Use of HF for Extended-Range Combat Net Radio (CNR)

Communications
E.N. Koski*
*Harris Corporation, RF Communications Division
1680 University Avenue
Rochester, NY 14610 USA
Abstract
In recent years military forces have become more mobile to
meet contemporary mission requirements, and are frequently
required to cover much larger geographic areas. As a result,
they frequently outrun the transmission range of traditional
VHF and UHF line-of-sight communication systems. Similar
issues can arise in non-mi Iitary situations such as border
security, drug interdiction, or natural disaster response. An
emerging solution to the range problems is the use of HF
radios for extended-range Combat Net Radio (CNR)
applications. HF signals have the ability to propagate to and
over the horizon using either sky wave or ground wave
modes. This unique ability has allowed HF radio to address
the coverage gaps not served by VHF/UHF radio and other
Beyond Line of Sight (BLOS) media. A key issue to be
considered when designing and using an HF radio network
for this purpose is that of network traffic capacity, due to the
delivery latency and transmission overhead of HF voice
communications waveforms and techniques. This paper
discussesthis issue and presentssome potential solutions.
1 Introduction
HF radio is an appealing potential medium for Combat Net
Radio communications due to its potential for range
considerably exceeding that of the Iine-of-sight
communications media typically used for these applications:
VHF and UHF radio. HF communications systems used for
'extended-range combat net radio' scenarios are likely to be
subjected to relatively heavy voice traffic loads in networks
containing larger numbers of users than are typically seen in
HF long-haul scenarios. In these extended-range CNR
scenarios, voice traffic capacity can be an important
performance attribute. This situation is compl icated by the
desire of many users for 'all-informed' voice
communications, in which each net member is capable of
hearing transmissions by any other net member. This
precludes increasing network capacity by spreading the voice
traffic across multiple frequencies, as would typically occur
in networks using Automatic Link Establishment (ALE).
2 Network voicetraffic capacity
When standard HF data waveforms are used for digital voice
communications, the resulting end-to-end voice latencies can
be relatively large [3]. These latencies result from design
38
attributes of the MIL-STD-188-110B data waveforms giving
them reliable performance on skywave channels: the initial
preamble used for acquisition and synchronization, and the
600 ms short interleaver used at data rates of 75 to 2400 bps
[6]. These waveform attributes also create transmission
overhead: an amount by which the RF transmission on the
channel exceeds in duration the voice audio payload being
conveyed. The entire 600 ms preamble used in MIL-STD-
188-110B short interleaving results in an equal amount of
overhead [6]. Additional overhead - roughly 300 ms on
average - results from the presence of an entire interleaver
block at the end of a transmission, even if the final data bits
only partially fill the last interleaver block. Still further
overhead can result from other data included in the
transmitted digital data stream.
HF voice communications networks can suffer from
diminished capacity because of the increased latency and
transmission overhead. Latency is probably the more
important of these two factors due to its creating opportunities
for transmission collisions in which two or more radios
transmit simultaneously, frequently causing neither
transmission to be received successfully. Networks
containing relatively large numbers of users (say, 10 or more)
with heavy traffic loads can suffer from frequent coil isions
between voice transmissions. When a collision occurs, the
receiving radio's modem is typically captured by the first
voice transmission to arrive; any later transmission causes
interference resulting in bit errors in the received data stream.
These can lead to corruption of the received voice signal or -
in secure communication modes - to loss of the entire
incoming transmission if bit errors cause crypto
synchronization to fail.
The increased latency of the HF voice communications
techniques is the principal factor responsible for the increased
occurrence of collisions. To understand this, it helps to view
an HF voice network as an example of a communications
network using a CSMA (Carrier Sense Multiple Access)
media access control (MAC) technique. Before transmitting,
radio operators listen for traffic on the channel in order to
avoid causing a collision by transmitting. Their listening to
the channel fi lis the role of a 'carrier sense' mechanism in a
CSMA system, so that an HF voice communications network
could be properly called a 'hand-operated CSMA' system.
Large end-to-end latencies such as can occur in an HF voice
0.7 .-- - - - - - - - - - - - - - - ---,
Figure 1. Throughput 5 vs, Offered Load G for 1-persistent
CSMA
ignores actuation of the handset key switch so as to
prevent the operator from causing a collision. This has
the benefit that it eliminates any possible human reaction
time [required to recognize the visible or audible
channel-busy indication) from the effective latency.
1
Offered traffic, G
0.1
0.6
Vl 0.5 +-
'5 ...... a =0.03
B- 0.4 -X-a=0.10
g> ...0- . . a =0.30
--+- a=l .00
I-
To understand the potential benefits and limitations of these
approaches, it is useful to have a reasonably accurate and
realistic model of an HF voice communications network from
which performance estimates can be derived by mathematical
analysis and/or simulation.
We can see that reducing the latency can yield a meaningful
improvement in network capacity; however, the upper limit of
53.8% seems disappointing. This limit is imposed by the 1-
persistent CSMA model's assumption that, whenever a
transmission is attempted while the channel is busy, the user
As a first rough approximation, an HF voice network can be
viewed as a 1-persistent CSMA system [1,2]. Kleinrock and
Tobagi [2] give a throughput analysis of 1-persistent CSMA,
for a model system with infinitely many nodes; such an
analysis provides useful approximate throughput estimates for
networks with 10 or more users. Figure 1 shows network
throughput 5 as a function of the offered load G for various
values of end-to-end latency, here normalized to a fraction of
the transmission duration (which is assumed to be constant).
G is the sum of the transmission durations of traffic arriving
at all nodes, divided by total time. 5 is the sum of the
durations of all successfully delivered traffic, divided by total
time. A fully-loaded ideal network might have six ten-second
voice messages arrive per minute for an offered load (G) of
1.0, and deliver them all successfully for a throughput (5) of
1.0. As we can see, network throughput is substantially
determined by the end-to-end latency. In the best case, with
zero latency, throughput reaches a maximum of
approximately 53.8%. For a transmission duration of 10
seconds and latency of three seconds, the value of a (latency
normalized to a fraction of the transmission time) is 0.3,
giving a maximum network capacity of approximately 36.4%.
The transmission overheads described above could reduce the
effective network throughput by a further 10%, resulting in a
maximum throughput of only about 32.7%.
Candidate methods to reduce the total voice latency might
involve providing an indication that the channel has become
occupied based on some event occurring earlier in the receive
process shown in the time-line. Two particular events could
be used: crypto sync (occurring when the receiving COMSEC
device detects its synchronization preamble), and modem
preamble sync (occurring when the receiving modem has
processed the entire modem preamble) . When the chosen
event occurs, we could either:
Attempt to reduce the likelihood of a collision by
providing some sort of warning to the receiving operator
that the channel is now busy. This could take the form of
a visual indication and/or an audible signal; or
Attempt to prevent a potential collision by prohibiting the
operator from transmitting. When the receiving radio
determines that the channel has become busy by
detecting either of the two events identified above, it
communications system limit the effectiveness of this manual
'Iisten-before-transmit' technique. The latency creates a
time-window in which an operator's pressing the radio's key
switch can cause a collision even though the operator has not
yet heard (or has not recognized) an incoming transmission.
If possible, we would like to addressthis problem by reducing
the end-to-end latency, preferably without changing the
underlying voice coding or modem waveform in order not to
forfeit their performance advantages. We also prefer not to
require development of additional voice coding techniques
and waveforms, which could be expensive to procure and
would complicate system operation and management [3].
Fortunately, significant latency reductions may be possible
without changing the underlying voice coding and waveform.
The total latency from the key switch actuation to the
receiving operator's hearing voice has both physical and
human elements. The latency can be expressed as + tprop +
t
rx
+ tsp + tn, where
ttxrepresents the latency from the transmitting operator's
keying the handset to the transmitting radio's beginning
to emit RF energy; this is determined somewhat by the
properties of the waveform and voice coding, but
primarily by the design of the transmitting radio.
t
prop
(propagation delay) should be less than 10 ms at
most for the extended-range CNR applications
considered here.
t
rx
represents the latency from the arrival of RF energy at
the receiving radio to the start of the audio signal played
to the receiving radio's handset; this is determined by the
waveform, voice coding, and radio design.
top represents the latency from the removal of hold-off
tone in the transmitting operator's handset to the
operator's beginning to speak: a human latency whose
duration can be as much as 500 to 750 ms or more.
t
n
representsthe latency from the arrival in the handset of
the start of the transmitted operator speech, to the
receiving operator's recognizing the speech so as to be
able to avoid transmitting and causing a collision. This is
another human latency (reaction time) whose duration
can be as much as 200 to 400 ms or more.
39
Figure 3. Impact of 'operator persistence' behaviour
To make our OPNET simulation model a more accurate
representation of a voice network of human operators, we
replaced the exponential distribution of 'persistence latencies'
in the Takagi and Kleinrock model [4,5] with a statistical
distribution matching the observed behaviour of human
operators in laboratory measurements: a random persistence
latency with a mean of approximately 660 rns,
Figure 3 compares the levels of throughput obtained with the
resulting model with those of 1-persistent CSMA for various
values of a. (The series labelled "oper" are for 'operator
persistence' behaviour based on our observations; those
labelled "1/p=0" are for 1-persistent CSMA.) For large or
moderate values of a, the 'operator persistence' behaviour
even reduces capacity. The random portion of the 'operator
persistence' delay duration is insufficient to appreciably
reduce the likelihood of collision, so the delay in transmitting
after receiving merely adds overhead. 'Operator persistence'
appearsto significantly reduce the frequency of post-channel-
busy collisions only once the value of a becomes very small:
0.02 or less, representing an extremely ambitious latency
reduction of more than 90% from our initial value of
approximately three seconds.
10
G
OL-_ _ _ _ __.....L.
0.1
0.7 r""' ......,...= "....-"....-......,......,..,..----,--= "".,..,..,.,.". "....--=",..,.., --a=0.32, IIp=O.OO
...[}.. a=0.32, oper
0.6 f---- - - - - - ---:-,...-)(;-- --I ----.- a=0.10, lI p=0.00
.../:; . .. a=0.10, oper
0.5 IIp=O.OO
.... .. .o-.. a=0.04, oper
0.4 l /p=O.OO
.. -X .. a=O.Ol , oper
x......
til
This leads one naturally to the thought: "If the existing p-
persistence isn't sufficient, why not make more?" A voice
communications system could add an explicit random ' dead
time' following each received transmission, to more
effectively prevent collisions among two or more users who
initiate transmission attempts during a time in which the
channel is busy. The receiving radio could implement a
precisely time-slotted ' persistence delay' scheme, benefiting
from the increased efficiency of such a scheme in preventing
collisions (in the same way that Slotted Aloha provides
greatly increased throughput relative to Un-slotted Aloha [1]).
A user wishing to transmit immediately following a channel-
busy period would be allowed to do so only starting in a
randomly-chosen time slot. The operator would press and
hold the key switch to transmit. The radio would either
commence transmission once the chosen time slot was
reached, or abort the transmission and process an incoming
transmission if one were to arrive earlier than the chosen time
slot. The slot duration would have to be at least equal to the
0.7
-o-a =0.10,1/p=0.00
.. .. a =0.10, 1/p=0.05
0.6
--+ .- a =0.10, 1/p=0.10
_.. - a =0.10, 1/p=0.15
......... ..
0.5 -o- " a =0.10,1 /p=0.20

--a =0.10,1/p=0.40
.......... +., ' . .......
. " " ".. ',.....
0.4 a =0.10, 1/p=0.60

\;.{. til
.. .... a=0.10, 1/p=0.80
\ . \. ' .
\ ... '. ... ".
0.3
/'
'it:'-.
" \':' "
"\'0. .....
...... .....
0.2

,\."'" <:
0.1



0
' ... ...:..:. :- .
0.1 1 10
G
Figure 2. Un-slottedp-persistent CSMA, 16 nodes.
waits until the channel becomes free and immediately
thereafter starts to transmit. Whenever two or more
transmissions are attempted during the same channel-busy
period, they inevitably collide once the channel becomes free.
However, this is not a completely accurate representation of
radio operator behaviour: there is some random time interval
after the channel becomes free before an operator attempts to
transmit. The difference between the times at which two
users attempt to transmit could conceivably give the later of
the two an opportunity to recognize the earlier one's
transmission on the channel and avoid a collision. To address
this possibility, we replace our 1-persistent CSMA model
with a p-persistent CSMA model [1,2,4,5].
Takagi and Kleinrock [4,5] provide an analysis of the
throughput of an un-slottedp-persistent CSMA system. For
our purposes, their model formulation has a number of
advantages: they provide analysis results for systems
containing finite numbers of nodes, and their un-slotted
model is a more accurate reflection of the behaviour of human
operators than a slotted p-persistent model would be.
However, the analysis formulas they give are not in closed
form, making direct computation of the network performance
estimates one might need a forbidding task. For this reason,
we prepared an OPNET simulation model of an un-slotted p-
persistent CSMA system conforming to their model definition
and obtained excellent agreement with the throughput values
presented in the paper. In the Takagi and Kleinrock papers,
1/p is the mean value of an exponentially-distributed
'persistence latency', the time that elapses between the
channel's becoming free and a waiting radio's starting to
transmit, so that p = infinity gives the equivalent of a 1-
persistent CSMA system.
Figure 2 shows the impact of different values for 1/p on the
throughput of a 16-user network with a =0.1. Only when the
value of 1/pappreciably exceedsthat of a does the addition of
p-persistence seem to provide a significant performance
benefit. This is as we would expect: for p-persistence to
prevent a collision after the channel becomes free, the
difference in the times at which two users attempt to transmit
must exceed the time required for the second user to detect
the transmission by the first.
40
10
-o-a=0.06, p=0.09, w=0.60
- v-X .. a=0.06, p=0.13, w=0.40
- . . - a=0.06. p=0.23, w=0.20
p=0.38, w=0.10
.. -+ .. a=0.06. p=1.00, w=o.oo .'
' ."" '+-' .
0.7
0.6
0.1 +r=-- - - - - - - - - - - - -----.:....n--- -----lI
o
0.1
0.8
Figure 5. Slottedp-persistent CSMA with reduced latency
G
0.3
0.2
Figure 6 shows the manner in which latency reduction and
explicit p-persistence combine to substantially improve
network capacity. Recall that with p = 1.0, maximum
capacity was limited to 53.8% even for a =O. Incorporating a
mean 'persistence wait' of w = 0.2 permits a maximum
throughput approaching 80% to be achieved.
collisions, the mean wait time wmust be so large as to create
a large amount of channel-idle time, reducing throughput so
much that the addition of explicit p-persistence yields no
significant increase in maximum throughput.
Figure 5 shows throughput as a function of offered load for a
similar network but with a = 0.06, representing a 600 ms
latency. Even with no p-persistence (p = 1.0), the greatly
reduced latency is obviously beneficial, yielding a maximum
throughput near 48%. Explicit p-persistence yields a further
substantial increase in throughput up to a maximum of
approximately 64%. The smallest p values (corresponding to
w values greater than 0.2) did not significantly increase
maximum throughput, although they did increase throughput
values for G 1. This was achieved at the price of some
reduction in throughput for G < 1. For a variety of latency
values (a), we have observed that fixed pvalues resulting in w
= 0.2 appear to yield a near-optimal increase in maximum
throughput without significantly reducing throughput for G <
1.
10

0.1
. . -+. . a=0.32, p=1.00, w=O.oo
p=0.76, w=0.10
- . -a=0.32, p=0.62, w=0.20
0.5 1--- - - - - - - - - ------1 " .x " a=0.32, p=0.44, w=0.40
-o-a=0.32, p=0.35, w=0.60
0.6
0.2
Cfl 0.3
0.1
effective latency plus the maximum propagation time, to
permit a transmission commenced in slot n to be detected
prior to the start of slot 11+1. Such a scheme would be an
application of a slottedp-persistent CSMA protocol [1,2,4,5].
In slottedp-persistent CSMA, the time following the end of a
received transmission is divided into a sequenceof time slots
of duration a. In each slot, if it has not detected a new
transmission on the channel, each station with new traffic
begins to transmit with probability p. (Note that this
definition of p differs from the definition employed in
discussing un-slotted p-persistent CSMA [4].) Slotted p-
persistent CSMA with a transmission probability p = 1 is
equivalent to un-slotted p-persistent CSMA with a mean
persistence latency 11p = O.
To investigate the level of capacity improvement that could
be achieved by means of an 'explicit p-persistence' scheme
for voice communications, we prepared an OPNET
simulation model containing a slotted p-persistence scheme
conforming to the model definition provided by Takagi and
Kleinrock [4]. Our throughput results matched very closely
the analysis results provided in their paper.
The end-to-end latency a would crucially determine the
effectiveness of such a scheme. A smaller latency value
would allow the slot duration to be small, permitting a smaller
slot transmission probability p so as to more effectively
prevent collisions, without excessively increasing the idle-
time overhead resulting from idle slots prior to the start of a
post-channel-busy transmission .
G
Figure 4. Slottedp-persistent CSMA with large latency
Figure 4 shows throughput as a function of offered load for a
16-node network with a= 0.32, corresponding to a 3.2 second
latency with a fixed transmission duration of ten seconds,
with a range of values of the transmission probability p. For
each p value, a corresponding value w is provided, which
represents the mean wait time once the channel becomes free
for each transmission initiated during a channel-busy period.
w = 0.2 indicates that the mean wait time for each
transmission is 0.2 times the transmission duration, or two
seconds. In this case, the slot size must be quite large (> a).
As a result, in order for the slot transmission probability p to
be low enough to significantly reduce the frequency of
Figure 7 shows the increase in throughput attributable to
explicit p-persistence for a values of 0.32,0.1, and 0.04. The
beneficial impact of explicit p-persistence increases markedly
as latency is reduced below a = 0.1. However, even for a =
0.1, the beneficial impact of adding explicit p-persistence is
much greater than that of reducing latency to as little as 0.04
without adding expl icit p-persistence. The data presented in
Figure 7 suggest what might be a feasible course of action to
significantly improve the capacity of an HF 'extended range
CNR' voice communications network. From an initial
latency near three seconds, a latency reduction to roughly one
second (a = 0.1) is probably feasible based on detecting the
MIL-STD-188-110A modem preamble as early as possible,
increasing network capacity to near 45%. Further latency
reductions have diminishing benefits because they do little to
41
reduce the occurrence of post-channel-busy collisions.
However, adding ' explicit p-persistence' immediately
increases network capacity to near 55%, and makes possible
significant further capacity increases through further latency
reduction.
0.9
" ' 0-" a =0.02, p=0.09, w=0.20
,x , a =0.04, p=O.17, w=0.20
0.8
---+- a =0.06, p=0.23, w=0.20 .Q.
- v ,
-v,
'0-,
0.7
" -0-" a =0.08, p=0.29, w=0.20 , ,-X - _ '.
--+--a=0.10, p=0.33, w=0.20
..
0.6
--a =0.16, p=0.44, w=0.20

0.5
" ' Is" a=0 .24, p=0.55, w=0.20
:__ i:.:' , . n -.
U1 x a=0 .32, p=0.62, w=0.20

.'8... ....,
0.4
x .. _ ... ",," 0. '"
0.3
'.. ..;;.". - ";': 1. " 'Y\ ". -,
<, -. "'" W ".
0.2
..)!C.... ' xl. .... ", ". .
/ .!f"
" A '" -.
0.1
' x...
...',6. "----'
a
'-i:':':""
0.1 1 10
G
Figure 6. Impact of latency reduction on capacity, slotted p-
persistent CSMA
0.9
__a=0.32, p=1.00, w=O.OO
" ' 0-" a=0 .32, p=0.62, w=0.20
0.8 ---+- a=0.10, p=1.00, w=O.OO
0.7
" ' 6'" a=0.10, p=0.33, w=0.20 . .. 0..... .
__a=0 .04, p=1.00, w=O.OO
. 0 "
0.6
"'0- " a=0.04, p=O.17, w=0.20 v-,
. > " 6 .... .. ..
11
. .
0.5
U1

'6-
0.4
.. D..\-""-
0.3
.s-:
'"
"".
0.2

-,

0.1
.. ..
a
0.1 1 10
G
Figure 7. Increase in capacity due to application of slotted p-
persistent CSMA for different latency values
3 Conclusions
HF communications systems used for 'extended-range
combat net radio' scenarios are likely to be subjected to
relatively heavy voice traffic loads in networks containing
larger numbers of users than are typically seen in HF long-
haul scenarios. In these extended-range CNR scenarios, voice
traffic capacity can be an important performance attribute, as
it can in similar non-military applications. The
communication techniques used in HF digital voice
communications systems give rise to relatively large end-to-
end voice latencies, which can frequently be on the order of
three seconds or more. By modelling an HF voice
communications network as a 'hand-operated CSMA' system,
we observed that this amount of latency can significantly
reduce network capacity by increasing the frequency of voice
traffic collisions. Reducing latency to one second or less
would be likely to improve network capacity; however, the
degree of improvement achieved in this way would be limited
42
by the occurrence of collisions whenever two stations both
attempt to transmit immediately following the end of a
preceding transmission on the channel. The addition of an
'explicit p-persistence' scheme could be used to further
significantly improve the capacity of an HF voice
communications network, but only after latency has been
reduced to a sufficient degree. Once 'explicit p-persistence'
is added, further latency reductions continue to yield
significant improvements in capacity which would not be
possible without use of this technique.
Acknowledgements
The author gratefully acknowledges the ingenious and
diligent assistance of Will Fletcher in preparing the OPNET
simulations for this paper, and of Don Hewitt, Tom Lochner,
and William Batts (and their unwitting test subjects) in
measuring the various forms of latency discussed in the paper.
References
[1] J. L. Hammond, P. J. P. O'Reilly. Performance Analysis
of Local Computer Networks. Addison-Wesley (1986).
[2] L. Kleinrock, F. A. Tobagi. "Packet Switching in Radio
Channels: Part I - Carrier Sense MUltiple-Access Modes
and Their Throughput-Delay Characteristics", IEEE
Transactions on Communications, Vol. COM-23 No.
12, pp. 1400-1416, (1975).
[3] J. W. Nieto. "Waveform Design Considerations for
Transmission of Digital Voice over HF", Proceedings of
the 2004 IEEE Military Communications Conference
MILCOM2004 (2004).
[4] H. Takagi, L. Kleinrock. ''Throughput Analysis for
Persistent CSMA Systems", IEEE Transactions on
Communications, Vol. COM-33 No.7, pp. 627-638,
(1985).
[5] H. Takagi, L. Kleinrock. "Correction to 'Throughput
Analysis for Persistent CSMA Systems''', IEEE
Transactions on Communications, Vol. COM-35 No.2,
pp. 243-245, (1987) .
[6] US MIL-STD-188-110B, "Military Standard:
Interoperability and Performance Standards for Data
Modems." US Dept. of Defense (2000).