SE507857C2 - Disturbance eliminating method e.g. pulses and linear chirps in radar of frequency modulated continuous wave type - Google Patents

Abstract

The method forms one wave for each target, the wave frequency, amplitude and phase containing target information. The frequency for a wave is dependent upon the distance to a target associated with the wave. The floating signal is subjected to time-frequency division for time-local frequency division of signals. Disturbances are detected and eliminated separately in each frequency band. Time signals freed from disturbances and their discrete Fourier transformation are calculated from time frequency division freed from disturbances. The time frequency division is effected by means of short time Fourier transformation. The disturbance detection is carried out in each frequency band with methods for detecting disturbances of short validity. The detection of linear chirp disturbances and pulse disturbances is carried out with methods for detecting straight lines in pictures. The radar employed has a transmitter (1) and receiver (2). An antenna (3) is connected to these two via circulator (4). In the transmitter is an oscillator control device (5) coupled to oscillator (6) with variable frequency. A frequency sweep from oscillator control device controls oscillator so that signal with periodically varying frequency is generated.

Description

507 857 linear frequency sweep. Therefore, since the received and transmitted frequency sweeps are parallel, time-shifted by a time equal to the running time, for a mål xt target, the difference in frequency between transmitted and received signal becomes constant. This constant frequency difference is given by the product between the travel time to the target and the slope of the frequency sweep expressed in frequency per unit time.

The signal processing in a linear FMCW radar essentially consists of mixing transmitted and received signals, so that the difference signal (the beat signal) is generated. This signal is the sum of a number of sine waves, where each sine wave represents a radar target. The sine waves have different frequency, amplitude and phase positions according to the principles that large amplitude corresponds to strong target, high frequency corresponds to target at large distance. The Doppler effect (mutual speed) mainly affects the phase modes.

To determine which targets including size and relative velocity are observed, the difference signal is frequency analyzed. The frequency analysis is advantageously done digitally by allowing the difference signal to pass an anti-alias filter and sampled at a constant sampling rate, after which the sampled signal is multiplied by a window function with the task of reducing the signal amplitude at the beginning and end of the sampling period and sent to a signal processor. a discrete Fourier transform, DFT, usually with a fast algorithm, so-called FFT, Fast Fourier Transform.

The Fourier transform is generally complex but has certain symmetry properties for a real time signal (difference signal). To be able to use FFT algorithms, the number of samples is usually chosen as a two power (256, 512, 1024, ...). 256 samples give 256 FFT coefficients, but if the signal is real, the symmetry properties give that of these 256 values only 128 (actually 129) are independent.

By Fourier transform, eg by FFl ", the signal is divided into a number of discrete frequency components, sines. Each frequency corresponds to a distance as above. The amount of a complex FFT coefficient is a measure of the radar measuring area (received power) for the target in the corresponding frequency slot The distance signal processing in the system is done digitally on the calculated FFT coefficients 507 857 It can be shown that the nominal width of a distance gap is inversely proportional to the frequency swing for the linear FMCW sweep during the sampling time.For a distance resolution of 1 m, the frequency swing 150 MHz is required.To change the distance resolution, for example, the slope of the frequency sweep can be changed while maintaining the same constant sampling time.

The sampling rate limits the frequencies of the attenuation signal that can be studied and thus the total observed distance range. The width of this "pay band", which is parallel to the linear FMCW sweep, is often less than 1 MHz.

A linear FMCW radar can be disturbed if it receives signals other than its own transmitted signal re-reflected from different targets. The radar can be disturbed by other radars, including pulse radars, pulse compression radars and other FMCW radars.

A pulse during the sampling time is very short in the time domain but thus very broadband in the frequency domain. A strong interference pulse affects only a few samples of the hovering signal but can affect all frequency slots in the Fourier transform. The "noise level" in the Fourier transform seems to be raised, whereby small targets can be masked by the disturbance.

A very common form of interference is a so-called chirp (chirp (eng) = chirping), as the interfering waveform moves with linear frequency through the FMCW radam's useful band. Such chirps are generated by a pulse compression radar, but also by another FMCW radar if its transmitted waveform during the rest or return phase enters the useful band of its own radar during the sampling time of its own radar. The third phase, the linear frequency sweep, can also generate a chirp disturbance if the frequency sweep of the interfering radar has a slope other than the frequency sweep of its own radar, for example because the interfering radar has a different distance resolution.

A disturbance in the form of a linear chirp is always broadband in frequency, but can also have a considerable extent in time, disturbing the entire FFT and affecting a very large part of the sampled time signal. 507 857 4 There are also short chirps, which can hardly be distinguished from pulse disturbances.

The chirps caused by the idle or return phase of interfering FMCW radar are of this type.

Short perturbations such as short pulses or fast chirps can generally be detected and eliminated in the sampled time signal, and an undisturbed FFT can then generally be reconstructed. On the other hand, a chirp interference with a large extent in both the time and Fourier domains cannot be eliminated by any simple manipulation of the time signal, without it having negative consequences for the FFT.

According to the present invention, a method for eliminating interference with FMCW-type radars is proposed which allows even interference to a large extent in both the time and Fourier domains to be eliminated.

The method according to the invention is characterized in that the attenuation signal is subjected to time-frequency division for time-local frequency division of the signal, that the disturbance is detected and that the disturbance is eliminated separately in each frequency band separately, after which the time-free signal, the interference-free time-frequency division.

The sampled hover signal, the time signal, lies entirely in the time domain.

The sample gives a resolution in time, but no resolution at all in frequency.

The FFT is a description of the same signal in the Fourier domain. FFf-n gives a good resolution in frequency, but no resolution at all in time. A disturbance, such as chirp disturbance, during part of the time signal is poorly seen in the Fourier domain. Information about the state of the disturbance is in fact mainly in the phases of the complex FFT values and not in the amounts.

A so-called time-frequency division makes it possible to have a certain (rough) resolution of the signal in the time domain and in the Fourier domain at the same time. A known time-frequency division is the Wigner-Ville transform, which is a so-called quadratic transform and thus creates false cross-modulation products, see Mayer, Wavelets, Algorithms & Applications, SIAM, Philadelphia, 1993. Another known time-frequency division is the so-called the wavelet transform, see Mayers' above-mentioned book or Rioul / Vetterli, Wavelets and Signal Processing, IEEE Signal Processing Magazine, October 1991, which makes a "musical" 507 857 frequency division. The frequency division is done in different scales, "octaves".

For high frequencies, the frequency resolution (expressed in Hz) is coarser but the time resolution is finer.

For application to interference suppression in FMCW radars, however, the simplest time / frequency division Short Time Fourier Transform, STFT, as described in the Rioul / Vetterli reference above, is mainly proposed.

The time signal is divided into short sections that can overlap. Each signal section is multiplied by a window function and discrete Fourier transform is calculated. After eliminating interference in each frequency band separately, the original time signal is calculated from the STFT. The STFT can then advantageously contain redundant (overlapping) information.

In this context, it is worth noting that the FMCW radar is the only common type of radar where a target responds to a standing wave with a certain frequency and thus meets the conditions to be able to apply the normal Fourier analysis with bandpass filter or DFT (FFI).

A form of signal processing in the time-frequency plane suitable for eg active sensors is known from EP 0 557 660, A2. In this case, an incoming broadband signal is separated by a bank of bandpass filters into a number of signals each assigned to a separate channel. After gain control and half-wave rectification, coherent components of the signals are integrated. In this way, a target can be detected simultaneously in olika your different frequencies and the best from each frequency is utilized. According to the mentioned EP document, it is thus avoided to use disturbed parts. On the other hand, in contrast to the claimed invention, there is no elimination of interference so that disturbed parts of the signal can be used.

Advantageously, the noise detection is performed in each frequency band with methods for detecting disturbances of short duration.

According to an advantageous embodiment of the method, the detection of linear chirp disturbances and pulse disturbances is performed with methods for detecting straight lines in images, for example so that disturbance patterns in the form of straight lines deviating from time axis parallel lines are identified, the identified lines are related to frequency bands and that the interference 507 857 6 is eliminated separately in each affected frequency band. Such methods are known from image processing, see for example Gonzalez / Woods, Digital Image Processing, Addison-Wesley 1992. In this case, Houghtransform can be used for detecting the straight lines.

According to another advantageous embodiment of the method according to the invention, the hovering signal in connection with the time-frequency division is narrowed in narrow frequency bands by the signal in order to increase the sensitivity of the detection.

The filter can then be determined by means of adaptive methods. According to a favorable embodiment, filters are then applied to one or more of the narrow-band frequency slots in the narrow-band frequency band of the time-frequency division.

According to a further advantageous embodiment of the method according to the invention, the hovering signal or the useful signal is reconstructed after interference elimination by extrapolation from undisturbed samples, in one or more of the narrowband frequency slots in the narrowband frequency band of the time-frequency division.

STFT time-frequency division for interference detection, interference elimination and synthesis of the useful signal has many advantages, especially in chirp interference. The benefits generally depend on two features. One is that a chirp disturbance in each frequency slot in the STFT is of short duration and thus can be detected / eliminated with the same methods as, for example, pulse disturbances. The second is that the chirp interference is narrowband in each frequency slot in the STFT and can therefore be described (zeroed / extrapolated) with simple polynomials of a known structure.

The method according to the invention will be described in more detail below with reference to the accompanying figures, where: Figure 1 schematically shows the principle of how a linear FMCW radar works.

Figure 2 shows examples of suitable frequency sweeps in a time-frequency diagram.

Figure 3 shows samples of a simulated FMCW hover signal with Gaussian noise and a disturbance.

Figure 4 shows the absolute amount of the FFT-n for the hovering signal according to Figure 3. 507 857 Figure 5 shows the result of a time-frequency analysis of the hovering signal according to Figure 3.

Figure 6 shows the absolute amount of the FFT for a hover signal according to Figure 3 without interference.

The radam shown in Figure 1 comprises a transmitter part 1 and a receiver part 2.

An antenna 3 is connected to the transmitter part and the receiver part via a circulator 4. The transmitter part includes an oscillator control device 5 connected to an oscillator 6 with variable frequency. Frequency sweep from the oscillator control device 5 controls the oscillator 6 so that a signal with periodically varying frequency is generated, which signal is transmitted via a directional coupler 7 and the circulator 4 on the antenna 3. The period of a frequency sweep, see Figure 2, has essentially three parts in the form of a constant rest frequency 30, a linear frequency sweep 31 and a fast return 32 to the rest frequency.

Oscillator 6 operates with advantage in the Gigahertz area, eg 77 GHz. A reflected signal received by the antenna 3 is transmitted via the circulator to a mixer 8, where the reflected signal is mixed with the transmitted signal. After amplification in the amplifier 9 and filtering in the filter 10, a difference signal or hover signal is obtained which forms the basis for the further signal processing for detecting and eliminating interference and synthesis of the undisturbed useful signal in a processor block 11 which may also contain a so-called FFI 'processor 11' .

An example of how the time-frequency division can provide the possibility of analyzing a disturbed difference signal is shown in Figures 3-6.

Figure 3 then shows 1024 samples of an FMCW hover signal (time signal), which is simulated as a number of sine / cosine signals + Gaussian noise + a disturbance. With the naked eye, it is difficult to accurately locate and characterize the disturbance.

In Figure 4, where the absolute amount of the FFT for the hovering signal according to Figure 3 is shown, four distinct peaks 12, 13, 14 and 15 are seen protruding above a high noise mat. Each top 12-15 corresponds to one goal. Figure 4 also does not provide any possibility to characterize the disturbance. 507 857 Figure 5 shows the result of a time-frequency analysis of the hovering signal according to Figure 3. STFT of 64 samples at a time with overlap has been used.

Figure 5 shows directly through the V 16, which appears in the middle of the signal, that the disturbance is a linear chirp. The explanation for the chirp becoming a V and not just a line is that the FFT analysis used cannot distinguish between positive and negative frequencies. The dominant peaks in the spectrum appear as horizontal bands 17-21 corresponding to standing sine waves with constant frequency. In Figure 5, the frequency resolution is quite coarse and the two peaks 14, 15 in Figure 4 partially merge into a single wide band 19, 20. A horizontal band 21 at approximately 0.8 * The Nyqvist frequency does not correspond to any peak in Figure 4.

When studying the undisturbed simulated signal without chirps, shown in Figure 6 as the absolute amount of the FFT, a fifth peak 22 belonging to the band 21 corresponding to a fifth target appears. In Figure 4, this peak has been completely submerged by the disturbance.

One of the major advantages of time-frequency division (STFT) for interference elimination in FMCW signals emerges when comparing Figure 5 to Figure 3. In Figure 3, the interference is long. The length of the interference corresponds to the projection of the V on the horizontal time axis in Figure 5. In each frequency slot in Figure 5, on the other hand, the interference is relatively short.

By treating each frequency slot separately in an STFT, chirp disturbances can be detected and eliminated with the same methods used for short-term disturbances, pulse disturbances. The V 16 in Figure 5 can be detected and eliminated and the horizontal bands, the useful signals, can be reconstructed, after which reconstructed undisturbed time signal and undisturbed FFI can be calculated from the STFT. This is the principle behind the method according to the invention.

Figure 5 shows, as previously stated, that a chirp disorder appears as a V, denoted by 16, in an STFT division. In the same way, a pulse disturbance appears as a vertical line located in time but broadband. The useful signals, on the other hand, are horizontal lines. It is therefore possible to detect disturbances of the pulse or linear chirp type by looking in a STFT image for lines that are not parallel to the time axis. Such 507 857 methods are known from image processing, see for example the Gonzalez / Woods reference as above. A method suitable in this context is described in Chapter 7 of this reference and is based on the so-called Houghtransform.

In the following, the principles of filtering the useful signal are discussed in more detail.

The useful signal in an FMCW radar, ie the signal that corresponds to real targets, is a sum of sine waves. A signal consisting of a single sine wave, sampled at a constant frequency, has a simple linear relationship between the samples. Assume that the signal can be written as sin (m * t + q>).

Thus, between two samples, the phase angle of the sine wave changes with the angle cnT = 6, where T is the sampling interval. According to the trigonometric identity sin (cx + 6) + sin (a-6) = 2 * cos (6) * sin (oi) then applies to three successive samples of the signal that: x (n + 1) + x (nl) = 2 * cos (6) * X (n) Note that this applies regardless of the amplitude of the signal. This linear relationship can be interpreted in different ways: a) If the signal is run through a FlR filter (Einite lmpulse Response) with the coefficients [1 -2 * cos (6) 1], the output signal y of the filter will be identical O: y (n ) = x (n) - 2 * cos (6) * x (n-1) + x (n-2) It is thus possible to strongly suppress the signal with a simple FIR filter with constant coefficients. b) If the relation is instead written: x (n + 1) = 2 * cos (6) * x (n) - x (n-1) 507 857 10 it is seen that the next sample can be predicted well by a linear combination of the immediately preceding sample .

For a signal consisting of fl your sine waves with distinct frequencies, the corresponding filter can be created by multiplying the second-order FIR filter. A signal which is the sum of four different sine waves, ie an FMCW signal with four strong targets, can thus be zeroed exactly by an FIR filter of order 8, and a filter can be predicted linearly from the previous 8.

For a general FMCW signal, these relationships become approximate, but the following can be said to apply: 1. There is a great possibility of strongly suppressing an FMCW signal by a suitable linear FIR filter of a suitable order. 2. There is great potential for linearly predicting an FMCW signal through an appropriate linear relationship of appropriate order.

The application of paragraph 1 is that the sensitivity of a disturbance detection is greatly increased if the useful signal is filtered in an appropriate manner. In Figure 5, this corresponds to the horizontal bands being filtered out. Then only disturbances remain against a weak background. This makes it possible to detect disturbances whose amplitude is much lower than that of the useful signal, for example a signal which is absolutely invisible when analyzing the amplitudes in Figure 3 but which nevertheless raises the noise mat in the FPT in Figure 6.

Point 2 provides an opportunity to interpolate the useful signal past a short disturbed section, which will be described in more detail later.

A "suitable" filter can be calculated in different ways, or calculated as an adaptive filter. Both issues according to points 1 and 2 above are known from adaptive signal processing, see for example Haykin, Adaptive Filter Theory, 2nd Ed., Prentice-Hall 1991. The coefficients can be determined with common algorithms, eg LMS, standardized LMS, RLS etc, see especially Chapters 9 and 13 of the latter reference. 507 857 11 In adaptive determination of an .lter, it can often be used that the radar antenna rotates, but only a fraction of a beam width, since the previous FMCW frequency sweep. Consequently, the dominant sine waves in the signals from two consecutive FMCW sweeps have almost the same frequency and almost the same amplitude. The start values for the adaptation can therefore advantageously be selected as final values from the adaptation during the previous FMCW sweep.

An important observation is also that in each frequency slot in an STFT division of the signal, the filters become very simple. In each frequency slot, the signal is narrowband and the center frequency of the slot is known. This means that the phase shift between two successive samples is known, and already a second order filter has a very good effect.

In the following, the synthesis of the useful signal is discussed.

A common method of interference suppression is to detect a disturbance, eg a pulse, by the signal amplitude being unusually large, and then the clipping signal occurs, preferably to level 0. This can in itself eliminate the interference, but affects the FFT on an unfortunate way in that the useful signal is also affected.

The prerequisite for an FFI "is that the signal is equidistantly sampled for a suitable time. Cutting of the signal removes samples. It can be said that the usefulness time base is affected. A consequence is that distinct goals are broadened in the Fourier domain, which can reduce the resolution.

A much more advantageous method is to follow up the noise elimination with a synthesis of the useful signal. Here, point 2 as above can be advantageously applied. The synthesis can consist of an extrapolation (one-sided) or interpolation (two-sided) of the signal based on undisturbed values. Such a synthesis has been found to result in a dramatic improvement in the reconstruction of the undisturbed FMCW signal and its FFT. 507 857 12 As mentioned above, the interpolation / extrapolation polynomial can be determined adaptively or in another way. The interpolation is especially simple if the signal is narrowband, as low-order interpolation polynomials are often most complete.

The interpolation / extrapolation is numerically sensitive, partly because the roots of the extrapolation polynomial lie on or near the unit circle and numerical disturbances therefore do not die out, and can also for other reasons only be performed over short periods of time. It is therefore not possible to simply interpolate / extrapolate past a chirp disturbance of a certain length.

This problem can be solved by performing an STFT on the disturbed signal according to the invention. In each frequency slot, a chirp disturbance then becomes of only a short duration. In addition, the signal component in each frequency slot is narrowband, which according to the above makes interpolation / extrapolation much easier.

Claims (11)

507 857 13 Patent claims A method for eliminating interference, such as pulses and linear chirps, in an FMCW-type linear frequency sweep radar, in which transmitted and received signals are mixed to form a utility signal in the form of a difference signal, hover signal, with a wave for each target , which frequency, amplitude and phase of waves contain target information, the frequency of a wave depending on the distance to a target associated with the wave, characterized in that the hover signal is subjected to time-frequency division for time-local frequency division of the signal, that the disturbance is detected and that the disturbance is eliminated separately in each frequency band separately, after which the interference-free time signal and its discrete Fourier transform, DFI ", are calculated from the interference-free time frequency division. Method according to Claim 1, characterized in that the time-frequency division is carried out by means of so-called STFT LShort lime Fourier Iran form). Method according to one of the preceding claims, characterized in that the discrete Fourier transform, DFfzn, is performed with a fast algorithm called FFI ". Method according to one of the preceding claims, characterized in that interference detection is performed in each frequency band with methods for detecting disturbances of short duration. Method according to one of the preceding claims, characterized in that the detection of linear chirp disturbances and pulse disturbances is carried out with methods for detecting straight lines in images. Method according to claim 5, characterized in that disturbance patterns in the form of straight lines deviating from lines parallel to the time axis are identified, that the identified lines are related to affected frequency bands and that the disturbance is eliminated separately in each affected frequency band. 507 857 14 Method according to Claim 5 or 6, characterized in that Houghtransform is used for detecting the straight lines. Method according to one of the preceding claims, characterized in that the hovering signal in connection with the time-frequency division into narrow frequency bands is filtered by the signal in order to increase the sensitivity of the detection. Method according to Claim 8, characterized in that the filter is determined by means of adaptive methods. Method according to one of Claims 8 or 9, characterized in that filters are applied to one or more of the narrow-band frequency slots in the narrow-band frequency band of the time-frequency division. Method according to one of the preceding claims, characterized in that the hovering signal or the useful signal after interference elimination is reconstructed by extrapolation from undisturbed samples, in one or more of the narrowband frequency slots in the narrowband frequency band of the time-frequency division.