GB2299214A  Signal processing  Google Patents
Signal processing Download PDFInfo
 Publication number
 GB2299214A GB2299214A GB9506041A GB9506041A GB2299214A GB 2299214 A GB2299214 A GB 2299214A GB 9506041 A GB9506041 A GB 9506041A GB 9506041 A GB9506041 A GB 9506041A GB 2299214 A GB2299214 A GB 2299214A
 Authority
 GB
 United Kingdom
 Prior art keywords
 data points
 threshold
 signal
 input signal
 radar
 Prior art date
 Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
 Granted
Links
 230000001131 transforming Effects 0.000 claims abstract description 6
 238000002592 echocardiography Methods 0.000 claims abstract description 4
 230000005540 biological transmission Effects 0.000 claims abstract 2
 238000001228 spectrum Methods 0.000 description 32
 238000000034 methods Methods 0.000 description 10
 238000010586 diagrams Methods 0.000 description 6
 230000003595 spectral Effects 0.000 description 6
 230000000875 corresponding Effects 0.000 description 4
 230000000694 effects Effects 0.000 description 3
 230000001427 coherent Effects 0.000 description 2
 230000000051 modifying Effects 0.000 description 2
 230000001702 transmitter Effects 0.000 description 2
 230000001154 acute Effects 0.000 description 1
 230000001668 ameliorated Effects 0.000 description 1
 230000000737 periodic Effects 0.000 description 1
 238000003672 processing method Methods 0.000 description 1
Classifications

 H—ELECTRICITY
 H01—BASIC ELECTRIC ELEMENTS
 H01Q—ANTENNAS, i.e. RADIO AERIALS
 H01Q3/00—Arrangements for changing or varying the orientation or the shape of the directional pattern of the waves radiated from an antenna or antenna system
 H01Q3/26—Arrangements for changing or varying the orientation or the shape of the directional pattern of the waves radiated from an antenna or antenna system varying the relative phase or relative amplitude of energisation between two or more active radiating elements; varying the distribution of energy across a radiating aperture
Abstract
Description
SIGNAL PROCESSING This invention relates to a signal processing method for estimating missing data points in an input signal.
The method is of particular use in a radar receiver which comprises a multi channel array and uses digital receiving beam techniques. Within such systems the failure of receiving channels increases side lobe levels. It has in the past been proposed to substitute data from a failed channel with data from an adjacent channel or to simply interpolate from working channels on either side of the failed channel. However such methods have been found beneficial only at low operating frequencies, whilst at the medium and high frequencies the side lobe levels worsened as a result of inaccurate estimates.
This invention provides a method of estimating missing data points in an input signal which comprises the steps of; a) Fourier transforming the input signal from one of the frequency or time domains to the other, b) setting a threshold so as to separate at least one unambiguous component of the transformed signal from the remaining components, c) inverse Fourier transforming the separated component to the one domain, and d) replacing the missing data points in the input signal with data from the separated component to generate an improved signal.
The improvement in side lobe levels effected in any single iteration of the method depends on ihe proportion of missing data points. In an example with 109 of the data point missing a reduction of 20dB in side lobe levels was found as compared to the case when no correction was applied.
As mentioned above, depending on the proportion of missing data, steps (a) to (d) may be iteratively repeated on the improved signal, except that in step (b) a threshold is set which is lower than that in the preceding iteration so as to separate a greater number of unambiguous components.
In this way an improved estimate of the value of the missing data can be deduced after each iteration.
Depending on the processing power available, the threshold may be lowered in step (b) in each iteration until it is no longer possible to separate any further unambiguous components.
Preferably the threshold in step (b) is set by deducing an average value, i.e. a mean or median value, of the amplitude of the transformed signal and adding a margin to that average value.
If it is desired to reduce the number of iterations the threshold may be set by (i) deducing an average amplitude of the transform signal within a window (ii) adding a margin to generate a threshold within that window; (iii) moving the window and repeating steps (i) and (ii).
In this way a threshold may be set which more closely follows the spectrum and thus is capable of more easily separating out unambiguous components of the transformed signal.
Where a number of iterations are performed, the missing or estimated data points may simply be replaced by that data estimated by the latest iteration. However in an alternative method the value of the estimated data points may be weighted by a proportion of the value estimated in the preceding iteration. For example the value used may comprise one and a half times the value of an estimated data point generated by a current iteration subtracted from which is a half times the value of that estimated in the preceding iteration.
Although the invention has been described with reference to the generation of missing data points caused by channel failure in a multi channel radar receiving array, the same techniques can be used for a wide variety of purposes. For example an array could be designed with a smaller number of channels than would conventionally be required and a method according to the invention employed so as to create a "virtual" antenna array of larger size with a consequential saving in terms of hardware, albeit with a rise in the cost of processing. The advantage of creating new data points outside of the available data set is that the antenna aperture can be virtually enlarged, increasing the received bandwidth or the coherent integration time, all of which lead to enhanced resolution in the respective domain.
An alternative use of the method is when a high amplitude noise burst needs to be removed in the time domain, inevitably with the underlying signal, and when the original data must be recreated or restored. For example in HF radar, high amplitude noise bursts during local thunderstorms caused by lightning strikes can corrupt the received radar echoes and, for a finite but possibly vital period. can render the radar blind. In one use of the method the input signal is derived from a radar receiver, and the missing data points correspond to known periods in the time domain for which the receiver radar echo was blinded by a thunderstorm.
In order that the invention may be well understood, an embodiment thereof will now be described by way of example with reference to the accompanying drawings, in which Figure 1 is a diagram schematically illustrating a multi channel radar receiving array and showing how in azimuth signals arise in the frequency domain; Figure 2a to 2d are schematic time and frequency domain diagrams illustrating use of the method; Figure 3 is a flow chart showing the preiteration algorithm used for generating a threshold for use in a method according to the invention; Figure 4 is a flow chart showing the method according to the invention; Figure 5 is a diagram illustrating the spectrum of an example input test signal; Figure 6 is a diagram illustrating the spectrum of Figure 4, after a large proportion of the data points have been removed;; Figure 7 illustrates a spectrum of Figure 5 after a single iteration of the method; Figure 8 illustrates the spectrum after 17 iterations; Figure 9 illustrates of a running window method for generating a threshold level; and Figures 10 to 12 are time and frequency domain diagrams illustrating examples where the method may be employed.
Referring to Figure 1 a plane wave 1 is incident at an angle 6 onto a radar receiver 2 having a multiplicity of separate channels 3, some of which, e.g. 33, a,, have failed. It can be appreciated that the plane wave 1 strikes succeeding channels with a tirne or phase delay p. This phase delay is treated as a frequency, thus a spectral peak occurs in azimuth at an angle corresponding to the angle (3,the more acute the angle leading to a higher frequency signal. The effect of the finite number of data points, and indeed the missing data points, is to create uncertainty in the position of the main spectral peak S leading to side lobes Si on either side.
The above mentioned case is much simplified. In a practical system each channel is used simultaneously to retrieve information for each of a number of overlapping finger beams in azimuth and also in range and doppler.
For use of the method the input data is a finite set of real or complex numbers from which some data points are missing. The positions of the missing data are known, as in the example of the failed channels in Figure 1. It is assumed that the input signal is composed from the sum of periodic functions, sine or cosine waves, of arbitrary frequencies. However it is a prerequisite that the number of components (sine waves) is significantly less than the number of data points in the set. The randomly distributed missing or zero value points will, in effect, modulate the input signal with noise (multiplicative noise). Figure 2A shows schematically an output signal with time which could be generated by the multi channel array 2 shown in Figure 1. except that targets striking the array from a variety of different azimuth angles give rise to a multiplicity of different frequency components. As an example, noise is generated in time slots corresponding to failed channels 33 and 3n5 As a first step the time domain signal shown in Figure 2A is Fourier transformed into the frequency domain as shown in Figure 2B. Here the sine waves appear as spectral lines whilst the missing data produce an incoherent noise like spectrum. It is necessary to weight the input data with a window function, in order to produce a lower side lobe level than the permitted noise after data restoration.It is important that the peaks of the noise are less than the largest spectrum line in the input so that at least one component of the signal can be filtered out. Thus referring to Figure 2B a threshold level T is set such that at least one spectrum component fi lies above that threshold value, the remaining spectral components being discarded.
An inverse Fourier transform is then performed on the f, component leading to, for example, the sine wave shown in Figure 2C. From this an estimate for the value of the data signal at time slots corresponding to channel 33 and 3rF5 can be taken, the estimates now being referred to as 33' and 3n 1 respectively. These values are then reinserted into the waveform shown in Figure 2A so as to generate an improved time domain signal, as shown in Figure 2D.
It should be emphasised that signals shown in Figures 2A to 2D are purely schematic.
However Figure 2D is intended to show that in the "patched up" or updated set, the large signals are no longer modulated by the randomly distributed zeros and thus the overall noise power is reduced. Should further reduction of the noise be required, then the above steps can be iteratively repeated. Each iteration involves a reestimation of the noise level and a corresponding adjustment of the threshold level. The noise decreases monitonically with the ileralions. as more and more signal components contribute to the estimated values replacing the missing points and asymptotically approach a minimum. This noise floor is determined by parameters such as the ratio of the number of sine wave components to the number of data points, the external noise and the side lobe level of the window function used.Iterations may either be terminated after a fixed number of traversals or adaptively at that point where the noise no longer decreases in subsequent iterations.
Figures 3 and 4 show two flow charts illustrating the sequence of operations required in the method. Figure 3 shows the preiteration computation to estimate the first threshold level and to set up the arrays and pass parameters for the subsequent iterations. Firstly a window is applied to the input data. This weighted data is then kept in memory and updated in subsequent iterations. After that a complex Fourier transform is computed using conventional radix 2 or a mixed radix FFT technique. If the number of input data points happens to be a prime number, then a slower Discrete Fourier Transform DFT algorithm must be used instead. The vector of the complex spectrum is kept in the memory for passing to the iteration sub programme. After that the magnitude of the spectrum is computed and normalised to its maximum.This normalised vector of magnitudes is again a pass parameter to be updated during the iterations.
An estimated initial threshold (say between 0 and lOdB) can be used. After that those components of the signal which exceed this initial value are removed and the mean of the remaining signal plus any noise is then computed. This mean is then multiplied with a margin (or a margin is added if logorithmic units are used) to obtain the first estimated threshold.
Figure 4 is a flow diagram illustrating the functions performed within a single loop. In the first stage the current threshold and normalised spectrum magnitude is used to generate a new data set This contains only those complex spectrum lines whose magnitude exceeds the current threshold. Below the threshold the signal plus noise is rejected, i.e. set to zero. After that an inverse Fourier transform is performed on the chosen complex spectrum points to obtain an estimate of the input signal (when the noise below current threshold is excluded). From this estimate those points which are missing from the original input signal are selected and substituted in to the current input.
Once the input is ameliorated, the reduced noise level is estimated and a new threshold is set up using an identical method as in the preiteration step i.e. the new complex spectrum (CFFT) is computed. After that the old spectrum is updated or replaced. A normalised magnitude is then computed and this is then used update or replace the preceding normalised magnitude. The signal above the threshold value is rejected and the mean noise computed.
After that a margin is applied to the mean noise to obtain the new threshold. The new threshold is then compared with the preceding threshold and if the ratio is greater than a given constant then another iteration is performed. If the ratio, or the measure of improvement is less than expected, then computation is stopped and the latest updated input data is output. As an alternative a fixed number of iterations for terminating the computation can be provided.
A more detailed program for implementing the method will now be described with reference to Figures 5 to 8. Figure 5 shows the spectrum of several sine waves of different frequencies and amplitudes. Missing data is simulated by multiplying some of the data points with zeros, which are distributed randomly within one batch of data or waveform period. As will be described below 125 data points are removed from 240 data points  i.e. a reduction of some 52it. As has been described previously the method involves the application of forward and inverse Fourier transform pairs and a thresholding operation between the two. The process for generating the sample test signal is described below.
INPUT PARAMETERS FOR THE TEST WAVEFORM Number of data points N := 240 Number of sine waves in test waveform.... n:=11 #adB Amplitude difference (regular steps) #adB :=6 #a:=1020 #a=0.501 between sine waves Number of missing data points M:=125 mar dB/20 Margin of thresholding mar dB:=11 mar:= 10 mar = 3.548 DERIVED PARAMETERS Index numbers of data points i:=O..N1 Index numbers of frequencies k :=O..n 1 Index number of M missing data points I :=0..M I Randomising location of missing data r1:=floor(md(N I) + .5) r :=son(r) Generating a vector of 1 and Os. This will multiply the test signal so that the Os in random locations will represent the missing data. <img class="EMIRef" id="02694674100090001" />
Spacing between frequencies s:=N S=20 n+1 A vector listing the randomly spaced frequencies in the test signal <img class="EMIRef" id="02694674100090002" />
Amplitude vector of the sine waves ao:=1 ak+1:=ak.#a Reading in 240 DolphChebyshev weights COMPUTATIONS Generating equation for n sine waves at n frequencies and n amplitudes <img class="EMIRef" id="02694674100090003" />
Spectrum of the weighted " test signal <img class="EMIRef" id="02694674100090004" />
The effect on the spectrum of removing some 52% of the data points is illustrated in Figure 6 where it can be seen that very few spectral peaks appear unambiguous. the remainder being buried within noise. The preiteration computations are described below.
PRE ITERATION COMPUTATIONS Multiplying the decimated test signal <img class="EMIRef" id="02694674100100001" />
with a 90 dB DolphChebyshev window Tb Selecting an initial threshold Th 3 Th =1020 Th =0.708 (say3d8).
Fourier transform the weighted test signal Tw :=CFFT(tw) Normalised magnitude of the spectrum <img class="EMIRef" id="02694674100100002" />
Remove peaks above threshold <img class="EMIRef" id="02694674100100003" />
Compute the Mean of the remaining noise Nm :=mean(Tw) The mean multiplied by a "margin" Th :=Nm.mar will provide a new threshold It can be seen that an initial threshold level is set at lOdB.
After that the first iteration is performed as is described below.
FIRST ITERATION Cut out noise below new threshold Inverse FFT of thresholded spectrum From this "improved" signal select the data points missing in the original test signal and use them for replacement FFT of the patched up signal Normalised magnitude of the new improved spectrum Remove peaks above threshold <img class="EMIRef" id="02694674100100004" />
Mean c;' 'he remaining noise Nni :=mean(Twt) Nm =0.027 New threshold from the mean noise na :=Nmmar n. =0.097 Once transformed back into the frequency domain the spectrum has the appearance of that shown in Figure 7 where it can be seen that the level of the noise has been much reduced.
A further threshold is then set and the process repeated. After seventeen similar iterations, reducing the threshold level each time, the spectrum has the appearance of that shown in Figure 8. Comparing Figure 8 to Figure 5 it can be seen that, in spite of 52% of the data points being missing, the spectrum is faithful to the input test signal. The details of the seventeenth iteration are described below.
SEVENTEENTH ITERATION Cut out noise below new threshold <img class="EMIRef" id="02694674100110001" />
Inverse FFT of thresholded spectrum tw th:=ICFFT(Twth) From this improved signal select the data points missing in the original test signal and use them <img class="EMIRef" id="02694674100110002" />
for replacement FFT of the patched up signal Tw :=CFFl(tw) Normaiised magnitude of the new improved spectrum <img class="EMIRef" id="02694674100110003" />
Remove peaks above threshold <img class="EMIRef" id="02694674100110004" />
Mean of the remaining noise Nm :=mean(Twt) Nm =3.087108 New threshold from the mean noise Th :=Nm.mar Th = 1.0951410 In alternative variations of the method, the threshold value may be set predictively using a running window technique.Referring to Figure 9 a window W is moved across the Spectrum.
For each position of the window the average amplitude of the transformed signal is deduced and a margin added to generate a threshold T. By using this technique the threshold can more closely follow the contours of the spectrum and so separate more unambiguous spectral components during each iteration, thus reducing the number of iterations required.
As described, the missing or estimated data points are simply replaced by that data estimated by the latest iteration. However in an alternative method the value of the estimated data points may be weighted by a proportion of the value estimated in a preceding iteration. For example the value used may comprise one and a half times the value of an estimated data point generated by a current iteration subtracted from which is a one half times the value of that estimated in the preceding iteration.
Although the invention has been described with reference to the generation of missing data points caused by channel failure in a multi channel radar receiving array, the same techniques can be used for a wide variety of purposes. For example an array could be designed with a smaller number of channels than would be conventionally required and a method according to the invention employed so as to create a "virtual" antenna array of larger size with a consequential saving in terms of hardware. The advantage of creating new data points outside of the available data set is that the antenna aperture can be virtually enlarged, increasing the received bandwidth or the coherent integration time, all of which lead to enhanced resolution in the respective domain. Another use is explained with reference to Figure 10 showing a high frequency spectrum extending between 3 and 30MHZ.Certain parts of the spectrum C are used for communication purposes by other users. Conventionally only a part of the spectrum located between a single pair of communications bands C can be used for radar purposes. However using a technique according to the invention all parts of the spectrum identified by R may be used for radar purposes, and that part of the signal which would have been occupied by the communications bands C is generated using a method according to the invention. A high bandwidth signal allows hit resolution, and that if returning signals occupying the bands C weie sinlplv ignored then vilis would lead to large range side lobes.
An alternative use of the method is when a high amplitude noise burst needs to be removed in the time domain, inevitably with the underlying signal and when the original data must be recreated or restored. Figure 11 shows a received signal varying with time in an HF radar, a high amplitude noise burst B caused by a lightning strike can corrupt the received radar echo and for a finite but possibly vital period render the radar blind. The method according to the invention can be used to estimate the data which would have been received during this period of time. A similar technique may be used to restore scratched gramophone records, for example.
A monostatic radar operates by turning on a radar receiver, when the transmitter is turned off and vice versa. Figure 11 indicates the periods of time for which the receiver RX is switched on. A method according to the invention can be used to generate data which would have been received during the time that the transmitter TX is on, and the receiver RX is off, allowing the advantages previously described.
Claims (14)
Priority Applications (1)
Application Number  Priority Date  Filing Date  Title 

GB9506041A GB2299214B (en)  19950324  19950324  Signal processing 
Applications Claiming Priority (3)
Application Number  Priority Date  Filing Date  Title 

GB9506041A GB2299214B (en)  19950324  19950324  Signal processing 
AU51519/96A AU5151996A (en)  19950324  19960322  Method of estimating missing data points at known positions in an input signal 
PCT/GB1996/000691 WO1996030780A1 (en)  19950324  19960322  Method of estimating missing data points at known positions in an input signal 
Publications (3)
Publication Number  Publication Date 

GB9506041D0 GB9506041D0 (en)  19950510 
GB2299214A true GB2299214A (en)  19960925 
GB2299214B GB2299214B (en)  19990915 
Family
ID=10771826
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

GB9506041A Expired  Fee Related GB2299214B (en)  19950324  19950324  Signal processing 
Country Status (3)
Country  Link 

AU (1)  AU5151996A (en) 
GB (1)  GB2299214B (en) 
WO (1)  WO1996030780A1 (en) 
Cited By (3)
Publication number  Priority date  Publication date  Assignee  Title 

GB2458764A (en) *  20080318  20091007  Univ Manchester Metropolitan  Remote detection and measurement of a metallic or dielectric object 
US8103604B2 (en)  20080318  20120124  Manchester Metropolitan University  Remote detection and measurement of objects 
US9335407B2 (en)  20090917  20160510  Radio Physics Solutions Ltd  Detection of objects 
Families Citing this family (1)
Publication number  Priority date  Publication date  Assignee  Title 

JP2000193524A (en) *  19981228  20000714  Nippon Sanso Corp  Method and system for measuring light absorption spectrum 
Citations (1)
Publication number  Priority date  Publication date  Assignee  Title 

WO1995028766A1 (en) *  19940415  19951026  Gas Research Institute  Impulsive noise cancellation method and apparatus for low sample rate data 
Family Cites Families (5)
Publication number  Priority date  Publication date  Assignee  Title 

US4884248A (en) *  19880125  19891128  Mobil Oil Corporation  Method of restoring seismic data 
JPH0340181A (en) *  19890707  19910220  Sanyo Electric Co Ltd  Image correcting device 
JPH0349734A (en) *  19890718  19910304  Sanyo Electric Co Ltd  Restoring method for deteriorated mri image 
JP2636486B2 (en) *  19900925  19970730  日本電気株式会社  Image signal processing device 
SE503650C2 (en) *  19941110  19960722  Hans Hellsten  Put that on a radar improve radar data 

1995
 19950324 GB GB9506041A patent/GB2299214B/en not_active Expired  Fee Related

1996
 19960322 AU AU51519/96A patent/AU5151996A/en not_active Abandoned
 19960322 WO PCT/GB1996/000691 patent/WO1996030780A1/en active Application Filing
Patent Citations (1)
Publication number  Priority date  Publication date  Assignee  Title 

WO1995028766A1 (en) *  19940415  19951026  Gas Research Institute  Impulsive noise cancellation method and apparatus for low sample rate data 
Cited By (8)
Publication number  Priority date  Publication date  Assignee  Title 

GB2458764A (en) *  20080318  20091007  Univ Manchester Metropolitan  Remote detection and measurement of a metallic or dielectric object 
GB2458764B (en) *  20080318  20110316  Univ Manchester Metropolitan  Remote detection and measurement of objects 
US8103604B2 (en)  20080318  20120124  Manchester Metropolitan University  Remote detection and measurement of objects 
US9182481B2 (en)  20080318  20151110  Radio Physics Solutions Ltd.  Remote detection and measurement of objects 
US9746552B2 (en)  20080318  20170829  Radio Physics Solutions Ltd.  Remote detection and measurement of objects 
US10466351B2 (en)  20080318  20191105  Radio Physics Solutions Ltd.  Remote detection and measurement of objects 
US9335407B2 (en)  20090917  20160510  Radio Physics Solutions Ltd  Detection of objects 
US10067226B2 (en)  20090917  20180904  Radio Physics Solutions, Ltd.  Detection of objects 
Also Published As
Publication number  Publication date 

GB9506041D0 (en)  19950510 
AU5151996A (en)  19961016 
WO1996030780A1 (en)  19961003 
GB2299214B (en)  19990915 
Similar Documents
Publication  Publication Date  Title 

Dowling  Acoustic pulse compression using passive phase‐conjugate processing  
US10338217B2 (en)  Wideband sonar receiver and sonar signal processing algorithms  
Hassab et al.  Optimum estimation of time delay by a generalized correlator  
Goldstein et al.  Multistage partially adaptive STAP CFAR detection algorithm  
EP0396662B1 (en)  Adaptive waveform radar  
AU2003213763B2 (en)  A noise suppression system and method for phasedarray based systems  
Sahr et al.  The Manastash Ridge radar: A passive bistatic radar for upper atmospheric radio science  
US6362760B2 (en)  Method and apparatus for acquiring wideband pseudorandom noise encoded waveforms  
US6636810B1 (en)  Highresolution Radon transform for processing seismic data  
KR100241498B1 (en)  Digital signal coder  
Gupta  Highresolution radar imaging using 2D linear prediction  
Ching et al.  On wavelet denoising and its applications to time delay estimation  
US9806914B1 (en)  Successive signal interference mitigation  
EP1625420B1 (en)  Method for signaltonoise ratio enhancement of seismic data using amplitude noise attenuation  
US5127051A (en)  Adaptive modem for varying communication channel  
AU2003204265B2 (en)  Method for processing seismic data to attenuate multiples  
JP4751986B2 (en)  OFDM signal synchronization  
CA2625683C (en)  Sonar system and method providing low probability of impact on marine mammals  
Farina et al.  Coherent radar detection in lognormal clutter  
US7094204B2 (en)  Coded excitation imaging for use with bipolar, unipolar and other waveforms  
US5018088A (en)  Adaptive locallyoptimum detection signal processor and processing methods  
AU746030B2 (en)  A method of and an apparatus for training tap coefficients of an adaptive equalizer  
US7106250B2 (en)  Robust predictive deconvolution system and method  
Lucky et al.  An Automatic Equalizer for General‐Purpose Communication Channels  
Siggia et al.  Gaussian model adaptive processing (GMAP) for improved ground clutter cancellation and moment calculation 
Legal Events
Date  Code  Title  Description 

PCNP  Patent ceased through nonpayment of renewal fee 
Effective date: 20100324 