Such experiments illustrate that if you put a certain amount of charge into a less-than-ideal capacitor, you will get out a different amount of charge, depending on how long you wait. Thus, using low-soakage capacitors proves important in applications such as those involving high-resolution dual-slope integrating ADCs. And sure enough, many top-of-the-line digital voltmeters do use polypropylene (a low-soakage dielectric) devices for their main integrating capacitors.
But dielectric-absorption characteristics are most obviously detrimental in applications involving sample/holds. Manufacturers guarantee how fast these devices can charge a capacitor in their Sample mode and how much their circuits' leakage causes capacitor-voltage droop during the Hold mode, but they don't give any warning about how much the capacitor voltage changes because of soakage. This factor is especially important in a data-acquisition system, where some channels might handle small voltages while others operate near full scale. Even with a good dielectric, a sample/hold can hurt your accuracy, especially if the sample time is a small fraction of THOLD. For example, although a good polypropylene device can have only 1-mV hysteresis per 10V step if T/t=100 msec/10 msec, this figure increases to 6 mV if the T/t ratio equals 100 msec/0.5 msec. Because most sample/hold data sheets don't warn you of such factors, you should evaluate capacitors in a circuit such as Fig 5's, using time scaling suited to your application.
Fig 6 - Soakage can present problems when you're designing a fast-settling amplifier or filter. In the circuit shown here, for example, C1 can be a Mylar or tantalum unit, but making C2 a polypropylene device improves performance.
Other applications in which soakage can degrade performance are those involving fast-settling ac active filters or ac-coupled amplifiers. In Fig 6's circuit, C1 can be a Mylar or tantalum unit because it always has 0V dc on it, but making C2 polypropylene instead of Mylar noticeably improves settling. For example, settling to within -0.2 mV for a 10V step improves from 10 to 1.6 sec with the elimination of Mylar's dielectric absorption. Similarly, voltage-to-frequency converters benefit from low-soakage timing capacitors, which improve V/F linearity.
Some dielectrics are excellent at all speeds
Fortunately, good capacitors such as those employing polystyrene, polypropylene, NPO ceramic and Teflon dielectrics perform well at all speeds. Fig 7 shows the characteristics of capacitors using these dielectrics and others such as silver mica and Mylar. In general, polystyrene, polypropylene or NPO-ceramic capacitors furnish good performance, although polystyrene can't be used at temperatures greater than 80?C. And although NPO ceramic capacitors are expensive and hard to find in values much larger than 0.01 ?F, they do achieve a low temperature coefficient (a spec not usually significant for a S/H but one that might prove advantageous for precision integrators or voltage-to-frequency converters). Teflon is rather expensive but definitely the best material to use when high performance is important. Furthermore, only Teflon and NPO ceramic capacitors suit use at 125?C.
Fig 7 - Soakage-measurement results for a variety of capacitors illustrate the effects of tDISCHARGE values on dielectric-absorption-caused errors. Note the curves for two different samples of NP0 ceramic capacitors intersect.
If you look at Fig 7's dielectric-absorption values, you can see wide differences in performance for a given dielectric material. For example, polypropylene sample A is about as good as B at t=6 see, but B is four times better at high speeds. Similarly, NP0-ceramic sample A is slightly worse than NP0-ceramic sample B at low speeds, but A is definitely better at high speeds. And some Mylar capacitors (sample A) get better as speed increases from 1000 to 100 ?sec, but others (sample B) get worse. So if you want consistently good performance from your capacitors, evaluate and specify them for the speed at which they'll be used in your application. Keep in mind that because most sample/holds are used at much faster speeds than those corresponding to the 1- or 5-min ratings usually given in data sheets, a published specification for dielectric absorption has limited value.
In addition, other dielectrics furnish various levels of performance:
- Because any long word that starts with poly seems to have good dielectric properties, how about polycarbonate or polysulfone? No -- they are about as bad as Mylar.
- Does an air or vacuum capacitor have low soakage? Well, it might, but many standard capacitors of this type are old designs with ceramic spacers, and they might give poor results because of the ceramic's hysteresis.
- If a ceramic capacitor is not an NP0 device, is it any good? Most of the conventional high-K ceramics are just terrible -- 20 to 1000 times worse than NP0 and even worse than tantalum.
- Is silicon dioxide suitable for small capacitances? Although Fig 5's test setup, used in preparing Fig 7's chart, only measures moderate capacitances (500 to 200,000 pF), silicon dioxide appears suitable for the small capacitors needed for fast S/Hs or deglitchers.
Cancellation circuit improves accuracy
A practical method of getting good performance with less-than-perfect capacitors is to use a soakage-cancellation circuit such as one of the form shown in Fig 8, in which a capacitor of the type modeled in Fig 4 serves as an integrator. (Only the first two soakage elements are shown.) The integrator's output is inverted with a scale factor of -0.1, and this voltage is then fed through one or more experimentally chosen RC networks to cancel the equivalent network inherent in the capacitor's dielectric material.
Fig 8 - You can compensate an integrator for dielectric absorption by feeding its inverted output back to the input through one or more experimentally chosen RC networks, which cancel the equivalent network inherent in the capacitor's dielectric material.
Fig 9 shows a practical sample/hold circuit with an easily trimmed compensator. This network provides about a 10-fold improvement for sample times in the 50- to 2000-?sec range (Fig 10). Although this compensation is subject to limitations at very fast or slow speeds, the number of RC sections and trimming pots employed can be extended.
Fig 9 - Adding compensation circuitry to a sample/hold yields better-than-Teflon performance with a polypropylene capacitor. Using Teflon capacitors in such circuits can yield a 15- to 17-bit dynamic range.
Simple circuits similar to Fig 9's or Fig 8's have been used in production to let inexpensive polypropylene capacitors provide better-than-Teflon performance. In turn, using these compensator circuits with a good Teflon capacitor furnishes a dynamic range of 15 to 17 bits.
Fig 10 - Adding Fig 9's compensation network to a sample/hold circuit yields a 10-fold performance improvement for sample times of 50 to 2000 ?sec; additional RC networks and trimming pots can extend the time range. The short pulses represent normal S/H jumps and occur during the sample time. The exponentially rising waveform during the hold time results from soakage. Note that soakage effects are still visible during the second hold period.