Sensitivity Training

Andy Armutat
Keithley Instruments Inc.
Solon, Ohio

It's not uncommon for modern electrical measurements to involve extremely small quantities. Test methods that produce good results in such cases can't be done haphazardly. Designers accustomed to gathering quick readings with a multimeter must now think about techniques for highly sensitive measurements. They also must understand the limitations of what they can realistically expect for repeatability, resolution, and stability.

What is sensitive?

Generally, signal levels of less than 1 µV require sensitive measurement techniques. Attempting to measure 123 µV to five digits of resolution (123.45 µV) is certainly a sensitive measurement, because the least significant digit represents 10 nV of resolution.

A typical voltage source behaves as if it includes a series resistance, RSource, through which all current from the source flows. This is called source resistance or output resistance. Often, R_Source is less than 10k½ and can be ignored. For example, the source resistance of a thermocouple is usually less than 10 ½ and most sensors have output resistances less than 1k½. But there's a need for special care to avoid excessive loading of the source when a voltage is measured through a high source resistance.

In general, accurate low-level measurements require instruments that absorb as little energy as possible from the test circuit. For example, a sensitive voltage measurement might need R_In to be at least 10,000 times R_Source. In addition, every connector, switch, and relay contact in series with the voltage source has the potential to produce significant and unstable error voltage VError in the measurement circuit. Good system design and instrument selection will minimize such errors.

The right instrument

Using an instrument optimized for the type of measurement at hand is a good start. High-quality instruments use special components, conductors, insulation, and assembly techniques that minimize errors from noise, thermoelectric EMFs, and electrochemical effects.
Generally, there is a trade-off between measurement speed and resolution/noise suppression. And this trade-off becomes more pronounced at lower signal levels. Older generations of electrometers and related instruments were often limited to three to four readings per second. Today, sophisticated nanovoltmeters and electrometers can make hundreds of readings per second with even better stability and resolution. These instruments usually have features that help minimize the most common measurement problems associated with low-level signals.

One factor that low-level measurements must allow for is thermoelectric voltage. This is generated when dissimilar metals are joined together. This is the same principle as in thermocouple temperature sensors. These offset voltages can change dramatically with temperature, relative to the measured signal. The magnitude of a thermoelectric voltage depends on the type of metals at the junction. The thing to keep in mind is that every connector, switch, or relay in a test circuit is likely to create a thermoelectric voltage offset. These become troublesome in low-voltage measurements where they often show up as unstable readings. The signal path in a large test system with multiple connectors, cables, and relays can easily have an offset that totals several millivolts or higher.

Replacing the test voltage with a short circuit and recording the system offset voltage can correct for thermoelectric offsets. Subtracting this voltage from subsequent measurements will cancel out offset error for anywhere from seconds to hours, depending on the system's thermal stability and measurement resolution. Strategies for minimizing thermoelectric voltages include use of low thermal connectors and relays, differential signal switching, clustering instrument signal Hi and Lo connections together, keeping contacts oxidation free, and minimizing temperature gradients.

Dealing with electromagnetic noise

The most common electromagnetic noise radiated from circuits is the 50 or 60-Hz signal coming from ac power mains. However, higher frequency or random frequency noise can come from motors, welders, two-way radios, and other industrial sources.

Distance is one of the simplest but most effective tools for combating electromagnetic noise. Noise diminishes as the square of distance to the source, so simply relocating the measurement circuit can reduce noise pickup significantly. Another simple technique is to keep signal leads as short as possible, routed in straight lines and spread over as small an area as possible. This minimizes their antenna effect.

In general, higher frequency signals are more difficult to attenuate than lower ones.

Where moving is not possible, shielding can reduce noise pickup as well as signal radiation from conductors. However, all shielding is not created equal. Typically, shielded cables contain wire braid or a combination of wire braid and foil. Multilayer or multibraid shields are more effective in attenuating signal pickup or radiation than single-layer shields. But multilayer cables are stiffer and more difficult to position. To reject noise effectively, shields must be connected to quiet ground on the noise source. Typically, this is an earth ground.

Noise and error voltages can also result from ground loops. These form when both the signal source and the measuring instruments are connected to a common ground bus, but at different locations. A common scenario is when instruments and test fixtures are plugged into ac power strips on different test racks. Typically, the Lo input of the instrument connects to ground on the test rack or bench. The test fixture is grounded several feet away. There is typically a slight resistance between the chassis and earth ground at the two different power line outlets. The result is an electrical potential between them. The voltage difference causes current to flow around the ground loop, creating an offset voltage in series with the source signal.

Even when different grounding points are close together, ground loops may still be a problem. The solution is a single grounding point for all test equipment, connected to a solid earth ground. The preferred technique is to use isolated power supplies and measuring instruments. It can also be helpful to connect cable shields only at the measuring instrument end of the cable. But if an instrument is particularly sensitive, avoid connecting it to the same ground system used by electrically noisy machinery, high-power equipment, signal generators, and other such instruments.

Sensitive instruments often use integrating analog/digital conversion to improve noise rejection. An integrating a/d converter employs the charging and discharging of a capacitor for signal conversion. The capacitor is charged first by the unknown signal for a set interval, the integration period. Next, it is discharged back to zero at a fixed rate. The discharge time is a measure of the integrated input voltage and is used to deduce the unknown voltage.

Selecting the integration time to match the frequency of a known noise source makes the integrating a/d converter particularly effective in rejecting noise at that frequency. The greatest noise rejection happens when the integration period is an integral multiple of the frequency. This is sometimes referred to as line cycle integration. The abbreviation NPLC (number of power line cycles) often appears in the specifications of high quality instruments as a configurable measurement parameter. Devoting more line cycles to the integration period better rejects noise at the expense of measurement speed.

Signal averaging also aids in minimizing random noise and improving measurement integrity. The technique calculates an average based on up to 20 readings. If noise is truly random, the sum of all noise components will total zero and cancel each other. Averaging is relatively easy to perform in computer-controlled measurement systems.

A simplified low-voltage measurement circuit shows how connectors, switches, or relay contacts each add a small voltage to the voltmeter reading.