Resistance temperature detectors (RTDs) are among the most versatile and accurate of the commonly used temperature sensors. However, the typical RTD’s temperature response consists of resistance variations on the order of only tenths of ohms/°C. Therefore, strict attention must be paid in these designs to the effects of transducer lead-wire resistance if sensor accuracy is to be preserved. As a consequence, whenever the RTD can’t be located extremely close to the drive electronics, lead-wire resistance effects will be large and potentially problematic.
Differential input sensing arrangements are often necessary to sense and cancel out these spurious lead-voltage drops. For this reason, RTDs are routinely supplied in four-lead “Kelvin” packages (Fig. 1). Excitation current IR is applied through sensor lead 1, generating the desired temperature-related sensor voltage VT = IRRT.
Sensor lead-resistance RW however, contributes a parasitic lead-voltage term: VW = IRRW. So the voltages actually available at the ends of the sensor leads aren’t the desired VT. Instead, they are: V3 = VT + 2VW; V2 = VT + VW; V1 = VW. Voltage differencing (e.g., VT = V2 − V1) is needed to extract VT and accurately read out the actual sensor resistance and sensor temperature.
In systems using only one or two RTDs, this complication doesn’t present much of a problem either in terms of cabling or analog-to-digital converter (ADC) input costs. But for systems employing many sensors, things can get expensive rather quickly.
For example, this article resulted from an application that measures the infrared optical spectra of chemical aerosols in simulated planetary atmospheres. The apparatus used incorporates a liquid-nitrogen-cooled reaction tube instrumented with some 28 platinum RTDs. A conventional design for interconnection to the PRTD array would have required two ADC differential inputs for each sensor. In all, an unwieldy total of 56 ADC multiplexer channels would be needed.
Even worse is the fact that the reaction tube and the attached PRTDs are located inside a sealed and evacuated atmospheric chamber. Consequently, every conductor exiting the sensor array involves a separate hermetic “feedthrough” chamber-wall penetration. The cost for this arrangement is more than $50 for each wire and over $200 for every fourwire Kelvin sensor.
By contrast, the multiplex circuit shown in Figure 2 cuts the required number of ADC channels from 2N (where N = the number of RTDs) to only 2 + N. It also cuts the number of RTD array connections (feedthroughs) from 4N to 4 + N. For 28 sensors, the resulting savings are 2N − 2 − N = 26 inputs and 4N − 4 − N = 80 feedthroughs.
In Figure 2, the 28 four-wire PRTDs are connected in a single series string. The excitation current, IR, leaves the array via a parasitic feedthrough + interconnect cable resistance, RF. This generates offset voltage V1 = IRRF, sensed and presented to the ADC input V1. IR then passes through RTD1 and its paralleled leads to generate V2 = IR(R1 + RW/2 + RW/2) + V1 = IR(R1 + RW) + V1. It travels through RTD2 to generate V3 = IR(R2 + RW) + V2, etc., all the way to V29 = IR(R28 + RW) + V28. Finally, it produces V30 = IRRW + V29.
It’s reasonable to assume good matching between the RW values(given equal lead lengths and wire gauge). We can therefore write: VW = V30 − V29 and VT1 = IRR1 = V2 − V1 − VW; VT2 = V3 − V2 − VW, ..., VT28 = V29 − V28 − VW. Thus, 28 accurate, lead-resistancecompensated sensor measurements are made using only 32 feedthroughs and cable conductors and 30 ADC input channels.
Analog conversions of sensor voltages are made by sequentially “walking” the multiplexer address up the sensor array. For example, multiplexer input V2 serves first as the + input for the VT1 conversion, then as the - input for the VT2 conversion. Each multiplexer input does double duty in this way (except V1 and V30). To support this unusual addressing sequence, the multiplexer logic of some ADCs may require modification.