Approved for Public Release; Distribution Unlimited
Case # 07-0017
Coherent Integration With Range Migration
Using Keystone Formatting
R. P. Perry, (rpp@mitre.org), R. C. DiPietro, rcd@mitre.org , R. L. Fante,
The MITRE Corp., 202 Burlington Road, Rte. 62,
Bedford, MA 01730-1420, Ph: 781-271-7031 FAX 781-271-7045;
1. Introduction
Coherent integration has been used in
matched filter radar signal processing to
increase target signal to noise ratios.
Currently radar designers apply a rule of
thumb that a target must remain in the
same range resolution cell during the
coherent integration time. Thus Doppler
processing depends mainly on the change
in phase from pulse to pulse. This results
in limiting either the range resolution or
Doppler resolution of the radar enabling
very high range resolution or very high
Doppler resolution but not both. We
present here an approach to radar matched
filtering which does not have these
limitations and can produce simultaneous
high range and high Doppler resolution
matched filter outputs. The new matched
filter coherently integrates the radar data
even though the target scatterers move
through many range resolution cells
during the coherent integration time.
The new matched filter uses as its primary
component a new transformation which
we call Keystone formatting. Keystone
formatting provides the ability to match
filter all scatterers moving through range
cells with linear velocities regardless of
their individual velocity magnitudes or
1
directions. Intuitively, there is a direct
relationship between a scatterer’s Doppler
and linear range migration. The Keystone
formatting is based on this relationship.
The processing also includes the capability
to hypothesize range migration
acceleration, i.e. quadratic migration, and
higher order terms as well as to
accommodate under sampled radar PRF’s
producing folded Doppler spectrums. The
processing is in principal transparent to
any range migration. Although higher
order range migration can be
hypothesized, a practical approach would
be to limit the hypothesis to acceleration
and foldover. The Keystone formatting
can be applied prior to these hypothesis to
reduce computational redundancies.
2. Theoretical Background and the
Keystone Transformation
In this section we show how range
migration can be compensated, so that
coherent integration can be obtained of
target scatterers that have migrated
through many range cells during a
integration interval.
Consider a radar on an aircraft
emitting a series of pulses
 v(τ ) = p (t − kT1 )exp[− i 2π f c (t − kT1 )] (1) where P(f) is the Fourier transform of p(t').
Next, we can generalize Equation (4) from
where t = time, k = pulse number, T1 is the a single point scatterer to an assemblage of
interpulse period, fc is the carrier N moving point scatterers at ranges Rn(tk),
frequency and, typically*, p(t) is a chirp which can be used as a simplified model.
pulse, given by Then Equation (4) becomes
 πBt 2   t  N
p(t ) = exp − i rect   (2) S ( f , t k ) = P( f )∑ An
 To   To  n =1
. (5)
 4π
where To is the pulse length and B is the exp i ( f + f c )Rn (t k )
bandwidth. Now suppose this pulse train  c 
illuminates a moving point scatterer at a
range R(t), and the received signals are
recorded in a two-dimensional array The range Rn(tk) to scatterer n can be
s (t ′, t k ) where t ′ = t − kT1 is known as the expanded in a Taylor series about tk = 0 as
"fast time" and t k = kT1 is known as the rn 2
"slow time". Then, in terms of these Rn (t k ) = rn + rn t k + tk + " . (6)
2
variables the received signal after
downconversion, can be written as where rn = Rn (0 ), rn = R n (0 ) , etc. If
 2 R(t k )  4πf c 
Equation (6) is used in (5) we obtain
s (t ′, t k ) = A p t ′ −  exp i R(t k )
 c   c 
N
(3) S ( f , t k ) = P( f )∑ An
n =1
where A is a constant that depends on the (7)
 4π  
scatterer strength and the scatterer range expi ( f + f c )rn + rn t k + rn t k2  
has been assumed to vary negligibly  c  2 
during the pulse interval To.
Now recall that the matched filter in
The Fourier transform of Equation range-Doppler space is
(3) over the fast-time variable can be
∞ T 2
written as
I (ρ , f d ) = ∫ df ∫ dt S ( f , t )
k k
−∞ −T 2
(8)
 4π
S ( f , t k ) = A P( f )exp i ( f + f c )R(t k ) (4)  4πf
exp − i

ρ − i 2πf d t k 
 c   c 
where ρ is range, fd is Doppler,T is the
* Although p(t) is represented as a chirp coherent integration interval, and we have
pulse, our analysis is valid for arbitrary approximated the summation over pulses
p(t). by an integral. Then, if (7) is used in (8) it

2
is evident that the term exp[i 4πf rn t k c ] interval. We can remove the linear range
represents a coupling between range and migration if we rescale the time axis for
Doppler due to linear range migration of each frequency by the transformation
the scatterers. This range migration is the  fc 
dominant blurring mechanism that must be t k =  τ (9)
compensated in order to obtain a high-  f + fc 
resolution matched filter output. This If Equation (9) is used in (7) the rescaled
effect can be ignored if
data in the new (f, τ) domain, eliminating
πB∆R c << 1, where ∆R is the total range
higher than quadratic terms is:
migration during the coherent integration
N
S ( f ,τ ) = P( f )∑ An
n =1

 4π 
i ( f + f c )rn 
 c 
 4π  (10)
exp+ i rn f cτ 
 c 
 4π  
2

( f + f c )a n  f cτ  

+ i
 c  f + f c  

TimeExpansion=(f+fc)/fc
fc+ /2

fc
NewAzimuth
Samples

OldAzimuth
Sampl s

fc-B/2
Radar Pulses
Figure 1. Keystone Transformation of the 2-Dimensional Data

3
Thus, the coupling effect of the linear f dn = f bn + Ff 1 (14)
range migration has been removed
where F is the fold factor and f1 = 1/T1 is
Figure 1 shows the remapping process. the pulse repetition frequency. If Equation
We have called the process “Keystone (14) is substituted into (11) we obtain
Formatting” due to its Keystone shape.
The target scatterers can now be matched S ( f , τ ) = e i 2πf1Fτ S u ( f , τ ) (15)
filtered if the acceleration, i.e. a n = rn can
where
be estimated, and corrected by hypotheses
N

3. Keystone Formatting of S u ( f ,τ ) = ∑ G n ( f )exp(i 2πf bnτ ) (16)

n =1
Undersampled Data
is the undersampled data. Thus, the
undersampled data in the (f, τ) domain can
Range migration correction using
be corrected by multiplying by
Keystone formatting can still be done even
if the data is undersampled. Let us ignore exp(i2πf1Fτ) after Keystone formatting. If
acceleration and then rewrite Equation the foldover, F, is unknown it must be
(10) as hypothesized. The foldover correction can
also be applied before the Keystone
formatting.
N
S ( f ,τ ) = ∑ G n ( f )exp(i 2πf dnτ )
4. Examples Using Keystone Formatting
(11)
n =1 4.1 Moving Target Matched Filtering
where The effectiveness of the Keystone
 4π formatting can be shown by synthesizing
G n ( f ) = P ( f )An exp i ( f + f c )rn  (12) four simultaneous point targets whose
 c  range migration is shown in figure 3. All
And four targets have different velocities but
the same acceleration.
2 fc
f dn = rn (13)
c
is the true Doppler frequency of scatterer
n.

Now suppose the data is

undersampled and fbn is the folded Doppler
frequency of scatterer n. Then Equation
(13) can be rewritten as

4
 RANGE MIGRATION OVER CPI, 2 TARGETS RANGE MIGRATION AFTER KEYSTONE
25
1.6
10
20 1.4

20 1.2

15
1
RANGE CELLS

RANGE
30

0.8
10
40
0.6

5
50 0.4

0.2
0 60

100 200 300 400 500 600

PULSE NUMBER
-5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
SECONDS Figure 5 - Range Migration After
Figure 3 Keystone Formatting
Figure 4 shows the target data matched
FFT RANGE-DOPPLER IMAGE WITHOUT KEYSTONE
filtered in range for each pulse with no
Keystone formatting or acceleration 10 -5

hypotheses. Figure 5 shows the range

matched filter output after Keystone 20
-10

formatting and the correct acceleration

RANGE CELLS

-15
30
hypotheses. The range migration has been -20
eliminated for all four targets 40

simultaneously. 50
-25

-30
RANGE MIGRATION BEFORE KEYSTONE 60

-4 -3 -2 -1 0 1 2 3 4
TARGET VELOCITY m/sec
10 2.5

Figure 6
20 2

Figure 6 shows the 2D range/Doppler

RANGE

30
1.5
matched filter output with no Keystone
40
1 Formatting. Figure 7 shows the 2D
50 range/Doppler matched filter output after
0.5
Keystone formatting and acceleration
60
hypotheses. There is about a 24 dB
100 200 300 400 500 600
PULSE NUMBER improvement in coherent integration with
Figure 4- Range Migration With No Keystone formatting.
Keystone Formatting

5
 KEYSTONE RANGE-DOPPLER IMAGE KEYSTONE RANGE-DOPPLER IMAGE

10 -5 10 -5

-10 -10
20 20
RANGE CELLS

RANGE CELLS
-15 -15
30 30

-20 -20
40 40

-25 -25
50 50

-30 -30
60 60

-4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4
TARGET VELOCITY m/sec TARGET VELOCITY m/sec

Figure 7 Figure 9
Summary
4.2 Foldover Example
We have developed and demonstrated the
The target motion shown in figure 3 was Keystone format to simultaneously
increased by 8 m/sec producing a fold remove linear range migration for all
factor of one. Figure 8 shows the Keystone targets regardless of their velocities.
matched filter output with a foldover Higher order motion and under sampling
hypothsis of zero. foldover can be removed by hypotheses.
KEYSTONE RANGE-DOPPLER IMAGE

References
10 -5

20
-10
1. R. P. Perry, R. C. DiPietro, R. L. Fante,
“SAR Imaging of Moving Targets,” IEEE
RANGE CELLS

-15
30
Transactions on Aerospace and Electronic
-20
40 Systems, Vol. 35, NO. 1, January 1999
-25
50

-30
2. “Spaced-Based Bistatic GMTI Using
60 Low Resolution SAR”, DiPietro R. C. ,
-4 -3 -2 -1 0 1
TARGET VELOCITY m/sec
2 3 4
Fante R. L., Perry R. P. , Proc. 1997 IEEE
Aerospace Conf. , Feb 1-8 1997
Figure 8
Figure 9 shows the Keystone matched
3. “ Dim Target Detection Based on
filter with the correct fold factor
hypothesis. Keystone Transformation”, Zhang S.,
Zeng T., Long T. ,Yuan H. IEEE 2005
International Radar Conference, 9-12 May
2005