| A good sensor obeys the following rules: | | | | independent of the measured property, this is |
| | | | defined as drift. |
| 1. the sensor should be sensitive to the | | | | |
| measured property | | | | Long term drift usually indicates a slow |
| | | | degradation of sensor properties over a long |
| 2. the sensor should be insensitive to any | | | | period of time. |
| other property | | | | |
| | | | Noise is a random deviation of the signal |
| 3. the sensor should not influence the | | | | that varies in time. |
| measured property | | | | |
| | | | Hysteresis is an error caused by the fact |
| In the ideal situation, the output signal of | | | | that the sensor not instantly follows the |
| a sensor is exactly proportional to the value | | | | change of the property being measured, and |
| of the measured property. The gain is then | | | | therefore involves the history of the |
| defined as the ratio between output signal | | | | measured property. |
| and measured property. For example, if a | | | | |
| sensor measures temperature and has a voltage | | | | If the sensor has a digital output, the |
| output, the gain is a constant with the unit | | | | signal is discrete and is essentially an |
| [V/K]. | | | | approximation of the measured property. The |
| | | | approximation error is also called |
| If the sensor is not ideal, several types of | | | | digitization error. |
| deviations can be observed: | | | | |
| | | | If the signal is monitored digitally, |
| The gain may in practice differ from the | | | | limitation of the sampling frequency also |
| value specified. This is called a gain error. | | | | causes a dynamic error. |
| | | | |
| Since the range of the output signal is | | | | The sensor may to some extent be sensitive |
| always limited, the output signal will | | | | for other properties than the property being |
| eventually clip when the measured property | | | | measured. For example, most sensors are |
| exceeds the limits. The full scale range | | | | influenced by the temperature of their |
| defines the outmost values of the measured | | | | environment. |
| property where the sensor errors are within | | | | |
| the specified range. | | | | All these deviations can be classified as |
| | | | systematic errors or random errors. |
| If the output signal is not zero when the | | | | Systematic errors can sometimes be |
| measured property is zero, the sensor has an | | | | compensated for by means of some kind of |
| offset or bias. This is defined as the output | | | | calibration strategy. Noise is a random error |
| of the sensor at zero input. | | | | that can be reduced by signal processing, |
| | | | such as filtering, usually at the expense of |
| If the gain is not constant, this is called | | | | the dynamic behaviour of the sensor. |
| nonlinearity. Usually this is defined by the | | | | |
| amount the output differs from ideal | | | | Resolution |
| behaviour over the full range of the sensor, | | | | |
| often noted as a percentage of the full | | | | The resolution of a sensor is the smallest |
| range. | | | | change it can detect in the quantity that it |
| | | | is measuring. Often in a digital display, the |
| If the deviation is caused by a rapid change | | | | least significant digit will fluctuate, |
| of the measured property over time, there is | | | | indicating that changes of that magnitude are |
| a dynamic error. Often, this behaviour is | | | | only just resolved. The resolution is related |
| described with a bode plot showing gain error | | | | to the precision with which the measurement |
| and phase shift as function of the frequency | | | | is made. For example, a scanning probe (a |
| of a periodic input signal. | | | | fine tip near a surface collects an electron |
| | | | tunnelling current) can resolve atoms and |
| If the output signal slowly changes | | | | molecules. |