Capgo datalogger, data logger, datalogging and data logging.

  

Noise

 
 

Understanding Noise in Measurement

We will define noise as something that masks wanted signals and does not have a strong correlation to the signal. Generally the aim is to maximize the signal to noise ratio by minimizing the noise, maximizing the signal or a combination of both.

Noise can be added to a signal at various stages - the measures parameter can be noisy, the sensor can add noise in the sensing process, noise can be "picked up" in the cabling, there can be noise in the input stages of the measuring device, if the signal is being digitized there can be various sampling noises added. Once in the digital domain, a signal is more secure, but still such things as rounding errors in calculations and drop outs in communications can introduce more noise!

With all these potential noise sources, it is surprising it is possible to measure anything! In reality life is not that hard.

The topic of noise is a big one, but in practice you may never need to deal with the problem because the equipment you are using is effective in removing the noise or there is no significant noise present in you measurement. The two major noise sources are 50/60 Hz pickup and radio frequency interference - understanding these may be all you need to know. Check these out without looking at other noise issues if you like.

We shall start with the sources of noise and how to minimize them, then move onto Noise Filtering.

Internal Noise Sources

Noise sources have been divided into internal (to the measuring device) and external (but couple in some way). This distinction is a little arbitrary, but it does help in understanding the impact and preventative measures that may be effective.

Parameter Noise

The parameter being measured may be noisy - for example using a float system to measure river level can return a noisy signal due to waves. The float signal may well be accurate - too accurate! Think where a sensor is placed. If wave activity is not of interest, place the sensor where there is less wave activity or better still construct a mechanical shield to filter out the wave action.

Many parameter have the potential for being noisy due to gradients, nonhomogeneous mediums and other influences.

Solutions:

  • Understand the parameter being measured
  • Locate sensor carefully with regard to parameter variability
  • Use parameter filtering e.g. thermal mass, dampening etc.

Thermal Noise

The random movement of electrons in a material induces small temperature dependent currents in a conductor. The resulting voltage is dependent on the resistance value according to Johnson's noise equation:

Erms = Sqrt (4 .k.T.R.F )

where:

Erms = rms noise voltage (V)

Sqrt() = the square root function

k = Boltzmann's constant = 1.38x10-23 ( J K-1)

T = the absolute temperature (K)

R = the resistance of the circuit element (Ohms)

F = frequency bandwidth being considered = fmax- fmin (Hz)

 

In most systems, the minimum frequency fmin would be close to zero (DC) so the bandwidth is the maximum frequency fmax. At room temperature the peak to peak voltage (which is about five times the rms voltage) is given by the following equation

Ep-p = 6.5 x 10-10 Sqrt ( R .F )    volts

The magnitude of thermal noise is generally in the µV range and can often be ignored.

Solutions:

  • The above formula shows how to minimize the thermal noise by
  • Keep circuit series resistance as low as necessary (till thermal noise is about half other sources if possible)
  • Minimize the measuring device's bandwidth by input filtering and / or software filtering

Input Noise

The input circuit of a measuring device can introduce significant noise. This noise is has a number of components: thermal noise, excess current noise, flicker noise and shot noise. The latter two are characteristics of semiconductors with the flicker noise generally the most significant.

Solutions:

  • Use well specified measuring device
  • Minimize bandwidth by post input filtering - in hardware or software
  • Where the sampling speed is high compared to the low frequency noise, the noise can sometimes be treated as a rapid zero drift. Take zero readings between measurements and subtract the zero drift.

Thermal emf

Temperature gradients in conductors of different materials can induce small thermoelectric voltages by the Peltier effect. an often neglected source of this error is resistors - some of which have very high Peltier voltages (150µV/°C).

Solutions:

  • Keep same metals in positive and negative leads
  • Keep leads together so they are exposed to same temperatures
  • Minimize temperature gradients within measuring device

Dielectric Absorption

No insulator, other than a vacuum, is perfect. When measuring very small currents (say < pico Amp), the charge storage effects of the insulators in the system can become significant. This effect only occurs with high impedance (>100 Mohms) sensors such as pH electrodes and then only when the sensor output changes relatively quickly. Strictly, dialectic absorption is more of a signal distortion than superimposed noise, however the effect can amplify the impact of other noise sources.

Sensors that exhibit sensitivity to dielectric absorption noise are also likely to be susceptible to cable noise.

Solutions:

  • Use low absorption insulators such as polypropylene, polystyrene, Teflon.
  • Minimize magnitude of voltage changes to less than a few volts.

Audiophonic Noise

Physical vibration of the measuring system can induce measurable noise by several different mechanisms. The most common of these is the piezoelectric voltage generated by stress on ceramic capacitors constructed from materials with high dielectric constants. The problem arises when these capacitors are use in the signal path of sensitive amplifiers.

A second source of audiophonic noise is the strain gauge effect with resistors, especially surface mount types. Server stress may lead to ±0.5% or more variation with stress.

Solutions:

  • Avoid high value ceramic capacitors in input circuits. NPO types are OK.
  • Avoid component stress by restricting and damping vibration of circuit boards.
  • Place susceptible components such that the net effect cancels out. This is most easily done with resistors in bridge or divider networks.

External Noise Sources

External noise is usually introduced into a measuring device via sensor or communications wiring. Some times high frequency or high power radio frequency signals can interfere directly with the circuitry of a measuring device.

Electric Field Coupling

Varying electric fields such as from a power cable can be coupled into a measuring circuit by the "stray" capacitance between the two circuits. This capacitance is usually very small - 0.1 to 20 pF, but it is sufficient to introduce voltage noise into high impedance circuits.

Solutions:

  • Use a shielded cable connected to Ground at one end, preferably at the measuring device end.
  • Increase the distance between the electric field source and the measuring circuit.
  • Lower the measuring circuit impedance as much accuracy considerations allow.

Inductive Coupling (also see section on Magnetic Pickup)

Varying magnetic fields will induce a varying current into a measuring circuit. These are far more troublesome than electrically coupled voltages because the current tends to flow regardless of the circuit impedance and magnetic shielding is very expensive.

The magnitude of the induced current is proportional to the magnetic field strength, the rate at which it changes and and the area of pickup loop in the measuring circuit.

Solutions:

  • Minimize measurement circuit area.
  • Use twisted pair cables.
  • Use mu-metal shielding if affordable.
  • Place measurement circuit away from magnetic fields.
  • Avoid moving or vibrating magnetic materials near measurement circuit.
  • Tied down cables to prevent vibration.

Radio Frequency Coupling

Radio frequency coupling can be a difficult noise source to identify. The measurement circuit acts as an antenna to receive radio energy from radio transmitters (including mobile phones and radars). Radio energy is really a combination of electric and magnetic fields discussed above with the added complication of having short wavelengths. The inputs of many solid state circuits act like rectifiers to radio frequencies. The result is often an fluctuating DC input voltage.

Solutions:

  • If possible locate system away from sources of radio frequency energy such as radio transmitters, microwave systems, radar and inductive heaters.
  • Use shielded cable, grounding the shield to the case of the measuring device and capacitively coupling it via a 100nF capacitor to the case of the sensor. The reason for capacitively coupling at one end is to avoid ground loops.
  • Place ferrite beads over the signal wires including signal grounds, at both ends of the sensor cable.
  • Employ low pass RC or LC filters in the measuring device.

Ground Loops

Ground Loops are sufficiently common a problem that they are dealt in detail on a separate page. Most commonly, ground loops induce 50/60 Hz and related harmonics into a system. These can sometimes be filtered out. However ground loops can also induce unstable DC errors.

Solutions:

  • Avoid ground loops.
  • Use measuring devices with electrically isolated inputs
  • Use a distributed measurement system.

Common Mode Rejection Noise

Common Mode Rejection is discussed elsewhere. What is not discussed there is the noise impact that common mode voltages can generate when the common mode rejection is not adequate. Put simply, common mode voltages can cause the equivalent to a zero shift in the input circuit of a measuring device. The resulting noise added to the signal will tend to reflect the form of the common mode voltage.

Solutions:

  • Minimize common mode voltage, particularly AC voltages.
  • Select a measuring device with high CMRR
  • If the common mode noise is primarily AC, it may be possible to apply software filtering.

Cable Noise

An unexpected source of noise can be the cable between the sensor and the measuring device. The cable can exhibit piezoelectric charge generation when mechanically moved or stressed. The result is a most noticeable with high impedance sensors.

A similar but less well known cable induce noise is the triboelectric effect. This occurs when friction between the cable's insulator and conductor generate a surface charge. Again, the result is a most noticeable with high impedance sensors.

Both these noises can arise where sensors are place on a moving structure (e.g. force sensors on a robotic arm) or when cables are able to move for other reasons such as machine vibration.

  Electrical leakage currents in the cable can also inject noise currents into a measurement circuit. This can be particularly troublesome when the voltage between the cable conductors exceeds a volt and the circuit impedance is greater than say 100K Ohms.

Solutions

  • Avoid high impedance sensors or place signal conditioning circuit in sensor enclosure. This may only need to be a unity gain buffer amplifier to lower source resistance.
  • Minimize the length of cable between sensor and measuring device.
  • Minimize the cable movement opportunities - e.g. wind and vibration.
  • Use cable with quality insulating material (Teflon is excellent).
  • Select cable with lubricant between the various layers.
  • Ensure cable is internally dry. Water can "wick" many meters between the layers of a cable. A penetrating oil applied at each end can minimize this effect.

Noise Filtering

   See the Noise Filtering page for hints on improving the signal to noise ratio. Other relevant sections include Shielding, Magnetic Pickup and Ground Loops.