Laborelec/ULB research
	Using Labview for field applications: a front-end conditioner defeating ground-loops, with antialiasing filters, etc. for National MIO cards

Labview in the field

Field vs. Lab applications of virtual instrumentation

Card selection
 

If one decides to work with Labview, better use the National Instruments cards. They may turn out a little more expensive. However, Labview examples illustrate how to use their drivers efficiently. Chances are that you can start from them to write your application. Advanced applications involve ADC, DAC, counter triggered and/or double-buffered data transfer to memory or disc and many others. Experience has shown that the availability of reliable drivers for costlier cards far outweighs their marginal price increase.  The most popular series is the MIO family. You can find them at all prices and speeds for all sort of buses, from AT to PCI, PXI and even on cards with an independent processor running with a real-time OS. Their functions are (almost) alike for all of them. Well deserving hteir name they are really multiple input/output. They allow data acquisitions, counting, generating pulse trains, DAC, digital input/outputs, triggers of all kinds, etc., many of these acquisition or generation schemes being able to run concurrently even under Windows which is not particularly real-time.

Again the versatility of MIO cards and the wonderful list of examples how to use them  outweighs a few drawbacks like:  No simultaneous sample and hold operation. This can be partly overcome with the fast MIO cards with a very short interchannel delay between data acquisition channels. This was true in all applications of this site (vibration, vibroacoustics).   No antiliasing filter. This is a serious problem when dealing with frequency analyses. More modern solutions rely upon delta-sigma technology that operate in oversampling mode (typically 64) and have a built-in low-pass digital filtering scheme. However, these converters cannot operate at very high frequencies unfiltered and sometimes lack fast triggering functions. This basically rules out their use for such applications as gear noise where oversampling at more than 200 Ksamples/sec underlies synchronous sampling and subsequent phase-locked period averaging and is not conditioned by aliasing. At least the way we implemented it with a PCI MIO E 1 cards.   No ground-loop isolation which amounts to suicide in field applications if handled without care. This seems a drawback shared by most I/O PC plug-in cards on the market for a good reason: who wants to introduce common modes exceeding 500 V inside a PC slot? National Instruments supplies external modules for such isolation. Unfortunately, they aren't as flexible as MOI cards: gains with jumpers, no offset in front of the isolation amplifiers.   Then comes the ADC resolutions: 12 16 or more bits. This does not belong to the list of drawbacks but will be discussed below. Note that 12-bits MIO cards are supplied with a useful dithering capability that somewhat allows to operate with 12-bits ADC ad yet recover additional bits for slow measurements. Variable signals intrinsically contain dithering and their further processing like FFT recovers lots of missing bits.  provided one tosses in the follwing extra equipment

A 12-bit MIO card, preferably for PCI buses because AT ones consume lots of DMA channels and rule out their simultaneous use with such widespread cards as audio should be supplemented with.an external front-end signal conditioning fully controlled by the various I/O's of the MIO card (DIO, clock generation, etc.) through opto-couplers that performs ground-loop isolation with programmable floating gains and offsets to make use of the dynamic of the isolation amplifier in a transparent way for the user. Antiliasing 8-poles switched-capacitance filters either in front of after the isolation amplifier, programmable either as Bessel or Butterworth. When located in front of the isolation amplifier they further contribute to increase its dynamic range, e.g with accelerometers used for velocity or displacement measurements and containing high frequency contributions that might otherwise saturate the isolation amplifier whose S/N ratio is higher than the filter. Auto-calibration scheme Floating AC/DC coupling BNC buffered outputs of conditioned signals in parallel in the MIO input signal pins. It is not uncommon that one want to visualize, process or record these inputs by some other means than the PC vi's. So let us not be selfish. How would one proceed to get this feature with delta sigma converters?

With such a front-end module, all Labview applications of this site are safe for field use with an additional triggering unit module featuring: MIO gate and trigger inputs need to be driven with lower impedance that is mandated by TTL and or CMOS. Ground-loops remain a threat like for analog signals. The advent of thyristor drives often lead to noisy trigger signals and this must be dealt with. Rebounds like those occurring with a photocell picking the start of a reflecting strip may cause inaccurate speed measurements and this again must be counteracted. A fully Labview controlled trigger unit is being readied for the same module with an automatic setting of trigger levels. It also uses the resources of the MIO card. Ever tried to fight getting a proper trigger level for other tacho signals than TTL?

Applications
 

All applications of the present site qualify for this front end conditioning. However, it is best described by means of a digital recorder streaming transient signals directly onto the hard drive of a PC. Such a straightforward application is easy to understand without much background.

 

Wish List for aDigital Transient Recorder Virtual Instrument. What would one like as a device to record transients on multiple channels in a field application? Typically, one would wish not to bother about   ground-loops   aliasing no matter the data acquisition rate. Remember that traditional recorders implicitly filter the signal to avoid aliasing   setting proper offsets and proper gains for your inputs to exploit the full dynamics of the component with lowest S/N ratios (filters, isolation amps)   writing down your settings on some piece of paper etc. while   focusing on your sensors specs, the expected range of your inputs expressed in physical units   knowing you can visualize your records anytime.   having them immediately available on the hard disc of your favorite PC ready for Labview (if you are a fan of the product and wish to analyze them further) and for a final report in Office for example or exchanging them over Internet. etc

Example Suppose you want to record volts and bars on channels1 and 2. All you know about your sensors are:   their sensitivity in phys units/volt   the expected range expressed in physical units   a sensor offset, i.e. to how many physical units a zero volt output of the sensor corresponds. It remains to define at which sampling rate you want to record the data onto the disc. An antialiasing filter is automatically set to 1/3 (could be 1/4) of the sampling frequency and floating offsets and gains ensure that both the isolation amplifier and switched capacitance filters are operated at their full dynamics.

 

Vi for data recording

Typically asks you where to store data and associated configuration file Original signal from scope vi with a sawtooth signal (green for volts and red for bars) with large positive offset and small dynamic component. The scope operates from the same MIO card after setting offsets and gain for both channels at zero and one for both channels. The scope displays inputs in volts. Fig.1. Original signal for mixed volt bar record from Vi scope (another vi in your library for field use): a sawtooth around 8.5 V for both channels from a function generator acquired via the MIO card.

Now let us configure the digital recorder as follows. Suppose one wants to record a transient over two channels at 4000 samples/sec. Buffer sizes for the MIO card and lengths of data to stream to disc per access can be defined not to lose data to memory overflow or due to other disc accesses in Windows. The file *.DAT where to store data is selected with a prompt.  Fig.2. Setting up the digital recorder.

After hitting "SAVE", another vi panel comes up: If you hit new configuration, the vi asks for the number of input channels you want to record (not shown). In this case, one specifies 2 channels. A configuration table with 2 rows appears. In each row you can specify the physical label of the physical unit, the sensitivity of the sensor and the expected range expressed in physical units and finally the sensor offset. Then you may select whether you want a antialiasing filter and which type (more on this later) and eventually the kind of coupling you wish AC/DC. An optional text zone helps later identification of the record Fig.3. Configuring the gains and offsets for digital recoring the easy way when dealing with mixed sensors.

Push on "save configuration". The signal conditioning unit gets configured automatically and bargraphs appear telling how your actual signal compares with the dynamic range computed from your configuration. In this version they extend from zero to the min and max value. Better would be a bargraph extending from min to max. Fig.4. After seeting your channels, check overflows with current signals If there is saturation, press "not OK" otherwise "OK". In this case, recording starts. You stop it by pressing "stopping" button.

 

Action of the front-end signal conditioning

Let us now have a look on the scope at the input signals after conditioning. Since the first (green) channel was declared in [-10V, +10V] with a sensitivity equal to 1V/V, the signal conditioning does not alter the output signal and is confined to the upper zone between 8 and 10 V. In contrast to the first channel, the second (red) channel exploits the full dynamics of the [-10V, +10V] zone. In other words it exploits the full dynamic range of the upstream isolation amplifier and switched-capacitance antialiasing filter. The conditioning thus contributes to increase the signal-to-noise ratio of these two S/N weak components by introducing a correct offset and gain before the isolation amplifier. Fig.5. Full scope display of previous oscillogram

 

Playing back the digital record

After recording the volt/bar signals, one may wish to play it back. One must then start another vi for playback: Display12.vi. In its prototype version it enables to select the record file from a prompt. It associates the data file with the corresponding configuration file. In this way, it re-constructs the original signal for channel 1 (bar units) that was expanded for better S/N by the front-end conditioning unit. Most popular buttons of recorders and some others got implemented for easy use: Move across the record with cursors by pressing FF or RW buttons Extract portions of the record and further zoom on them display any combination of two signals for time comparisons (could be easily adapted to display more than a pair of signals) perform an FFT on any part of the time signal with(out) zooming. Better would be to implement time-frequency spectrograms in the future. In both cases and despite the tremendous speed of nowadays PCs and their RAM sizes, it is advisable not to select too lengthy time records to process. This is partiuclarly true when Joint-Time-Frequency methods are invoked to obtain spectrograms.

Playing back the previous record typically looks like the Figure below. Fig.6 Original front panel of vi for playback. Channel zero is masked by the (optional) display of the configuration. Label is in V. It is confined in the upper region of the [-10, 10] V region as prescribed in the configuration. Channel 1 was reconverted in physical units (bar) from the output of the font-end conditioner (cf Fig.5)  with a good S/N ratio and from the associated configuration file. It lies between 75 and 95 bars according to the configuration. For people lacking the basics of Labview, this may be interesting news. Suppose you zoom on any interesting part of the record and get a display of data you would like to analyze further with a new method. Then simply select the zoomed oscillogram, copy it and paste it in a new vi of your creation. Declare it as control instead of an indicator. Data are then ready for use. Indicators are not mere graphics displays. they are objects holding the data as well. This feature was extensively used along the digital recorder to devise special data processing in a railways application.

Filter issues

Background information When using an ordinary tape recorder, signals are implicitly filtered according to the tape speed. The S/N may not be great at times (at least in FM mode at low speed). At any rate, one need other devices to perform FFTs and other frequency-related data processing. These devices are fit with antialiasing filters. Trouble with digital recorders is that they must eliminate aliasing first priot to digitizing signals. If that is the case, playing back the record and performing FFTs is a child's game in Labview. Since the AD of the MIO card are not supplied with antialiasing filters, one must supply them prior to recording in digital mode. The front-end conditioning provides such a filtering function with a cutoff frequency depending on the scan rate (data acquisition rate per channel). As mentioned earlier, it is set to 1/3 of this scan rate and could be somewhat lowered to 1/4 to make sure that the cutoff is complete at the Shannon frequency (half scan rate). The next example shows the action of these filters by feeding channels 1 and 2 with a square wave whose frequency is swept from almost zero Hz to 4000 Hz. The data acquisition rate is 4000 samples/sec. Channel 0 is unfiltered. Channel 1 is Buttterworth filtered at 1200 Hz cutoff frequency. One records both channels for about 60 sec and plays it back until one reaches the point when the frequency of the square wave passes through the Shannon frequency. The playback with Display12.vi is shown below with the progressive attenuation of channel 1 by the 8-poles Butterworth filter between 19 and 36 seconds into the recording. Fig.7. Original vi panel

FFT on filtered signal Let us then zoom on the start of this transient at about 19.2 sec into the recording phase and perform an FFT on channel 0 (filtered) and 1 unfiltered. At 19.25 sec approximately after strating the recorder, the fundamental of the square wave is shown in the frequency amplitude spectrum below obtained with Display12.vi by selecting the first trace as belonging to the channel 1 and pressing the "FFT(x)" button. It is at circa 900 Hz. Fig.8. Original vi panel

FFT on unfiltered signal In contrast aliasing is severe on the frequency spectrum obtained for the unfiltered square wave at the same instant (channel 0). The amplitude of the fundamental is somewhat higher since there is no Butterworth filter to attenuate channel 1. Fig.9. Original vi panel.

Butterworth vs. Bessel filters When using switched-capacitance IIR filters mimicking analog filters, one must ask the question what the filters are for. Are these simply meant to clean high-frequency noise from time signals and to perform subsequent frequency analyses forgetting about the shape of the time signals? In the first case, better use Bessel while Butterworth filters are better choice for frequency analyses. This dilemma does not exist when using FIR filters as implemented in DSP. We'll see why later when discussing the transfer functions of the filters inside the front-end signal conditioning unit. Records of a square wave with Bessel and Butterworth filters A school case to compare Bessel and Butterworth filters is to feed them both with the same square wave and see what happens. Again the responses below were obtained with the digital recorder and playback vi's. Fig.10. Original vi panel Channel zero shows a square wave with shaven high frequencies. As a result, the corners of the square waves are rounded as one would expect intuitively. Butterworth filters tend to cause overshoots at the square wave transitions. This is no problem in subsequent frequency analyses but quite annoying if one wants to determine crest values for the time. Some clients have ruled out Butterworth filters in the past for that reason. On the other hand, frequency rolloff is cleaner with Butterworth filters, what makes them prime candidates for frequency analyses with IIR filters. This is all described next with the transfer functions of the Bessel and Butterworth filters used for digital recording.

 

Transfer functions of the signal conditioner antialiasing filters
 

Free multichannel Fourier analyzer with MIO cards In case one overlooks it, many intermediate and advanced MIO cards have two DAC channels. These can be used to generate pink noise to obtain transfer functions in the classical way through averaging of the auto- and cross- power spectra. Adapting some examples from Labview to the MIO card hardware, one managed to get a multi-channel Fourier analyzer. This was then applied to obtain the transfer functions of the Butterworth and Bessel antialiasing filters of the front-end conditioner. For the frequency bands of interest here (up to 20 KHz), the fact that the FET scanner of the MIO ADC are not provided with simultaneous S/H technology is no big drawback at least for the top MIO cards with tiny interchannel delays. The pahse shift this causes is negligible. This Fourier analyzer named FullDynamicAnalyzer.vi yields the following transfer functions for the Bessel and Butterworth filters with the 1200 Hz cutoff frequency of a digital recorder at a 4000 samples/sec scan rate. Before starting the process, one starts another vi called GeneratePulseTrains.vi with a control specifying the cutoff frequency. Once launched, this vi does not consume any processor time (seen from Profile). One assumes that the front-end conditioning unit was configured for either Butterworth or Bessel filtering with gain 1 and zero offset.

Bessel transfer function Nice for cleaning time signal from HF noise because the phase shift in linear in frequency in the bandpass region. This almost amounts to a distortion-free time lag of the original signal like with FIR DSP-based filters. Fig.11. Original vi panel.

Butterworth Transfer Function Roll-off is much cleaner in the bandpass portion where the gain remains close to unity, but the phase lag is not linear in frequency causing distortion on the time signal, i.e. no constant group delay. Most FIR filters escape this problem and even more interesting for FIR antialiasing filters is the fact that the equivalent time lag amounts to half the number of taps times the inverse of the scan rate. Bessel filters almost behaves with a constant group delay, but they exhibit a progressive attenuation in the bandpass region. Fig.12. Original vi panel.

 

12- or 16-bits successive approximation ADC?

Background info In some people's mind, selecting a 16-bits successive approximation ADC (SA/ADC in short) is the panacea for forgetting about signal conditioning altogether. To some extent they are right but they may fream in other respects.. First never infer from a fast SA/ADC spec with a FET scanner that its useful scan rate is merely the speed of the SA/ADC divided by the number of channels. This assumes that the preamps upstream from the sampling capacitance and often combined in a single chip are fast enough to ensure that their settling time ensures that the sampling capacitance is loaded within at least half one least significant bit of the SA/ADC. Therefore, serious manufacturers (National Instruments among others) often specify different guaranteed scan rates for such SA/ADC for one or more channels operations. Settling times are longer for high gains for preamps. The reason is that average levels for different channels may markedly differ and the scanner must switfly switch between these at high scan rates with a resulting loss of precision. The front end conditioner developed here doesn't dynamically switch channel gains at hte scan rate. Furthermore, the preamp gains are at their minimum value due to the automatic rescaling occurring in the conditioner as mentioned in the configuration file. Second, even if settling time is no issue, you face a more serious one for quasi DC tiny signals. Measurements are generally noisy and it takes some averaging to get an accurate value. That is plain theory of measurement. In that case manufacturers supply well-known solutions for such measurements as (intrinsically noisy from their physics) thermocouples, RTDs and other slow signals. There are the traditional single or double ramp integrators over a few line frequency periods or lately delta sigma converters with built-in low-pass filters to eliminate the line frequencies. All of them perform some implicit averaging (a low-pass filter is an alias thereof amounting to not quite an exponential average). Such averaging defeats the use of such converters for fast signals where either fast delta sigma or SA/ADC rule. The latter are more adequate than delta-sigma converters for fast signals beyond the audio range, provided one knows about the approximate frequency contents or controls it with proper antialiasing techniques. That is one reason among others why the present conditioning has been devised. It covers the whole spectrum of applications plus the ground-loop isolation for field applications. What about the small quasi DC signals if one uses a 12 bits SA/ADC converters. A partial and mostly adequate answer lies in the dithering technique of MIO 12-bits cards. It is used for precise autocalibration of the front-end conditioner. Let us a have a closer look at the Labview driver of the conditioner in case of a slow input signal close to zero.

Front conditioner driver and dithering and autocalibration Let us have a 12-bit MIO SA/ADC connected to the front end conditioner. The latter outputs a small quasi dc signal to the first channel of the FET scanner of the moi card. The signal is close to zero and varies due to drift within two offset values. Using the built-in dithering feature of this card, on can follow the drift between two successive approximation of the ADC by averaging the dithered input over four line periods. Successive reading of dc signal with and without dithering 1st reading With dither: 1.45 mV. without: 4.88 mV, i.e. the resolution of the 12-bits ADC for the range [-10V, 10 V]. 2nd reading With dither :1.45 mV without: 0 mV. How to proceed with the driver? Just select GND/IN to GND (red button) and DC/AC to AC (green) and a gain close to one (DAC2=314). This amounts to ground the input while disconnecting the external signal and then to set the offset to zero with AC coupling. For more detail about the electronics and drivers of the conditioner, mail dany carlens from the ULB. Why the difference? Presumably because the prototype of the conditioner was not properly shielded. In the first reading the SA/ADC happened to pick up the signal with some noise at the line frequency. This raised the sample by one step of the resolution of the ADC. Dithering is implemented for autocalibration purposes to get precise spline interpolations for gain and offset controls (DAC1 and DCA2). It consists of setting dithering on for the MIO card and averaging 80 samples over four line periods. This statistically eliminates common mode and enhances the 12-bits SA/ADC resolution.

 

Physical Layout of the conditioner prototype The results presented above were obtained with a two-channel prototype that helped devise the drivers and applications like the digital recorder and playback vi's. The front panel of the front-end conditioner would look like the design below and accommodate 16 inputs with isolated BNC plugs. Each plug-in card conditions two inputs with ground-loop isolation, filtering and enhanced S/N through floating gains and offsets. The enclosure is shielded for EMC. A few LEDS monitor the operation of the conditioner. Power supply is either 110-240 V ac or optionally from a 12 V battery with voltage stabilization. The rear panel has two sockets for the replication of the MIO card pin plus two isolated BNC plugs for DAC functions either ground loop isolated and thus not connected to the pin replicator or non isolated. This may be helpful for tests on machinery drives where groundloop issues might be crucial (for automatic control engineers). The empty left-hand side panel is reserved for advanced triggering units, where MIO cards and their drivers show some stamina). The reason for a pin replicator is not to foreclose subsequent applications for such a versatile card as the MIO's, assuming that knowledgeable people are in command and avoid output conflicts with the conditioner. Also for routing the conditioned outputs to other devices (FFT, recorders, etc.) concurrently used with present and future vi's in Labview.

 

Temporary Specifications Input ranges: -10 to 10 V with protection from transients (e.g. electrostatic discharge) and moderate differential levels. Ground-loop isolation: 1000 V peak. S/N ratio 72 dB (exploiting the full dynamics of switched capacitance filters and isolation amplifiers) Floating gain: from 1 to 50 with variable pitch. Selection from spline interpolation. Offset: -10 to 10 V in 4096 almost equal steps. Also spline interpolation. Drift: being investigated with low-drift opamps and resistances.

 

For the reader of this web page. In case you are interested in this conditioner and its applications please mail me or Serge Bassem @ ACT'L, the company that might build the hardware for you. More production volume means lesser unit cost. .  We hope that this page amply demonstrates how versatile the MIO cards from National Instruments may be and how to extend their use to field applications. Then Labview which has been so good to enhance productivity in a lab environment stands a better chance to make inroads in the field where one can move the lab experience switfly and without dangers and without worries about ground-loops and without cergoes of devices to carry along. In a forthcoming page, we'll demonstrate how easy it is to perform synchronous sampling with MIO cards phase-locked with a phase reference generating one top/rev without loss of data, an hitherto difficult task reserved to costly devices and perhaps too demanding for delta-sigma converters. Synchronous sampling is almost mandatory for vibrations analysis of all sorts of rotating machinery. National Instruments markets a front-end conditioner to avoid ground loops. It lacks proper antialias filters, programmable floating gains and has no floating offsets. For more detail, click here.