A criterion is established for determining the limit on radar resolution enhancement with Burg algorithm (maximum entropy method (MEM)) by comparing the radar range and Doppler resolution limits of MEM with those obtained by the Fourier transform (FT). Also examined are errors in range and Doppler estimation due to MEM and in Doppler space as contiguous pulses are independently extrapolated to form a new train of pulses prior to Doppler processing. It is found that, for two identical point-scatterers, the MEM (Burg algorithm) has approximately a factor of 2.6 advantage over the FT. Data extrapolation by a factor of 2.6 could thus result in improved range and Doppler resolution; beyond that, the effect is primarily due to the increased size of the data window, with a cosmetic improvement in the appearance of the plotted data. Selection of the filter order is also discussed. A practical technique is proposed to facilitate selection of the filter order when a large number of radar pulses have to be extrapolated pulse-by-pulse, thus ensuring a near-optimum result. The two point-scatterer model is used to characterize performance even as the complexity of the target increases from two point-scatterers to two spheres and to a circular cone and finally to a range-Doppler image of a circular cone. Errors in range and Doppler are also characterized by the two point-scatterer model. Doppler error can cause distortion or even sign change in some cases, as exemplified by a conical target with a small surface scatterer The amplitude and the phase errors for each of the two scatterers are also evaluated as a function of the signal-to-noise (SNR). Both are sensitive to the initial phase of the interfering sinusoid. Thus, for a SNR of less than 15 dB, the amplitude error becomes too large to be acceptable. In addition, spurious artifacts may occur due to large pulse-to-poise phase errors. Despite the potential for errors, the comparison clearly establishes the intrinsic gain in the resolution of up to 2.6/spl times/2.6 for a typical range-Doppler image. It is recommended that the methodology discussed here be applied in order to quantify the relative merits of other algorithms that may be used for resolution enhancement. >
[1]
G. R. Stegen,et al.
Experiments with maximum entropy power spectra of sinusoids
,
1974
.
[2]
P. Jackson,et al.
Truncations and phase relationships of sinusoids
,
1967
.
[3]
J. P. Burg,et al.
Maximum entropy spectral analysis.
,
1967
.
[4]
N. Andersen.
On the calculation of filter coefficients for maximum entropy spectral analysis
,
1974
.
[5]
Ta-Yung Liu,et al.
Superhigh image resolution for microwave imaging
,
1990,
Int. J. Imaging Syst. Technol..
[6]
E. Jaynes.
On the rationale of maximum-entropy methods
,
1982,
Proceedings of the IEEE.
[7]
Eric K. Walton.
Comparison of Fourier and maximum entropy techniques for high‐resolution scattering studies
,
1987
.
[8]
Arthur B. Baggeroer,et al.
Confidence intervals for regression (MEM) spectral estimates
,
1976,
IEEE Trans. Inf. Theory.
[9]
S.M. Kay,et al.
Spectrum analysis—A modern perspective
,
1981,
Proceedings of the IEEE.
[10]
H. Akaike.
Power spectrum estimation through autoregressive model fitting
,
1969
.
[11]
J. Makhoul,et al.
Linear prediction: A tutorial review
,
1975,
Proceedings of the IEEE.
[12]
K. Toman,et al.
The spectral shifts of truncated sinusoids
,
1965
.
[13]
R. S. Berkowitz,et al.
Characteristics of autoregressive spectral estimation in range-Doppler imaging
,
1988,
Proceedings of the 1988 IEEE National Radar Conference.
[14]
Thomas M. Sullivan,et al.
High-resolution two-dimensional spectral analysis
,
1979,
ICASSP.