Vous êtes sur la page 1sur 6

Mismatched Filters for Frank Polyphase Codes

via Sidelobe Inversion

Edwin A. De Roux Fuentes Adly T. Fam


Department of Electrical Engineering Department of Electrical Engineering
State University of New York at Buffalo State University of New York at Buffalo
215 Bonner Hall, Buffalo, NY 14260 USA 332 Bonner Hall, Buffalo, NY 14260 USA
ead23@buffalo.edu afam@eng.buffalo.edu

Abstract—A computationally efficient mismatched filter for The proposed filter lies in the class of mismatched filters
Frank polyphase codes is proposed. It is based on an efficiently where an FIR filter is cascaded with the matched filter.
constructed matched filter followed by one or more stages of a
sidelobe inversion (SLI) filter. This sidelobe inversion filter An efficient structure for the matched filter of Frank codes
inherits the computationally efficient structure used for the is introduced in section II. This matched filter is equally valid
matched filter. With the addition of few extra multipliers, a for the periodic and aperiodic cases. The concept of sidelobe
significant increase in the main lobe to peak sidelobe ratio is inversion (SLI), which is the basis of the computationally
achieved. The proposed filter is area and power efficient, as well. efficient mismatched filter presented in this work, is discussed
The total delay is more than that of the length-optimal in section III. Section IV gives details on how SLI filters for
mismatched filter, but is of no significant consequence in most Frank codes are cascaded with the matched filter. Since the
applications. SLI technique uses the matched filter or the ACF as a building
block, this imparts the matched filter efficiency to the SLI
I. INTRODUCTION filter. In section V the hardware complexity of the
mismatched filter and its effect on the signal to noise ratio are
Pulse compression codes are used in many applications presented. Recommendations follow in the conclusion.
such as radar, wireless communication, geophysical
exploration, sonar, and biomedical sensing and imaging. A II. EFICIENT REPRESENTATION OF FRANK POLYPHASE
good pulse compression code is preferably a constant energy CODES AND THEIR MATCHED FILTERS
envelope wave (continuous wave CW) and should have an
autocorrelation function (ACF) with high main lobe to peak The aperiodic Frank codes [8] were shown to have good
sidelobe ratio (MSR) and a narrow main lobe. For binary main to peak sidelobe ratio (MSR). They are generated by
aperiodic codes the best possible is achieved by Barker codes concatenating of the rows of the DFT matrix of size N × N
which attain a maximum length of 13 in the odd case, and are and
conjectured to have no even length larger than 4 [1]. For
higher signal to noise ratio (SNR), longer codes are needed. w = exp( j 2π p / N ) ,
This could only be achieved in the CW case by polyphase (1)
codes. Generalized Barker codes are known up to length 64 at
this time [2]. Longer codes require relaxing the MSR where N and p are relatively prime. This aperiodic code,
condition. Frank polyphase codes and related perfect which is just one period of the periodic one, is of length N2. It
sequences [3] offer ideal periodic ACF, and there are sets that is suggested in [8] that the period should start by
have good crosscorrelation function (CCF), as well. concatenating the complete set of rows in their natural order as
Aperiodic Frank polyphase codes and Chu codes [4] offer in the DFT matrix
good MSR and are available at a variety of length.
1 1 1 … 1 
Even though aperiodic polyphase codes have good ACF, it  
is desirable in most cases to further increase their MSR. Some 1 w w2 … w N −1

 2 2(N − 1) 
methods have been developed in order to accomplish this 1 w w4 … w 
objective. Classical amplitude weighting [5], least square      
amplitude and phase weighting [6], sliding windows [7],  2

inverse filtering method [2], and FIR filters are some of them. 1 w N −1 w 2(N − 1) … w (N − 1) 
 . (2)
since this seems to produce the best MSR. The Z transform of used with the switching and resetting technique [9] to in effect
this code could be represented by the efficient form stabilize pole-zero cancellation.

III. SIDELOBE INVERSION TECHNIQUE


N −1
z −rN
(
X (z ) = 1 − z −N ) ∑ 1−w z
r =0
Mismatched filters based on sidelobe inversion are
proposed r −1
in [10] as a new technique to achieve good sidelobe
, (3)
suppression per arithmetic operation at the cost of increasing
which is valid for any N and P without requiring them to be the length of the filter. The work in [11, 12] represent a
relatively prime. This structure contains 2N adders and compromise between the optimized SLI approach presented in
this work and the length optimal filter. In [11, 12], the extra
N − 1 multipliers as is clear from Fig. 1.
delay is less than that of the SLI filter but more than that of the
length optimal one, while the computation per output is more
1
x(n) than that of SLI filter, but less than that of the optimal length
1 − z −N −1 one. The sidelobe inversion concept is defined as following:
1−z
z −N Let the incoming code be X (z ) , then the autocorrelation
1
function is given by:
1 − wz −1
z −N
1 R(z ) = X (z )X * (z −1 ) . (5)
2 −1
1−w z
  X (z ) represents the received code but it is also referred in
* −1
1
z −N the sequel as the conjugate of the matched filter X (z ) . In
1 − w N − 1z −1
practice the system should be made causal by adding the
Figure 1. Bank filter generator for aperiodic Frank codes. appropriate delays, that is

The matched filter for the periodic and aperiodic Frank 2 − 1)


codes can be implemented using the same filter bank proposed Rd (z ) = X (z ) ⋅ X * (z −1 )z −(N , (6)
above. The equation of the matched filter combined with
(N 2 − 1) delay to render it causal is: but here this will be omitted for simplicity. The
autocorrelation function can be rewritten as follow:
−  N 2 − (r + 1)N 
N −1
z   R(z ) = M + S (z ) .
−(N 2 −1)
( )
X * z −1 z (
= z −N − 1 )∑ r =0 z −1
−w −r
(7)
, (4)
Where M = N2 is the length of the code and
where N is the size of the DFT matrix. This efficient S (z ) represents the sidelobes which are either symmetric for
structure, shown in Figure 2, is in IIR form, and its marginally real valued codes, or have conjugate symmetry for polyphase
stable poles are canceled by some of its zeros. codes. For the SLI to be of practical significance, either the
matched filter or the ACF should have a computationally
x(n) 1 efficient representation. From (7), S (z ) can be represented
z−N − 1
z −1 − 1 by
z−N
1
z −1 − w −1
S (z ) = R(z ) − M = X (z )X * (z −1 ) − M . (8)
z−N
The transfer function of the first stage SLI filter is given
1
by:
z −1 − w −2
z−N


H 1 (z ) = M − S (z ) = 2M − X (z )X * (z −1 ) . (9)
1 r(n)
z −1 − w −(N −1) And its output is:
2
Figure 2. Matched filter for Frank codes of length N .
Y1 (z ) = R(z ) ⋅ H 1 (z ) = M 2 − [S (z )]2 . (10)
Due to finite word length effect this cancellation might no
be exact. In such case, two copies of each block should be
A second stage SLI filter, following the same principle, is Where X (z ) is given by (3). The block diagram of the
defined by: first stage H 1 (z ) is shown in Fig. 3.

2 2 − 1)
2N 2
z −(N
H 2 (z ) = M 2 + [S (z )]2 = M 2 + X (z )X * (z −1 ) − M  R(z )z −(N
2 − 1)
Y1(z)
  . (11) +
-
2 − 1)

And its output is given by:


X (z ) ( )
X * z −1 z −(N

Figure 3. First stage SLI filter for Frank code of length N2.
Y2 (z ) = Y1 (z ) ⋅ H 2 (z )
Adding one extra optimized multiplier µ1 to the first stage
(
= M 2 − S (z )2 )(M 2
)
+ S (z )2 = M 4 − [S (z )]4
. (12) as is shown in Fig. 4 results in

The main idea of the SLI technique is that if the MSR of Y1 (z ) = R(z ) ⋅ H 1(z ) + µ1 ⋅ R(z )
the ACF is high enough then the SLI filters will improve it.
However, an optimized multiplier is added to each stage to = M 2 − [S (z )]2  + µ1 M + S (z )
 
further increase the MSR. This modification is considered in
the following section. = M 2 + µ1M  − [S (z )]2 − µ1S (z )
   . (18)
For  ≥ 2 , the th stage SLI filter and its output are
given respectively by: R(z ) Y1 (z )
H 1 (z )

 −1 2 − 1
H  (z ) = M 2 + S (z ) µ1

 −1 2 − 1
= M2 + X (z )X * (z −1 ) − M  Figure 4. Modified first stage SLI filter.
  (13)
Optimal values of µ1 are obtained via computer search
and using a discrete grid and are shown in Table I for codes of
length up to N2=322. The resulting increase on the output
[Y1(z)] MSR over the input [R(z)] MSR of the modified first
2 2 . (14) stage SLI filter is shown in Figure 5. Here the output [Y1(z)]
Y (z ) = Y −1 (z ) ⋅ H  (z ) = M − [S (z )]
MSR is, on average, 4.2534dB larger than the input [R(z)]
MSR with a standard deviation of 0.3676dB.
In the final implementation of the SLI filters a normalized
structure should be used, as following
TABLE I. OPTIMIZED MULTIPLIER FOR THE 1ST SLI
STAGE UP TO LENGTH 322.
H1(z ) X (z )X * (z −1 )
=1−
2M 2M (15) N µ1 N µ1
3 2.8 18 305.6
and 4 7 19 346.2
5 32.4 20 381.9
2 − 1 6 31.7 21 423.8
H  (z )  X (z )X * (z −1 ) 
=1−  − 1 7 46.4 22 463.3
 −1
M2  M  ;for  ≥ 2 . (16) 8 57 23 512.6
9 76.8 24 555.3
IV. SIDELOBE INVERSION FILTER 10 90.1 25 606
11 113.2 26 654.4
A. First Stage SLI Filter
12 132.8 27 711.8
The technique of SLI permits the utilization of the
matched filter efficiency to built cascaded SLI filter stages for 13 159.3 28 762.2
increasing the MSR. The first stage derived from (7) is: 14 181.3 29 821.8
15 213.1 30 879
* −1 (17) 16 241.5 31 943.4
H 1 (z ) = 2M − X (z )X (z )
17 274.9 32 1002.7
Now the output of the second stage is given by,
output
Y2 (z ) = Y1 (z ) ⋅ (H 2 (z ) + µ2 )
 
 
= M (M + µ2 ) − S (S + µ2 )
. (20)
input
Where,


M = M 2 + µ1M
(21)

and


S = [S (z )]2 − µ1S (z )
Figure 5. MSR of the input and output of the modified first stage SLI filter. . (22)
An average increase of 4.2534dB in the MSR is obtained
The optimum multipliers for codes of length N2=32 up to
2
32 are shown in table II. A search step of 0.1% and 1% of the
B. Second Stage SLI filter final value of the multiplier µ1 and µ2, respectively, were
found to be adequate to converge to their optimal values. Fig.
Further sidelobe suppression can be achieved by using an
8 shows the MSR of the input signal R(z) and the output signal
additional SLI filter stage, which takes the following form,
Y2(z) of two modified SLI stages filter of length up to 322.
based on the first stage with the multiplier µ1.

H 2 (z ) = [M 2 + µ1M ] + [S (z )2 − µ1S (z )] output


= [M 2 + µ1M ] + [(R(z ) − M )2 − µ1 (R(z ) − M )]
(19)

The block diagram of the second stage is shown in Fig. 6.


The first multiplier µ1 was optimized to get the best MSR input
output at the first stage. If it is desired to optimize the MSR at
the output of the second stage, the multiplier µ1 needs to be
recalculated. However, it is found that better results are
obtained by adding a second multiplier µ2 to the second stage
and optimizing both µ1 and µ2 for best MSR. This modified
second stage is shown in Figure. 7.

Y1 (z ) Y2 (z )
H 2 (z )
Figure 8. MSR of the input and output of two modified stages SLI filter.
An average increase of 14.5815dB in the MSR is obtained.

µ2
Here an average increase of 14.5815dB in the MSR of the
output is obtained over the MSR of the ACF with a standard
Figure 7. Modified second stage SLI filter for Frank codes of length M. deviation of 1.9346dB.

2 − 1)
z −2(N
M (M + µ1 ) +
Y2 (z )
+

Y1 (z ) 2 − 1)
M - 2 − 1) -
z −(N + z −(N +
M + µ1

2 − 1) 2 − 1)
X (z ) ( )
X * z −1 z −(N X (z ) ( )
X * z −1 z −(N

Figure 6. Second stage SLI filter for Frank codes of length M.


TABLE II. OPTIMUM MULTIPLIERS FOR BEST MSR OF THE filter for one-stage and two-stages SLI filter with optimum
MISMATCHED FILTER WITH TWO-STAGES SLI FILTERS UP TO multiplier for 3 ≤ N ≤ 32 .
LENGTH 322.

N µ1 µ 2 [Ku] N µ1 µ 2 [Ku]
3 3.8 0 18 -44.1 -41 One-stage
4 6.3 0 19 -42.5 -52
5 7.7 0 20 -63.7 -60
6 19.9 -1 21 -48.6 -71 Two-stages
7 -4.9 -1 22 -58.8 -92
8 -2.6 -2 23 -87 -103
9 -5.8 -3 24 -78.9 -125
10 -15.8 -4 25 -109 -137
11 -12.2 -6 26 -82.6 -175
12 -24.6 -8 27 -73.9 -205
13 -25.6 -11 28 -109.5 -229
Figure 9. SNR gain of the whole mismatched filter with one-stage and
14 -26.1 -16 29 -102.6 -266 two-stage modified SLI filter for Frank code up to length 322.
15 -38.9 -18 30 -139.3 -292
16 -40.1 -25 31 -137 -340 TABLE III. SNR GAIN OF THE MISMATCHED FILTER WITH
17 -54.8 -29 32 -125 -390 ONE-STAGE AND TWO-STAGE MODIFIED SLI FILTER FOR
FRANK CODE UP TO LENGTH 322.

V. SIGNAL TO NOISE RATIO AND HARDWARE COMPEXITY SNR gain (dB) SNR gain (dB)
N
SNR gain (dB) SNR gain (dB)
N
one stage only two stages one stage only two stages
A. Signal to noise ratio 3 -0.35102158 -0.49348834 18 -0.0217718 -0.13610924
It is desirable to increase the MSR as much as possible. 4 -0.26734253 -0.45832054 19 -0.02008881 -0.1280762
Additional SLI stages could improve MSR, but at the cost of
5 -0.06478711 -0.33184785 20 -0.01927225 -0.12213831
longer delay and further reduction in SNR. Considering an
additive white Gaussian noise, the impulse response or noise 6 -0.08194503 -0.51267926 21 -0.01806364 -0.11134211
energy gain of the system is 7 -0.06067401 -0.36799689 22 -0.01740257 -0.11029496
8 -0.05726184 -0.33463105 23 -0.01632499 -0.10541123

2 9 -0.04492416 -0.28852935 24 -0.01578117 -0.10033826


noise energy gain = ∑ h(n ) , (23)
10 -0.04352322 -0.25750734 25 -0.01497906 -0.09598876
n
11 -0.03672403 -0.2244387 26 -0.01447597 -0.09231585
12 -0.03446116 -0.20219653 27 -0.01372678 -0.08788128
where h(n ) is the impulse response of the system. In this
13 -0.03046005 -0.18886223 28 -0.01335728 -0.08570602
scenario, the SNR gain of this system is
14 -0.02908186 -0.18053948 29 -0.01277429 -0.08198614
15 -0.02599234 -0.16182981 30 -0.0123858 -0.07976241
  16 -0.02470397 -0.15383701 31 -0.0118822 -0.07705253
 Eo 
SNRgain dB  = 10 log   17 -0.02273345 -0.14253179 32 -0.01156778 -0.07407018
2
 E ⋅ ∑ h(n ) 
 i , (24)
B. Hardware complexity
where Ei is the energy of the desired part of the input The direct implementation of the matched filter of a
signal. polyphase code of length N2 requires N2 complex
The noise energy gain and the output energy Eo are found multiplications and N2-1 complex additions per output. The
from the impulse response of the whole system and the proposed representation of the matched filter requires only
magnitude of the output signal mainlobe. Table III and Fig. 9 N − 1 complex multiplications and 2N complex additions per
show the SNR of the mismatched filter including the matched output.
R(z ) Y1 (z ) Y2 (z ) Y (z )
... H  (z )
X * (z −1 ) H 1 (z ) + H 2 (z ) + +

µ1 µ2 µ

Figure 10. Mismatched filter with  -stages modified SLI filter

The SLI technique uses the matched filter as a building components of the matched filter where  is the number of
block in each stage. From (9) it can be shown that one stages. In this paper we have considered the optimization of
efficient structure of the matched filter and its conjugate is the mismatched filter for various values of N, but for P equal
used as a building block of the first SLI stage. The hardware unity in the aperiodic Frank polyphase code. Others values of
complexity of the first SLI stage is approximately twice that of P in (1) that are relative primes to N could produce different
the matched filter. This represents 2N-1 complex ACF sidelobes and requires different optimum multiplier and
multiplications and 4N+1 complex addition per output. could results in a different MSR and SNR. The investigation
For the second stage, as implied by (11), the same of Frank code with others values of P as well as the
principle is noticed but now with two building blocks. Since application of SLI mismatched filters for other polyphase
each SLI stage roughly doubles the number of components of codes such as P1, P2, P3, and P4 are a topic for further
the previous one, the whole system, comprised of the matched research.
 +1
filter and  SLI stages, has (2 − 1) times the numbers of
REFERENCES
components of the matched filter, more precisely
[1] N. Levanon and E. Mozeson, “Radar signals,” Wiley-Interscience,
2004.
(2 +1 − 1)N − ( + 1) (25) [2] Turyn, R. “On Barker codes of even length,” Proceedings of the IEEE,
Vol. 51, Issue 9, p.p. 1256- 1256, Sept. 1963.
[3] Frank, R.; Zadoff, S.; Heimiller, R. “Phase shift pulse codes with good
complex multipliers and periodic correlation properties,” IRE Transactions on Information
Theory, Vol. 8, Issue 6, p.p. 381 – 382, October 1962.
[4] P.Z. Fan and M. Darnell, “The synthesis of perfect sequences,”
(2 +1 − 1)(2N + 1) − ( + 1) (26)
Proceeding on Cryptography and Coding, 5th IMA Conference,
Cirencester, UK, December 1995.
complex adders. [5] F. Kretschmer and L. R. Welch, “Sidelobe reduction technique for
polyphase pulse compression codes,” IEEE 2000, International Radar
Conference, 2000, pp 416-421.
The optimized multipliers (Fig. 10) result in one extra
[6] M. H. Ackroyd, and F. Ghani, ‘‘Optimal mismatched filters for
multiplication and one extra addition per output per stage. sidelobe suppression,” IEEEE Transactions on Aerospace and
Electronic Systems, Mar 1973, pp 214-218.
VI. CONCLUSION [7] Benard L. Lewis, “Range-time sidelobe reduction technique for FM-
derived polyphase PC codes,” IEEEE Transactions on Aerospace and
A mismatched filter for aperiodic Frank polyphase codes Electronic Systems, Vol.29, No. 3 , Jul. 1993.
based on sidelobe inversion (SLI) is proposed and [8] Frank, Robert L. “Polyphase codes with good nonperiodic correlation
implemented via a number of stages in cascade with the properties,” IEEE Transaction on information theory, pg. 43–45, 1963.
matched filter. The improvement in MSR due to the SLI [9] T. Saramanki and A.T. Fam, “Properties and structures of linear-phase
filter, in comparison to the matched filter output, is an average FIR filters based on switching and resetting of IIR filter,” IEEE
International Symposium on Circuit and Systems, Vol. 4, pp. 321-3274,
of 4.2534dB for one SLI stage and 14.5815dB for two stages May 1990.
for codes of length up to 322. A very efficient implementation
[10] Adly T. Fam, Indranil Sarkar, and Thomas Poonnen, “Area and power
of the matched filter is also introduced. Its hardware efficient mismatched filters based on sidelobe inversion.” 2008 IEEE
complexity is proportional to the square root of the length of Radar Conference, Rome, Italy, pp. 1262-1267, May 25-29, 2008.
the code while that of the straightforward implementation is [11] Indranil Sarkar, Adly T. Fam. “Multiplicative Mismatched Filters for
proportional to the length of the code. Since the SLI Sidelobe Suppression in Barker Codes.” IEEE Transactions on
technique uses the matched filter or the ACF as a building Aerospace and Electronic Systems Vol. 44, No. 1 January 2008.
block, this imparts the efficient implementation of the [12] Adly T. Fam, Indranil Sarkar. “Multiplicative mismatched filters for
matched filter to the SLI filter. Each SLI stage approximately optimum range sidelobe suppression in barker code reception.” Issued
US patent #7,492,312, February 17, 2009.
doubles the numbers of components of the previous one
giving a total cost that is (2 + 1 − 1) times the number of

Vous aimerez peut-être aussi