Correction of I/Q channel errors without calibration

[0003] (Not Applicable)

[0004] In modern coherent sensor systems such as radar and sonar, it is advantageous to represent the received signal as complex numbers. Complex signals are conventionally derived by separating real signals into in-phase (I) and quadrature (Q) components; a representation that is conducive to efficient digital signal processing. This technique is known as quadrature demodulation.

[0005] Typically, an intermediate frequency (IF) signal is split and mixed with coherent I and Q reference sources given by 2 cos ωt and −2 sin ωt, respectively. After mixing, the products are low pass filtered to retain the difference frequency terms. As a result, the signal is split into I and Q terms which ideally combine to provide a complex output Y(t)=I(t)+jQ(t)=A(t)·e^jΘ(t).

[0006] This ideal signal is obtained only if the gains of the I and Q. paths are equal and if the phase difference between channels is 90°. If the ideal signal is not attained, the resulting data is corrupted, ultimately producing ghosts and other undesirable artifacts in a radar image.

[0007] Practically, balancing of the I and Q channels is difficult to achieve and maintain, especially considering the wide bandwidths and high frequencies of modern high-performance radar systems. Existing techniques fall into one of two broad categories:

[0008] The first category attempts to measure the imbalance, and then correct the data accordingly. These techniques either have a limited capability to mitigate any but the simplest of phase and/or amplitude errors, or they impose a significant processing load on the system. (An imaging radar, such as a synthetic aperture radar (SAR), imposes a tremendous data processing requirement on a system even in the ideal situation. A requirement for additional processing of large SAR signals does not result in rapid image acquisition, and may be prohibitive in many applications.)

[0009] F. Churchill et al., “The Correction of I and Q Errors in a Coherent Processor”, IEEE Transactions on Aerospace and Electronic Systems,Vol. AES-17-1, January 1981, pp. 131-137, discusses one of the first of such techniques. They used a test signal to determine the amount of correction to be applied to the channels. However, such a system is not dynamic in that the system can change as it operates, and this change would only be detected by degrading performance. Otherwise, to prevent this, periodic recalibration periods would need to be incorporated within the normal operation of such a system, resulting in missed data collection opportunities.

[0010] Another system using a test signal is discussed by N. Halwani et al., U.S. Pat. No. 5,315,620, which aims to avoid the need for high data sampling rates. The quadrature phase error is detected using a test signal and an error signal is generated. The received I and Q signals are applied to a correction network, along with the error signal, to minimize quadrature phase error between the I and Q signals.

[0011] The second category attempts to generate I and Q data using only a single analog to digital converter (ADC). These systems either require excessive video bandwidth (by a factor or two or more), or they suffer a need for excessive power (by a factor of two or more).

[0012] K. Ho et al., “A Digital Quadrature Demodulation System”, IEEE Transactions of Aerospace and Electronic Systems,Vol. 32, No. 4, October 1996, pp. 1218-1227, discloses an I-Q, demodulator based on low pass filtering of the input samples. They show that with a sampling frequency at least equal to 2B, the I components are the decimated input samples with appropriate sign changes and the Q samples are the low pass filter output of a frequency shifted input sequence. A finite impulse response low pass filter minimizes computation. A sampling scheme is also provided that downconverts without IF mixing.

[0013] The video-frequency offset technique involves using a single channel to process digital I and Q data in one channel. FIG. 1 shows a prior art system 2 where the received radar signal X_IF(t)=cos(ω_IFt+Φ(t)) is mixed in mixer 4 with a local oscillator signal of cos((ω_IF−πf_s/2)t) down to a frequency f_s/4, passed through a low pass filter 5 to limit the signal to IF bandwidth B_IF, and sampled in analog to digital converter (ADC) 6 at rate f_s, where f_s≧2B_IF. The samples are then digitally mixed in mixer 7 to their final form with a digital local oscillator (DLO) signal of exp{−jnπ/2}, which provides a repetitive input to mixer 7 of 1, −j, −1,j.

[0014] One problem with this scheme is that it requires f_s≧2B_IF, whereas separate IVQ channels and ADCs require only that f_s≧B_IF. (In a typical radar for which this circuit may be used, B_IFmay range from several MHz to several hundreds of MHz, where an additional factor of two for the ADC sampling frequency might be particularly problematic.) Consequently, this circuit requires more expensive ADCs and accompanying digitial circuits, which may not even be available, or must contend with the performance limits of a lesser IF bandwidth.

[0015] FIG. 2A shows the IF signal location in the 2-sided frequency domain at the input of filter 5 for the circuit of FIG. 1. As shown in FIG. 2B, at the output of filter 5, the real-valued signal is symmetric about DC. FIG. 2C shows the output signal after mixing with the DLO. This signal can be digitally filtered and decimated, if desired, to the equivalent of conventional I/Q channel data.

[0016] According to this invention, the spectrums of the desired balanced signals are separated from the error imbalance signals in a manner similar to the second category, but two ADCs are used to avoid the shortcomings denoted above.

[0017] To achieve the foregoing and other objects, and in accordance with the purpose of the present invention, as embodied and broadly described herein, the present invention may comprise a method of providing a balanced demodular output comprising providing an analog pulsed input; adding a variable phase shift as a function of time to the input signal, applying the phase shifted input signal to a demodulator; generating a baseband signal from the input signal and low-pass filtering the result; converting the baseband signal to a digital output signal; and removing the variable phase shift from the digital output signal to form a complex data output representative of the output of a balanced demodulator.

[0018] Additional objects, advantages, and novel features of the invention will become apparent to those skilled in the art upon examination of the following description or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained as particularly pointed out in the appended claims.

[0019] The accompanying drawings, which are incorporated in and form part of the specification, illustrate an embodiment of the present invention and, together with the description, serve to explain the principles of the invention.

[0020] FIG. 1 shows a prior art circuit for video-frequency offset phase-history collection.

[0021] FIGS. 2A-2C show a signal at various locations on the circuit of FIG. 1.

[0022] FIG. 3 shows a doppler offset phase-history collection circuit.

[0023] FIGS. 4A-4C show phase-history plots for the circuit of FIG. 3.

[0024] FIG. 5 shows a phase modulated I/Q demodulator of the invention.

[0025] The prior art circuit of FIG. 1 maintained a video-frequency separation between the signal and its image-signal prior to ADC sampling. A similar circuit can be utilized to maintain Doppler frequency separation prior to sampling, as shown in FIG. 3. A Doppler signal is one which may embody a pulse to pulse phase shift. A Doppler frequency spectrum is a collection of such signals.

[0026] The operation of this circuit of this invention is based on the determination that if a Doppler frequency spectrum is modified by a pulse to pulse rolling phase shift which is cancelled at the output of the circuit, then the imbalance errors are separated from the input Doppler spectrum, allowing the imbalance errors to be cancelled.

[0027] The circuit 10 of FIG. 3 would typically be used to process a Doppler signal of the form X(t)=cos[ω(t)+Φ(t)], where Φ(t) represents a phase shift that has been applied to the signal, and may vary over time. In a typical radar, Φ(t) originates with a chirp where each transmitted pulse varies in frequency over a predermitted range in a manner well known in the art, which signal is subsequently received and processed. This signal is typical of the signal received by a modern radar such as a synthetic aperture radar (SAR) and mixed down to intermediate frequency ω_IFin a manner well known to those of ordinary skill in the art. In a typical prior art system, this signal would be processed in a syncronous quadrature demodulator to yield I/Q outputs, which outputs would be as accurate as the balance between channels in the demodulator, as discussed above.

[0028] In accordance with a first embodiment of this invention, the IF input signal of FIG. 3 also includes a pulse-to-pulse rolling phase shift of jkπ/2 radians, so the input signal to mixer 14 is X_IF(t,k)=cos(ω_IFt+Φ(t,k)+kπ/2), where k is the pulse index (an integer that is incremented by 1 with each pulse) and the pulses occur at a frequency f_p, the frequency at which the signal pulses were transmitted. Doppler frequency, for practical purposes, is the pulse-to-pulse phase change, and hence can be modeled as the derivative of the phase with respect to k. For each four counts of k, the input function is incremented by a phase shift of 0, 90°, 180°, and 270°. On the fifth count, the phase shift is back to 0.

[0029] This rolling phase shift is easily generated by a direct digital synthesizer such as disclosed in copending Ser. No. 09/238,762, filed Jan. 28, 1999, by Peter Dudley and Bertice Tise, still pending and assigned to the assignee of this invention. This phase shift may be applied to the transmitted pulse, or to any of the local oscillators that are conventionally mixed with the received signal to bring it down to the intermediate frequency that is input into the demodulator of FIG. 3.

[0030] After being mixed with cos(ω_IFt) at mixer 14 and fed through low pass filter 15 (with a cutoff of B_IF/2), the signal of FIG. 3 may be described as:

[0031] X_LPF(t,k)=½(exp{jΦ(t,k)}exp{jkπ/2}+exp{−jΦ(t,k)}exp{−jkπ/2}).

[0032] The positive video-frequencies have a positive Doppler shift, but the negative video-frequencies have a negative Doppler shift. This signal can be sampled at the reduced rate of f_s≧B_IF.

[0033] FIGS. 4A-4C shows the signal spectrums from the circuit of FIG. 3 in the 2-Dimensional phase-history plane, where the horizontal axis shows Doppler frequency and the vertical axis shows carrier frequency. In FIG. 4A, the signal at the input to the mixer 14 is centered at f_IFand the Doppler is centered at f_p/4. FIG. 4B shows the signal at baseband after passing through LPF 15, and FIG. 4C shows that the desired signal has been shifted to the center point by the digital Doppler shift from mixer 17, while the undesired image shifted to −f_p/2 can be removed by digital filtering of the data, or as part of a typically subsequent transform via digital filtering of the data after the digital Doppler shift.

[0034] Analysis of the circuit of FIG. 3 shows that the digital Doppler shift is accomplished by pulse-to-pulse multiplication with the sequence exp{−jkπ/2}=1,−j,−1, j, . . . This signal is the complement of the phase shift that was applied to the input signal to the circuit 10 of FIG. 3, and serves to cancel, on a pulse-by-pulse basis, the input phase shift. That is, an input pulse that was shifted 90° is shifted by −j at the output; an input pulse that was shifted 180° is shifted by −1, and an input pulse that was shifted 270° is shifted by +j. It also implies that every other radar pulse is purely I-channel with the in-between pulses being purely Q-channel. The single ADC alternates between I and Q on a pulse-to-pulse basis. For each of the I- and Q-channel pulse trains, every other pulse has its data negated, which is easily compensated for with programmable logic or other digital circuitry.

[0035] Whereas the circuit of FIG. 1 requires twice the ADC sampling frequency to maintain separation of the desired spectral region from its image, the circuit of FIG. 3 requires twice the pulse rate frequency (or azimuth sampling rate) to maintain separation. That is, the PRF must satisfy f_p≧2B_Doppler, the Doppler bandwidth of the IF signal. However, the ADC sampling frequency, f_s, is half that required for the other circuit.

[0036] The Doppler offset technique requires two pulses to gather data to make an I/Q data pair, whereas the video-frequency offset technique requires only one pulse; therefore, the Doppler technique has a 3 dB signal-noise-ratio disadvantage compared to the video technique.

[0037] Both of these techniques also offer unwanted spectral components centered at their respective fold-over frequencies. With enough separation they can be filtered, after which the data can be decimated into their respective dimensions. FIG. 5 shows another embodiment of this invention that uses the same input signal, X_IF(t,k)=cos(ω_IFt+Φ(t,k)+kπ/2), to the quadrature demodulator as the embodiment of FIG. 3, discussed previously.

[0038] This signal is input to each of mixers 24, 34 where the I/Q channels are formed. In particular, X_IF(t,k) is mixed with 2 cos(ω_IFt) in mixer 24 and with −2 sin(ω_IFt) (the result of shifting 2 cos(ω_IFt) by 90° in phase shifter 32) in mixer 34. These components are standard RF hardware, and quadrature mixers comprising the phase shifter 32 and mixers 24 and 34 are commonly available in a single package.

[0039] After passing through low pass filters 25, 35 with a cutoff of B_IF/2 to remove unwanted mixing products, the signal at point i₁is described by X_i(t,k)=cos(Φ(t,k)+kπ/2) and the signal at point q_iis described by X_q(t,k)=sin(Φ(t,k)+kπ/2). These are perfect signals; i.e., the signals that would be present if no imbalance exists.

[0040] Since, in reality, even the best systems have some imbalance, these will be modeled by a transfer function H_i(t) in the I channel and a function H_q(t) in the Q channel. For the purpose of analysis, the imbalances are split between these transfer functions such that at the respective outputs of these transfer functions, the signals take on the form: $X_{i_2}\left(t,k\right)=\left(1+a_{ϵ}\right)\cos \left(\Phi \left(t,k\right)+\frac{k\pi }{2}+{\theta }_{ϵ}\right),$

[0041] and $X_{q_2}\left(t,k\right)=\left(1-a_{ϵ}\right)\sin \left(\Phi \left(t,k\right)+\frac{k\pi }{2}-{\theta }_{ϵ}\right),$

[0042] where the error terms a_ε and θ_ε are typically small and video-frequency dependent.

[0043] Each of these analog signals is now converted to a digital form by one of AID converters 26 and 28 in a manner well known in this art. Each of ADC 26, 28 is sampled in a conventional fashion with a sampling function, taking samples at times described by $\delta \left(t-\frac{n}{f_s}\right),$

[0044] where n is a fast-time sample number and f_sis the sample frequency. The digital output of ADC 28 is at this point merely data and is declared the imaginary component, thereby effectively causing it to be shifted 90° as represented by block 30. The outputs of ADCs 26, 28 are appropriately added, and the sampled signal at point y₀is: $Y_0\left(n,k\right)=\left[\begin{array}{c}\left(1+a_{ϵ}\right)\cos \left(\Phi \left(n,k\right)+\frac{k\pi }{2}+{\theta }_{ϵ}\right)+\\ j\left(1-a_{ϵ}\right)\sin \left(\Phi \left(n,k\right)+\frac{k\pi }{2}-{\theta }_{ϵ}\right)\end{array}\right],$

[0045] which equation may be manipulated to $Y_0\left(n,k\right)=\left[\begin{array}{c}\cos \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\cos {\theta }_{ϵ}+j\sin \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\cos {\theta }_{ϵ}-\\ \sin \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\sin {\theta }_{ϵ}-j\cos \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\sin {\theta }_{ϵ}+\\ a_{ϵ}\cos \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\cos {\theta }_{ϵ}-j{a}_{ϵ}\sin \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\cos {\theta }_{ϵ}-\\ a_{ϵ}\sin \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\sin {\theta }_{ϵ}-j{a}_{ϵ}\cos \left(\Phi \left(n,k\right)+\frac{k\pi }{2}\right)\sin {\theta }_{ϵ}\end{array}\right],$

[0046] which yields $Y_0\left(n,k\right)=\left[\begin{array}{c}\cos {\theta }_{ϵ}\exp \left\{\mathrm{j\Phi }\left(n,k\right)\right\}\exp \left\{j\frac{k\pi }{2}\right\}-\\ \sin {\theta }_{ϵ}\exp \left\{-j\left(\Phi \left(n,k\right)-\frac{\pi }{2}\right)\right\}\exp \left\{-j\frac{k\pi }{2}\right\}+\\ a_{ϵ}\cos {\theta }_{ϵ}\exp \left\{-\mathrm{j\Phi }\left(n,k\right)\right\}\exp \{-j\frac{k\pi }{2}+\\ a_{ϵ}\sin {\theta }_{ϵ}\exp \left\{j\left(\Phi \left(n,k\right)+\frac{\pi }{2}\right)\right\}\exp \left\{j\frac{k\pi }{2}\right\}\end{array}\right].$

[0047] For small error terms, this last equation may be approximated as $Y_0\left(n,k\right)=\left[\begin{array}{c}\exp \left\{\mathrm{j\Phi }\left(n,k\right)\right\}-\\ \sin {\theta }_{ϵ}\exp \left\{-j\left(\Phi \left(n,k\right)-\frac{\pi }{2}\right)\right\}\exp \left\{-jk\pi \right\}+\\ a_{ϵ}\exp \left\{-\mathrm{j\Phi }\left(n,k\right)\right\}\exp \left\{-jk\pi \right\}\end{array}\right]\exp \left\{jk\frac{\pi }{2}\right\}.$

[0048] This signal is fed through mixer 27 for modulation by a complementary signal, i.e., a pulse-dependant corrective signal with opposite phase than that was applied to X_IF. This constitutes a digital Doppler shift, yielding: $Y_1\left(n,k\right)=\left[\begin{array}{c}\exp \left\{\mathrm{j\Phi }\left(n,k\right)\right\}-\\ \sin {\theta }_{ϵ}\exp \left\{-j\left(\Phi \left(n,k\right)-\frac{\pi }{2}\right)\right\}\exp \left\{-jk\pi \right\}+\\ a_{ϵ}\exp \left\{-\mathrm{j\Phi }\left(n,k\right)\right\}\exp \left\{-jk\pi \right\}\end{array}\right]$

[0049] The term exp{jΦ(n,k)} is the desired ‘error-free’ term, while the 2^ndand 3^rdare the result of I/Q imbalance. If these terms are removed, the I and Q channels are ‘balanced.’

[0050] Since each of lines 2 and 3 in the equation for Y₁are modulated by exp{−jkπ}, these error terms are centered at Doppler frequency f_p/2, as were the unwanted spectral components for Doppler offset phase-history collection of FIGS. 3 and 4. Therefore, for a large enough pulse rate frequency f_p≧2B_Doppler), the ‘imbalance’ is separated in Doppler frequency from the desired signal, and can therefore be filtered.

[0051] Unlike the previous embodiment, each radar pulse contributes both I-channel and Q-channel data. Consequently; there is no 3 dB SNR penalty as there was in the circuit of FIG. 3. In addition, ADC sampling frequency must satisfy f_s≧B_IF, only half the value needed in the video frequency embodiment of FIG. 1.

[0052] The mixing operation to provide the complementary output phase shift and thereby remove the rolling phase shift in the data comprises multiplication of appropriate values from 1,−j,−1,j, and may be accomplished by appropriately trading of, and sign changes to, the initially generated I and Q data. One of ordinary skill in the art will appreciate that this phase shift may be implemented in the ADC board or elsewhere.

[0053] The particular sizes and equipment discussed above are cited merely to illustrate a particular embodiment of this invention. It is contemplated that the use of the invention may involve components having different sizes and shapes as long as the principle, that applying controlled phase shifts to the signal prior to I/Q demodulation, followed by corresponding removal of the phase shifts in the data after the ADCs, in an effort to separate in Doppler the balanced signals from imbalance errors between the I and Q channels, is followed. For example, other phase shift functions besides the particular rolling phase shift discussed might be employed to the same advantage, as long as the input phase shift has enough variation in phase to cause the system to operate, and the output phase shift mirrors and cancels the input phase shift. And while the input signal, X_IF, is denoted as an intermediate frequency signal of a digital radar receiver, the invention is applicable to any pulsed signal that may be derived as a summation of sinusoids. Furthermore, one of ordinary skill in the art recognizes that this technique is ‘linear’, that is, the property of superposition holds for input signals and output signals. Consequently, a scaled input signal would yield a like scaled output signal, and a sum of input signals would yield an output signal that is the sum of output signals if the inputs were applied individually. It is intended that the scope of the invention be defined by the claims appended hereto.