MULTI-SIGNAL DIGITAL DATA ACQUISITION SYSTEM BACKGROUND OF THE INVENTION

FIELD OF THE INVENTION

This invention relates to a system for acquiring and digitizing multiple analog signals.

DESCRIPTION OF RELATED ART

Over the past two decades the use of analog signal recording has been drastically reduced as high performance digital data acquisition electronics have been developed. Almost all systems now use digital data acquisition and analysis. An essential difference between analog recording and digital recording is that in analog recording a signal is recorded continuously for example on a magnetic tape or is analyzed in analog electronic circuits. The frequency content that can be retained in such recordings or analysis is a function of the recording medium capability and the ability of the electronics to follow the time variations in the signal. Equipment such as tape recorders are capable of recording very high signal frequencies. For example an analog tape recorder used in music recordings can record frequencies of 40,000 Hz and higher. A tape recorder for video signals will typically be able to record signal variations in the tens of MHz. When such devices are used to record signals that a have much lower frequency content, the fast response of the recording devices is able to record the signal without distortion of the low frequency portion of the signal since any high frequency signals mixed in with the signal of interest will be captured correctly. In analog recording the signal will be continuously recorded in time, capturing essentially the entire signal. With digital recording only a small fraction of a signal is sampled and stored or analyzed. The signal is passed to an analog to digital converter (ADC), which samples the instantaneous value of the signal only at discrete times as controlled by the sampling electronics. For example a particular signal may be sampled at 1000 times per second at time intervals spaced apart by 0.001 second. With modern ADCs, the sampling process typically takes microseconds or less.

An ADC converter generates an approximation of the continuously varying analog signal that discretizes the signal in two ways. First, the digital representation of the analog signal can represent the originally continuously

varying analog signal only in the form of discrete steps, i.e. the digital approximation of the analog signal can take on only specific discrete values. For example if an 8-bit ADC is used to sample an analog signal that could range between 0 to 10V, the only levels that could be stored would be multiples of the minimum voltage change that could be detected, which would be 10 V/2 ⁸ = 0.039063 V. Hence any signal between 0 V and 0.039063 V would be represented as 0; a signal between 0.039063 V and 0.078126 V would be represented as 0.039063 V, a signal between 0.078126 V and 0.117189 V, would be represented as 0.078126 V, etc. The digitizing process thus loses some of the information in the original signal. The number of bits in the digital representation of the signal defines the resolution of the ADC. A 10-bit ADC will have four times the resolution of an 8-bit ADC. A 16-bit ADC will have 256 times better resolution than an 8-bit ADC. For a 16-bit ADC the discrete step size in the above example becomes 0.000152 V, which clearly will have much less error than an 8-bit representation.

The second discretization occurs in time. As described above, an ADC will sample a signal only at specific time intervals. The signal will be sampled, for example over small time intervals of 1 microsecond or less. The sampling process will be repeated at fixed time intervals. As described above, for a sampling rate of 1000s/s, the interval between samples will be 0.001s. Hence only a small fraction of the original signal is sampled. In the example given, 99.9 % of the signal is ignored, while only 0.1 % is sampled. Depending on the nature of the signal unless the proper analog signal conditioning is applied prior to digitizing, this can lead to significant distortion in the representation of the signal of interest.

If in the time interval between samples the signal does not change by more than the discretization resolution, no information is lost due to the missing segment of the signal. However, if there is variation in the signal level in between the samples that exceeds the discretization resolution, there is a loss of information about the signal. Typically, in most real situations, a signal will be composed of a possibly slowly varying output from some measurement device plus an electrical noise signal that can have both well defined frequency content as well as random noise. The noise components can arise either due to the random input that drives a device producing a signal, or from

noise added to the output signal of the device or picked up along the transmission path prior to the ADC. For example, the noise could come from pickup of the electromagnetic signals coming from the numerous radio and television signals. These are in the frequency range 500 kHz to GHz. Unless care is taken to protect the signal conditioning electronics from such noise it will be added to any real signal coming from the device generating the signal of interest.

Since the noise portion of the signal is effectively random, the exact timing of the sampled portion of the signal will be uncorrelated with the occurrence of the noise signal. The effect of sampling a noisy signal at sampling rates that are lower than the frequency of the noise component is that the noise component will appear mixed in with the desired signal.

When a signal is sampled at a given rate, the maximum frequency that can be identified in the discretely sampled version of the signal is given by the Nyquist frequency, which is one-half the sampling rate. For example, if a signal is sampled at 1000 s/s, the Nyquist frequency will be 500 Hz. No signal above this frequency can be detected in the discretely sampled version of the signal even if the original analog signal contains frequency components above the Nyquist frequency (500 Hz in this example). However, the energy associated with signal components above the Nyquist frequency will appear in the sampled signal but at frequencies below the Nyquist frequency. This shifting of high frequencies into the frequency range that is preserved in the discretely sampled signal is called aliasing. The frequency band that is retained in the discretely sampled signal will include the band

fio _W < f< fhi _gh or (/, < / < /„ )

where the limiting frequencies are given by

f, = - and /, = —

' T ^{J h} 2δt where T is the total time over which the signal was sampled and δt is the time interval between sampling of the signal. All frequency components that

are in the original signal prior to digitizing will appear within this band. This includes both signal components that are within this band as well as signal components that have been aliased into the frequency band. This aliasing phenomenon is well known and is described in many books on digital data acquisition.

To ensure that the signal within the retained frequency band is a true representation of the portion of the original signal that is within the frequency band, the signal must be filtered prior to the digitizing process to remove frequency components that are at frequencies above the band of interest. Hence digital data acquisition systems must include an electronic analog low pass filter circuit ahead of the ADC to attenuate signal frequencies above the frequencies of interest. This filter is typically referred to as a low pass filter (LPF). Many designs and devices for an LPF are available. However, there is a common error made in selecting the cutoff frequency for the LPF in a digital data acquisition system. In many data acquisition systems the filter is set at or near the Nyquist frequency described above based on the assumption that this frequency is the highest one that can be identified within the retained signal. However, realistic low pass filters act to slowly attenuate the signal amplitudes above the LPF cutoff frequency. By convention, the cutoff frequency is defined as the frequency where the signal has been attenuated by 3 dB (30% of amplitude). If the cutoff frequency is set to the Nyquist frequency, there will typically be aliasing of signals into the frequency band of interest, since above the cutoff frequency the attenuation increases slowly with frequency. Hence, proper selection of the filter should place the LPF cutoff frequency at approximately one-fifth the Nyquist frequency if a fourth order filter is used. The attenuation at the Nyquist frequency will then be quite high, limiting aliasing into the frequency band of interest, which is then defined as / _; to 0.2/ _λ . Note that the digitized signal includes frequencies up to f _h . However, there can be increased error in the sampled signal as this Nyquist frequency is approached.

The foregoing considerations result in the typical configuration for a data acquisition system illustrated in Fig. 1. The system includes a buffer amplifier 1 , which could be a differential amplifier, for receiving an analog

signal, an LPF 2, a driver amplifier 3 and an ADC 4. It should be noted that these are the basic elements that are required. A practical system may include various additional elements such as multiplexers, band pass filters and precision voltage reference for calibration. The LPF in this system configuration must be an analog device and the cutoff frequency must be set consistently with the sampling interval and the ADC resolution. For a single purpose data acquisition system this poses no problem, because once the frequency band of interest is selected, the filter frequency and the sampling rate can be selected. However, data acquisition systems are typically designed for multiple applications, and thus are typically capable of a wide range of sampling rate. However, the LPF cutoff frequency must still be matched for each desired sampling rate. This requires that if the sampling rate is changed then the LPF cutoff frequency must also be changed. There are devices that allow for this. One option is to have the LPF as a plug in unit that can be physically changed according to the required sampling rate which is cumbersome and requires the purchase or manufacture of new LPF modules for each sample rate. There are also signal conditioning electronics available that include complex filter stages where the filter cutoff frequency can be set within some range or where one of several preset filter cutoff frequencies can be selected. This requires complex and costly electronic circuitry. One variant of this approach that has modest cost is to use switched capacitive filters. However, this type of filter has relatively poor performance compared to filters with dedicated cutoff frequencies.

Digital data acquisition systems are typically designed to sample multiple input signals. To limit the requirement for multiple ADCs, which are typically the most expensive components, the configuration shown in Figure 1 is altered as shown in Figure 2 by adding several additional analog channels and a multiplexer 5 which can select each analog signal sequentially and pass that on to a single ADC. This requires that the ADC sample at a much faster rate than is required for any single channel. It also results in non- simultaneous sampling of the signals, since the analog signals are passed in sequence to the ADC. As illustrated in Fig. 3, this can be corrected by adding a set of sample and hold (S&H) devices 6 that can be triggered to sample the analog signals simultaneously and hold them fixed while the multiplexer 5

sequences through the signals and passes them on to the ADC 4. This approach requires a number of electronic components to implement.

When the signals from multiple sensors are digitized a configuration similar to that shown in Figure 4 is required. In Fig. 4, the sensors 8 are configured as Wheatstone bridges. This is a common approach which can be used for sensing strain, pressure, temperature, acceleration, etc. The sensors 8 are typically driven by a common drive voltage which could be a steady DC voltage or an AC voltage. The output of each sensor 8 has to be transmitted along a dedicated signal line to the input amplifier 9 and then to a low pass filter 10 and a buffer 11 of the data acquisition electronics. The signal lines are typically twisted pairs of wires, i.e., each sensor requires two signal leads running from the sensor to the input stage of the acquisition electronics. In addition to these signal wires a ground wire and a signal shield is required. Similarly, to limit noise pickup the leads providing the drive voltage to a sensor 8 will have a shield and possibly a ground wire. Hence each sensor 8 requires at least two wires running from the drive voltage source to the sensor and typically at least four leads from the sensor to the data acquisition electronics. A system with multiple sensors thus becomes quite complex, requiring a great number of electrical wires from the drive and sensing electronics to the systems being monitored.

BRIEF SUMMARY OF THE INVENTION In accordance with the present invention, use is made of signal modulation and digital processing to achieve multiplexing of several analog signals onto one stream such that one transmission path can be used to carry the output of several sensors 8 to the data acquisition electronics and such that one physical wire pair can drive several sensors. The output of multiple sensors is transmitted along a single wire pair to one signal conditional channel and to one analog to digital converter such that several signals are sampled simultaneously without the need for either a physical multiplexing device or sample and hold devices to achieve high performance and low cost for the system. The individual signals are recovered through digital signal processing in a field-programmable gate array (FPGA) or a similar digital signal processor.

More specifically, in accordance with one aspect, the invention relates to a system for acquiring and digitizing analog signals from a plurality of sensors comprising:

(a) a digital signal processor means for generating a set of orthogonal code signals and forming a signal comprised of the sum of the orthogonal code signals;

(b) a set of filter means, each adapted to isolate one unique code, which is used to modulate the output of an individual sensor:

(c) a follower stage for passing the summed signal from the digital signal processor to the set of filter means;

(d) a buffer and summing amplifier stage for receiving the coded output of each sensor; and

(e) a signal conditioning and digitizing stage including an analog to digital converter for receiving the output of the signal conditioning and summing amplifier stage, wherein the output of the analog to digital converter is passed to the digital signal processor, and, within the digital signal processor the orthogonal codes are used to recover the digital versions of each of the individual sensor signals. This invention also relates to a method for acquiring and digitizing multiple analog signals from multiple individual sensors comprising the steps of:

(a) coding the output signal of each individual sensor with a unique code; (b) summing the coded output signals from all sensors to generate one analog signal defined by the sum of the coded individual signals;

(c) conditioning and digitizing the summed signal; and

(d) applying the suitable algorithms to the digitized signal to recover each of the individual sensor signals. BRIEF DESCRIPTION OF THE DRAWINGS

Figures 1 to 4 are schematic block diagrams of prior art digital data acquisition system;

Figure 5 is a schematic block diagram of a digital data acquisition system in accordance with the present invention;

Figure 6 is a graph of velocity versus time for three unmodulated signals;

Figure 7 is a graph of velocity versus time for the signals of Fig. 6 after modulation; Figure 8 is a graph similar to Fig. 7 showing the sum of the modulated signals of Fig. 7.

Figure 9 is a graph showing spectral densities of the three signals of Fig. 6.

Figure 10 is a graph showing the spectral density for the sum of the modulated signals;

Figures 11 to 13 are graphs showing the individual, original and recovered signals of Fig. 6.

DETAILED DESCRIPTION OF INVENTION With reference to Fig. 5, three separate sensors 8 (identified as

Sensors 1 , 2 and 3) are excited by one common source driving a complex wave form comprised of a combination of signals which can be separated in the frequency domain. Band pass filters 13 isolate different components of a drive signal from FPGA 14 and a complex wave follower stage 15, passing a different frequency limited signal through each of the sensors 8. The output of each sensor 8, shown in Figure 5 as differential outputs, is amplified by differential or instrumentation amplifiers 16 to produce a single sided output signal for each sensor. The single sided signals are summed with a summing amplifier 18. The output of the summing amplifier 18 is transmitted through a single signal path to the signal conditioning electronics, which includes again buffer stage 19, a low pass filter 20 and a buffer amplifier 21 , and is digitized in ADC 22.

The digitized signal is transmitted from the ADC 22 to the FPGA 14. Within the FPGA 14 the signal is treated as follows: the signal is passed through a low pass filter to recover the output signal from Sensor 1 ; multiplied by a sine wave with a frequency corresponding to that passed through Sensor 2 and low passed filtered to recover the output of Sensor 2; and multiplied by the sine wave driving Sensor 3 and low pass filtered to recover the output of

Sensor 3. This process can be applied to many more channels than the three shown.

The advantages of the system described above are: only one source signal is required to drive several sensors, only one pair of wires are required from the source voltage to drive many sensors; and only one wire pair is required to transmit the signals from several or many sensors to the signal conditioning and digitizing electronics. This simplifies greatly the complexity of installing a complex system requiring data acquisition from multiple sensors, and assures simultaneous sensing of all the signals. The system does require some electronics close to the sensors.

However, the electronics are quite compact. The overall system complexity and cost will be reduced compared to the typical architecture shown in Figure 4, while maintaining high performance.

Figure 6 shows three example signals, each with comparable bandwidth.

The signals are modulated according to the approach shown in Fig. 5 as follows:

y ^m ι (t) = yι {t)

where y _λ it) , y ₂ (t) and y ₃ (t) are the original signals, ym ₂ {t) and ym _i it) are the modulated version of signals 1 and 2 and ω _x and ω ₂ are the first and second harmonics of the modulating complex drive signal. The amplitudes for the modulation signals are each set to one, reflecting appropriate amplification in the band pass filter stages between the follower and the multiplier stages. The modulated signals are shown in Figure 7 and the summed modulated signals are shown in Figure 8. Figure 9 shows the power spectral densities for the original signals and Figure 10 shows the power spectral density for the sum of the modulated signals.

The signal shown in Figure 8 is sampled by the single ADC 22. The output of the ADC 22 is passed to the FPGA 14. In the FPGA 14, the summed signal is passed through a digital low pass filter to recover original signal 1 (from Sensor 1 ); the summed signal is multiplied by the first harmonic of the square wave and then low pass filtered to recover signal 2 (from Sensor 2); and the summed signal is multiplied by the second harmonic of the square wave and low pass filtered to recover signal 3 (from Sensor 3).

The original and recovered signals are shown in Figures 11 , 12 and 13 for signal 1 , 2 and 3, respectively. These show that the original signals are recovered with good accuracy.

The system has clear advantages over current assemblies for multichannel digital data acquisition electronics. The system reduces the number of source leads required to drive active sensors and reduces the number of output leads required to transmit the output signals to the data acquisition electronics and does not require physical multiplexers or sample-and-holds. Simultaneous sampling is still achieved as is high performance. While the example shows three input signals, many more can be accommodated with a single ADC by using additional harmonics of the modulating complex drive signal. An alternative approach to the electronics is to have the FPGA output directly sine waves at the desired harmonics. This eliminates the need for band pass filters ahead of the sensors but would require individual drive paths to the sensors.

While the modulation signals shown in the example are sine waves, there are alternatives to the modulation. For example coded bit patterns could be used. Recovery of these within the FPGA can make use of Phase Lock Loop (PLL) techniques.

In summary, the system of the present invention multiplexes multiple signals through a single signal transmission line and through a single analog to digital converter (ADC) simplifying system installation by reducing substantially the number of physical electrical wires required to transmit signals from multiple sensors to the data acquisition electronics while maintaining high performance. The approach makes use of FPGA controlled modulation of several active sensors, transmission of the output of multiple

sensors on one shared analog line such that one ADC can be used to simultaneously sample several independent signals with the individual signals recovered subsequently within the FPGA through digital processing. The approach avoids use of the standard switching multiplex approach thereby limiting noise generated after the analog signals pass through the anti-aliasing filters. This reduces the requirement for complex and expensive electronics components and increases the data quality that is possible compared to other data acquisition designs. For comparable performance, the new system will reduce the manufacturing cost.