Fundamental Limitations On Repeater Jamrmng Frequency-Hopping Communications
Fundamental Limitations On Repeater Jamrmng Frequency-Hopping Communications
Fundamental Limitations On Repeater Jamrmng Frequency-Hopping Communications
receiver during a certain portion of each frequency-hop- This inequality provides a fundamental upper bound on
ping pulse. From the definitions given previously, it fol- the processing time for a repeater that is to respond rap-
lows that the duration of this portion is idly enough to disable a victim receiver. The first two
terms of the upper bound are equal to qTd.
As an example, suppose that the targeted frequency-
hopping system uses continuous-phase frequency-shift
keying (CPFSK) as the data modulation, and the devia-
Consider the repeater jamming of a digital communi-
tion ratio is 0.7. It is assumed for simplicity that the fre-
cation system for which a suitable performance measure
quency estimation is perfect so that 6 = F, ( p , , ) , and that
is the average probability of an error in a channel symbol
the thermal noise in the victim receiver is negligible so
P,. Let ‘Pjdenote the probability that a symbol is jammed
that Fnj = 0. The jamming waveform is modeled as band-
given that the processing time and geometry are favorable
limited white Gaussian noise. Assuming that pt, = - 3 dB
for jamming the symbol. Among the reasons why Pi may
be less than unity are: a) the interception receiver fails to
FJ
and using the results given in [l], we obtain F, = ( p I , )
monitor some of the frequency channels used by the com-
z 0.23. It is assumed that Pso = 0.1, Pi = 1, d2 d3 - +
dl = 30 km, and Td = 1.6 ms. Using ( 6 ) , we then find
municators, b) the repeater jammer lacks the power to jam
that Tprmust not exceed 0.904 ms. If the jamming power
all intercepted frequency-hopping communications, and
so ignores some intercepted pulses, and c) the intercep-
is increased so that p,, = - 6 dB, then F, =
0.29, and we
then find that TPrmust not exceed 1.048 ms.
tion receiver fails to detect some of the pulses. If there
are many symbols per pulse, then the fraction of the sym- 111. FREQUENCY ESTIMATION
bols that are geometrically susceptible to jamming is well Let B denote the bandwidth of a frequency channel, and
approximated by ( Td - Tnj) / T d . Among the susceptible let ufdenote the standard deviation of an estimate of the
symbols, the symbol error probability is PjF, (1 - + carrier frequency of an intercepted frequency-hopping
P i ) Fnj, where F, is the conditional symbol error proba- pulse. From Chebyshev’s inequality, it follows that the
bility given that a symbol is jammed, and Fnj is the con- probability that a random variable is within three standard
ditional symbol error probability given that a symbol is deviations of its mean value is at least 8/9. Thus, if the
not jammed. Among the symbols that are not geometri- frequency estimate is unbiased and 3uf IB / 2 , the prob-
cally susceptible to jamming, which constitute the frac- ability is at least 8 / 9 that the frequency estimate and,
tion TnJ/Tdof the symbols, the symbol error probability hence, the carrier frequency of the transmitted jamming
is Fnj. Thus, after regrouping terms, we find that the aver- waveform will be within the same frequency channel as
age symbol error probability is the intercepted signal. If the repeater responds quickly
Td - Tnj Td - Tnj
enough, then with high probability most of the jamming
P, = ~
Td
PjF,+
( 1--
Td
power will be in the instantaneous passband of the re-
ceiver. The foregoing reasoning implies that a frequency
Let p,, denote the signal-to-jamming ratio in the pass- estimator with
band of the victim receiver. If p l , were deterministic, then B
Uf I- (7)
FJ would be a function of p,,; that is, F, = ( p , . ) . How- FJ 6
ever, the interception receiver must estimate the carrier is a reasonable requirement for an interception receiver to
frequency of each pulse and then transmit a jamming support a repeater jammer.
waveform with the same carrier frequency. If the estimate We model an intercepted frequency-hopping pulse as a
is inaccurate, less jamming power will be transmitted in tone with an unknown initial phase. The Cramer-Rao in-
the receiver passband. Thus, p I , is a random variable, and equality for an unbiased frequency estimate in the pres-
ence of white Gaussian noise is [l]
4 = io
( P I )mdx
F,(ptJ)f(p,,)dpI,7 (5)
UfL
2a2T2E
(T)
’ (8)
where f ( p I , ) is the probability density of p,, and ( p,,)max
is the maximum value of p , , , which is realized when the where T is the duration of the observation interval, E is
frequency estimation is perfect and any other sources of the energy in the intercepted pulse during the observation
random variation are negligible. interval, and N 0 / 2 is the two-sided noise power spectral
Suppose that it is necessary that P, 2 P , , for a com- density. Although it is difficult to accurately determine ut’
munication system to be disabled. We assume that F, 1 for a particular frequency-estimation system, this bound
PTO> FnJsince P, > F, cannot be achieved by a repeater provides at least a rough measure of what might be
and P , = FnJin the absence of repeater jamming. There- achieved in a well-designed system.
fore, P, 2 PTO requires that TnJ< Td. Equations (3) and Let pi denote the signal-to-noise ratio in one frequency
(4) then imply that the processing time must satisfy channel at the input to the frequency-estimation system of
the interception receiver. Thus, E / N , = piBT. Combin-
ing this relation with (7) and (8), we obtain
T L 1J6B-I ( pi ) - ‘ I 3 , (9)
TORRIERI FUNDAMENTAL LIMITATIONS O N REPEATER JAMMING 57 I
* [Td -
Td(psO - F ~ J-) d2
' J ( 4- ' n , )
+ d3 - dl
C
1'.
( 16) I '.. ---_--_---
-- --__
This inequality explicitly indicates the importance of the .,' J I N = 0.07 ---___
-----__-----__
normalized target sector, 2 e / Q , which might be deter- 5 1 '\..,
(22)
0 1 2 3 4 5 6 1
Td (msl
Fig. 4. Required signal-to-noise ratio for time-of-arrival sorting as func-
tion of pulse dwell time. Parameter values are P , , , = 0. I , F, = 0.29, F,,,
= 0, dz + d, - d , = 30 km, and B = 25 kHz.
(23)
which is valid provided that Td ( P , - Fri,) < ( Td - TriJ) V O W E L L TIME-:
I I
( F , - FIv ). If the latter inequality is not satisfied, then
neither (19) nor (23) can be satisfied.
As an example, consider an extension of the examples ,
1 '
RISE
1
'
,
1TIME
FALL IC
I
+
Flil = 0, d2 d3 - d l = 30 km, and B = 25 kHz. Fig. I TIME ?-
-HOP DURATION-
4 shows the lower bound on pI as a function of T,/ for
various combinations of N and Tpr, The curves indicate Fig. 5 . Time durations associated with a pulse.
that a large value of T,:, which is necessary to accom-
modate a long observation interval, is useful to the re- no output. The switching time is the dead time plus the
peater only if T,Iexceeds some minimum value. rise time and the fall time of a pulse. The nonzero switch-
The ideal time-of-arrival sorting described in this sec- ing time decreases the duration of transmitted symbols,
tion depends upon a fixed interval between the frequency- which in turn affects the transmitted spectrum. The dwell
hopping pulses generated by a single transmitter. If the time of a pulse is the duration of the dwell interval during
targeted communication system uses a variable interval, which the pulse has its full amplitude and the channel
time-of-arrival sorting is much more difficult. symbols are transmitted. A buffer circuit is required to
Most of the preceding results are derived under the as- store code or data symbols during the switching time. The
sumption that there are many symbols per pulse. The contents of the buffer are extracted and transmitted during
modifications for other cases, such as the transmission of the dwell time.
one symbol per pulse, are straightforward. The switching time, T,,,., plus the dwell time, Td,,, is
VI. LIMITATIONS O N HOPPINGRATE
equal to the hop duration, Th;that is,
The obvious response to a threat of repeater jamming Th = Td,, + T,,,. . (24)
is to increase the hopping rate. However, the following The rise time, T,, plus the fall time, T f , plus the dead
three major problems arise as the hopping rate increases: time, Tde,is equal to Tsw;that is,
1) the cost of the frequency synthesizer increases and its
reliability decreases, 2) synchronization becomes more T,, = T, Tde , (25) + +
difficult, and 3) limitations on the switching time become Let y denote the required transmission rate of channel
obstacles. The third problem, which usually poses the symbols. The number of symbols that must be transmitted
most fundamental limitations on the hopping rate, is ana- during a hop interval is yT,, if this number is constant
lyzed in this section. from hop to hop. Therefore,
The various time durations associated with each pulse
of a frequency-hopping signal are illustrated in Fig. 5. y Th Ts, Tdw =(26)
The hop duration is the duration of the time interval be- where T, denotes the duration of a channel symbol trans-
tween changes in frequency, which is called the hop in- mitted during the dwell time. Combining (24) to (26), we
terval. The dead time is the duration of the part of the hop obtain
interval during which the frequency synthesizer produces Th( 1 - y T , ) = T, Tf Tde. (27) + +
514 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS. VOL. I. NO. 4. MAY 1989
This equation indicates that y I 1 / T 5 , which is obvious For voice communications, the lowering of y requires an
from the definitions. increase in the implementation complexity and possibly a
The switching time usually cannot be arbitrarily de- loss of voice quality. A substantial increase in rr may re-
creased for the following reasons. quire an increase in the bandwidth of a frequency channel
1) To limit the transmission of spurious frequencies beyond the conventional 25 or 50 kHz.
when an indirect frequency synthesizer is used, it may be Consider a frequency-hopping system for which there
necessary to inhibit the transmitter output during fre- are no problems of multiple-access interference or spec-
quency transitions. tral splatter. If a direct frequency synthesizer [6] is used,
2) If a pulse rises too abruptly, ringing in the output of it is possible to adequately suppress spurious frequencies
the receiver’s intermediate-frequency (IF) filter may se- while keeping T,, negligibly small relative to Td,,..Sup-
riously degrade the first symbol of the dwell interval. pose that there is no code symbol transmitted during the
3) Time may be needed for the IF filter to recover from first T, seconds of the dwell interval. This tactic prevents
one pulse before the next one is received. the amplitude changes, phase changes, and effects of the
4) Spectral splatter, which is the spectral overlap in ex- IF filter memory from affecting the first code or data sym-
traneous frequency channels produced by a time-limited bol in the dehopped signal. Therefore, Th = TdM= yTh T,
transmitted pulse, increases as the rise and fall time de- + T,. Using rh = 1/ T h and r, = 1/ T , , we obtain
crease.
rh = r, - y. (30)
The first symbol of the dwell interval may be seriously
degraded by the phase discontinuities in the dehopped sig- This requirement is clearly much different from that given
nal. A phase discontinuity may be caused by 1) a Doppler by (29).
shift, 2) frequency-selective fading, 3) a nonzero dead
VII. COMMENTS A N D CONCLUSIONS
time, or 4) imperfect synchronization in the receiver.
For combat net radios [3]-[5], which often operate in Since the very beginning of the use of frequency-hop-
close proximity to combat net radios in different nets, the ping communications, the specter of repeater jamming has
most serious limitation on the switching time is often due loomed in the minds of system developers. In this paper,
to the spectral splatter. The spectral splatter from one fre- the Cramer-Rao inequality and other fundamental rela-
quency-hopping signal can cause errors in the reception tions are used to probe the ultimate limits of the effec-
of another frequency-hopping signal even though the two tiveness of a repeater jammer. A lower bound on T, the
signals have widely separated instantaneous carrier fre- duration of the observation interval required for frequency
quencies [ 13. To reduce the amount of splatter into nearby estimation, is derived in Section 111. In Section IV, a lower
nets and to limit the other deleterious effects of a small bound is derived for J , the number of frequency-hopping
switching time, it is necessary that neither T,. nor Tr be signals that must be simultaneously jammed when direc-
much less than T,, which is itself largely determined by tion-of-arrival sorting is used. In Sections IV and V, lower
the bandwidth of a single frequency channel and the need bounds are derived for p I , the required signal-to-noise ra-
to limit the splatter. Thus, we require that tio in one frequency channel at the input to the intercep-
tion receiver, when either direction-of-arrival or time-of-
min (T,., T,-) > CYT,, (28) arrival sorting is used.
where CY is a constant that typically satisfies 1 ICY I4. Considerably larger values of p l , J, and T than are in-
The exact value of CY is a function of the shapes of the dicated by the bounds may be required for the following
pulse edges. reasons.
Let rh = 1/ Th denote the hopping rate and r, = 1/T, 1) The Cramer-Rao bound is known to be usually loose
denote the symbol rate during the dwell interval. Substi- whenE/No < 10dB.
tuting (28) into (27), we obtain 2) Even if the Cramer-Rao bound is tight, it remains a
r - ,. bound that can only be attained by an ideal estimator. A
practical estimator, such as an adaptive thresholding sys-
tem for arrival-time estimation [I], may well exhibit a
This inequality must be satisfied if the spectral splatter, significantly larger standard deviation than is indicated by
and possibly the other problems arising from a short the Cramer-Rao bound.
switching time, are to be accommodated. 3) Various idealizations, such as the perfect estimation
As an example of the application of (as), consider dig- of the hopping rate in time-of-arrival sorting, are made to
ital communications with y = 16 kbits/s, which is the keep the analysis tractable.
standard rate in combat net radios. Let Tde = 0, r, = 20 4) The time required for processing may be signifi-
kbits/s, and CY = 1. Inequality (29) then yields rh < 2 cantly larger than the length of the observation interval;
khops/s. The fraction of the time that symbols are trans- that is, T 5 Tprmay provide a loose bound.
mitted is Td,/Th = y T , = y / r s = 0.8. If CY = 2, then rh 5) Multipath signals degrade both direction-of-arrival
< I khop/s. From these results, it appears that combat and time-of-arrival sorting.
net radios cannot have hopping rates greater than 1 to 2 Despite the potential looseness of the derived bounds,
khops/s unless either 1) serious spectral splatter and other they indicate important irreducible limitations on the ex-
problems are tolerated, 2) y is lowered, or 3) r, is raised. tent of the repeater threat.
T O R R I E R I : F U N D A M E N T A L L I M I T A T I O N S ON R E P E A T E R J A M M I N G 575
The parameter values chosen in the numerical examples 131 P. J . Munday and M . C . Pinches, "Jaguar-V frequency-hopping radio
system,'' IEE Proc., vol. 129. pt. F , p. 213, June 1982.
of Sections 11, IV, and V are those typical of a combat net 141 D. M. Russell, "Tactical radios designed to elude enemy jammers."
radio and a high-power jammer. If the radio hops at a rate Dejense Elecrrori., vol. I I O , June 1983.
of 500 hops/s or more and transmits energy during ap- 151 M. Kachmar, "SINCGARS from ITT-The Army's radios grow up,"
Microwaves & RF. vol. 29. Apr. 1984.
proximately 80 percent of a hop interval so that T,, 5 1.6 161 U. Rohde. Digital PLL Frequency Syrirhesi;crs. Englewood Cliffs.
ms, the analysis indicates that a repeater will be faced NJ: Prentice-Hall, 1983.
with a severe challenge when many frequency-hopping
communicators are present. However, if T(, > 3 ms,
which is true for all or nearly all currently used or planned
combat net radios, then signal sorting and successful re-
peater jamming are potentially much simpler. In Section Don J . Torrieri received the B S degree from the
Massachusetts Institute of Technology, Cam-
VI, the limitations on the hopping rate are examined. bridge, the M S degree from the Polytechnic
From the results, it appears that a combat net radio that University, New York, and the Ph.D. degree i n
hops at 500-1000 hops/s is a viable possibility, but one electrophysics from the University of Maryland.
College Park.
that hops at more than 1 khop/s may create unacceptable He ha5 analyzed electronic systems at the Nd-
levels of spectral splatter and other practical problems. val Research Laboratory, Washington, DC. and
since 1977 he has been employed by the United
REFERENCES States Army His primary interests are the analy-
sis and assessment of communication systems and
[ 11 D. Torrieri, Principles of Secure Communication Systems. Dedham. electronic countermeasures He is the author of Princrples of Secure Coni-
MA: Artech House, 1985. munication Systems (Dedham, MA. Artech House, 1985) and a contributor
[2] L. Van Brunt, Applied ECM, Vol. I . Dunn Loring, VA: EW Engi- to Acousro-Optic Signal Processrng (N. J . Berg and J N. Lee, Eds , New
neering, 1978. York: Marcel Dekker, 1983)