An integrating ADC is a type of analog-to-digital converter that converts an unknown input voltage into a digital representation through the use of an integrator. In its basic implementation, the dual-slope converter, the unknown input voltage is applied to the input of the integrator and allowed to ramp for a fixed time period (the run-up period). Then a known reference voltage of opposite polarity is applied to the integrator and is allowed to ramp until the integrator output returns to zero (the run-down period). The input voltage is computed as a function of the reference voltage, the constant run-up time period, and the measured run-down time period. The run-down time measurement is usually made in units of the converter's clock, so longer integration times allow for higher resolutions. Likewise, the speed of the converter can be improved by sacrificing resolution.
Converters of this type can achieve high resolution, but often do so at the expense of speed. For this reason, these converters are not found in audio or signal processing applications. Their use is typically limited to digital voltmeters and other instruments requiring highly accurate measurements.
The basic integrating ADC circuit consists of an integrator, a switch to select between the voltage to be measured and the reference voltage, a timer that determines how long to integrate the unknown and measures how long the reference integration took, a comparator to detect zero crossing, and a controller. Depending on the implementation, a switch may also be present in parallel with the integrator capacitor to allow the integrator to be reset. Inputs to the controller include a clock (used to measure time) and the output of a comparator used to detect when the integrator's output reaches zero.
The conversion takes place in two phases: the run-up phase, where the input to the integrator is the voltage to be measured, and the run-down phase, where the input to the integrator is a known reference voltage. During the run-up phase, the switch selects the measured voltage as the input to the integrator. The integrator is allowed to ramp for a fixed period of time to allow a charge to build on the integrator capacitor. During the run-down phase, the switch selects the reference voltage as the input to the integrator. The time that it takes for the integrator's output to return to zero is measured during this phase.
In order for the reference voltage to ramp the integrator voltage down, the reference voltage needs to have a polarity opposite to that of the input voltage. In most cases, for positive input voltages, this means that the reference voltage will be negative. To handle both positive and negative input voltages, a positive and negative reference voltage is required. The selection of which reference to use during the run-down phase would be based on the polarity of the integrator output at the end of the run-up phase.
The basic equation for the output of the integrator (assuming a constant input) is:
Vout=-
| Vin | |
| RC |
tint+Vinitial
Assuming that the initial integrator voltage at the start of each conversion is zero and that the integrator voltage at the end of the run down period will be zero, we have the following two equations that cover the integrator's output during the two phases of the conversion:
Vout-up=-
| Vin | |
| RC |
tu
Vout-down=-
| Vref | |
| RC |
td
Vout-up+Vout-down=0
The two equations can be combined and solved for
Vin
Vin=
| -V | ||||
|
From the equation, one of the benefits of the dual-slope integrating ADC becomes apparent: the measurement is independent of the values of the circuit elements (R and C). This does not mean, however, that the values of R and C are unimportant in the design of a dual-slope integrating ADC (as will be explained below).
Note that in the graph, the voltage is shown as going up during the run-up phase and down during the run-down phase. In reality, because the integrator uses the op-amp in a negative feedback configuration, applying a positive
Vin
The resolution of the dual-slope integrating ADC is determined primarily by the length of the run-down period and by the time measurement resolution (i.e., the frequency of the controller's clock). The required resolution (in number of bits) dictates the minimum length of the run-down period for a full-scale input (e.g.
Vin=-Vref
td=
| 2r | |
| fclk |
During the measurement of a full-scale input, the slope of the integrator's output will be the same during the run-up and run-down phases. This also implies that the time of the run-up period and run-down period will be equal (
tu=td
2td
tm=2
| 2r | |
| fclk |
Typically the run-up time is chosen to be a multiple of the period of the mains frequency, to suppress mains hum.
There are limits to the maximum resolution of the dual-slope integrating ADC. It is not possible to increase the resolution of the basic dual-slope ADC to arbitrarily high values by using longer measurement times or faster clocks. Resolution is limited by:
The basic design of the dual-slope integrating ADC has a limitations in linearity, conversion speed and resolution. A number of modifications to the basic design have been made to overcome these to some degree.
The run-up phase of the basic dual-slope design integrates the input voltage for a fixed period of time. That is, it allows an unknown amount of charge to build up on the integrator's capacitor. The run-down phase is then used to measure this unknown charge to determine the unknown voltage. For a full-scale input equal to the reference voltage, half of the measurement time is spent in the run-up phase. Reducing the amount of time spent in the run-up phase can reduce the total measurement time. A common implementation uses an input range twice as large as the reference voltage.
A simple way to reduce the run-up time is to increase the rate that charge accumulates on the integrator capacitor by reducing the size of the resistor used on the input. This still allows the same total amount of charge accumulation, but it does so over a smaller period of time. Using the same algorithm for the run-down phase results in the following equation for the calculation of the unknown input voltage (
Vin
Vin=
| -V | ||||
|
| td | |
| tu |
Note that this equation, unlike the equation for the basic dual-slope converter, has a dependence on the values of the integrator resistors. Or, more importantly, it has a dependence on the ratio of the two resistance values. This modification does nothing to improve the resolution of the converter (since it doesn't address either of the resolution limitations noted above).
The purpose of the run-up phase is to add a charge proportional to the input voltage to the integrator to be later measured during the run-down phase. One method to improve the resolution of the converter is to artificially increase the range of the integrating amplifier during the run-up phase. One method to increase the integrator capacity is by periodically adding or subtracting known quantities of charge during the run-up phase in order to keep the integrator's output within the range of the integrator amplifier. The total accumulated charge is the charge introduced by the unknown input voltage plus the sum of the known charges that were added or subtracted.
The circuit diagram shown to the right is an example of how multi-slope run-up could be implemented. During the run-up the unknown input voltage,
Vin
Rp
Rn
Ri
To the right is a graph of sample output from the integrator during such a multi-slope run-up. Each dashed vertical line represents a decision point by the controller where it samples the polarity of the output and chooses to apply either the positive or negative reference voltage to the input. Ideally, the output voltage of the integrator at the end of the run-up period can be represented by the following equation:
Vout=-
| 1 | \left( | |
| C |
| NVint\Delta | |
| Ri |
+
| NpVreft\Delta | |
| Rp |
-
| NnVreft\Delta | |
| Rn |
\right)
where
t\Delta
Np
Nn
N
The resolution obtained during the run-up is given by the number of periods of the run-up algorithm. The multi-slope run-up comes with multiple advantages:
While it is possible to continue the multi-slope run-up indefinitely, it is not possible to increase the resolution of the converter to arbitrarily high levels just by using a longer run-up time. Error is introduced into the multi-slope run-up through the action of the switches controlling the references, cross-coupling between the switches, unintended switch charge injection, mismatches in the references, and timing errors.
Some of this error can be reduced by careful operation of the switches. In particular, during the run-up period, each switch should be activated a constant number of times. The algorithm explained above does not do this and just toggles switches as needed to keep the integrator output within the limits. Activating each switch a constant number of times makes the error related to switching approximately constant. Any output offset that is a result of the switching error can be measured and then numerically subtracted from the result.
The simple, single-slope run-down is slow. Typically, the run down time is measured in clock ticks, so to get four digit resolution, the rundown time may take as long as 10,000 clock cycles. A multi-slope run-down can speed the measurement up without sacrificing accuracy. By using 4 slope rates that are each a power of ten more gradual than the previous, four digit resolution can be achieved in roughly 40 clock ticks—a huge speed improvement.[2]
The circuit shown to the right is an example of a multi-slope run-down circuit with four run-down slopes with each being ten times more gradual than the previous. The switches control which slope is selected. The switch containing
Rd/1000
Vin
Rd/1000
Rd/1000
Rd/100
Rd
The multi-slope run-down is often used in combination with a multi-slope run-up. The multi-slope run-up allows for a relatively small capacitor at the integrator and thus a relatively steep slope to start with and thus the option to actually use much more gradual slopes. It is possible to use a multi-slope rundown also with a simple run-up (like in the dual-slope ADC), but limited by the already relatively small slope for the initial phase and not much room for much smaller slopes.
In the example circuit, the slope resistors differ by a factor of 10. This value, known as the base (
B
The basis of this design is the assumption that there will always be overshoot when trying to find the zero crossing at the end of a run-down interval. This will be true due to the periodic sampling of the comparator based on the converter's clock. If we assume that the converter switches from one slope to the next in a single clock cycle (which may or may not be possible), the maximum amount of overshoot for a given slope would be the largest integrator output change in one clock period:
V\Delta=
| Vref | |
| RC |
| 1 | |
| fclk |
To overcome this overshoot, the next slope would require no more than
B
Tfirst=\left\lceil
| VmaxCRs1fclk | |
| Vref |
\right\rceil
where
Tfirst
Vmax
Rs1
The remainder of the slopes have a limited duration based on the selected base, so the remaining time of the conversion (in converter clock periods) is:
Td\leB(N-1)
where
N
Converting the measured time intervals during the multi-slope run-down into a measured voltage is similar to the charge-balancing method used in the multi-slope run-up enhancement. Each slope adds or subtracts known amounts of charge to/from the integrator capacitor. The run-up will have added some unknown amount of charge to the integrator. Then, during the run-down, the first slope subtracts a large amount of charge, the second slope adds a smaller amount of charge, etc. with each subsequent slope moving a smaller amount in the opposite direction of the previous slope with the goal of reaching closer and closer to zero. Each slope adds or subtracts a quantity of charge
Qslope
Qslope=\pm
| VrefTslope | |
| Rslopefclk |
Tslope
B
Rd/100
Qslope2
| 100Vref | |
| Rdfclk |
\leQslope2\le
| 1000Vref | |
| Rdfclk |
| 100Vref | |
| Rdfclk |
That is,
B
N
M
N=logBM
Substituting this back into the equation representing the run-down time required for the second and subsequent slopes gives us this:
Td\leB(logB(M)-1)
Which, when evaluated, shows that the minimum run-down time can be achieved using a base of e. This base may be difficult to use both in terms of complexity in the calculation of the result and of finding an appropriate resistor network, so a base of 2 or 4 would be more common.
When using run-up enhancements like the multi-slope run-up, where a portion of the converter's resolution is resolved during the run-up phase, it is possible to eliminate the run-down phase altogether by using a second type of analog-to-digital converter. At the end of the run-up phase of a multi-slope run-up conversion, there will still be an unknown amount of charge remaining on the integrator's capacitor. Instead of using a traditional run-down phase to determine this unknown charge, the unknown voltage can be converted directly by a second converter and combined with the result from the run-up phase to determine the unknown input voltage.
Assuming that multi-slope run-up as described above is being used, the unknown input voltage can be related to the multi-slope run-up counters,
Np
Nn
Vout
Vin=
| 1 | |
| Nt\Delta |
Ri\left(-\dfrac{NpVreft\Delta}{Rp
This equation represents the theoretical calculation of the input voltage assuming ideal components. Since the equation depends on nearly all of the circuit's parameters, any variances in reference currents, the integrator capacitor, or other values will introduce errors in the result. A calibration factor is typically included in the
CVout
Instead of being used to eliminate the run-down phase completely, the residue ADC can also be used to make the run-down phase more accurate than would otherwise be possible. With a traditional run-down phase, the run-down time measurement period ends with the integrator output crossing through zero volts. There is a certain amount of error involved in detecting the zero crossing using a comparator (one of the short-comings of the basic dual-slope design as explained above). By using the residue ADC to rapidly sample the integrator output (synchronized with the converter controller's clock, for example), a voltage reading can be taken both immediately before and immediately after the zero crossing (as measured with a comparator). As the slope of the integrator voltage is constant during the run-down phase, the two voltage measurements can be used as inputs to an interpolation function that more accurately determines the time of the zero-crossing (i.e., with a much higher resolution than the controller's clock alone would allow).
By combining some of these enhancements to the basic dual-slope design (namely multi-slope run-up and the residue ADC), it is possible to construct an integrating analog-to-digital converter that is capable of operating continuously without the need for a run-down interval. Conceptually, the multi-slope run-up algorithm is allowed to operate continuously. To start a conversion, two things happen simultaneously: the residue ADC is used to measure the approximate charge currently on the integrator capacitor and the counters monitoring the multi-slope run-up are reset. At the end of a conversion period, another residue ADC reading is taken and the values of the multi-slope run-up counters are noted.
The unknown input is calculated using a similar equation as used for the residue ADC, except that two output voltages are included (
Vout1
Vout2
Vin=
| 1 | |
| Nt\Delta |
Ri\left(-
| NpVreft\Delta | |
| Rp |
+
| NnVreft\Delta | |
| Rn |
-C\left(Vout2-Vout1\right)\right)
Such a continuously-integrating converter is very similar to a delta-sigma analog-to-digital converter.
In most variants of the dual-slope integrating converter, the converter's performance is dependent on one or more of the circuit parameters. In the case of the basic design, the output of the converter is in terms of the reference voltage. In more advanced designs, there are also dependencies on one or more resistors used in the circuit or on the integrator capacitor being used. In all cases, even using expensive precision components there may be other effects that are not accounted for in the general dual-slope equations (dielectric effect on the capacitor or frequency or temperature dependencies on any of the components). Any of these variations result in error in the output of the converter. In the best case, this is simply gain and/or offset error. In the worst case, nonlinearity or nonmonotonicity could result.
Some calibration can be performed internal to the converter (i.e., not requiring any special external input). This type of calibration would be performed every time the converter is turned on, periodically while the converter is running, or only when a special calibration mode is entered. Another type of calibration requires external inputs of known quantities (e.g., voltage standards or precision resistance references) and would typically be performed infrequently (every year for equipment used in normal conditions, more often when being used in metrology applications).
Of these types of error, offset error is the simplest to correct (assuming that there is a constant offset over the entire range of the converter). This is often done internal to the converter itself by periodically taking measurements of the ground potential. Ideally, measuring the ground should always result in a zero output. Any non-zero output indicates the offset error in the converter. That is, if the measurement of ground resulted in an output of 0.001 volts, one can assume that all measurements will be offset by the same amount and can subtract 0.001 from all subsequent results.
Gain error can similarly be measured and corrected internally (again assuming that there is a constant gain error over the entire output range). The voltage reference (or some voltage derived directly from the reference) can be used as the input to the converter. If the assumption is made that the voltage reference is accurate (to within the tolerances of the converter) or that the voltage reference has been externally calibrated against a voltage standard, any error in the measurement would be a gain error in the converter. If, for example, the measurement of a converter's 5 volt reference resulted in an output of 5.3 volts (after accounting for any offset error), a gain multiplier of 0.94 (5 / 5.3) can be applied to any subsequent measurement results.