Title of Invention  A METHOD FOR NOISE REDUCTION IN DATA ACQUISITION SYSTEMS 

Abstract  The invention relates to a method for eliminating noise in data acquisition systems. The signal to be acquired is sampled and digitised at a rate that is several times the required sampling rate. Then the sample values are averaged to derive a single representative value that is free of the influence of high frequency random noise. To eliminate the effect of low frequency systematic noise, the number of samples to be averaged must be an integral multiple of fs/fn where fs is the sampling frequency and fn is the noise frequency. Over sampling of the input signal simplifies the antialias filter requirements. System calibration techniques may be employed to further reduce the effect of component nonlinearities, offsets, reference errors etc. and a digital multiplier, to enhance the measurement range. 
Full Text  A Method for Noise Reduction in Data Acquisition Systems Technical field The invention relates to a method for eliminating or reducing the effect of noise in data acquisition systems. The present invention particularly relates to a method for eliminating the high frequency, random noise as well as for eliminating low frequency, systematic noise. The present invention further relates to a method to implement a digital multiplier to improve the measurement range to any desired value. Background and prior art With the growth of Electronics Industry more and more dependence is on electronic data. As such this data has to be secure and free from any external disturbance. One of the main problems faced in this area is that of Noise. The noise is any unwanted signal that may interfere with the useful signal thereby obscuring/corrupting the acquired data and making its reconstruction difficult. Thus it is required to minimise the effect of noise for meaningful reconstruction/interpretation of the acquired data. There are various methods for eliminating noise in data acquisition systems. US Patent 6,295,331 on 'Method and apparatus for noise compensation in imaging systems' explains an imaging system generally referred to as a computed tomography (CT) system in which an xray source projects a fanshaped beam which is collimated to lie within an XY plane of a Cartesian coordinate system and generally referred to as the "imaging plane". The xray beam passes through the object being imaged, such as a patient. The beam, after being attenuated by the object, impinges upon an array of radiation detectors. The intensity of the attenuated beam radiation received at the detector array is dependent upon the attenuation of the xray beam by the object. Each detector element of the array produces a separate electrical signal that is a measurement of the beam attenuation at the detector location. The attenuation measurements from all the detectors are acquired separately to produce a transmission profile. US Patent 5,619,040 on 'Data acquisition system' relates to a data acquisition circuit for a particle detection system that allows for time tagging of particles detected by the system. The particle detection system screens out background noise and discriminates between hits from scattered and unscattered particles. The detection system utilizes a particle detection pixel array that is in electrical contact with a data acquisition pixel array. In response to a particle hit, the former generates a current which is detected and integrated by the latter to produce a voltage across a capacitor that is related to the amount of energy deposited in the pixel by the particle. The current is also used to trigger a read of the pixel hit by the particle. Another 'System and method of eliminating systematic noise in stimulusresponse systems is described in US Patent 5,640,124 which details a method and a system of eliminating systematic noise in a stimulusresponse system. A data acquisition time frame is triggered during which a desired response signal is acquired. The triggering initiates the systematic noise, which is stored along with the desired response signal. A stimulus signal is provided to the stimulusresponse system in order to induce the desired response signal. A variable delay is introduced between the triggering of the data acquisition time frame and the provision of the stimulus signal. Individual sets of response signals are acquired which respectively correspond to each introduced variable delay. The sets of acquired response signals are shifted so that the stimulus signals are equivalent in time with respect to each set of acquired response signals. The shifted sets of acquired response signals are summed in order to convolute in time the systematic noise. 'Noise reduction method, apparatus and program for digital image processing' has been described in many other patents such as US PCT application No: PCT/US00/27502 which relates to a method of processing a digital image channel to remove noise that includes the steps of: identifying a pixel of interest; calculating a noise reduced pixel value from a single weighted average of the pixels in a sparsely sampled local region including the pixel of interest; replacing the original value of the pixel of interest with the noise reduced pixel value; and repeating these operations for all of the pixels in the digital image channel. But none of these above prior art references explains how averaging techniques can be used to eliminate the effect of low frequency systematic noise. Further, the above prior art references do not teach how over sampling and system calibration techniques can be used in conjunction with averaging to simplify the component requirement of the system as well as enhance the quality of measurement. In addition, employment of digital multiplier for an improved measurement of the signals was not resorted to in the prior art. Further, nonideal behaviour of components constituting the data acquisition system results in nonlinearities and offsets, producing an output disproportionate to the input at the output of intermediate devices as well as the final output of the system. The errors caused by these are generally collectively referred to as gain and offset errors respectively. Any measurement system requires a reference against which the input is scaled. The voltage reference that is employed for this purpose (i.e. to set the fullscale of the measurement range) in a data acquisition system may introduce errors due to inaccuracy and drift in its value. The error caused by this in the system output is generally referred to as fullscale error. Accordingly, a method wherein the quality of measurement is improved by reducing the effect of component nonlinearities was felt necessary. Objects of the invention The primary object of the present invention is to provide a method for eliminating the effect of high frequency random noise and low frequency systematic noise in data acquisition system using averaging techniques. An object of the present invention is to adopt over sampling techniques by estimating the number of samples of analogue signals to be averaged so as to eliminate the effect of low frequency systematic noise. Another object of the present invention is to make use of system calibration techniques together with averaging to reduce hardware requirements and to enhance the quality of measurement. Yet another object of the present invention is to improve the measurement range by making use of a digital multiplier. Further object of the present invention is to provide a method wherein the quality of measurement is achieved by reducing the effect of component nonlinearities, offsets, reference errors etc. Summary of the invention The present invention provides a method for eliminating both high frequency random noise and low frequency systematic noise in data acquisition systems. In this method the input signal to be acquired is sampled and digitised at a rate that is several times the required sampling rate. The digitised sample values are averaged to derive a single representative value that is free of the influence of high frequency, random noise. In order to eliminate the effect of low frequency systematic noise, the number of samples to be averaged is an integral multiple of fs/fn where fs is the sampling frequency and fn is the noise frequency. Brief Description of the Drawings Fig 1 shows the functional block diagram of a method for eliminating the effect of noise in the data acquisition system. Fig 2 (a) shows a signal Vi, which is a time invariant signal. The term 'time invariant' signifies that the signal magnitude remains constant and does not vary with time. Such a 'constant' signal is chosen only to make the illustration simpler. The noise reduction method described here is general and applies equally well to signals of any type. Fig 2(b) is a graphical depiction when the White Noise (random high frequency noise) added onto the above signal Vj. This corrupted signal Vc and its digitized samples are shown in this figure. Fig 2 (c) shows the processed signal by adopting the method of present invention. Fig 3 (a) shows a signal, Vs that is corrupted by systematic noise Fig 3 (b) shows the recovered signal Vi that is free from the influence of noise. Fig 4 is a flow diagram depicting the steps involved in the removal of high frequency random noise. Fig 5 is a flow diagram depicting the steps involved in the removal of low frequency systematic noise. Fig 6 is a flow diagram depicting the steps involved in averaging by using over sampling of analog signals. Fig 7 is a flow diagram depicting the steps involved in system calibration techniques. Fig 8 is a flow diagram depicting the steps involved in employing a digital multiplier in a data acquisition system. Detailed description of the invention Preferred embodiments of the present invention are explained initially by referring to Fig 1 of the accompanied diagrams. The signal input (1) is a signal corrupted with low and/or high frequency noise. This signal input (1) is given to a sampler (2). The sampler (2) also receives another input in the form of sampling frequency fs (3), which is predetermined to be a rate that is several times the required minimum sampling rate. The output of the sampler (2) is sampled and the data provided to the digitizer (4). The digitizer (4) also has another input from the reference (5). The digitizer (4) works typically using the successive approximation method where the signal (output of 2) and its residues are compared in steps to a progressively smaller binary fraction of a set reference voltage (5). The output of the digitizer (4) is given to several signal processors (6, 7, 8 & 9) for further processing of signals. The signal processor (6) is the processor used to eliminate random or high frequency noise. The functions of the signal processor (6) are described in the flowcharts of Fig 4. The output of signal processor (6) is a signal which is free from the effect of high frequency noise. The signal processor (7) is the processor used to eliminate systematic or low frequency noise. The signal processor (7) has two inputs, one from the digitizer (4) and the other input is the expected noise frequency, fn which is estimated. The functions of the signal processor (7) are described in the flowcharts of Fig. 5. The output of the signal processor (7) is a signal free from the effect of low frequency noise. The signal processor (8) is the processor used to eliminate the effect of component nonlinearities, offsets, reference errors etc. using System calibration on data samples. The signal processor (8) has two inputs, one from the digitizer (4) and the other input is gain, offset or fullscale coefficient. The output of the signal processor (8) is a signal free from the effect of component nonlinearities, offsets, reference errors, etc. The functions of signal processor (8) are described in the flowcharts of Fig 7. The signal processor (9) is used to improve measurement range. The signal processor (9) has two inputs, one from the digitizer (4) and the other is the measurement range factor. The output of the signal processor (9) is a signal with improved measurement range. The functions of signal processor (9) are described in the flowchart of Fig 8. The signal processors (6, 7, 8, and 9) can be used independently or there can be a combination of two or more or all of these processors, to achieve the desired result. Accordingly, the present invention provides a method for reducing or eliminating the effect of high frequency and random noise in data acquisition systems. The signal selected from analogue voltage or current signals such as those corresponding to physical parameters of pressure, temperature, flow rate etc. to be acquired is sampled typically by closing a switch and transferring the signal charge on to a capacitor at the sampling rate so that the capacitor holds the signal voltage periodically and digitised typically by using a successive approximation method where the signal and its residues are compared in steps to a progressively smaller binary fraction of a set reference voltage at a rate that is several times the required sampling rate. Further, the averaging of sample values is derived from the acquired signals that are over sampled. An embodiment of the present invention wherein the low frequency systematic noise is reduced or eliminated by averaging the number of samples of the input signals (Vi). This averaging must be done on an integral multiple of fs/fn samples where fs is the sampling frequency and fn is the noise frequency. This requires apriori information on the nature of the noise. Apriori information means knowledge on the frequency and amplitude of the noise that may be present in the system output before the measurement is done. This prior knowledge is vital to decide the number of samples that is required to be averaged in the process. It may be noted that the over sampling of the input signal simplifies the antialias filter requirements and by employing system calibration techniques the effect of component nonlinearities, offsets, reference errors etc are further reduced. The electrical signal corresponding to the physical parameter to be measured is sampled and then digitised in data acquisition systems. The digital samples are essentially in the form of numbers and are processed by a logic circuit that may consist of an embedded processor. This invention describes a method wherein two schemes of averaging the required number of samples are employed to reduce/eliminate the effect of noise that can corrupt the acquired signal. The number of digital samples to be processed is independent of noise frequency in one case and dependent on noise frequency in the other. In the scheme of eliminating high frequency random noise the number of samples does not depend on noise frequency. A sufficiently high number (say 10 to 100) can be chosen according to the convenience and time allowed for processing. On the other hand, in the scheme of eliminating low frequency systematic noise, the number of samples to be averaged has to be calculated based on the noise frequency. Therefore in the present invention two kinds of noise signals are considered: a random, high frequency noise and a systematic, low frequency one. Further, the present invention also explains how techniques like system calibration can be used in conjunction with averaging to enhance the quality of measurement by reducing the effect of component nonidealities. Now by referring to Figures 23 of the accompanied diagrams, wherein sampling and digitisation methods of the present invention are depicted in the form of an illustration. Fig 2 (a) shows a signal Vj, which is a time invariant signal. The term 'time invariant' signifies that the signal magnitude remains constant and does not vary with time. Such a 'constant' signal is chosen only to make the illustration simpler. The noise reduction method described here is general and applies equally well to signals of any type. Figure 2 (b) represents a signal that has been corrupted by high frequency random noise. The signal is sampled at the time instants where the vertical lines are drawn. For instance, if a sampling frequency fs = 1 KHz, these samples are apart by Ts = 1/fs = 1 ms. The height of vertical lines indicates the signal sample values. These sample magnitudes are then converted to a number with discrete steps in the digitisation process. One way to do is by 'coding' to represent null (zero) signal magnitude with number 0 and the maximum possible signal magnitude (say 5 V) by the 'full scale number' (11111111 in 8 bit coding). It may be noted that such an 8 bit coding gives 28 (256) steps, each corresponding to (5 V / 256) i.e. 19.5 mV. The process of digitisation maps each signal sample to one of these 256 values. Similar explanation holds for the signal in figure 3 (a) that has been influenced by low frequency systematic noise. The sample values for both these signals are tabulated as below. Then the sample values are averaged to derive a single representative value that is free of the influence of high frequency, random noise. The steps involved in this operation are shown in the flow diagram of Fig 4. The averaging of samples to eliminate the effect of low frequency systematic noise by using fs/fn samples is represented graphically in Figure 3 a. Here Ts is the sampling period i.e. the time interval between two successive signal samples and is the reciprocal of fs i.e. 1/fs. Tn is the period of low frequency noise that rides over the signal i.e. the time period over which the noise cycle repeats and is the reciprocal of fn i.e. 1/fn. Number of samples Nl=fs/fh and N2=2fs/fh are also shown in the figure. The figure uses fs/fn=10 for illustrative purposes so that Nl=10 or N2=20 or any integer multiple of 10 (such as 30, 40 etc.) represent the number of samples to be averaged to achieve the elimination of noise. The steps involved in this operation are shown in the flow diagram of Fig 5. This requires apriori information on the nature of the noise. Apriori information means knowledge on the frequency and amplitude of the noise that may be present in the system output before the measurement is done. This prior knowledge is vital to decide the number of samples that are to be averaged in the process. A method of reducing or eliminating random high frequency noise from data acquisition systems Consider a signal, Vj as shown in Figure 2a (A time invariant signal is shown for the purpose of simplicity). Assume that 'white' noise (random, high frequency noise) is added onto the signal. In a data acquisition system this can be due to device noise of components, noise getting coupled from power supply, noise picked up by wires and Printed Circuit Board (PCB) traces etc. The corrupted signal, Vc and its digitised samples are shown in Figure 2b. Now consider that a data processing circuit is used to average every set of ten samples of the signal. The processed signal samples thus derived are shown in Figure 2c. Here every set of ten samples in the acquired signal is replaced by a single representative value that is the average of the samples in that set. It is clear that this processed signal, V0 is free of the influence of 'white' noise and closely resembles the original analog signal at the system input. The number of samples that is averaged may vary from 2 to 10000 (100 being a typical value) so that the method can be applied with an over sampling ratio of 1:2 to 1:10000. The signal frequency may vary from 1 Hz to 100 MHz (1 KHz being a typical value) and the signal amplitude from 1 mV to 100 V (5 V being a typical value). The amplitude of the interfering noise signal may vary from 1% to 100% (10% being a typical value) of the original signal amplitude. A method of reducing or eliminating low frequency systematic noise from data acquisition systems As an embodiment of the present invention, next consider a signal, Vs that is corrupted by systematic noise as shown in Figure 3a. Such low frequency noise may be caused by ac power line feedthrough, by dcdc converter switching or by electromagnetic coupling of signal from a neighbouring channel in a data acquisition system. Here also the averaging can prove effective if the number of samples to be averaged is chosen in such a way that integer number of cycles of noise signal is included. From the figure it can be seen that the number of samples can be either Nj or N2 to satisfy this. The recovered signal, Vi after averaging is shown in Figure 3b and is free of the influence of noise. The criterion for effective elimination of noise in this case is that the number of samples to be averaged must be an integral multiple of fs/fn where fs is the sampling frequency and fn is the noise frequency. The corresponding time periods are indicated by Ts (=l/fs) and Tn (=l/fn) in Figure 3a. It may be noted that this scheme requires apriori information on the nature and frequency contents of the noise signal. The signal frequency may vary from 1 Hz to 100 MHz (1 KHz being a typical value) and the signal amplitude from 1 mV to 100 V (5 V being a typical value). The amplitude of the interfering noise signal may vary from 1% to 100% (10% being a typical value) of the original signal amplitude. The noise frequency can vary from 0.1 mHz to 50 MHz (50 Hz being a typical value). The number of samples that is averaged may vary from 2 to 10000 (100 being a typical value) so that the method can be applied with an over sampling ratio of 1:2 to 1:10000. It may be noted that in both the above methods the signal is sampled at a frequency much higher than the minimum that is required (referred to as Nyquist frequency). Usually the Nyquist frequency is two times the maximum signal frequency but in the schemes explained above the sampling is done at least 5 times this rate. This over sampling of the signal also simplifies the requirements of antialias filter at the frontend of the data acquisition system. The antialias filter is generally used to limit the bandwidth of input to the system and its order and complexity becomes less when the sampling frequency increases beyond the Nyquist frequency. A single pole filter may be adequate to band limit the signal if the over sampling ratio (ratio of sampling to Nyquist frequency) is high. Also if the over sampling ratio and the number of samples to be averaged is chosen to be a power of 2, then the division operation for extraction of mean can be realised simply by shifting the binary representation of accumulated result. This is because in digital domain any binary number when shifted to the right by one bit position becomes half its original value thus producing the effect of dividing by 2. Similarly shifting right by more number of bit positions results in the value being divided by higher powers of 2 (4, 8, 16 and so on). The steps involved in this operation are shown in the flow diagram of Fig 6. Further, techniques like system calibration may be used in conjunction with the above signal processing methods to reduce the effect of component nonlinearities, offsets, reference errors etc. System calibration refers to the technique where an input signal of known magnitude is applied to the system and the output is observed to derive a set of coefficients. These calibration coefficients are then made use of to adjust/correct the output when a measurement is taken subsequently. This procedure helps to eliminate the influence of noise and other errors due to system nonidealities such as gain, offset and fullscale errors. For instance, assuming that input signal amplitude of Voff produces an output null and Vref, an output fullscale in a data acquisition system, then an output value of Vout may be calibrated to correspond to an input signal of (VoutVoff)/(VrefVoff)*Vref. This may be done by table lookup or analytical software/hardware methods. Here Voff is the system offset voltage by which the output is shifted over the entire range i.e. it is the offset error. Vref is the system fullscale output voltage and contains the gain error factor. Once these errors are known by prior calibration procedure, any measurement that is taken subsequently has to be corrected as shown by the formula. This corrected output can be derived from the actual measured output, Vout by various methods. For example, in table lookup method the corrected output corresponding to various measured outputs are stored in a table (in say, the memory of the system). Then for a particular measurement the corrected output is read from this table. The calculation may also be performed by a processor executing a programme in which case it is done through software or by a dedicated digital hardware means. Here it may be noted that system nonlinearity is corrected by using a gain factor, system offset is corrected by incorporating an offset factor and reference errors are eliminated by having a variable full scale. The quality of measurement in terms of accuracy and repeatability are significantly enhanced by applying the calibration steps on a processed data as described previously. The method is effective for system full scale voltages from 0 to 100 V (5 V being a typical value), offset voltages from 0 to 50 V (0.1 V being a typical value) and reference voltages from 0 to 100 V (5 V being a typical value). The steps involved in this operation are shown in the flow diagram of Fig 7. A digital multiplier implemented in either hardware or software can be used to amplify the signal by a preset factor. This can be realised by suitably shifting and adding the digitised samples of the input signal. For example, shifting a sample by two bit positions to the left gives multiplication by a factor of 4. Adding the original sample to this result gives a multiplication by 5. Shifting this result one bit position to the right gives a division by 2 thus giving an amplification of the original sample by a factor of 2.5. It is to be noted that the input measurement range can be significantly enhanced (by factors from 1.1 to 1000) by incorporating this method provided the input dynamic range and the system resolution are sufficiently high to realise the extra bits needed for this processing. The steps involved in this operation are shown in the flow diagram of Fig 8. Thus the methods described above help in eliminating the effect of random high frequency noise as well as systematic low frequency noise in data acquisition systems. They also enhance the quality of measurement by using techniques such as calibration in conjunction with the averaging processes. Data processing units such as multipliers further help to enhance the measurement range. The invention is further explained in the form of following examples. These examples illustrate how digital codes corresponding to the samples are processed to remove the noise and to recover the original signal. These examples, however, should not be construed as limiting the scope of the invention. Example 1 By considering the original signal amplitude of 2.5 V dc for the signals as provided in figures 2 (b) and 3 (a) and further with reference to Table 1, it can be seen that the noise influences (i.e. changes the magnitude of) all the sample values. The digital codes corresponding to the fourth column of the table are fed to the data processor  1 block in figure 1. By estimating the average of the 10 binary numbers, which works out to 10000000 (as shown in the last row of the Table 1) and further by replacing the 10 samples by a single sample of the corresponding value i.e. 2.5 V as shown in figure 2 (c). Thus the effect of high frequency, random noise is removed. Example 2 Similarly, the digital codes corresponding to the fourth column of the table are fed to the data processor  2 block in figure 1 along with the noise frequency value, fn=100 Hz in this case. The ratio, fs/fn is calculated as 1 KHz/100 Hz = 10 and therefore the block knows that the number of samples that it has to process is 10, 20 or 30 and so on. Taking the case where the 10 digital codes shown in table are processed, the processor computes their average, which works out to 10000000 (as shown in the last row of the Table 1) and replaces the 10 samples by a single sample of the corresponding value i.e. 2.5 V as shown in figure 3 (b). Thus the effect of low frequency, systematic noise is eliminated. In the above two examples, it is noted here that the signal is over sampled by 10 times so that replacing the 10 raw samples by one processed sample does not hamper the reconstruction of the signal. Advantages 1. The hardware requirement for implementing the method of the present invention is simplified by the use of over sampling techniques, thereby eliminating the use of complex antialias filter. 2. The adoption of system calibration in conjunction with the signal processing method of the present invention reduces the effect of component nonlinearity, offset, reference error etc. 3. Measurement of range of a given signal to any desired value can be improved by implementing a digital multiplier in either hardware or software. We claim 1. A method of eliminating/reducing the effect of high frequency random noise and low frequency systematic noise in data acquisition systems, said method comprising; over sampling and digitization of analog input signals that are corrupted by high and/or low frequency noise, averaging the digitized corrupted and over sampled high frequency noise signal samples, estimating the frequency of low frequency noise signals, predetermining the sampling frequency for signals corrupted by low frequency noise, averaging said low frequency corrupted signals with number of samples being integer multiple of ratio of sampling frequency to noise frequency, deriving a single representative value free from both high and low frequency noise, and multiplying the signal samples by means of a digital multiplier and calibrating the signals to remove component nonidealities. 2. The method of claim 1, wherein the high frequency random noise is in the frequency range of 1 KHz to 100 MHz. 3. The method of claim 1, wherein the low frequency systematic noise is in thejrequency range of 0.1 mHz  50 MHz. 4. The method of claim 3, wherein the low frequency systematic noise is in the preferred frequency range of 10 Hz  1 KHz. 5. The method of claim 1, wherein the over sampling of analog inputs signals is done at a rate that is several times the required minimum sampling rate (Nyquist frequency). 6. The method of claim 5, wherein the over sampling of analog input signals is performed in the ratio of 1:2 to 1:10000 over the minimum required sampling rate of 2 times the signal frequency. 7. The method of claim 5, wherein the over sampling of analog input signals is performed in the ratio of 1:100. 8. The method of claim 1, wherein the averaging of digitized sampled signals consisting of high frequency random noise is performed to obtain a single representative value that is free from the high frequency random noise. 

048che2004claims duplicate.pdf
048che2004claims original.pdf
048che2004correspondneceothers.pdf
048che2004correspondnecepo.pdf
048che2004description(complete) duplicate.pdf
048che2004description(complete) original.pdf
Patent Number  204289  

Indian Patent Application Number  48/CHE/2004  
PG Journal Number  26/2007  
Publication Date  29Jun2007  
Grant Date  13Feb2007  
Date of Filing  22Jan2004  
Name of Patentee  M/S. DEPARTMENT OF SPACE, ISRO  
Applicant Address  ANTRIKSH BHAVAN,NEW B.E.L ROAD,BANGALORE 560094  
Inventors:


PCT International Classification Number  G10L 21/02  
PCT International Application Number  N/A  
PCT International Filing date  
PCT Conventions:
