Attenuator pads: pi, T and minimum-loss resistor values
Pi and T pad resistor values from the attenuation in dB, the 50 to 75 Ω minimum-loss pad, the power each resistor must dissipate and what chip attenuators add.
A 10 dB pad for a 50 Ω system is a pi of 96.25 Ω shunts with a 71.15 Ω series resistor, or a T of 25.97 Ω series arms with a 35.14 Ω shunt. Both give exactly 10 dB, both present 50 Ω at each port, and in both the resistor at the input dissipates 52 % of the power the source has available. Matching 75 Ω to 50 Ω takes a different pad: 43.3 Ω in series on the 75 Ω side and 86.6 Ω across the 50 Ω side, the minimum-loss pad, which costs 5.72 dB of power whichever way the signal travels.
Every value follows from one number, the voltage ratio K = 10A/20, and from equations published by Skyworks, by Yeh et al. and in Maxim’s AN972. The sections below cover where those equations come from, which topology suits which attenuation, how the power divides among the resistors, and what a real chip attenuator’s datasheet adds that the ideal arithmetic leaves out. The attenuator pad calculator solves every case below as a resistor network, and its numbers are the ones quoted here.
What a matched attenuator pad does that a divider does not
A resistive divider drops the voltage and changes the impedance each side sees. A matched pad drops the power by a fixed number of decibels while the source still sees Z0 and the load, looking back, still sees Z0. Nothing is reflected from the pad itself.
That second property is often the reason a pad is fitted at all. A reflection from a mismatched load passes through the pad twice before it reaches the source, so the pad’s loss counts double against it. A 75 Ω load on a 50 Ω source has a return loss of 13.98 dB; put a 10 dB, 50 Ω pad in front of it and the source sees 33.98 dB. Even an open circuit behind a 10 dB pad looks like a 20.00 dB return loss from the front. The VSWR and return loss calculator converts those figures to reflection coefficient and VSWR.
The attenuation is defined from the voltage ratio. Yeh et al. state it as “K = Vin/Vout is the ratio of the input voltage to the output voltage”, with K² the power ratio:
For 10 dB, K = 3.1623 and K² = 10. Using the power ratio where K belongs gets every resistor wrong at once: feed K = 10 into the 10 dB equations and the result is a correct 20 dB pad, with 61.11 Ω shunts and a 247.5 Ω series resistor. The decibel calculator keeps the two ratios apart.
Pi attenuator resistor values from the attenuation
Skyworks gives the pi pad in its PIN diode application note, where the diodes are variable resistors but the equations are those of any resistive pi. With RS1 = RS2 the two shunts and RS3 the series element, p.15 prints:
The first equation runs from resistor to attenuation, which is the wrong way round for design. Solved for the shunt and substituted into the second, it gives the pair the calculator uses:
At 10 dB in 50 Ω that is 96.25 Ω and 71.15 Ω; at 20 dB, 61.11 Ω and 247.5 Ω. The shunt can never reach Z0, because Equation 38 has RS1 − Z0 in its denominator, and Skyworks says so of its own plotted curve:
Note that the minimum value for RS1 and RS2 is 50 Ω.
Every value is Z0 times a function of K alone, so a pad scales with its impedance. The 10 dB pi for 75 Ω is 144.4 Ω and 106.7 Ω, exactly 1.5 times the 50 Ω values.
T pad resistor values, and the two-shunt trap
The T equations used here are from Yeh et al., who built thin-film attenuators to keep thermal noise away from superconducting qubits. On p.2 of their paper:
with the worked case:
For a 10 dB attenuator sub-stage (K² = 10) connected to a characteristic impedance of Z0 = 50 Ω, one finds R1 = R2 = 26.0 Ω and R3 = R4 = 70.3 Ω.
The calculator gives 25.97 Ω and 70.27 Ω, agreeing to the digit printed. The catch is the shunt. Their Fig. 1(a) draws it as two equal resistors, R3 and R4, side by side to ground, and the layout in Fig. 1(b) shows why: in coplanar waveguide the shunt runs from the centre conductor to the ground planes on both sides, one resistor to each. Equation 2 is each of those two. An ordinary T with one shunt resistor needs their parallel value:
which is 35.14 Ω at 10 dB. Copy Equation 2 as the single shunt and the pad built from it gives only 7.39 dB, presents 62.48 Ω to the source (a return loss of 19.1 dB), and is in fact a 7.26 dB pad for 65.77 Ω.
Pi or T: choosing by the resistor values
Between equal impedances the pi and the T give identical attenuation, match and, as the power section shows, identical power split. What separates them is the values, and the values diverge at both ends of the range.
At 1 dB the pi needs 869.5 Ω shunts around a 5.77 Ω series resistor, while the T needs 2.88 Ω arms and a 433.3 Ω shunt. At 40 dB the pi series resistor is 2.5 kΩ and the T shunt is 1 Ω. In each case one element stays close to Z0 (the pi shunt approaches 50 Ω from above, the T series arm from below) and the other runs away. A resistor of an ohm or two is where the pad’s connections and the part’s own inductance matter most against its resistance; a resistor of kilohms is where its capacitance does. So the usual choice is the topology whose values stay nearest Z0 at the attenuation needed, or two pads in cascade.
Yeh et al. chose for a different reason. In their thermal simulations the T-pad had “a larger cooling power” than the pi, so they built the T, and they built 20 dB as two 10 dB cells after finding that splitting the attenuation differently made little difference to the heat.
The bridged-T and variable attenuators
The bridged-T keeps two series arms equal to Z0 at every attenuation and varies only a shunt from their junction and a bridge across them. Skyworks gives it on the same page:
At 10 dB in 50 Ω the shunt is 23.12 Ω and the bridge 108.1 Ω, and their product is always 2500 Ω². That product rule is what makes the bridged-T the variable attenuator’s topology. Skyworks uses two PIN diodes as the shunt and the bridge:
The relationship between the forward resistance of the two diodes ensures maintenance of a matched circuit at all attenuation values.
The note also states the condition that makes it hard: “the PIN diodes are biased at two different resistance points simultaneously, which must track to achieve proper attenuator performance.” Its frequency claim for these circuits is modest too, “the low RF (1 MHz) through UHF frequency range”. The network adds one property none of the sources mentions: when a bridged-T is matched, no current flows in its output arm. At 10 dB the input arm takes 46.75 % of the available power, the bridge and the shunt 21.62 % each, and the output arm nothing.
The minimum-loss pad between 50 and 75 Ω
A pi or T is symmetric, so it cannot match two different impedances. Maxim’s AN972 does it with an L-section of two resistors: R1 in series on the higher impedance side and R2 in shunt across the lower one. For RS > RL, its Equations 1 and 2 are:
For 75 Ω and 50 Ω the note works them to R2 = 86.6 and R1 = 43.3, and the calculator agrees. The note takes the design from its first reference, Weinschel’s chapter in Reference Data for Engineers, and does not show why the loss is a minimum; what it does show is that the loss is fixed by the impedance ratio alone. Substituting Equations 1 and 2 into Maxim’s Equation 18 gives, with r = Rhigh/Rlow:
That closed form is derived here and checked against Equation 18 in the calculator’s tests; it is not in the note. For common pairs:
R_high → R_low series R1 shunt R2 power loss
75 Ω → 50 Ω 43.3 Ω 86.6 Ω 5.72 dB
100 Ω → 50 Ω 70.71 Ω 70.71 Ω 7.66 dB
150 Ω → 50 Ω 122.5 Ω 61.24 Ω 9.96 dB
300 Ω → 50 Ω 273.9 Ω 54.77 Ω 13.42 dB
600 Ω → 50 Ω 574.5 Ω 52.22 Ω 16.63 dB
100 Ω → 75 Ω 50 Ω 150 Ω 4.77 dB
The note calls the method “a broadband method to measure voltage and power accurately”, and broadband is the trade. An L-network of an inductor and a capacitor, which the impedance matching calculator designs, matches the same pair with no loss but only over a band. The resistive pad matches from DC upward and pays 5.72 dB for it. Using an ordinary 50 Ω pad instead does not help: a 6 dB, 50 Ω pi between a 75 Ω source and a 50 Ω load loses 6.18 dB and still shows the source 50 Ω, a return loss of 14.0 dB.
Rounding costs little here. Built from E24 values, 43 Ω and 91 Ω, the pad loses 5.58 dB and keeps return losses of 54.9 dB on the 75 Ω side and 37.3 dB on the 50 Ω side.
Power loss vs voltage loss in a minimum-loss pad
A minimum-loss pad has two losses, and quoting the wrong one is an easy mistake in a CATV measurement. Maxim’s introduction puts it plainly:
The amount of power loss with a MLP differs from voltage loss due to the impedance transformation. Specifically, power loss, which is used for power measurements, is the same when transforming in either direction. Voltage loss, which is used for voltage measurements, differs depending on the direction of impedance transformation.
For 75 Ω to 50 Ω, its Table 1 gives a power loss of 5.72 dB both ways, a voltage loss of 7.48 dB from 75 Ω to 50 Ω and 3.96 dB from 50 Ω to 75 Ω. The difference is the impedance ratio term, 10 log(75/50) = 1.76 dB, added in one direction and subtracted in the other. The same table shows a consequence worth knowing for back-to-back pads (75 Ω device, 50 Ω instrument at both ends): the two voltage losses add to 11.44 dB, exactly twice the power loss.
The low-to-high voltage loss has a ceiling the note does not mention. As the ratio grows, R1 becomes nearly all of Rhigh and the loss approaches 20 log 2 = 6.02 dB; even the 600 Ω to 50 Ω pad loses only 5.83 dB of voltage in that direction. The rule for which to use is the note’s last paragraph:
Use voltage loss when measuring in dBmV, dBµV, or dBV, or for any voltage-related measurement. Use power loss when measuring in dBm or for any power measurement.
One printed intermediate does not quite match. Equation 29 writes 86.6 Ω ∥ 50 Ω as 31.69, where the parallel is 31.698 Ω; the figure is truncated rather than rounded, and the −5.72 dB result is unaffected.
How much power each resistor must dissipate
A matched pad presents Z0 at its input, so all the available power enters it, and the output voltage is Vin/K. That fixes the split as fractions of the available power: the input resistor (the pi’s input shunt or the T’s input arm) takes (K − 1)/(K + 1), the load 1/K², the output resistor (K − 1)/((K + 1)K²), and the middle resistor the rest. The split is the same for a pi and a T.
A input middle output load
3 dB 17.1 % 24.2 % 8.6 % 50.1 %
10 dB 52.0 % 32.9 % 5.2 % 10.0 %
20 dB 81.8 % 16.4 % 0.8 % 1.0 %
30 dB 93.9 % 5.9 % 0.1 % 0.1 %
40 dB 98.0 % 2.0 % 0.0 % 0.0 %
So a 30 dB pad fed 1 W is not three resistors sharing a watt: its input resistor dissipates 0.939 W on its own and must be rated for that. Yeh et al. found the same in their thermal model, where “the resistor R1 within each 10 dB cell dissipates the most power and has the largest temperature”. Splitting the attenuation also splits the heat. One 20 dB pad puts 81.8 % of the input into its first resistor; two 10 dB pads in cascade put 52.0 % into the first pad’s input resistor and 5.2 % into the second’s. The same applies to the minimum-loss pad, where the resistor next to the source takes 57.7 % in either direction and the other 15.5 %.
What a chip attenuator’s datasheet adds
A chip attenuator is the same pi in one package, and its datasheet shows what the ideal formulas cannot. Vishay’s CZA is a “Thick Film Chip Attenuator, Surface Mount, Unbalanced π Type”, and its ratings are small: “Rated dissipation at 70 °C” of 0.075 W for the CZA06S and 0.040 W for the CZA04S, “DC to 3 GHz”, and a VSWR of “1.2 max.”.
Power. A 10 dB pad dissipates 90 % of its input, so the CZA06S’s 0.075 W is reached at 83.3 mW available, 19.2 dBm, and the CZA04S’s at 44.4 mW, 16.5 dBm. The rating is for the package; the datasheet does not say how it is shared among the three film resistors, while the network puts 52 % of the input into the input shunt alone. The derating curve runs from 100 % at 70 °C to zero at 150 °C for the CZA06S (125 °C for the CZA04S). Read as a straight line, at 100 °C the CZA06S is down to 46.9 mW and 17.2 dBm of input.
Match. A VSWR of 1.2 is a return loss of 20.8 dB. The ideal arithmetic promises far better: a discrete 10 dB pi rounded to E24 values (100 Ω, 68 Ω, 100 Ω) still gives 49.6 dB. The datasheet gives one VSWR limit for the whole DC to 3 GHz range and does not break it down by frequency.
Tolerance. The table splits it by attenuation: ±0.3 dB (code L) for 1 to 5 dB and ±0.5 dB (code H) for 6 to 20 dB. The datasheet is not consistent with itself in places. Its impedance table lists 50, 75, 100, 300 and 600 Ω for the CZA06S, but the part-number key has codes only for 50, 75 and 100 Ω; its attenuation list jumps from 6 dB to 10 dB, with other values by consulting the factory; and the endurance table heads its first column “0.5 dB to 5 dB” although the range starts at 1 dB. The whole document is marked “End of Life June-2021”, so it serves as a reference point for what a small pad is rated for rather than as a part to design in.
Temperature and power coefficients of attenuation
Resistors drift with temperature and with applied power, and Spectrum Control’s note on Weinschel attenuators is about what that does to a pad. It quotes the military requirement: “Military Standard, MIL-A-3933 for fixed attenuators calls for a TCA of 0.0004dB/dB/°C”, which for a 30 dB attenuator over 100 °C allows 30 × 100 × 0.0004 = 1.2 dB. Weinschel’s own figure, “0.0001dB/dB/°C”, gives 0.3 dB. Power has its own coefficient: “a typical 2 W attenuator has a PCA of < 0.005 dB/dB/W”, or 0.3 dB for 30 dB at 2 W.
The note’s central point is about matched coefficients:
As long as the shunt and series resistive elements of an attenuator have the same TCR the attenuation will always increase at DC, independent of the temperature and the magnitude of the TCR.
The network bears it out. Scale every resistor of a 10 dB, 50 Ω pi by +1 % and the attenuation rises by 0.00019 dB; scale them by −1 % and it also rises, by 0.00020 dB. A pad whose resistors move together stays a perfect 10 dB pad, but for 50.5 Ω instead of 50 Ω, and the small rise is the mismatch. The match is what suffers: at ±10 % the input resistance is 54.48 Ω or 45.48 Ω and the return loss falls to 27.4 dB or 26.5 dB, which matches the impedance curve the note plots in its Figure 3.
The note’s Figure 2 is the attenuation of a 10 dB distributed attenuator over the same ±10 %. Uniform scaling treats every resistive topology alike, so the lumped pi can be compared with it directly. The plot reads about 10.010 dB and 10.012 dB at its extremes. The calculator’s figures are 10.0177 dB and 10.0217 dB, because it reports the loss from the power the source has available, reflected power included. Taking the power that enters the pad instead gives 10.0098 dB and 10.0119 dB, which matches the plot. Both are honest definitions of attenuation; they differ by the mismatch loss.
The same curve shows how little of a TCA specification the DC arithmetic explains. Its worst point, 0.0119 dB of change on 10 dB over 100 °C, is about 0.00001 dB/dB/°C, a tenth of Weinschel’s quoted 0.0001. The note says where the specified figure comes from: “The worst-case coefficient arrived at is based on the maximum attenuation change over the frequency band.”
The caution is for discrete pads. Spectrum Control warns that “the discrete shunt and series chip resistors may have different TCRs”, and the arithmetic shows the cost: if only the series resistor of the 10 dB pi rises 1 %, the attenuation rises 0.045 dB, over two hundred times the common-drift figure. That is the reason Vishay lists “Tolerance matching and temperature tracking superior to individual components” as a feature of the one-package part.
Where the resistor model stops working
Every number above treats the resistors as pure resistance with zero-length connections. Yeh et al. ran into the limit while sizing their thin-film resistors:
For a large resistor, its parasitic capacitance and inductance become important, and a lumped element model is not valid. Microwave simulations using ANSYS’s high frequency software simulator (HFSS) confirmed that the response of the attenuator degraded at high frequencies when the physical size of the resistors got too large.
They state the constraint in thermal terms too: “While physically larger resistors are thermally advantageous, the maximum size of the shunt resistors is limited by the microwave response.” Their target was “a flat microwave response (less than 3 dB change) up to 12 GHz in the simulations”, a loose definition of flat by instrument standards, and the built 20 and 30 dB parts came within 3 dB of design from 1 to 10 GHz at room temperature. The equations give the DC and low-frequency answer. Above that, the resistor size and the layout decide, which is why a chip attenuator states a frequency range and a VSWR limit rather than the arithmetic’s perfect match.
Sources
- Skyworks 200312E, Design With PIN Diodes: p.15, bridged-T Equation 37 with Z0² = RS1 × RS2, pi Equations 38 and 39, the 50 Ω minimum shunt, and the tracking requirement on the two diodes.
- Yeh et al., Microwave attenuators for use with quantum devices below 100 mK, J. Appl. Phys. 121, 224501 (2017): p.2, the T pad Equations 1 and 2 with the shunt drawn as two resistors, the 10 dB, 50 Ω example and the parasitic limit on resistor size; p.4, the power in R1; p.5, the measured response.
- Maxim AN972, CATV Minimum Loss Pad for 75 Ω Measurements: Equations 1 and 2 for any two impedances, the voltage and power loss Equations 6, 18 and 22, Table 1 and the 75 Ω to 50 Ω example.
- Spectrum Control, Understanding Temperature & Power Coefficient in Attenuators: TCA and PCA, the MIL-A-3933 figure, matched TCR and the DC attenuation and impedance plots for a 10 dB attenuator.
- Vishay CZA chip attenuator datasheet: power rating and derating, DC to 3 GHz, VSWR, attenuation tolerance by range, and the end-of-life notice.