PHASE ESTIMATION ALGORITHM FOR FREQUENCY HOPPED BINARY PSK AND DPSK

WAVEFORMS WITH SMALL NUMBER OF REFERENCE SYMBOLS

Benjamin R. Wiederholt
The MITRE Corporation
Bedford, MA

and
Mario A. Blanco
The MITRE Corporation
Bedford, MA
ABSTRACT
In military satellite communication systems that use fre-
quency hopped waveforms, it is difficult to coherently de-
tect phase shift keying (PSK) (binary or M-ary) signals
unless many reference symbols are used to aid with phase
estimation of the received signal. For this reason, many
systems use differential phase modulations, such as differ-
ential phase shift keying (DPSK), which can be detected
noncoherently without the use of phase information. How-
ever, the use of DPSK over PSK results in reduced power
efficiency. This paper presents a novel approach to phase
estimation that provides improved power efficiency
through coherent detection of phase modulated signals
(with and without differential encoding) using few refer-
ence symbols. More specifically, the algorithm presented
herein uses signal processing techniques to estimate the
phase of each hop by using both reference and information
symbols in a hop. This paper focuses on binary PSK and
DPSK waveforms where data is transmitted in blocks with
only one or two reference symbols per block. The per-
formance of this algorithm was evaluated for the case of
additive white Gaussian noise (AWGN) and Rayleigh fad-
ing channels via Monte Carlo simulations. At bit error
rates (BER) of interest, the results indicate that, depending
on the environment and modulation used, performance
gains of up to 3 dB were realized when hops contain as
few as two reference symbols. The results also show that
coherent detection of binary phase shifting keying (BPSK)
and DPSK modulated waveforms containing one reference
symbol per hop performed more efficiently than noncoher-
ent detection in all channels considered
1. INTRODUCTION
Frequency hopped waveforms in military satellite commu-
nication systems are characterized by the number of refer-
ence and information symbols contained in each transmit-
ted block of data. When these blocks of data hop from one
carrier frequency to the next, the unknown phase intro-
duced by the channel also changes. To efficiently detect
the received signal using coherent methods, phase modu-
lated waveforms utilize the known reference symbols to
estimate the phase of each hop. In order for these wave-
forms to perform near theoretical coherent detection, hops
must contain a sufficiently large number of reference sym-
bols. Since phase information cannot be accurately deter-
mined from waveforms containing few reference symbols,
systems typically overcome the phase ambiguity by em-
ploying power inefficient differential phase modulations
that can be noncoherently detected. As a result, depending
on the environment and modulation used, such systems
can suffer power losses of up to 3 dB.
This paper presents two phase estimation algorithms that
provide improved power efficiency through coherent de-
tection of binary phase modulated waveforms containing a
small number of reference symbols. Both algorithms use
signal processing techniques to estimate the phase of each
hop using the reference and information symbols. The
algorithms performance was evaluated through Monte
Carlo simulations of BPSK and DPSK waveforms in
AWGN and Rayleigh fading channels. Within each modu-
lation and channel, hops containing only one and two ref-
erence symbols were considered. The advantage of the
phase estimation algorithms was quantified by comparing
the performance of coherent BPSK and DPSK detection to
noncoherent DPSK detection. This paper concludes by
demonstrating that the algorithms presented can be ex-
tended to estimate the phase of quadrature phase shift key-
ing (QPSK) and eight-phase shift keying (8PSK) modu-
lated waveforms.
2. ALGORITHM DESCRIPTION
The algorithms presented here overcome the limitations of
phase modulated waveforms containing few reference
symbols by using a unique approach to estimate the phase
of each hop. The foundation of each algorithm is a simple
squaring technique which modifies all symbols within a
hop so that reference and information symbols can be used
to estimate the phase. By making use of the inherent prop-
erties of differential decoding, this technique alone pro-
duces estimates which support coherent DPSK detection.
However, additional corrective measures using knowledge
from the reference symbols are required to estimate the
phase of BPSK modulated waveforms.
2.1 DPSK Algorithm
1 of 6
Approved for Public Release; Distribution Unlimited
Case #05-0778
Figure 1 illustrates the DPSK phase estimation algorithm
block diagram.

Figure 1: Block Diagram for DPSK Phase Estimation
Algorithm

To demonstrate the sequence of this algorithm, the base-
band signal received at the input of Figure 1 is shown as
] [
] [ ] [ ] [
n j
e n n x n y
u
k
'
=
(1)
where is the differentially encoded data ( ] [n x 1 ] [ = n x )
and ] [n k and ] [n u ' are the noise corrupted amplitude and
phase rotation of each symbol.
The first step of the algorithm is to square the input y[n] to
remove the data component so that all symbols represent
one phase. As a result, the squaring produces
] [ 2 2 2
] [ ] [
n j
e n n y
u
k
'
=
(2)
where the 180 phase shift is removed at the expense of
each symbol being shifted to ] [ 2 n u ' . The squaring also
loses the location of the true phase offset due to the
modulo-2t addition. With all samples representing one
phase offset, the average rotation of L symbols in a hop is
determined by
u

2
1
2
] [
j
L
n
e
n y
=
=

=
(3)
Since the summation determines the average rotation of
twice the desired phase, is corrected by dividing the
phase by two. The phase division is performed by adding
a vector of equal magnitude and zero phase. The proof of
this division technique is shown as
u
u
u

2
)

cos( 2
j
j j
e
e e
=
+ = + =
(4)
Since the rotation of is only of interest, the magnitude
u)

cos( 2 is ignored and the phase estimate becomes .

u-
lated wave r
true offset or a 180 shift from the offset. Consequently,
ing and
n of all symbols in a hop. How-
en-
ic
u

The properties of differential decoding allow DPSK mod

forms to be coherently detected using eithe the
this algorithm is limited to coherent DPSK detection since
the true offset is lost by the squaring in (2).
2.2 BPSK Algorithm
The BPSK algorithm employs the same squar
ed by the DPSK algorithm to phase division approach us
find the average rotatio
ever, since BPSK modulated waveforms are not differ
tially encoded, the true phase offset must be determined
through an additional phase correction process. The bas
structure of this algorithm is shown in Figure 2.


Figure 2: Block Diagram for BPSK Phase Estimation
Algorithm

For illustrative purposes, the phase estimate in (4) is
shown as

u
|

j
e =
(5)
where the denotes the 180 phase ambiguity that results
from the squaring in ( ).
ef
he first step in correcting the
2
Since the r erence symbol(s) represent the true offset, the
phase ambiguity is corrected using the average rotation of
the reference symbols. T
phase is to find the phase error of | by
( )

= ) (
*
s symbol ref | c
(6)
The hard decision to shift | by 180 or 0 is made based
on the angle of the phase error. This decision becomes
( )
|
c c c
) 1 (
1 ) Re( 90 90
=
= < > sign and if
(7)

( )
|
c c c
) 1 (
1 ) Re( 90 90
=
= < > sign or if
(8)
This correction recovers the information lost by the squar-
ing in (2) and produces an estimate which is near th
true offset.
e
2 of 6
2. WAVEFORM MODEL
The performance of the phase esti ation algorithm w
evaluated thr
m as
ough Monte Carl simulations of BPSK and
DPSK wavefo h fading, and rms in AWGN, fast Rayleig
slow (flat) Rayleigh fading channels. The simulations
were developed using the discrete baseband receiver
model shown in Figure 3.

Figure 3: BPSK and DPSK Baseband Receiver with Phase
Estimation Algorithm

at the output of ma
The received signal with L symbols per hop was modeled
tched filter as
L n N e
N
n x n y
j
o
b
... 3 , 2 , 1 ) 1 , 0 ( ] [ ] [ = + =
u
o
(9)
where
E 2
o is the amplitude variation dictated by the channel
and is the binary data or differentially encoded binar
data ( ) [1]. The model uniformly generate
] [n x y
d the 1 ] [ = n x
unknown phase u and assumed it remains constant over
all L mbols. The Gaussian noise was normalized with
) ( = N E and 1 ) ( = N Var so that the signal to noise ratio
(SNR) could be varied directly.
The AWGN channel was modeled assuming the signal was
ctral noise. Therefore, the ampli-
tude variation in (9) was set to
sy
only corrupted by the spe
0
1 = o (10)
By contrast, the two Rayleigh fading channels were mod-
eled assuming the signal corruption was due to amplitude
degradation as well as additive noise. For these channels,
o became a Rayleigh distribute
predetermined mean and variance so that the model simu-
lated an average SNR of
d random variable with a
o
b
N
E
. This condition was satisfied
when
4
) (
t
o = E and
4
1 ) (
t
o = Var so that when moved
inside the square root, o be
tion with [2].
Within this Rayleigh fading model, two fading examples
came an exponential distribu-
were considered. The first case was fast Rayleigh fading
where
1 ) (
2
= o E
o v om s aried fr ymbol-to-symbol. As a result, a
Rayleigh distributed vector of length L was generated by
L n n u n ... 3 , 2 , 1 ]) [ 1 ln( ] [ = = o
(11)
where u represents a random variable uniformly distrib-
uted over (0,1).
tude variation was equally affected by all
The other case considered was slow (flat) Rayleigh fading
where the ampli
frequency components. This channel was modeled by
generating a single o for each hop by
) 1 ln( u = o
(12)
The received signal in (9) was passed to the phase estima-
tion algorithm where the phase of the hop was estimated
using all L symbols. Using the phase estimate, the output
of the matched filter was coherently detected and hard de-
cisions were made as
0 *} ] [ Re{ ] [
>
<
aveform, the coherently detected
data was differentially decoded.
er demonstrate three
different en ation in
a

ll number of reference symbols is shown here


mance of the BPSK and
he AWGN channel when each hop
= n y n x
(13)
In the case of the DPSK w
3. PERFORMANCE RESULTS
The channels discussed in this pap
vironments which affect phase estim
satellite communications. The simulations of the wave-
form and channel models presented in Section 2 provide
measure of the phase estimation algorithm performance.
The advantage of phase estimation was quantified by com-
paring the algorithm BER versus SNR waterfall curves to
the noncoherent DPSK waterfall curves at BER=10
-2
. The
difference between these curves signifies the power sav-
ings achieved through coherent detection using phase es-
timation.
The algorithms ability to estimate the phase of each hop
using a sma
through simulation results of two different hop structures
for each channel and modulation. The first structure con-
sidered had one reference symbol at the beginning of each
hop followed by eight information symbols. The other
contained two reference symbols at the beginning followed
by sixteen information symbols.
3.1 AWGN Channel
Figure 4 demonstrates the perfor
DPSK algorithms in t
contained only one reference symbol. These results show
that the DPSK and BPSK algorithms performed similarly.
At BER=10
-2
, both waveforms performed 0.68 dB more
efficiently than noncoherent DPSK detection. These gains
illustrate that the DPSK algorithm approached near theo-
retical performance while the BPSK algorithm under-
achieved by comparison. Using only one reference sym-
bol to perform phase correction, the BPSK algorithm was
3 of 6
unable to achieve large performance gains due to bursty
errors produced by estimates with phase errors near 180 .
1E-02
1E-01
1E+00
0 1 2 3 4 5 6
Eb/No (dB)
B
E
R
BPSK Clairvoyant Coherent Detection
DPSK Clairvoyant Coherent Detection
DPSK Differentially Coherent Detection
DPSK Algorithm
BPSK Algorithm

Figure 4: AWGN channel - 1 Reference Symbol and 8
Information Symbols

o demonstrate the influence of the reference symbols, the
erformance of BPSK and DPSK waveforms with two
T
p
reference symbols and sixteen information symbols in
each hop are shown in Figure 5. In comparison to the first
hop structure considered, these results show that the BPSK
algorithm performed quite differently than the DPSK algo-
rithm. The DPSK algorithm outperformed noncoherent
DPSK detection by 0.71 dB while the BPSK algorithm
realized a 1.5 dB improvement at BER=10
-2
. The rela-
tively large performance advantage of the BPSK algorithm
was attributed to the improved phase correction using two
known reference symbols. More specifically, the two ref-
erence symbols accurately resolved the 180 phase ambi-
guity that results from the squaring operation in (2).
1E-02
1E-01
1E+00
0 1 2 3 4 5 6
Eb/No (dB)
B
E
R
BPSK Clairvoyant Coherent Detection
DPSK Clairvoyant Coherent Detection
DPSK Differentially Coherent Detection
DPSK Algorithm
BPSK Algorithm

Figure 5: AWGN Channel - 2 Reference Symbols and 16
Information Symbols

3.2 Fast Rayleigh Fading Channel
The fast Rayleigh fading channel is an alternative envi-
ronment in which waveform and algorithm performance is
affected. The performance of the BPSK and DPSK algo-
rithms in this channel are shown in Figure 6. These results
demonstrate that at BER=10
-2
, the algorithms experienced
a small power improvement of 0.10 dB over noncoherent
DPSK detection when hops contain one reference symbol.
In this channel, the DPSK algorithm was limited to small
gains since theoretical coherent DPSK detection and non-
coherent DPSK detection converged at high SNR. How-
ever, as seen in the AWGN channel, the BPSK algorithm
underperformed when phase correction was performed
with only one reference symbol.
1E+00
BPSK Clairvoyant Coherent Detection
1E-02
1E-01
0 2 4 6 8 10 12 14 16 18
Eb/No (dB)
B
E
R
DPSK Clairvoyant Coherent Detection
DPSK Differentially Coherent Detection
DPSK Algorithm
BPSK Algorithm

Figure 6: Fast Fading Channel - 1 Reference Symbol and 8
Information Symbols

Even though the DPSK algorithm achieved near theoreti-
cal performance with only one reference symbol per hop,
adding an additional reference symbol aided the perform-
ance of the BPSK algorithm. The results in Figure 7 show
that when hops contain two reference symbols, the BPSK
algorithm performed 3.05 dB better than noncoherent de-
tection. Again, the improved performance was credited to
reducing the phase errors near y using two refer-
ence symbols to perform the phase correction.

180 b
4 of 6
1E-02
1E-01
B
E
R
1E+00
0 2 4 6 8 10 12 14 16 18
Eb/No (dB)
BPSK Clairvoyant Coherent Detection
DPSK Clairvoyant Coherent Detection
DPSK Differentially Coherent Detection
DPSK Algorithm
BPSK Algorithm

Figure 7: Fast Fading Channel - 2 Reference Symbols and
16 Information Symbols

3.3 Slow (Flat) Rayleigh Fading Channel
In comparison to the fast fading channel, the slow fading
channel results in Figure 8 show that theoretical coherent
and noncoherent DPSK detection maintained a constant
difference over the entire range of SNR. As a result, the
DPSK and BPSK algorithms with one reference symbol
per hop achieved a sizeable 0.55 dB advantage over non-
coherent detection. However, the performance of both
algorithms represents a relatively small percentage of the
total possible power savings. Even for the DPSK algo-
rithm, a 0.35 dB margin exists between theoretical coher-
ent detection.

1E-02
1E-01
B
E
R
1E+00
0 2 4 6 8 10 12 14 16 18
Eb/No (dB)
BPSK Clairvoyant Coherent Detection
DPSK Clairvoyant Coherent Detection
DPSK Differentially Coherent Detection
DPSK Algorithm
BPSK Algorithm

Figure 8: Slow Fading Channel - 1 Reference Symbol and
8 Information Symbols

In the AWGN and fast fading channels, the underperfor-
mance of the BPSK algorithm was the result of poor phase
correction with one reference symbol. However, the re-
sults in Figure 9 demonstrate that with improved phase
correction using two reference symbols, the BPSK algo-
rithm was unable to achieve near theoretical performance
in the slow fading environment. Even though this algo-
rithm experienced an improved performance advantage of
1.75 dB over noncoherent DPSK detection, a considerable
1.40 dB difference remains between theoretical coherent
detection. In contrast to the first two channels, the remain-
ing difference was the consequence of phase errors uni-
formly distributed between and . 90 180
1E+00
BPSK Clairvoyant Coherent Detection
DPSK Clairvoyant Coherent Detection
DPSK Differentially Coherent Detection
1E-02
1E-01
0 2 4 6 8 10 12 14 16 18
Eb/No (dB)
B
E
R
DPSK Algorithm
BPSK Algorithm

Figure 9: Slow Fading Channel - 2 Reference Symbols
and 16 Information Symbols

4. FUTURE RESEARCH
The basic phase estimation squaring technique presented
in this paper can be extended to QPSK and 8PSK modu-
lated waveforms. Algorithm structures for these modula-
tions are shown in Figures 10 and 11 (signal constella-
tions?). Currently, the performance of these algorithms is
being investigated in the three channels discussed in this
paper.


Figure 10: Block Diagram for QPSK Phase Estimation
Algorithm
5 of 6


Figure 11: Block Diagram for 8PSK Phase Estimation
Algorithm
5. CONCLUSION
This paper presented a unique approach to phase estima-
tion that minimizes the power loss of binary PSK wave-
forms containing a small number of reference symbols. It
was shown that in the three channels discussed, both the
DPSK and BPSK algorithms performed more efficiently
th -
tection when hops con one reference symbol.
lso, it was shown that when an additional reference sym-
bol was added to each hop, the BPSK algorithm achieved
significant power savings of up to 3 dB. Currently, we are
investigating whether these performance gains can be real-
ized when the algorithm is extended to QPSK and 8PSK
modulations.
6. REFERENCES
[1
cGraw-Hill, 1983.
[2] A. Papoulis, Pr ariables, and Sto-
chastic Processes. New York: McGraw-Hill, 1965.



an differential phase modulations with noncoherent de
tained only
A
] J. G. Proakis, Digital Communications. New York:
M
obability, Random V
6 of 6