AN-86: A Standards Lab Grade 20-Bit DAC with 0.1ppm/°C Drift

Introduction

Significant progress in high precision, instrumentation grade D-to-A conversion has recently occurred. Ten years ago 12-bit D-to-A converters (DACs) were considered premium devices. Today, 16-bit DACs are available and increasingly common in system design. These are true precision devices with less than 1LSB linearity error and 1ppm/°C drift.¹ Nonetheless, there are DAC applications that require even higher performance. Automatic test equipment, instruments, calibration apparatus, laser trimmers, medical electronics and other applications often require DAC accuracy beyond 16 bits. 18-bit DACs have been produced in circuit assembly form, although they are expensive and require frequent calibration. 20 and even 23+ (0.1ppm!) bit DACs are represented by manually switched Kelvin-Varley dividers. These devices, although amazingly accurate, are large, slow and extremely costly. Their use is normally restricted to standards labs.² A useful development would be a practical, 20-bit (1ppm) DAC that is easily constructed and does not require frequent calibration.

20-Bit DAC Architecture

Figure 1 diagrams the architecture of a 20-bit (1ppm) DAC. This scheme is based on the availability of a true 1ppm analog-to-digital converter with scale and zero drifts below 0.02ppm/°C. This device, the LTC^®2400, is used as a feedback element in a digitally corrected loop to realize a 20-bit DAC.³

In practice, the “slave” 20-bit DAC’s output is monitored by the “master” LTC2400 A-to-D, which feeds digital information to a code comparator. The code comparator differences the user input word with the LTC2400 output, presenting a corrected code to the slave DAC. In this fashion, the slave DAC’s drifts and nonlinearity are continuously corrected by the loop to an accuracy determined by the A-to-D converter and V_REF⁴. The sole DAC requirement is that it be monotonic. No other components in the loop need to be stable.

This loop has a number of desireable attributes. As mentioned, accuracy limitations are set by the A-to-D converter and its reference. No other components need be stable. Additionally, loop behavior averages low order bit indexing and jitter, obviating the loop’s inherent small-signal instability. Finally, classical remote sensing may be used or digitally based sensing is possible by placing the A-to-D converter at the load. The A-to-D’s SO-8 package and lack of external components makes this digitally incarnated Kelvin sensing scheme practical.⁵

Circuitry Details

Figure 2 is a detailed schematic of the 1ppm DAC. The slave DAC is comprised of two DACs. The upper 16 bits of the code comparator’s output are fed to a 16-bit DAC (“MSB DAC”), while the lower bits are converted by a separate DAC (“LSB DAC”). Although a total of 32 bits are presented to the two DACs, there are 8 bits of overlap, assuring loop capture under all conditions. The composite DACs’ resultant 24-bit resolution provides 4 bits of indexing range below the 20th bit, ensuring a stable LSB of 1ppm of scale. A1 and A2 transform the DAC’s output currents into voltages, which are summed at A3. A3’s scaling is arranged so that the correction loop can always capture and correct any combination of zero- and full-scale errors. A3’s output, the circuit output, feeds the LTC2400 A-to-D. The LT^®1010 provides buffering to drive loads and cables. The A-to-D’s digital output is differenced against the input word by the code comparator, which produces a corrected code. This corrected code is applied to the MSB and LSB DACs, closing a feedback loop.⁶ The loop’s integrity is determined by A-to-D converter and voltage reference errors.⁷ The resistor and diodes at the 5V powered A-to-D protect it from inadvertent A3 outputs (power up, transient, lost supply, etc.). A4 is a reference inverter and A5 provides a clean ground potential to both DACs.

Linearity Considerations

A-to-D linearity determines overall DAC linearity. The A-to-D has about ±2ppm nonlinearity. In applications where this error is permissible, it may be ignored. If 1ppm linearity is required, it is obtainable by correcting the residual linearity error with software techniques. Details on LTC2400 linearity and this feature are presented in Appendices D and E.

DC Performance Characteristics

Figure 3 is a plot of linearity vs output code. The data shows linearity is within 1ppm over all codes.⁸ Output noise, measured in a 0.1Hz to 10Hz bandpass, is seen in Figure 4 to be about 0.2LSB.⁹ This measurement is somewhat corrupted by equipment limitations, which set a noise floor of about 0.2µV.

Dynamic Performance

The A-to-D’s conversion rate combines with the loop’s sampled data characteristic and slow amplifiers to dictate relatively slow DAC response. Figure 5’s slew response requires about 150 microseconds.

Figure 6 shows full-scale DAC settling time to within 1ppm (±5µV) requires about 1400 milliseconds. A smaller step (Figure 7) of 500µV needs only 100 milliseconds to settle within 1ppm.¹⁰

Conclusion

Summarized 1ppm DAC specifications appear in Figure 9. These specifications should be considered guidelines, as the options and variations noted will affect performance. Consult the appropriate appendices for design specifics and trade-offs.

Notes

This Application Note was derived from a manuscript originally prepared for publication in EDN magazine.

Note 1: See Appendix A, “A History of High Accuracy Digital-to-Analog Conversion,” for a review of high accuracy digital-to-analog conversion.

Note 2: Consult Appendix C, “Verifying Data Converter Linearity to 1ppm,” for discussion on Kelvin-Varley dividers. Also, see Appendix A, “A History of High Accuracy Digital-to-Analog Conversion.”

Note 3: The LTC2400 analog-to-digital converter is profiled in Appendix B, “The LTC2400—A Monolithic 24-Bit Analog-to-Digital Converter.”

Note 4: D-to-A converters have been placed in loops to make A-to-D converters for a long time. Here, an A-to-D converter feeds back a loop to form a D-to-A converter. There seems a certain justified symmetry to this development. Turnabout is indeed fair play.

Note 5: One wonders what Lord Kelvin’s response would be to the digizatation of his progeny. Such uncertainties are the residue of progress.

Note 6: The code comparator is detailed in Appendix D, “A Processor Based Code Comparator.”

Note 7: Voltage reference options are discussed in Appendix I, “Voltage References.” For tutorial on the LTC2400, refer to Appendix B.

Note 8: Establishing and maintaining confidence in a 1ppm linearity measurement is uncomfortably close to the state of the art. The technique used is shown in Appendix C, “Verifying Data Converter Linearity to 1ppm.”

Note 9: Noise measurement considerations appear in Appendix H, “Microvolt Level Noise Measurement.”

Note 10: Measuring DAC settling time to 1ppm is by no means straightforward, even at the relatively slow speed involved here. See Appendix G, “Measuring DAC Settling Time.”

Appendix A

A History of High Accuracy Digital-to-Analog Conversion

People have been converting digital-to-analog quantities for a long time. Probably among the earliest uses was the summing of calibrated weights (Figure A1, left center) in weighing applications. Early electrical digital-to-analog conversion inevitably involved switches and resistors of different values, usually arranged in decades. The application was often the calibrated balancing of a bridge or reading, via null detection, some unknown voltage. The most accurate resistor-based DAC of this type is Lord Kelvin’s Kelvin-Varley divider (Figure, large box). Based on switched resistor ratios, it can achieve ratio accuracies of 0.1ppm (23+ bits) and is still widely employed in standards laboratories.1 High speed digital-to-analog conversion resorts to electronically switching the resistor network. Early electronic DACs were built at the board level using discrete precision resistors and germanium transistors (Figure, center foreground, is a 12-bit DAC from a Minuteman missile D-17B inertial navigation system, circa 1962). The first electronically switched DACs available as standard product were probably those produced by Pastoriza Electronics in the mid 1960s. Other manufacturers followed and discrete- and monolithically-based modular DACs (Figure, right and left) became popular by the 1970s. The units were often potted (Figure, left) for ruggedness, performance or to (hopefully) preserve proprietary knowledge. Hybrid technology produced smaller package size (Figure, left foreground). The development of Si-Chrome resistors permitted precision monolithic DACs such as the LTC1595 (Figure, immediate foreground). In keeping with all things monolithic, the cost-performance trade off of modern high resolution IC DACs is a bargain. Think of it! A 16-bit DAC in an 8-pin IC package. What Lord Kelvin would have given for a credit card and LTC’s phone number.

Note

Note 1: See Appendix C, “Verifying Data Converter Linearity to 1ppm,” for details on Kelvin-Varley Dividers.

Appendix B

The LTC2400—A Monolithic 24-Bit Analog-to-Digital Converter

The LTC2400 is a micropower 24-bit A-to-D converter with an integrated oscillator, 4ppm nonlinearity and 0.3ppm RMS noise. It uses delta-sigma technology to provide extremely high stability. The device can be configured for better than 110dB rejection at 50Hz or 60Hz ±2%, or it can be driven by an external oscillator for a user defined rejection frequency in the range 1Hz to 120Hz.

This ultraprecision A-to-D converter in an SO-8 pin package forms the heart of the 20-bit DAC described in the text. It is significant that the device is used here as a circuit component rather than in the traditional standalone role accorded precision A-to-D converters. This freedom, in keeping with the IC’s economy and ease of use, is a noteworthy opportunity. Alert designers will recognize this development and capitalize on it. Key specifications for the A-to-D are given in Figure B1.

Appendix C

Verifying Data Converter Linearity to 1ppm

Help from the Nineteenth Century

Introduction

Verifying 1ppm linearity of the DAC and the analog-to-digital converter used to construct it requires special considerations. Testing necessitates some form of voltage source that produces equal amplitude output steps for incremental digital inputs. Additionally, for measurement confidence, it is desirable that the source be substantially more linear than the 1ppm requirement. This is, of course, a stringent demand and painfully close to the state of the art.

The most linear “D to A” converter is also one of the oldest. Lord Kelvin’s Kelvin-Varley divider (KVD), in its most developed form, is linear to 0.1ppm. This manually switched device features ten million individual dial settings arranged in seven decades. It may be thought of as a 3-terminal potentiometer with fixed “end-to-end” resistance and a 7-decade switched wiper position (Figure C1).

The actual construction of a 0.1ppm KVD is more artistry and witchcraft than science. The market is relatively small, the number of vendors few and resultant price high. If $13,000 for a bunch of switches and resistors seems offensive, try building and certifying your own KVD. Figure C2 shows a detailed schematic.

The KVD shown has a 100kΩ input impedance. A consequence of this is that wiper output resistance is high and varies with setting. As such, a very low bias current follower is required to unload the KVD without introducing significant error. Now, our KVD looks like Figure C3. The LT1010 output buffer allows driving cables and loads and, more subtly, maintains the amplifier’s high open-loop gain.

Approach and Error Considerations

This schematic is deceptively simple. In practice, construction details are crucial. Parasitic thermocouples (Seebeck effect), layout, grounding, shielding, guarding, cable choice and other issues affect achievable performance.¹ In fact, as good as the chopper-stabilized LTC1152 is with respect to drift, offset, bias current and CMRR, selection is required if we seek sub-ppm nonlinearity performance. Figure C4, an error budget analysis, details some of the selection criteria.

The buffer is tested with Figure C5’s circuit. As the KVD is run through its entire range, the floating null detector must remain well within 1ppm (5µV), preferably below 0.5ppm. This test ensures that all error sources, particularly I_B and CMRR, whose effects vary with operating point, are accounted for. Measured performance indicates the sum of all errors called out in Figure C4 is well within desired limits.

In Figure C6, we add offset trim, a stable voltage source and a second KVD to drive the main KVD. Additionally, an ensemble of three HP3458A voltmeters monitor the output.

The offset trim bleeds a small current into the main KVD ground return, producing a few microvolts of offset-trim range. This functionally trims out all sources of zero error (amplifier offsets, parasitic thermocouple mismatches and the like), permitting a true zero volt output when the main KVD is set to all zeros.

The voltmeters, specified for <0.1ppm nonlinearity on the 10V range, “vote” on the source’s output.

Circuitry Details

Figure C7 is a more detailed schematic. It is similar to Figure C6 but highlights issues and concerns. The grounding scheme is single point, preventing mixing of return currents and the attendant errors. The shielded cables used for interconnections between the KVDs and voltmeters should be specified for low thermal activity. Keithley type SC-93 and Guildline #SCW are suitable. Crush type copper lugs (as opposed to soldered types) provide lower parasitic thermocouple activity at KVD and DVM connection points. However, they must be kept clean to prevent oxidation, thus avoiding excessive thermal voltages.² A copper deoxidant (Caig Labs “Deoxit” D100L) is quite effective for maintaining such cleanliness. Low thermal lugs and jacks, preterminated to cables, are also available (Hewlett-Packard 11053, 11174A) and convenient.

Thermal baffles enclosing KVD and DVM connections tend to thermally equilibrate their associated banana jack terminals, minimizing residual parasitic thermocouple activity. Additionally, restrict the number of connections in the signal path. Necessary connections should be matched in number and material so that differential cancellation occurs. Complying with this guideline may necessitate deliberate introduction of solder-copper junctions (marked “X” on Figure C7) to obtain optimum differential cancellation.³ This is normally facilitated by simply breaking the appropriate wire or PC trace and soldering it. Ensure that the introduced thermocouples temperature track the junctions they are supposed to cancel. This is usually accomplished by locating all junctions within close physical proximity.

The noise filtering capacitor at the main KVD is a low leakage type, with its metal case driven by the output buffer to guard out surface leakage.

When studying the approach used, it is essential to differentiate between linearity and absolute accuracy. This eliminates concerns with absolute standards, permitting certain freedoms in the measurement scheme. In particular, although single-point grounding was used, remote sensing was not. This is a deliberate choice, made to minimize the number of potential error-causing parasitic thermocouples in the signal path. Similarly, a ratiometric reference connection between the KVD LTZ1000A voltage source and the voltmeters was not utilized for the same reason. In theory, a ratiometric connection affords lower drift. In practice, the resultant introduced parasitic thermocouples obviate the desired advantage. Additionally, the aggregate stability of the LTZ1000A reference and the voltmeter references (also, incidentally, LTZ1000A based) is comfortably inside 0.1ppm for periods of 10 minutes.⁴ This is more than enough time for a 10-point linearity measurement.

Construction

Figures C8 and C9 are photographs of the voltage source and the reference-buffer box internal construction. The figure captions annotate some significant features.

Results

This KVD-based, high linearity voltage source has been in use in our lab for nearly two years. During this period, the total linearity uncertainty defined by the source and its monitoring voltmeters has been just 0.3ppm (see Figure C10’s measurement regime). This is more than 3 times better than the desired 1ppm performance, promoting confidence in our measurements.⁵

Acknowledgments

The author is indebted to Lord Kelvin and to Warren Little of the C. S. Draper Laboratory (née M. I. T. Instrumentation Laboratory) standards lab. Warren taught me, with great patience, the wonders of KVDs some thirty years ago and I am still trading on his efforts.

Notes

Note 1: See Appendix J, “Cables, Connections, Solder, Layout, Component Choice, Terror and Arcana,” for relevant tutorial.

Note 2: See above Footnote.

Note 3: See Appendix J, “Cables, Connections, Solder, Layout, Component Choice, Terror and Arcana,” for further discussion.

Note 4: The LTZ1000A reference is detailed in Appendix I, “ Voltage References.”

Note 5: The author, wholly unenthralled by web surfing, has spent many delightful hours “surfing the Kelvin.” This activity consists of dialing various Kelvin-Varley divider settings and noting monitoring A-to-D agreement within 1ppm. This is astonishingly nerdy behavior, but thrills certain types.

Appendix D

A Processor-Based Code Comparator

The code comparator enforces the loop by setting the slave DAC inputs to the code that equalizes the user input and the LTC2400 A-to-D output. This action is more fully described on page one of the text.

Figure D1 is the code comparator’s digital hardware. It is composed of three input data latches and a PIC-16C5X processor. Inputs include user data (e.g., DAC inputs), linearity curvature correction (via DIP switches), convert command (“DA WR”) and a selectable filter time constant. An output (“DAC RDY”) indicates when the DAC output is settled to the user input value. Additional outputs and an input control and monitor the analog section (text Figure 2) to effect loop closure. Note that although a total of 32 bits are presented to the two 16-bit slave DACs, there are 8 bits of overlap, allowing a total dynamic range of 24 bits. This provides 4 bits of indexing range below the 20th bit, ensuring a stable LSB of 1ppm of scale. The 8-bit overlap assures the loop will always be able to capture the correct output value.

The processor is driven by software code, authored by Florin Oprescu, which is described below.

;20bit DAC code comparator
;
;***************************************************
;                                                  *
;    Filename:          dac20.asm                  *
;    Date         12/4/2000                        *
;    File Version:      1.1                        *
;                                                  *
;    Author:            Florin Oprescu             *
;    Company:           Linear Technology Corp.    *
;                                                  *
;                                                  *
;***************************************************
;
; Variables
;============
; uses 17 bytes of RAM as follows:
;
; {UB2, UB1, UB0} user input word buffer
;———————————————————————————————————————
; 24 bits unsigned integer (3 bytes):
;
; The information is transferred from the external input register
; into {UB2, UB1, UB0} whenever a “user input update” event
; is detected by testing the timer0 content. Following the data
; transfer, the UIU (“user input update”) flag is set and the DAC
; ready flags RDY and RDY2 are cleared. UB0 uses the same physical
; location as U0. The user input double buffering is necessary
; because the loop error corresponding to the current ADC reading
; must be calculated using the previous user input value.
; The old user input value can be replaced by the new user input
; value only after the loop error calculation. Read full article

Appendix E

Linearity and Output Range Options

The LTC2400 used as the feedback A-to-D element in the DAC has a typical ±2ppm residual nonlinearity. Figure E1’s lower curve shows this, along with the first order correction necessary (upper curve) to get nonlinearity inside 1ppm (center curve). If true 1ppm performance is necessary, the software based correction described in Appendix D can be utilized. The software generates the desired “inverted bowl” correction characteristic. The correction may be set to complement the residual nonlinearity characteristics of any individual LTC2400 via DIP switches at the code comparator.

The LTC2410 offers another approach to improved linearity. This LTC2400 variant has improved linearity but specifies a maximum 2.5V input range. Figure E2 divides the DAC output with a precision resistor ratio set, allowing LTC2410 use while maintaining the 5V full-scale output. The disadvantage of this approach is the ratio set’s additional 0.1ppm/°C and 5ppm/year error contribution.¹ Figure E3 is similar, although the ratio set’s new value permits a 10V full-scale output.

Note

Note 1: The strata is becoming rarified when “error contribution” is delineated in fractional parts-per-million and the yearly drift rate noted.

Appendix F

Output Stages

Some applications may require outputs other than the text circuit’s 0V to 5V range. The simplest variation is a bipolar output, shown in Figure F1. The circuit, a summing inverter, subtracts the DAC output from a reference to obtain a bipolar output. Resistor and reference values may be varied to obtain different output excursions. The LT1010 output buffer provides drive capability and the chopper stabilized amplifier maintains 0.05µV/°C stability. The resistors introduce a 0.3ppm/°C error contribution.¹

Figure F2 yields voltage gain by dividing the DAC output prior to its application to the feedback A-to-D. In this case, the 1:1 divider ratio sets a 10V output, assuming an A-to-D reference of 5V. As in Figure F1, the resistors add a slight temperature error, about 0.1ppm/°C for the ratio set specified.²

Figure F3 uses active devices for voltage outputs as high as ±100V. The discrete high voltage stage is driven in closed-loop fashion by a chopper stabilized amplifier. Q1 and Q2 furnish voltage gain, and feed the Q3-Q4 emitter follower outputs. Q5 and Q6 set current limit at 25mA by diverting output drive when voltages across the 27Ω shunts become too high. The local 1M-50k feedback pairs set stage gain at 20, allowing LTC1152 drives to cause full ±120V output swing. The local feedback reduces stage gain-bandwidth, making dynamic control easier. This stage is relatively simple to frequency compensate because only Q1 and Q2 contribute voltage gain. Additionally, the high voltage transistors have large junctions, resulting in low f_ts, and no special high frequency roll-off precautions are needed. Because the stage inverts, feedback is returned to the amplifier’s positive input. Frequency compensation is achieved by rolling off the amplifier with the local 0.005µF-10k pair.

Heating and voltage coefficient errors are minimized in the feedback term by using four individual resistors. Trimming involves selecting the indicated resistor for exactly 100.0000V output with the DAC at full scale.

Figure F4 increases output current capability with a current gain stage inside the DAC output amplifier’s feedback loop. This stage replaces the LT1010 150mA buffer shown in the text. The figure shows two options, differing in output capacity. It is worth noting that as output current rises, wiring resistance becomes a large potential error term. For example, at only 10mA output, 0.001Ω of wiring resistance introduces 10mV drop—a 2ppm error. Because of this, heavy loads should be supplied via short, highly conductive paths and remote sensing employed.

Notes

Note 1: See Note 1 in Appendix E.

Note 2: See above footnote.

Appendix G

Measuring DAC Settling Time

Measuring the 20-bit DAC’s output settling time is a challenging task. Although the time scale involved is relatively slow, the 5µV LSB step size presents problems. The issue reduces to obtaining a great deal of gain without inducing overdrive in the monitoring oscilloscope. Such overdrive will corrupt the measurement, rendering displayed results meaningless.

Figure G1 is a solution. The DAC output is resistively balanced against a precision variable reference supply, adjustable in 0.5µV steps.¹ The circuit’s remainder constitutes a clamped, distributed gain of 2000 amplifier. Diode clamping, placed at each gain stage input, prevents saturation from occurring even with large DAC-reference supply imbalances. The distributed gain allows 10kHz bandwidth while maintaining clamping effectiveness. The monitoring oscilloscope, operating at 5mV or 10mV/DIV (5µV to 10µV at the DAC output), can readily discern 5µV settling without incurring deleterious overdrive.

Layout and construction of this circuit requires care. Figure G2 shows construction details. A linear layout minimizes parasitic feedback paths, preventing oscillation. The DAC input step is fully shielded, preventing feedthrough to various sensitive points within the amplifier. Finally, the entire circuit is built into a shielded enclosure to minimize effects of stray RF and pick up.

The circuit is tested by applying a test step that settles much faster than the DAC. Figure G3 uses a mercury wetted reed relay based pulse generator to supply the step. The unit noted is commercially produced, although similar results are obtainable with standard mercury based reed relays. When the relay opens the circuit’s output settles essentially instantaneously (Figure G4) relative to DAC speed and settling time amplifier bandwidth.

Figure G1’s response is tested by grounding one of its inputs and driving the other with the pulse generator. Figure G5 shows settling to within 1ppm (±5µV) in 2ms. This is much faster than the DAC settles, lending confidence to text Figures 6 and 7 indicated results.

Note

Note 1: See Appendix C for details on such a supply.

Appendix H

Microvolt Level Noise Measurement

Verifying DAC output noise requires a quiet, high gain amplifier at the oscilloscope. Figure H1 shows one way to take the measurement. The input preamplifier, operating at a gain of 1000, supplies a high pass cutoff at 0.1Hz. It drives the oscilloscope via a 10Hz discrete low pass filter. The oscilloscope, set to 1mV/DIV, indicates 1µV/DIV referred to the preamplifier input. Figure H2 indicates DAC output noise well below an LSB, about 0.9µV. Equipment limitations set measurement noise floor at 0.2µV.

Figure H3 shows the noise measurement test setup. Note that the signal levels involved dictate a completely shielded, coaxial path from breadboard to oscilloscope.

Figure H4 lists some applicable high sensitivity amplifiers suitable for the noise measurement. Current generation oscilloscopes rarely have greater than 2mV/DIV sensitivity, although older instruments offer more capability. The figure lists representative preamplifiers and oscilloscope plug-ins suitable for noise measurement. These units feature wideband, low noise performance. It is particularly significant that many of these instruments are no longer produced. This is in keeping with current instrumentation trends, which emphasize digital signal acquisition as opposed to analog measurement capability.

The monitoring oscilloscope should have exceptional trace clarity. In the latter regard high quality analog oscilloscopes are unmatched. The exceptionally small spot size of these instruments is well-suited to low level noise measurement.¹ The digitizing uncertainties and raster scan limitations of DSOs impose display resolution penalties. Many DSO displays will not even register the fine structure of the noise waveform.

Notes

Note 1: In our work we have found Tektronix types 453, 453A, 454, 454A, 547 and 556 excellent choices. Their pristine trace presentation is ideal for discerning small signals of interest against a noise floor limited background.

Appendix I

Voltage References

Figure I1 lists some voltage reference options for use with the DAC. The self-contained types are convenient and easily applied. The LM199A and the LTZ1000A require external circuitry but offer higher performance. All choices must be trimmed to establish absolute DAC accuracy. The LTZ1000A offers the highest stability and is discussed below.

Figure I2 shows the LTZ1000A and its support circuitry. A1 senses LTZ1000A die temperature and accordingly controls the IC heater via the 2N3904. A2 controls reference current. The Zener reference is sensed via Kelvin connections, minimizing voltage drop effects. A single point ground eliminates return current mixing and the attendant errors that would be produced.

Figure I3 offers choices for reference buffering. All employ a chopper stabilized amplifier augmented with a buffer output stage. Buffer error is extremely low, as noted in Appendix C’s discussion. I3a, a simple unity-gain stage, transmits the input to the output with low error and minimal reference loading. I3b takes moderate gain, allowing a 7V reference input to produce (in this case) 10V at the output. I3c offers two ways to get 5V from the nominal 7V input. A precision divider lightly loads the reference in one case; the 5V output is taken at the LT1010. Reference loading is avoided by placing the divider at the output (optional case shown) and driving the A-to-D reference input from the divider output, which is permissible.

Appendix J

Cables, Connections, Solder, Component Choice, Terror and Arcana

Subtle parasitic effects can have pronounced and seemingly inexplicable effects on low level circuit performance. Perhaps the most prevalent detractor to microvolt level circuitry is unintended thermocouples. Considerable discussion for dealing with thermocouples appeared in Appendix C and should be considered preliminary to this section’s material.

In 1822, Thomas Seebeck, an Estonian physician, accidentally joined semicircular pieces of bismuth and copper (Figure J1) while studying thermal effects on galvanic arrangements. A nearby compass indicated a magnetic disturbance. Seebeck experimented repeatedly with different metal combinations at various temperatures, noting relative magnetic field strengths. Curiously, he did not believe that electric current was flowing and preferred to describe the effect as “thermomagnetism.” He published his results in a paper, “Magnetische Polarisation der Metalle und Erze durch Temperatur-Differenz” (see References).

Subsequent investigation has shown the “Seebeck Effect” to be fundamentally electrical in nature, repeatable and quite useful. Thermocouples, by far the most common transducer, are Seebeck’s descendants. Unfortunately, unintended and unwanted thermocouples are also Seebeck’s progeny.

In low drift circuits, unwanted thermocouples are probably the primary source of error. Connectors, switches, relay contacts, sockets, wire and even solder are all candidates for thermal EMF generation. It is relatively clear that connectors and sockets can form thermal junctions. However, it is not at all obvious that junctions of wire from different manufacturers can easily generate 200nV/°C—four times a precision amplifier’s drift specification! Figure J2 shows a plot obtained for such a wire junction. Even solder can become an error term at low levels, creating a junction with copper or Kovar wires or PC traces (see Figure J3). Figure J4 lists thermocouple potentials for some common materials found in electronic assemblies.

The unusually energetic response of Cu-CuO necessitated the treatment described in Appendix C (Figure C7 and associated text) for cleaning DVM and Kelvin-Varley divider connections. Readers finding this figure’s information seemingly academic should be awakened by Figure J5. This chart lists thermoelectric potentials for commonly employed laboratory connectors. Thermocouple activity of some types is more than 20 times greater than others. Be careful!

Connection Type	Description	Thermoelectric Potential (μV/°C)
BNC-BNC Mate		0.4
BNC-Banana Adapter		0.35
BNC-BNC "Barrel" Adapter		0.4
Male/Female Banana Mate Sample #1		0.35
Male/Female Banana Mate Sample #2		1.1
Male/Female Banana Mate (Type Specified for Low Thermal Activity) Sample #3		0.07
Copper Lug-Copper Banana Binding Post		0.08
Copper Lug-Standard Banana Binding Post		0.5
Plated Lug-Copper Banana Binding Post		1.7

Minimizing thermal EMF induced errors is possible if judicious attention is given to circuit board layout. In general, it is good practice to limit the number of junctions in the signal path. Avoid connectors, sockets, switches and other potential error sources to the extent possible. In some cases this will not be possible. In these instances, attempt to balance the number and type of junctions in the signal path so that differential cancellation occurs. Doing this may involve deliberately creating and introducing junctions to offset unavoidable junctions. This can be a tricky procedure. Repeated deliberate temperature excursions may be necessary to determine the optimal number and placement of added junctions. Experimentation, tempered by a healthy reserve of patience and abundance of time, is required. This practice, borrowed from standards lab procedures, can be quite effective in reducing thermal EMF originated drifts. Figure J6 shows a simple example where a nominally unnecessary resistor is included to promote such thermal balancing. For remote signal sources connectors may be unavoidable. In these cases, choose a connector specified for relatively low thermal EMF activity and ensure a similarly balanced approach in routing signals through the connector along the circuit board and to circuitry. If some imbalance is unavoidable, deliberately introduce an intentional counterbalancing junction. In all cases maintain the differencing junctions in close physical proximity, which will keep them at the same temperature. Avoid drafts and temperature gradients, which can introduce thermal imbalances and cause problems. Figure J7 shows the LTC1150 set up in a test circuit to measure its temperature stability. The lead lengths of the resistors connected to the amplifier’s inputs are identical. The thermal capacity each input sees is also balanced because of the symmetrical connection of the resistors and their identical size. Thus, thermal EMF induced shifts are equal in phase and amplitude and cancellation occurs. Very slight air currents can still affect even this arrangement. Figure J8 shows a strip chart of output noise with the circuit covered by a small styrofoam cup (HANDI-KUP^™ Company Model H8-S) and with no cover in “still” air. This data illustrates why it is often prudent to enclose low level circuitry inside some form of thermal baffle.

Thermal EMFs are the most likely, but not the only, potential low level error source. Electrostatic and electromagnetic shielding may be required. Power supply transformer fields are notorious sources of errors often mistakenly attributed to amplifier DC drift and noise. A transformer’s magnetic field impinging on a PC trace can easily generate microvolts across that conductor in accordance with well known magnetic theory. The circuit cannot distinguish between this spurious signal and the desired input. Attempts to eliminate the problem by rolling off circuit response may work, but often the filtered version of the undesired pickup masquerades as an unstable DC term. The most direct approach is to use shielded transformers but careful layout may be equally effective and less costly. A circuit that requires the transformer to be close by to achieve a good quality grounding scheme may be disturbed by the transformer’s magnetic field. An RF choke connected across a scope probe can determine the presence and relative intensity of transformer fields, aiding layout experimentation.

Another source of parasitic error is stray leakage current. Such leakage currents must be prevented from influencing circuit operation. The simplest way to do this is to connect leakage sensitive points via teflon standoffs. Because the points never contact the PC board, stray leakage currents do not affect them. Although this approach is effective, its implementation may not be acceptable in production. Guarding is another technique for minimizing board leakage effects. The guard is a PC trace completely encircling the leakage sensitive points. This trace is driven at a potential equal to that of the point, preventing leakage to the “guarded” point. On PC boards, the guard should enclose the node(s) to be protected. Guarding was used to eliminate the effects of capacitor surface leakage in Appendix C’s Figure C7.

参考电路

1. Linear Technology Corporation, “LTC2400 Data Sheet,” Linear Technology Corporation, January 1999.

2. Linear Technology Corporation, “LTC2410 Data Sheet,” Linear Technology Corporation, April 2000.

3. Keithley Instruments, “Low Level Measurements,” Keithley Instruments, 1984.

4. Williams, J., “Testing Linearity of the LTC2400 24-Bit A/D Converter,” Linear Technology Corporation, Design Solution 11, November 1999.

5. Seebeck, T. Dr., “Magnetische Polarisation der Metalle und Erze durch Temperatur-Differenz,” Abhaandlungen der Preussischen Akademic der Wissenschaften (1822–1823), pp. 265–373.

6. Williams, J., “Component and Measurement Advances Ensure 16-Bit DAC Settling Time,” Linear Technology Corporation, Application Note 74, July 1998.

7. Lee, M., “Understanding and Applying Voltage References,” Linear Technology Corporation, Application Note 82, November 1999.

8. Williams, J., “Applications Considerations and Circuits for a New Chopper-Stabilized Op Amp,” Linear Technology Corporation, Application Note 9, August 1987.

9. Huffman, B., “Voltage Reference Circuit Collection,” Linear Technology Corporation, Application Note 42, June 1991.

10. Spreadbury, P. J., “The Ultra-Zener—A Portable Replacement for the Weston Cell?” IEEE Transactions on Instrumentation and Measurement, Vol. 40, No. 2, April 1991, pp. 343–346.

11. Williams, J., “Thermocouple Measurement,” Linear Technology Corporation, Application Note 28, February 1988.

12. Hueckel, J. H., “Input Connection Practices for Differential Amplifiers,” Neff Inst. Corporation, Duarte, California.

13. Gould Inc., “Elimination of Noise in Low Level Circuits,” Gould Inc., Instrument Systems Division, Cleveland, Ohio.

14. Williams, J., “Prevent Low Level Amplifier Problems,” Electronic Design, February 15, 1975, p. 62.

15. Pascoe, G., “The Choice of Solders for High Gain Devices,” New Electronics (Great Britain), February 6, 1977.

16. Pascoe, G., “The Thermo-E.M.F. of Tin-Lead Alloys,” Journal Phys. E, December 1976.

17. Brokaw, A. P., “Designing Sensitive Circuits? Don’t Take Grounds for Granted,” EDN, October 5, 1975, p. 44.

18. Morrison, R., “Noise and Other Interfering Signals,” John Wiley and Sons, 1992.

19. Morrison, R., “Grounding and Shielding Techniques in Instrumentation,” Wiley-Interscience, 1986.

20. Vishay Inc., “Vishay Foil Resistors,” Vishay Inc., 1999.

作者

Florin Oprescu

Josh Guerrero

Josh Guerrero于1999年11月加入凌力尔特公司，担任混合信号部门的助理设计工程师。多年来，他有幸与许多知名的模拟权威专家一起工作。目前在ADI公司担任精密转换器技术部的评估工程师，工作地点在圣克拉拉。

Patrick Copley

James Brubaker

Jim Williams

James M. Williams（1948年4月14日-2011年6月12日）是一名模拟电路设计人员兼技术文章作者，先后就职于麻省理工学院(1968–1979)、Philbrick、National Semiconductor (1979–1982)和凌力尔特公司(LTC) (1982–2011)。[1]他撰写了350多篇有关模拟电路设计的论文[2]，包括5本书、21篇National Semiconductor应用笔记、62篇凌力尔特应用笔记以及超过125篇《EDN》杂志文章。 Williams于2011年6月10日中风，6月12日去世