Dealing with noise
In a previous post I explained the sources of (electrical) noise in test systems. Noise introduces error into measurements and will cause control systems to malfunction by responding to erroneous inputs. This post covers noise suppression strategies.
Have you ever stepped into a crowded restaurant and been hit with the cacophony of a mass of humanity all speaking at once? Like in the video below.
If you are far enough away a from a large crowd all you hear is an unintelligible din. But you are able to enter the crowd and carry on a conversation with individuals, while filtering out all of the garbled background noise. Why? Well, you’d have to ask a neuroscientist for a precise answer, but essentially the brain is able to “lock on” to a specific voice while filtering out all of the background noise.
The environment a test system operates in is no different. The world is awash in electronic noise. The test engineer is required to design a test system that will “lock on” to the signal of interest while filtering out all of the useless background noise. This is accomplished through a choice of transmission signal, cabling and wiring techniques, and using specialized circuit components such as signal conditioning modules (SCM).
Transmission Signal
As mentioned in the post on test engineering fundamentals there are two types of signals: digital and analog. When selecting the devices or transducers to use in a test system design, the designer usually has a choice in the type of output signal to use. These choices are:
- Voltage output: milliVolt output, 0–5V or 0–10V.
Pros:
— Easy to read/interpret.
— Easy to wire to voltage inputs on DAQ.
— Easy to debug using handheld multi-meter.Cons:
— Voltage drop over long lengths of wire.
— Susceptible to EMI: Voltage outputs are connected to high impedance inputs at the DAQ/DMM end. This means that any small noise current induced in the circuit will be amplified (V = I*R).
- Current output (mA output): 4–20 mA current loop. Detailed explanation in this excellent white paper by Acromag. Summary diagram below. A current loop transducer requires a power supply, and the transducer acts as a variable resistor and regulates current in the loop such that the current is proportional to the quantity being measured. However, DAQs usually require voltage inputs, and are not designed to sink currents.
- The current in the current loop is dropped through a 250Ohm precision resistor and the voltage across the resistor is wired to the voltage input on a DAQ.
Pros:
— High noise immunity. Immunity does not mean that noise currents and voltages aren't induced in the wire, but that the effects of these induced currents are negligible since the current loop is a low impedance system.
— Lossless transmission.Cons:
— Wiring is more complicated than voltage output mode.
— Each transducer requires a separate current loop. An independent pair of wires is required for each device, multiple devices cannot use the same pair of wires, since the current has to be dropped over a 250ohm resistor before being wired to the DAQ. This is not solely a drawback of current loop transducers; it also applies to voltage output models. The alternative is to use digital signaling like serial communication or HART protocol.
- Digital output: RS-232, RS-485, HART, Counter output (rotary encoders). RS-232/485, HART and others are digital communication protocols. In analog devices the output signal is a variable quantity (0–5V, 4–20mA) that has to be read by an instrument and then scaled to convert it to engineering units (pressure, temperature).
- Digital protocols send information encoded as binary signals. Decoding the signal yields the measured value, no scaling/conversion is required.
- Counter output devices such as encoders output square voltage pulses whose frequency is proportional to the rotation of the input shaft. This is not a digital communication protocol; the pulses have to be counted using a digital counter and then converted to RPM.
- Digital signals are not affected by additive noise because it generally does not affect the information encoded within the signal. In the image shown below the ideal signal shown in blue, and the overlaid noisy signal in yellow both encode the same information, because both signals are high and low at the same points for the same duration. It doesn’t matter that the noisy yellow signal has high frequency noise since it doesn’t impact converting the signal into bits (1’s and 0’s).
Pros:
— Signal is transmitted digitally and is highly immune to noise. This is because in digital communication the signal waveform does not have to be precise, and low levels of additive noise does not affect the interpretation of the digital waveform.
— Signal conditioning and DAQ is not required at the receiver side since the device is sending measurement values directly.Cons:
— Sensors without analog outputs are only good for "point-read" applications and are unsuitable for continuous waveform acquisition applications.
Current loop is better than Voltage output for accuracy, and noise suppression. Digital output transducers reduce system cost and complexity and have excellent noise suppression characteristics but are not suitable for continuous data acquisition applications.s
Select the transducer with an output type that is best suited for the application. Current loop transducers are generally the most accurate option.
Cabling
Twisted pair: Large AC currents generate strong variable magnetic fields which can then induce voltages in low voltage DC signal wires. Analog signals from instruments are usually transmitted over a pair of wires: power/ground pair for 2-wire 4–20mA or +ve Output/ -ve Output for 4-wire differential voltage output.
By twisting the pairs of wires together the noise voltage induced on the pair of wires will be equal and can be rejected by the DAQ through common-mode voltage rejection. What is common mode voltage? It is voltage that is common to both terminals of an electrical device (such as a DAQ). If there are two wires and one is at +5V and the other is 0V, the voltage difference between the two wires is +5V.
But why twist the wires? What if the wires were simply laid side by side, wouldn’t common mode rejection should still work? It would work, but there would be a small difference in the induced voltage on the two conductors because one conductor would invariably be closer to the noise source than the other and thus have a higher induced voltage. By twisting the two conductors each conductor is at the same average distance from the noise source.
Twisted wires also reduce the magnitude of the noise current induced by inductive coupling. This is because the direction of induced current is opposite in each half loop and cancels out.
But why is the induced current in the red and green wires opposite? This is entirely due to the position of each wire in the magnetic field. The direction of current induced in a wire is perpendicular to the magnetic field causing the current (Lenz’s law). This is taught as the “right hand rule”.
To use the right-hand grip rule, align your thumb in the direction of the induced magnetic field (pointing down in Figure 5) and curl your fingers. Your fingers will point in the direction of the induced current (clockwise). The image above shows a coil, but if instead of a coil there were two parallel conductors attached to a source and a load, there still exists a complete circuit and this a “loop”. And as seen above the current will be flowing in opposite directions on either side of the loop.
The noise currents in two parallel wires will be amplified in a high impedance system (any voltage input on a DAQ), but if the wires were to be twisted together then the noise currents induced in each small section of the wire would have opposite directions and there would be no net current in the wire. This is the beauty of twisted pair cabling, and it is used in USB and Ethernet cables.
Use cables with twisted pairs for all differential signals.
Shielded cable: A cable shield is a sleeve of conductive material (copper, aluminum) that is wrapped around wires in a cable. The individual wires within the cable are usually twisted pairs.
The shield provides a low impedance path to ground for any currents induced in the cable due to capacitive coupling. The drain wire provides a low resistance connection to the cable’s metallic shield and should be grounded at both ends (instrument and receiver ends).
Conventional industry wisdom dictates that the shield should only be grounded at one end to prevent a ground loop forming through the shield causing ground currents to flow in the shield. A ground current flowing in the shield is undoubtedly bad because it is defeating the purpose of a shield. However not connecting the shield to ground at both ends to avoid a ground current simply IGNORES the real problem. Ignoring a problem is worse than isolating it and trying to fix it.
The real problem is that the so called “ground points” are not at the same voltage potential. The designer should solve the problem that is causing the distributed grounds to not be at the same potential, and then ground the shield at both ends and at any intermediate points. This is the standard recommendation of Profinet International.
Always use shielded cables for signals unless there is a good reason not to. Cost isn’t a good enough reason to not use shielded cables. A test system that cannot produce accurate measurements only has worth as scrap metal.
In addition to using shielded, twisted pair cables, good layout is also important. Do not lay signal and power cables side by side in the same conduit. Use separate cabinets for AC and DC sources and loads.
Signal Conditioning Modules
Signal conditioning is the process of filtering, isolating, converting and amplifying a signal to remove noise. Dataforth is a leading supplier of SCMs. SCMs use an isolated dc-dc converter to remove high frequency noise from signals. Since SCMs provide signal isolation they also protect the inputs of the DAQ from damage due to over-voltage conditions.
There are claims that analog signal conditioning can be replaced with oversampling and software filtering. Omitting SCM is simply a risky gamble. An SCM guarantees that HF noise is rejected, and thus NOT sampled. You can’t sample noise, that’s not there. Oversampling isn’t without consequence; it increases analysis and processing time. Test time is a key performance metric for automated test systems. It increases data storage capacity requirements if the raw input waveform from the DAQ is to be stored for audit/archival purposes.
Use input signal conditioners on all critical signals used for pass/fail determination on the test system.
DAQ, sampling and other topics will be covered in a future post.
