The residual microwave error terms are tested with precision airlines or low frequency resistance. Page 45 . These IF errors are then combined together in an IF model to calculate the complete dynamic accuracy. The frequency errors and various A/D measurements are easily traced back to NIST.

This same verification kit can then be measured in the field on the customers system and the results compared. Page 46 . The process does not verify the individual error sources but provides an excellent final system performance check. A verification kit made up of airlines and pads is measured on a factory system and the uncertainty is calculated and documented.SLIDE 28 The actual system specs are generated as explained in the past two slides but an additional method is provided to verify measurement integrity.

The key advances will now be discussed. Page 47 .SLIDE 29 Many improvements in the RF hardware have occurred over the past few years. Sometimes these advances get lost in the fervor over error correction.

The front-end test set components. and switches. Page 48 . step attenuators. bias networks. The IF system must be free from drift and nonlinearities so they will not degrade the tracking error term. need to be stable and have excellent "raw" uncorrected performance. such as the directional couplers.SLIDE 30 After years of improvements in the performance of calibration standards and the development of more powerful calibration algorithms. the key now to improving the performance of modern network analyzers is back to improving the quality of hardware components. The front-end mixers and samplers must track each other with temperature and time so that their errors are ratioed out.

Page 49 . Connectors must be repeatable and rugged. The test ports must be able to stand abuse as mechanical stress is applied without changing electrically. Cables must be low loss and stable as they are flexed and as the temperature changes.SLIDE 31 The calibration standards must be machined to the state-of-the-art and modeled accurately.

As the test frequency continues to increase. uncorrected performance. these stability issues and uncorrected hardware performance will determine the final outcome of the measurements. it does not have good broadband performance. To illustrate further. or. inherent. consider the vector diagram shown. a directional coupler cannot have good broadband performance if the coupler hardware is unstable. the change in the measured reflection coefficient (Γm) is decreased. If the system directivity error (D) caused by the coupler is reduced as shown in Figure b. if the coupler is typically unstable. then.SLIDE 32 If the test set components do not have good. For example. conversely. the stability is typically degraded. for the same percentage instability in the coupler. Page 50 .

probes. that the sensitivity of Γm is increased to changes in the directivity error term (D). The test port of the network analyzer needs to be as close as possible to the test device. cables. The necessity for stability of the directivity term is even more important at higher frequencies. or fixtures after the directional device. This sensitivity increase is a particular problem at higher frequencies where the losses are the greatest from coax cables. and other components. Page 51 . Obviously the losses due to components and test port cables need to be kept as low as possible. Note in Figure b that if the tracking term (1+T1) is decreased.SLIDE 33 A problem exists if there is loss caused by adapters. This can be illustrated by referring to the vector diagram showing that changes in the system directivity is more critical if there is loss.

The error.02 (34 dB return loss). will be ±0.11 dB).SLIDE 34 The error in the transmission tracking term is M1µ2+M2µ1.0126 (±0. low-loss measurements. which is very significant for precision.316 (10 dB return loss) and the residual error corrected port match (µ1 and µ2) reflection coefficient is 0. Two ways to reduce this error are to achieve a better residual port match using higher quality standards and error-correction methods. or to improve the uncorrected "raw" performance of the test ports. Page 52 . in linear terms. To see the effect of this analysis let's consider a system where the raw port match (M1 and M2) reflection coefficient is 0.

5 mm or SMA connectors. The basic definition of the connector interface is also an issue. how do you model slots and gaps in connectors. Many of these are due to mechanical issues involving tight tolerances. And connector repeatability determines an absolute performance floor.SLIDE 35 If the standards were perfect. If each of the standards does not have a consistent flush reference plane the resultant calibration will not calibrate out the test port gaps. A 1 to 5 mil gap is not uncommon in normal 3. with a 5 mil gap the error is only down 32 dB. Others are due to modeling errors. If there is not a clean definition of the connector interface it is very difficult to cascade devices Sparameters accurately. Skin loss also affects the characteristic impedance of the airline. A major improvement is to define standards that are flush with no gaps. Unfortunately there are numerous errors that potentially can cause difficulties. For example. the only errors would be due to the network analyzer. These gaps can have significant error contributions. Page 53 .

but it remains a major error. It is a very high life low impedance contact. The resultant simple mechanical structure is easy to trace to primary national mechanical standards and the measurement system calibrated with these standards can be certified. The inductive component added by the slots can be partially compensated for.SLIDE 36 The slotless contact is an inner contact that does not change the mechanical outer detail as connection is made. Page 54 . The error due to slots changes as the male pin diameter changes over its tolerance range. Connector repeatability is better since there are no center conductor fingers to change as the male pin is rotated during reconnection.

The outer conductor extends beyond the center conductor to provide shielding so there is no radiation. The open center conductor is extended beyond the reference plane to hide the difficult to model male pin. slotless and very rugged. This simple structure is easier to model accurately.SLIDE 37 The open circuit calibration standard is flush. Note that even if the phase model of the open is in error less than a degree that there is still significant reduction in the performance of the calibrated test port match. Page 55 .

Slide 38 The sliding load also incorporates the slotless and flush contact. And the ability to easily connect the load by pushing the center conductor forward to make contact. Page 56 . Then the back stop allows the center conductor to return to the flush position after connection.

It should be noted that 40 micro-inches is about one micron. Page 57 . In addition to the errors listed in the table there are additional errors caused by concentricity and eccentricity. The mechanical tolerances require the same dimensional control as state of the art microwave FET gate widths.SLIDE 39 This table illustrates the tight tolerances required to achieve high performance for airlines used in sliding loads or TRL line standards.

SLIDE 40 Skin loss in non perfect conductors causes the characteristic impedance to change. Page 58 . Pure gold 7 mm airlines at 100 Mhz have a theoretical return loss limit of 55 dB.

and Line. and Sliding Load. This is just a sample listing of the most often used methods. Short. LRL = Thru Line. What is the best method to use? When should it be applied? These questions depend on many factors and considerations that will now be discussed.SLIDE 41 There are a number of different calibration methods. TRM = Thru. and Matched Load. Short. Three Two-Port Standards Methods: TRL = Thru Reflect. Short. and Fixed Load. Reflect. and Matched Load. SSLo = Offset short. SSS = Three Different Offset Shorts. There are numerous other techniques that are used in the industry. Reflect. Page 59 . SSLs = Offset Short. and Offset Load used mainly in wavequide. Four Known Two-Port Standards Methods: OSLf = Open. LRM = Thru Line. Reflect. and Sliding Load used mainly in waveguide. OSLs = Open. and Line. De Embed = Characterizing a fixture and mathematically removing it. Short.

What frequency range is being used? Low frequency calibrations are easier to use and more accurate. The calibration time is much shorter for some calibration methods than for others. The connectors need to be kept clean and in good repair for best results. What is the cost of a calibration kit? Are there calibration kits available for this media? Environment: Is the temperature stable? The calibration will hold much better if the temperature is controlled. Also if the environment is stable the cal will hold longer and the cal kit will last longer. Cables present a unique situation because they are not always stable as flexed and as the temperature changes. Accuracy: What accuracy a particular measurement requires varies a great deal depending on the application. Do the measurements need to be traceable? What connector family is being used? Some utility type connectors do not have well defined calibration kits. Page 60 . If high accuracy is not needed the calibration standards can cost less and may be easier to use. coax.SLIDE 42 There are many considerations to balance when selecting a particular calibration method. Can fixed loads be used as impedance standards or are higher quality sliding loads or transmission line standards required? How accurate do the standards models need to be? Media and Standards: Is the transmission media in waveguide. or some type of microstrip structure? Perhaps it is some special media that is unique.

These choices will now be discussed. Also de-embedding techniques can be used to remove previously characterized fixture errors. Special considerations arise when making fixtured or wafer probing measurements.SLIDE 43 DUT Issues: If the test device is non-insertable or transitional then special calibration steps need to be taken. Different calibration methods work better for measuring high reflections than they do for low reflections. What is the training investment? All these questions can seem overwhelming! But there is really an optimum choice for most of the measurements being made today. Page 61 . Ease of Use: One of the most important questions is how easy is the calibration method to use? How long does it take to calibrate? How many steps are involved in the process? If only 1-port measurements are being made only certain calibration methods will work.

The example is for 7mm but can be scaled to other connector types. The tradeoff is that it has the lowest accuracy. It is fairly easy to use but reasonably expensive. All of the previous methods will work for one-port calibrations. The OSLf (open. but this may be fine for many measurements. The SSLo (offset short. The TRL calibration method needs a two port system in order to calibrate. TRL provides the best accuracy and particularly for the match and tracking terms. It is fairly expensive and sometimes the sliding load is not as easy to use. load-sliding) is the traditional calibration that has been used for many years for accurate calibrations. load-fixed) cal is the least expensive and usually the easiest to use.Slide 44 This table gives a tradeoff in accuracy for various calibration methods in coax. The OSLs (open. short. Page 62 . The short and open determine most of the match and tracking error and this error does not change dramatically with the improved directivity values. short. short. offset-load) provides better directivity than the sliding load but is not available in all connector types. The relative differences stay the same.

LRL is best if the desired reference impedance is the transmission line and LRM if the desired reference impedance is a resistor. Page 63 . The most accurate calibration for coax is TRL. For one port waveguide calibration use the SSLo (offset short.Slide 45 This table is an attempt at grading various calibration methods for different applications. offset load) method. easy to use coaxial calibration is the OSLf (open. For fixtured or wafer probe systems the most accurate methods are either LRL or LRM. short. The best method for low cost. A quick and to the point summary is next. The easiest method to use for fixtured of wafer probe systems is LRM. The best low cost and accurate method for waveguide is TRL. load-fixed) method. short.

U.K.) and John Barr. Then the question is to select the best calibration method to meet the measurement needs. And the measurement accuracy is well understood for coax and waveguide. program manager of the 8510 Network Analyzer family.SLIDE 46 There are some very nice calibration methods that exist today. With this combination excellent results are obtained. Roger Pollard (also of Leeds University. I would like to acknowledge the contributions of the following peers from Hewlett-Packard: Dr. The most important criteria for improved performance and ease of use is RF hardware that is designed and optimized for today's measurement needs. The test equipment must be stable and have good uncorrected performance. Page 64 . The standards need to be mechanically precise and accurately modeled.

frequency range) is defined. This calibration type is particularity suited to very broadband wafer-probing measurements. LRM: A calibration procedure using a transmission line. Error correction is also called accuracy enhancement. MEASUREMENT CALIBRATION: A process performed on a network analyzer once a measurement setup (cables. arbitrary identical reflection and a known load (Line. and arbitrary length transmission line of known Zo appropriate for the frequency range (Thru. Load. The process computes the "error coefficients" which are subsequently used to "error correct" measured data. Match). Examples of imperfect hardware include couplers with less than infinite directivity and test ports that are not matched exactly to 50 ohms. Most microwave network analyzer measurements would simply not be possible without these computational procedures.ERROR CORRECTION GLOSSARY ERROR CORRECTION: A complex mathematical computational process built into most network analyzers introduced since 1984 which "corrects" for imperfect hardware. ERROR COEFFICIENTS: A mathematical array of complex (magnitude and phase) numbers stored within a network analyzer. arbitrary identical reflection. and Thru calibration standards. Page 65 . TRL: An alternative calibration procedure to OSLT using a thru connection. test fixture. Reflect. OSLT: An abbreviation for a two-port error-correction calibration procedure that determines the "error coefficients" with subsequent measurements of precisely known Open. This calibration type is particularity suited to calibration in non-coaxial mediums. One set of numbers (typically 24 coefficients) is required for each measurement frequency. Short. Reflect Line). It provides the best accuracy for metrology grade measurements in coax and waveguide.

Speciale. MTT. [2] S. IEEE Trans. Pollard.. 23rd ARFTG Conference Digest. MTT-37-11. Spring 1984. [10] C. J. Rehnmark.Port Calibration Techniques and Some Applications". Page 66 . November 1989. Proc. [9] D. "A Unified Mathematical Approach to Two. "A Generalized Vector Network Analyzer Calibration Technique".668. J. 18th European Microwave Conf. Dec 1989. U. Dec. "Basic Theory of Waveguide Junctions and Introductory Microwave Network Analysis". 1987. June. Microwave Journal. 5th European Microwave Conf. Schiek. pp 665 . pp 1669 . "A New Procedure for System Calibration and Error Removal in Automated Sparameter Measurements". 34th ARFTG Digest .K. T. on Microwave Theory and Techniques. pp 38-46.Port For Measuring Two-Ports With Any Connectors". April 1974. F. Lauderdale. B. Hoer. "On the Calibration Process of Automatic Network Analyzer Systems". IEEE Trans. A. Soares et al. on Microwave Theory and Techniques. May 1968.1674. "Advances in Microwave Error Correction Techniques". Eul. 1967. W. Engen. [6] R. "Calibrating a Dual Six-Port or Four. "An Automatic Network Analyzer System". pp 909 . A.S International Microwave Symposium Digest.458. [8] D. pp 45 . Franzen. R. [5] H.REFERENCES [1] R. [4] G. June. Hoer. R. [7] J. Proc. M. G. pp 69 . IEEE Trans. Barr. Pergamon Press. M. "Thru-Match-Reflect: One Result of A Rigorous Theory For De-Embedding and Network Analyzer Calibration". Rytting. 1975. [11] Roger D. Engen. A. Pervere. Kerns. 1986 IEEE MTT. pp 73. [3] N. on Microwave Theory and Techniques. 1986... K.27-12.914.987. pp 983 . Sept 1988.73. Oxford. "Verification of System Specifications of a High Performance Network Analyzer". Hackborn. "Thru-Reflect-Line: An Improved Technique for Calibrating the Dual 6-Port Automatic Network Analyzer". C. pp 457 . F. A. Hewlett-Packard RF & Microwave Symposium.52. A. R. Beatty.Ft. 1979.