Spatial Frequency
Spatial Frequency
Mark A. Richards
July 7, 2007
(1)
(2)
t=
2R c
(3)
Suppose x(t) is the output of a radar receiver as a function of time; we can use (3) to relabel the time axis in units of range, giving a range profile of the scene viewed by the
Temporal and Spatial Frequency in Radar Imaging Page 1 of 4 July 7, 2007
radar. We would like to have a relationship between the range profile and a spectrum expressed in spatial frequency units. Using the relationship (3), we can re-write (1) as
2 R j 4 RF c 2 X (F ) = x dR e c c
(4)
Note that the quantity 4RF/c often appears as 4/, but we will leave it in terms of frequency here. Now define
2 2R x c c
(5)
( R ) e j 2 ( 2 F c ) R dR x
(6)
Note that this is exactly of the form of a Fourier transform (compare to Eq. (1)) of a function of range R provided we identify the quantity 2F/c as the range frequency variable Fr. That is,
(F ) = X r
( R ) e j 2 F R dR x
r
(7)
To emphasize, the range frequency variable Fr in m-1 or cycles/m is related to temporal frequency F in s-1 or cycles/s (Hz) according to
Fr m-1 = 2 F s-1 c (8)
Note that range frequency Fr is not the same as spatial frequency Fy. In particular, Fr = 2Fy.
4 Consequences
Equation (7) shows that a range profile forms a Fourier transform pair with a spectrum expressed in range frequency units. This means that we can reason about signal properties such as relationships between bandwidth and resolution, sampling and aliasing, and so forth using the same rules that apply to temporal signals and their Fourier transforms. In particular,
Page 2 of 4
July 7, 2007
Analogous to the fact that a resolution of t seconds requires a signal bandwidth of F = 1/t Hz, a range resolution of R m requires a range frequency bandwidth of 1/R cycles/m. Note that the range frequency bandwidth of 1/R is equivalent to c/2R Hz in temporal units, which is the usual formula relating range resolution to temporal frequency bandwidth [1]. Sampling a range profile at intervals of R m will cause its range frequency spectrum to replicate at intervals of 1/R cycles/m in range frequency. If a signal has a range frequency bandwidth of Br cycles/m, the Nyquist theorem requires a sampling rate of at least Br samples/m, corresponding to a sampling interval of 1/Br m, to avoid aliasing. If a range frequency spectrum is sampled at an interval Fr in range frequency (for instance, through computation of a discrete Fourier transform), the range profile will be replicated at intervals of 1/Fr m in range. Thus, the spectrum of a range profile of length Rmax m should be sampled in range frequency at intervals of 1/Rmax or less to avoid aliasing.
(9)
Spatial resolution, replication interval, and sampling can be reasoned about in terms of cross-range spatial frequency Fcr in the same manner as described above for range spatial frequency Fr.
Page 3 of 4
July 7, 2007
x R
target
direction of flight
6 References
[1] M. A. Richards, Fundamentals of Radar Signal Processing. McGraw-Hill, New York, 2005.
Page 4 of 4
July 7, 2007