ON-OFF KEYING ASSISTED ACQUISITION SCHEME FOR BURST MODE
DS/SS PACKET RADIO t
Desiiiond Yan and Paul Ho
Sinion Fraser University, Buniaby, B.C., Canada. V5A 1S6
email: desmondQcs .sfu.ca,paulQcs .sfu.ca
A l s f r a e f - Code Division Multiple Access (CDMA) is a promis-
ing technology for future digital cellular and personal coinn~uui-
cations. To despread the pseudo noise (PN) code and thus under-
stailding the message sent, syncluonieation (acquisitionand track-
ing) is necessary in priori. As a matter of fact, the throughput
of a packet inode CDMA system is directly related to the perfor-
mance of its fraue syiicllroilizatioii algoritlun. hi this paper, we
propose a uew acquisition tecluuque for packet mode CDMA op-
erating in the Rayleigh fading clmmiel. The teclucque is based on
on-off keying (during the preamble section) followed by a Markov
chain search. The leiigtli of the preamble is a design parameter
and the perfonumice measure is the tluougliput of packets. We
have tested our new scheme on a flat fading and a selective fading
cliaiuiel, based on a fixed blocking probability criterion. Specifi-
cally for the latter case, we adopt a FLAKE receiver to combat the
iiiultipatli effect, and introduce a windowing tecluiique wllicl~we
found effective in overcoilling the teiuporary increase in noise level direct sequence (DS) spread spectrum packet radio system
caused by partial correlation of the PN sequence used. Results in
both cases show that the proposed on-offkeying tecluuque is iiiore
efficient than many others in the literature such as the coincidence
detector or simple Markov search.
I. INTRODUCTION
Advantages of spread spectrum or code division multiple
access (CDMA) such as high capacity or multipath miti-
gation have made it a potential candidate for future packet
radio system [l]. To despread the received pseudonoise ( P N )
code and thus understanding the message sent, however, syn-
chronization is necessary in priori. T h e process of synchro-
nization usually involves two stages: acquisition and track-
ing. T h e foriner aims to detect the start of transmission (the
presence of signal) and find a coarse alignment between the
incoming signal and the locally generated PN code, while
the latter focuses on fine adjusting the phase, usually by a
delay-locked loop or a tau-dither loop.
A common practice for setting the correlator threshold is
by Neyman-Pearson statistics to ensure a chosen false alarm
probability, Pfa.In addition, a second verification proce-
dure is usually performed to enhance performance such as
the coincidence detector [a, 31, and searching for additional
threshold crossings [4, 51.
In a packet radio system, the preamble length is fixed
and the acquisition time [3]-[SI may not be an appropriate
measure. We believe that throughput is.a more desirable
measure of performance. In our study, we use a fixed block-
ing probability (&, probability of finding the receiver busy
'This work was supported i n part by the NSER(: of Canada and
in p a r t by t h e Canadian Institute for Telecommunications Research
under t h e Canadian government's Networks of Centers of Excellence
program
582
O-7803-1927-3/94/$4.00 @ 1994 IEEE
Fig. 1. Packet Data Format.
when a packet arrives) criterion, in contrast to the constant
false alarm rate. Based on a Poisson distribution of false
alarm arrival, we derive equations for f b for various meth-
ods and use them to set the thresholds accordingly. In this
way, the throughput is given by (1 - pb) x pd, where pd is
the detection probability.
In this paper, we propose a new acquisition strategy in
which utilizes on-off keying during the preamble section.
Simulation results show that this method outperforms the
others in both a flat fading and a frequency selective fading
channels.
I I. SYSTEMDESC:RIPTION
In a mobile radio packet system, d a t a is sent in pack-
ets, with the beginning of each packet being a fixed length
of preamble used for synchronization. Figure 1 shows the
packet data format where the preamble consists of L sym-
bols.
We have made the assumption here that the length of each
symbol equals one PN code.
A . Flat Fading Channel Model
Assuming BPSK used for the modulation of data, the
transmitted DS-SS signal is
~ ( t=) A . b ( t ) . P N ( 1 ) . C O S ( W+~e),
.~ (1)
where A is the amplitude of carrier; b ( t ) is a sequence of
{ f l } d a t a symbols; wr is carrier frequency; P N ( 1 ) is PN
sequence of intended receiver; and B is phase offset. During
acquisition, a stream of bits 1 are sent and we can neglect
b ( t ) . In a Rayleigh fading channel with a maximum doppler
frequency of fn, the received signal is
~ ( t=) A . g ( t ) . P N ( t ) c o s [ ~ c +,B
t + 4(t)] + ILW(~), (2)
where g ( t ) is the Rayleigh fading envelope, d ( t ) is a uni-
formly distributed random phase, and n w ( t ) is the channel's
additive white Gaussian noise (AWGN). It is assumed that
g ( l ) has a mean square value of 2.: and the AWGN has a

Fig. 2. Noiicohereiit Detector Structure
two-sided power spectral density of N 0 / 2 . T h e received sig-
nal is assumed to be first brought down to baseband before
correlation and despreading takes place; see figure 2. The
resultant complex baseband signal can be written as
w(t) = u ( t ) +j .v ( t ) ,
where
u(t) = A z ( t ) P N ( t ) nr(t),+
+
~ ( t )= A y ( t ) P N ( t ) r c y ( t ) ,
and z ( t ) = g(t)cos[+(t) +e], y(t) = g(t)sin[d(t) B ] are two +
independent zero mean Gaussian processes that represent
fading; while n r ( t ) and n ~ ( t are
) the equivalent low-pass
AWGN. For typical mobile radio channel, both z ( t ) and y(l)
have an autocorrelation function of
p(T) = I ( t ) .I(t + T) = y(t). Y(t + T) = U;.Jo(2XfDT), (4)
where Jo(.) is the zero order Bessel function and as men-
tioned before, fo is the doppler frequency. On the other
hand, the power spectral density of the low-pass AWGN,
n r ( t ) and tiy(t), are both N O .
When the local PN sequence is in synchronization with
the transmitted one, the output of the energy detector, E l ,
follows an exponential distribution
P(El)= T
1
C
-2,
2ai
where U: = Au~T+ N O T . When the transmitted and local
PN waveforms are not in synchronization, or when there is
no signal present, the received signal energy is again expo-
nentially distributed
with U: = N O T . From the pdf in equation (6), we can select
appropriate thresholds for a given false alarm rate, which in
turn is determined by the desired blocking probability, Pb.
In the above model, we do not account for power control
which is possible to eliminate some fading effect. Perfect
power control can maintain the received signal at a constant
SNR and the channel is reduced to a AWGN channel. In
real situations, however, non-ideal power control will result
in performance in between an AWGN and a fading chan-
nel. Although not presented in this paper, simulations on a
AWGN channel also prove the on-off schemes superior per-
formance to others.
I V
Fig. 3. On-Off patteni in Preamble
(3)
HIT MLSS
Fig. 4. On-Off Keying Assisted Search Strategy
111. ON-OFFKEYING ASSISTED DS/SS ACQ~JISITION
SYSTEM
In the On-Off Keying Assisted DS/SS Acquisition Sys-
tem, we vary the power in the preamble section following an
on-off pattern. The power varying capability is necessary in
any (:DMA system to overcome the near-far problem, thus
this scheme appears to impose no additional hardware re-
quirements. For simplicity, lets assume that the length of a
single on state equals one PN sequence length. Figure 3
shows an overall view of the new acquisition system, assum-
ing a correlator length of one PN sequence. Since an on-off
scheme is adopted, we can double the power during the on
state, while still maintaining the same average power as in
other schemes. Depending on the false alarm rate needed
(calculated from the blocking probability), we set different
thresholds for the on and off periods, say TI and so re-
(5)
spectively. For the receiver, who does not know whether
the current pulse sent is on or o f f , keeps on tracking
threshold T I in the first place. Once the first threshold is
detected, the algorithm continuously looks or alternate TO
and T I until either acquisition is reached or the search being
terminated as the number of state transitions exceeds the
preamble length L (explained in later sections). Figure 4
shows the search strategy used. T h e TI or TO in each node
indicates the current threshold in each state. A HIT r e p
resents correlator output exceeding the threshold, while the
MISS is the opposite. Note that a MISS (not a HIT) will
cause a transition from state 2 to state 3 because falling be-
low the threshold means a high probability for the presence
of noise.
Intuitively speaking, the benefits of this method are three-
fold: first, a higher power in the on state increases its resis-
tance to Rayleigh fading; second, the uoff state which now
consists of only AW(:N, is not subject to any fading; third,
we gam an extra degree of freedom in setting the threshold
for the off state - since only noise is present, we can set
the threshold as high as desired so that the probability of
going from state 2 to state 3 (ref. Figure 4) is very high,
approaching more towards acquisition state 4.
...
 Fig. 5 . On-Off Keying
A . Probability of False Alarm
We will use the Markov chain approach t o find Pf,. Fig-
ure 5 shows the Markov chain model. In the figure, Poo(P0l)
denotes the probability of crossing T O ( T : ) by noise. Further-
more, a reject state is introduced because returning to state
1 will not cause a false alarm. We implement an additional
mechanism in our acquisition: a counter is kept for the num-
ber of transitions since the first T I threshold crossing. This
counter will be increased by one for each additional transi-
tion and the search will terminate if the counter exceeds L.
T h e mechanism is able to lower the total number of com-
binations to reach false alarm, thus producing a lower false
alarm rate, given particular TI and TO values.
T h e probability of false alarm is
where Q is the transition matrix of the Markov chain in
figure 5. Subscript (1,4) in the above equation denotes the
first row, fourth column element in the matrix Q raised to
power L.
El. Blocking Probability
To derive the blocking probability, we use a similar ap-
proach as those in [9, IO]. T h e blocking probability is given
bv
where A is the expected time measured from the end of a
busy period to the next false alarm arrival, and B is the ex-
pected busy time due to false alarm. Putting various possible
busy time into consideration, R can be expressed as
where L D is the d a t a length, and T is one symbol (bit) du-
ration, and z = (1 - Poo)(l - POI).
The first term in the equation (9) accounts for the time
decoding the whole packet due to false alarm; and the second
term accounts for various possible busy time periods for the
receiver to reach the Reject state when the verification pro-
cedure is not passed. T h e expected false alarm inter-arrival
time A can be obtained based on a Poisson distribution and
given by equation
A = -,1,
T, is one chip duration.
Po 1
Having chosen the desired value of Po0 (TO is fixed), and with
a particular value of blocking probability Pb, the value of TI
can be calculated with equations (8) - (10).
When t h e preamble length is set t o L , t h e r e is n o reason t o search
for more t h a n L symbols.
Fig. 6. Siiiiulatioii Results for a Blocking Probability of
As seen from the expression for the blocking probability,
it accounts not only the packet decoding time with. a false
alarm, but also various other busy periods possible when not
passing the verification; therefore, it is a more appropriate
criterion in the packet radio environment, while the constant
(7) false alarm rate (CFAR) assumes a fixed penality price.
CI. Other Techniques f o r Comparison
To put our results in perspective, we will compare our
algorithm with the followings: the coincidence detector, and
two other techniques [4, 51 relying on additional threshold
testings.
Iv. SIhWLATION RESULTSO N FLATFADING
C:HANNEL
In our simulations, we have used a fade rate of IOOHz, and
a data rate of 16kbit/s. The fading is assumed to stay con-
stant during one symbol. Figure 6 shows simulation results
for a blocking probability of IOv5. Note that the SNRs refers
to the correlator output, not the received signal.
We can make two observations from the results. First, the
on-off assisted keying scheme always performs better than
the others, regardless of the SNR. Second, increasing the
preamble length will cause increase in throughput, except
for the conincidence detection where the extra busy time
encountered with increasing preamble length offsets the gain
in longer coincidence period.
v. FREQUENCY
SELECTIVEFADnvc; CHANNEL
In a frequency selective channel, we receive mutipath sig-
nals each of them is subject to Rayleigh fading3. A RAKE
receiver has been found effective to combat the multipath
effect.
A. R A K E Receiver Structure
In theory, two correlators are needed for each tap of the
RAKE receiver. However, when the digital matched filter (a
passive device) is used, the RAKE receiver can be obtained
(IO)
from the flat fading channel receiver by adding a memory
array and a processing unit, figure 7.
When using t h e CFAR criterion, t h e conincidence d e t e c t o r will
also show a n increase in detection probability with preamble length.
3Simulations o n a multipath AWGN channel also proves t h e on-
off schemes superior performanc t h a n o t h e r s based o n ideal power
control

U - I PROCESSINGUNIT
I
OUTWl
Fig 7 RAKE receiver Implemented with Digital Matched Filter
The sum of output of the correlators is shifted into an
array of memory locations with length equals the tap-delay
line. In this way, each value in the array represents one cor-
relator output, with the leftmost being the most recent one.
T h e processor unit is responsible for summing up values in
the memory array according to the particular RAKE receiver
algorithm designed.
A. R A K E Receiver Algorithm
Assume that the received signal consists of several path
components. The correlator output will become high the
first time when the signal from the shortest path arrives,
which is shifted into the first memory location. The receiver
algorithm works in two stages as follows, see figure 8: i.)
the processing unit continuously nionitors the value shifted
into the first memory location. When this value exceeds a
particular threshold, ~ ~ ~ ( a lower
~ threshold
l , set~ for
only one tap output), enough additional correlator outputs
are collected (more samples collected) until the value origi-
nally in the first location has been shifted to the end loca-
tion. Now, the memory array will contain all the different
path components with the first ray located a t the rightmost
memory location, provided that the memory length ,is longer
than the delay spread. ii.) T h e processing unit then selects
the largest five from the array, s u m them up and test the
value against another threshold which is set according to the
scheme (on-off, coincidence ...) and the particular blocking
probability required. Passing both stages makes declaration
of presence of signal. If the second stage is not passed, we
scan backwards from the last location to check if any exceeds
~ ~ ~ and repeat
~ l the ,second
~ stage , if necessary.
~ ~ If~ no
such value exists, the algorithm restarts again in checking the
first location continuously. T h e purpose of the first stage is
to for aligning the first ray to the last memory location, so
that the RAKE receiver covers the whole delay spread, while
the second stage does the actual threshold testing.
C. Problem of Partaal (:orrelation with PN Sequence
For energy detection, we are faced with an abrupt change
from nothing to having a signal. Before the first symbol
(first P N code) has been shifted entirely into the correlator,
we have partial overlap between the incoming and the locally
generated PN sequence - partial correlation. Figure 9 shows
the correlator output for a train of bits 1 received which have
passed through a two-ray channel without any noise. We
can see that in a period of 1 symbol (bit) before the first
match (high output), some additional noise caused by the
partial correlation is found, which can be as high as 15% of
the signal strength. This temporary increase in noise occurs
within one symbol (bit) before the first ray and may cause
the RAKE receiver to align incorrectly. In o u r experiments,
we introduce a simple wiudowing technique to overcome this
(a.) R e a m d signal s h d b i into memory locaoon
I (b ) Fusl ray ets ulb me firs1 memory laavon
and the prehmmnafy Ihmrhold ks aossed
(c.) Gel additional shifts unI3 the first ray llignr
with the last memory losation. In this way.
me lap-delay Line will COVCT all me pvhs
(d.) Test agmst the second threshold
Fig. 8. FLAKE Receiver Algorithm
051 I
0 100 200 300 400 So0 600 700 800
Samples
Fig. 9. Partial Correlation.
, ~ ~ ~ ~
difficulty.
I). Windowing Technique
Let us denote the RAKE receiver output as a sequence
of [. . . , 0 - 2 , 0 - , , 0 0 , 0 1 , 0 2 , . . .] in which 00corresponds to
the correct acquisition. Suppose ~ ~ is crossed ~at ~ l
index -2 and 0-, is also larger than r. Instead of treating -2
as the correct acquisition, we will look forward the outputs
[O-,+, , . . . , 0-,+~] and see if any of them is greater than 0-,.
If 01;is the largest value in the window that satisfies ( O k >
0-,,-i+ 1 5 k 5 z + l ) , and the single correlator output a t
index k is greater than rprellmtnaryr k will be treated as the
correct acquisition.
~ T h e underlying working principle is as follows: The
~ ~ crossing may ~ be due~ to noise ori partial corre- , ~ ~
lation. In the former case, 0-, will be unlikely greater than
T , while in the latter case, the first ray will be within one
symbol. As long as the value 1 is large enough, the window
will cover the actual acquisition position and the correspond-
ing RAKE output will be the largest because it sums up all
the rays. Due to the finite length of the tap-delay line, the
value of 1 is restricted to (1 symbol - tap-delay line) in
other methods, but (2 symbols - tap-lelay line) in the on-
off scheme (this is possible because the next symbol is noise
only). A larger 1 will imply higher chance of covering the
actual acquisition and thus making the on-off scheme more
effective in using the windowing technique.
VI. SIMLJLATION
RESULTSON FREQUENCY
SELECTIVE
FADINGCHANNEL
In our simulation studies, we use a two-ray frequency se-
lective Rayleigh fading channel [Ill. Three directions are
studied: 1 . ) various SNR, 2.) preamble length, and 3.)
different distribution among the two rays. Parameters are
same as used in the flat fading channel and the two paths
585
 smi
~SrrbS- ..-Cain&W b a I 0.-011

Fig. 10. Simulation Results for: Pb = L = 5, Q u a l Power Fig. 12. Simulation Results for: pb = L = 5, Varying Power
Distribution. Distribution.

.. .......... ...
...
0.4
5 6 7 8 9 10 I1
-w
5 6 7 8 9 10 I1
-Lml*
=9 . r s h s w _-_c&mwa ...
Fig. 11. Simulation Results for: f i = IO-, Equal Power Distribu-
tion.
are separated by 6ps.
Figure 10 shows the simulation results where only the SNR
is changed. T h e preamble is fixed a t 5 and the power is
distributed equally among two rays. As one expects, the
trend is increase in thronghput with increasing SNR.
In the second study, m fix the SNR, while varying the
preamble length. Figure 11 shows simulation results for
two SNR vdaes, 15dB and 20 dB. Generally speaking, the
throughput increases with increasing preamble length.
At last, in figure 12, we perform simulations where we vary
the distribution of energy among the two paths.
Results show that the on-& scheme outperforms the oth-
ers in all situations. Compare with the flat fading chan-
nel, we discover t h a t the RAKE receiver can achieve simi-
lar throughput, given the s a n e SNR. This implies t h a t the
RAKE receiver has actually performed its job of frequency
selectivity and combining energy from all paths.
V I I. CONCLUSIONS
Synchronization in DS/SS communications is very impor-
tant because it is the process necessary in priori before any
d a t a demodulation possible. In this paper, we have pro-
posed a new method for the acquisition procedure in DS/SS
..... ..... CDMA packet radio system. T h e method involves overlaying
On-Off keying in the preamble section. Simulation results
have shown that this method outperforms the other existing
ones, both in a flat fading and a frequency selective fading
channel. In the latter caae, we propose a particular RAKE
receiver algorithm to overcome the multipath effect and also
I2 13 14 I5 a windowing technique to eliminate partial correlation prob-
lem.
REFERENCES
[l] W. C. Y. Lee, Overview of Cellular CDMA, IEEE Trans-
aciions on Vehicular Tcchnologg, Vol. 40,No.2, May 1991.
[2] A. Polydoros, C. L. Weber, A Ullified Approach to Se-
-
rial Search Spread-Spectrum Code Acquisition Part 11: A
I2 13 14 15
MatQed-Filter Receiver, IEEE hnractions on Commu-
nications, Vol. COM-32, pp.550-560, May 1984.
[koLI [3] E. Sourour,S. C. Gupta, Diuect-Sequence Spread Spectrum
Parallel Acquisition hi Nouselective and Fkequeucy-Selective
Hjcian F&g Channels, IEEE Trowactions on Communi-
cations, Vol. 10, pp.535-544, April 1992.
[4] P. M. Hopkins, A llnified Analysis of Pseudonoise Synchro-
uization by Envelop Correlation, IEEE Transactions on
Communicationr, Vol. COM-25, pp.770-778, August 1977.
[5] S. M. Pau, D. E. Dodds, S. K u a r , Acquisition Time Dis-
tribution for Spread-Spectrum Receivers, IEEE Journal on
Selecied Areas in Communications, Vol. 8 , pp.800-807, Juue
1990.
[6] J. K. Holmes, C. C. Cheu, Acquisition Time Perforniaice
of PN Spread-Spectrum Systems, IEEE Tmnaactions on
Communications, Vol. COM-25, August 1977.
[7] Y. T. Su, Rapid Code Acquisition Algoritluns Employhg
PN Matched Filters, IEEE Transactions on Communica-
tions, vo1.36, June 1988.
[SI L. B. Milstein, J. Gevargiz, P. K. Das, Hapid Acquisition
for Direct Sequence Spread-Spectrum Couuuunicatiolls Us-
iiig Parallel SAW Convolvers,IEEE Transactions on Com-
munications, Vol. COM-33, July 1985.
[9] P. F. Driessee, Z. L. Shi, A New Rapid Acquisition Scheme
For Burst Mode DS Spread Spectnuu Packet Radio, MIL-
COM 91 Proceedings, pp. 809-813.
[lo] Z. L. Shi,P. F. Driessen, Automatic Tlueshold Control For
Acquisition in Spread Spectrum Packet W O Coiiuuullica-
tiou, ICC 93 Proceedings, pp. 478-402.
[ll] T. Chan, P. Ho, Bit-Error Probability of Uncoded QPSK
Trausiuitted Over a 2-Hay Frequency Selective Haylei& Fad-
ing Chauiel, ICC 92 Proceedings, Vol. 1, pp.311-315.