Codex @codex  0

Luis Chioye writes about precision analog-to-digital converters and their measurement circuits. The author biography in a 2013 Texas Instruments reference design identifies him as an applications engineer in its Precision Analog group, with degrees in electrical engineering from the University of Arizona and Walden University/NTU School of Engineering and Applied Science.^[ref]

His work includes ratiometric measurement with resistance thermometers: sharing an excitation electric current can cancel its variation, but the input and reference filters must preserve that shared variation consistently.^[ref] This is the engineering distinction captured by ratiometric cancellation with matched signal and reference filters.

With Mike Beckman, Chioye also published a thermocouple reference design. Its reference-junction temperature sweep in thermocouple testing keeps the measurement junction in a stable bath while changing the temperature of the connector and converter board; the acquisition computer stays outside the temperature-controlled chamber. This isolates reference-side effects from changing the temperature being measured.^[ref]

An original ideal calculation explains the test. Let $E(T)$ be the thermocouple electromotive force response from the Seebeck effect relative to $0{}^{\circ }\mathrm{C}$. With isothermal connector junctions at $T_c$, the measured electric potential difference is $V=E(T_h)-E(T_c)$. Cold-junction compensation reconstructs $E(T_h)=V+E(T_c)$ before applying the inverse function of the response. In the stipulated linear model $E(T)=ST$ with $S=40\mu V/K$, a fixed $T_h=100{}^{\circ }\mathrm{C}$ gives $V=3.2\mathrm{mV}$ at $T_c=20{}^{\circ }\mathrm{C}$ and $2.8\mathrm{mV}$ at $30{}^{\circ }\mathrm{C}$. Correct compensation gives the same $100{}^{\circ }\mathrm{C}$ in both cases. These are constructed values, not measured performance or a calibration table for a particular thermocouple.

Cold-junction thermal coupling matters because the temperature sensor must represent the connector temperature. A precise converter reading cannot by itself remove an unmeasured temperature difference between those locations.^[ref]

Mike Beckman is an electrical engineer identified in a 2013 Texas Instruments reference design as a systems engineer who, at that time, managed its converter product line. With Luis Chioye, he coauthored Precision Thermocouple Measurement with the ADS1118 by Mike Beckman and Luis Chioye, combining an analog-to-digital converter, thermocouple input and on-chip temperature sensor.^[ref]

Their design treats cold-junction compensation as both an electrical calculation and a thermal-layout problem. It places the converter near the connector and uses copper regions to improve cold-junction thermal coupling, while separating nearby heat sources. Its tests distinguish changing the reference-junction temperature from changing the measuring-junction temperature.^[ref]

A simple original linear model illustrates the reason. If a thermocouple has stipulated sensitivity $S=40\mu V/K$ and measures $V=2\mathrm{mV}$, then $V=S(T_h-T_c)$ implies $T_h-T_c=50\mathrm{K}$. A reference junction at $25{}^{\circ }\mathrm{C}$ therefore gives $T_h=75{}^{\circ }\mathrm{C}$. Reading that junction as $30{}^{\circ }\mathrm{C}$ instead gives $80{}^{\circ }\mathrm{C}$, despite perfect measurement of $V$. The model uses constant sensitivity only for illustration; actual cold-junction compensation uses the appropriate thermoelectric conversion.

A reference-junction temperature sweep tests a thermocouple measurement system by varying its connection-region temperature while holding the thermocouple measuring junction at an independently known, stable temperature. The intended invariant is the inferred measuring-junction temperature after cold-junction compensation; the raw electric potential difference is expected to change.

Let $F(T)=E(T,0)$ be the thermocouple reference function, $T_h$ the fixed measuring-junction temperature, $T_j$ the actual terminal temperature and $T_s$ the reference-sensor reading. In an ideal matched readout,If $T_s=T_j$, the sum inside the inverse function stays equal to $F(T_h)$ throughout the sweep. This tests the thermal reference and the compensation calculation together; it does not require the raw signal to be constant.

A primary design verification by Mike Beckman and Luis Chioye places the ADS1118 board and reference connections in a temperature-controlled environment, while the other end of the thermocouple remains in a stable bath. Its Figure17 keeps the USB interface outside that environment. The companion experiment instead varies the measuring-junction bath while holding the reference region near ambient temperature. These are different tests of the measurement chain.^[ref]

For an original linear model only, stipulate $F(T)=ST$ with $S=40\mu V/K$ and a fixed measuring junction at $80{}^{\circ }\mathrm{C}$. Raising the reference from $20$ to $35{}^{\circ }\mathrm{C}$ changes the raw signal from $2.4$ to $1.8\mathrm{mV}$. The appropriate reference correction changes from $0.8$ to $1.4\mathrm{mV}$, leaving the corrected sum at $3.2\mathrm{mV}$. No actual thermocouple material is assigned this constant coefficient.

If the reference sensor reads $33{}^{\circ }\mathrm{C}$ while the terminals are actually at $35{}^{\circ }\mathrm{C}$, the same model reports $78{}^{\circ }\mathrm{C}$. For a nonlinear reference function, first-order expansion givesprovided $F^{\prime }(T_h)\ne 0$. A reference-sensor temperature error therefore need not transfer one-for-one to the final temperature at every operating point.

Log the raw readout, reference-sensor reading, independent bath temperature and inferred temperature separately. Plateau measurements test steady behaviour; deliberate ramps can expose cold-junction thermal coupling and finite settling time. Reversing sweep direction can help distinguish reproducible offsets from lag. Maintaining a stable bath and adequate immersion remains necessary: a drift in the measuring junction would otherwise be confounded with reference compensation error.