The Communications Edge ™

Tech-note
Author: J. Mark Steber

PSK Demodulation (Part 2)

SYMBOL TIMING RECOVERY er, a delay and multiply circuit, or a thresh- aggravate acquisition time. However, a PLL
The second subsection required for PSK old detector. Typically, the methods shown is often used as a compromise between rea-
demodulation is the symbol timing recovery, in Figure 2 have gained favor due to their sonable acquisition characteristics and excel-
or STR circuitry. All techniques discussed simplicity and low cost. Versions A and B lent filtering tracking performance for mod-
here are data-derived; i.e., the timing infor- are equivalent, and analyzed by Feher [12]. erate baud-rate systems.
mation required to optimize for correct bit They are primarily low-to-moderate speed
An STR narrowband filter employing an
decisions is derived from the raw data itself. realizations, but can be used in systems where
analog PLL is shown in Figure 3. The pre-
Additionally, only suboptimal techniques the baud rate approaches the switching lim-
filter is excited with a pulse train derived
will be discussed. Optimal techniques devel- its of the digital hardware, roughly 10 Mbs.
from the raw data stream. Any missing pulse
oped for deep space telemetry applications Version C, a delay and multiply technique,
transitions between two adjacent time slots,
[3] can be slow in acquisition when used for is preferred for high-speed application [11].
as in a series of consecutive “ones” or “zeros,”
burst-mode modem operation, and are diffi- To complete the STR, the desired spectral will be interpreted by the loop as a gross fre-
cult and costly to implement. line at quency error for which the VCO tries to
Ideally, the STR circuitry provides clock compensate. This results in a hunting effect
f = 1/Ts
pulses to the bit sampler at the midpoint which prevents the loop from locking to the
between data transitions; i.e., in the middle must be recovered, while residual spectra is nominal clock frequency. The prefilter is
of each hit’s time slot. The STR tracks any suppressed to minimize timing jitter. High used to introduce these missing pulse transi-
fluctuation in frequency and/or phase due to data-rate systems will opt for straight-for- tions [12]. A prefilter is simply a high-Q fil-
transmitting clock oscillator drift or jitter. ward bandpass filtering over a PLL to elimi- ter that, upon loss of input (transitions), will
nate loop feedback time delays, which would continue to oscillate (ring) at the desired
The typical STR used in uncoded PSK sys-
tems is comprised of two blocks; a nonlinear
element and a narrowband filter, as is shown
POWER SPECTRAL DENSITY G(F) (DB)

in Figure 1. Consider the typical case in BPSK

which the raw data are formatted as non- 0 QPSK
return to zero (NRZ) pulses. An NRZ pulse -10
can assume either of two allowable levels, and -20
is held there for the duration of the signaling
-30
interval [11]. The power spectral density for
-40
such a bipolar NRZ data stream takes the
form of that shown in Figure 1, Part 1, for
BPSK, shifted down to fc = 0 [The figure is
-3.0 -2.0 -1.0 0 1.0 2.0 3.0 f-fc/Tb
repeated below for convenience.-Ed.].
With a spectral null at the symbol frequency, Figure 1. (Part 1) BPSK and QPSK spectra.

nonlinear processing is required to extract

timing information; that is, to generate spec-
tral lines at multiples of the symbol rate. The RAW DATA
BIT
DEMODULATED
DEMOD DATA
narrowband filter then tracks the line at SAMPLER
(±180° φ AMBIGUITY)

f = 1/Ts
X(t) Z(t) NARROW-
NONLINEAR
The residual lines contribute to the recov- BAND
ELEMENT
FILTER RECOVERED
ered clock jitter. CLOCK

The nonlinearity can be implemented in

many forms, such as a differentiator, a squar- Figure 1. Block diagram of symbol timing recovery.

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com
 The Communications Edge ™
Tech-note
Author: J. Mark Steber

clock frequency any detail; a survey of variations is provided compared with a divided-down clock refer-
in [13]. ence. Sequential loop filtering is performed
f = 1/Ts
by the K-counter, which is steered via the
for a few clock periods. This electrical fly- The lead lag DPLL or (LL)-DPLL has been
phase detector to count up from zero to K
wheel allows the PLL to remain locked. analyzed by Cessna [14], and forms the basis or down from K to zero. The stochastic
of a commercially available IC, the Texas process which describes the K-counter out-
A method of filtering tracking which is Instruments SN74LS297. [These are still put is termed, random walk. Carry or bor-
moderately insensitive to data transition loss
commercially available as of August 2000 - row pulses are generated when the K-
is that of the digital phase-locked loop, or
Ed. ] Model block diagrams are shown in counter recycles, and are used in conjunc-
DPLL, which allows discrete phase adjust-
Figure 4. tion with the increment-decrement circuit
ment of the output clock frequency. Since
to add or delete one-half cycle from the ID
there are many variations of DPLLs, only Basic circuit operation is as follows: The
output. In this way, the frequency (fout)
one, the (LL)-DPLL, will he described in incoming data stream (raw data) is phase- and phase ([FL1](out) are continuously
adjusted in discrete steps. Loop parameters
such as bandwidth and center frequency can
+1 he adjusted by the judicious selection of
X(t) clock reference frequencies and divider
DUAL
0 Z(t)
ONE SHOT moduli. A complete description of the TI
A
5N74L5297, including pertinent design
equations, can be found in reference [15].
+1 A digital PLL should he considered for low
X(t) baud-rate STR.
0 Z(t)

Td
CHANNEL FILTERING
B
The last major factor degrading demodula-
tor performance is intersymbol interference,
or ISI, a term used to describe pulse smear-
X(t) X Z(t) ing between time slots in a band-limited
channel. Consider the idealized impulse
response at the output of the data channel
Td
as the summation of
C
sin x
Figure 2. Practical nonlinear processors (A and B: threshold detectors; C: delay and multiply) x
waveforms, as depicted in Figure 5. Clearly,
if hit sampling occurs at multiples of Ts sec-
onds, the sin x/x waveshape guarantees that
LINES AT
HI Q
f = nfs
PREFILTER φ F(s)
adjacent pulse “tails” pass through zero; i.e.,
FROM NONLINEAR
PROCESSOR
f = 1/Ts zero ISI. Noting the Fourier pair relation-
ship shown in Figure 6, it is seen that the
optimal impulse shaping filter is that of an
unrealizable ideal low-pass filter.
VCO
Nyquist developed criteria for realizable fil-
ters which ensure controllable zero cross-
RECOVERED CLOCK @ f = 1/Ts ings. One commonly used class is that of
the raised cosine family, shown in Figure 7
Figure 3. STR narrowband filter employing an analog PLL. with the required

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com
 The Communications Edge ™
Tech-note
Author: J. Mark Steber

Although the expressions for optimal pulse

shaping were given at baseband, the designer
ADD OR STABLE
RECOVERED N FIXED
has the option of channel filtering at IF via
DELETE
TIMING SIGNAL
ONE CYCLE FREQ. OSC an LP-BP transformation. Both techniques,
RETARD
while equivalent, lead to different hardware
LEAD implementations.
PHASE SEQUENTIAL
RAW DATA
DETECTOR LOOP FILTER
LAG ADVANCE FILTERING AT IF
Quite often the required amplitude response
Figure 4A. The (LL)-DPLL. is approximated by a conventional passive
filter; e.g., Butterworth, Gaussian, elliptic,
followed by group delay equalization, as
shown by Figure 8.
Mfc K CLK CA
K COUNTER
D/U BQ All pass filters, such as the bridged-tee net-
work, are used to flatten the group delay in
DECR INCR the passband. This is tantamount to ensur-
fin, φin PHASE I/D
CLK 2Nfc ing linear phase, since
DETECTOR 2
OUT
H(jω) = |H(jω)| e-jωτ
fout, φout
N for a linear phase network
COUNTER
θ (jω) = -ωτ
-dθ
Group delay = =τ
Figure 4B. T.I. SN74LS297 block diagram. dω
a constant
Typically, these channel filters are designed
using a CAD program, and are iteratively
optimized to a prescribed group delay roll-
off factor requirement. Typical systems usu-

t
h(t)
Ts

Figure 5. Ideal impulse response. 1 1

B B
t
1 1
x B = single-sided baseband band- 2B 2B
sin x width in hertz
BT = single-sided double-sideband rf
amplitude equalization necessary for pulse Figure 6A. Desired data channel impulse response.
bandwidth in hertz
transmission [2].
α = roll-off factor 0 < α < 1
The raised-cosine filters are parameterized by
a roll-off factor, α, which can vary from If adjacent channel interference requirements H(jω)

zero, the ideal LPF case, to one for full are strict, the designer has only two parame-
raised-cosine shaping. For a desired pulse ters available for adjustment. The symbol
rate rate can be reduced, resulting in decreased
ω
link efficiency, or the roll-off factor can be -2πB 2πB
2B = BT = Rs (1 + α)
decreased, resulting in increased filter design
where, Rs = pulse rate in symbols/sec complexity. Figure 6B. Required filter.

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com
 The Communications Edge ™
Tech-note
Author: J. Mark Steber

ally require multi-pole filters and several

stages of equalization. [H(jf)]
(LINEAR SCALE)
Another method of pulse shaping involves
the use of surface acoustic wave (SAW) fil- 1.6
ters. The basic SAW filter consists of a piezo- 1.4
electric substrate upon which fingered metal-
1.2
ized transducers have been implanted. These
1
transducers serve as convertors between elec- 0.9
0.8 α = 1
trical and acoustic signals. By varying the 0.7 α = 0.5
spacing, length, and 0.6 α = 0.3
0.5 α = 0
0.4
width of the transducer fingers, the impulse 0.3
0.2
response can he carefully controlled. 0.1
f/fn
Advantages of SAW filtering include direct 0 0.2 0.4 0.6 0.8 1 1.3 1.5 2
analysis in the time domain, linear phase,
flexible fractional bandwidths, small size, (ω Ts/2)

{
π
0≤ω≤ (1 - α)
and high repeatability. Sin (ω Ts/2) Ts
H (jω) (ω Ts/2) π (1 - α)
Ts
FILTERING AT BASEBAND Sin (ω Ts/2)
Cos2 { [
4α
ω-
Ts ]} π
Ts
(1 - α) ≤ ω ≤
π
Ts
(1 + α)

π
Baseband channel filtering techniques, men- 0 ω > (1 + α)
Ts
tioned only in passing, are transversal filter- θ (jω) = -ωτ LINEAR PHASE
ing and digital (FIR) filtering. These tech-
niques are prevalent in low-speed demodula- FROM FEHER WITH PERMISSION © 1981 PRENTICE HALL
tion systems and those which are computer
intensive. Both methods require a somewhat Figure 7. Nyquist channel-pulse transmission.

lengthy explanation; the interested reader is

referred to the literature, [16, 17, 18].
ALL-PASS RAW
IF INPUT CONVENTIONAL
ISI DEGRADATION BPSK-QPSK PASSIVE FILTER
GROUP DELAY
EQUALIZER
DEMODULATOR DATA
OUT
MEASUREMENT
A simple but effective method of measuring Figure 8. Nyquist filter approximation.
the peak intersymbol interference degrada-
tion of a band-limited channel is by examin-
ing its eye diagram [19]. The test setup is
shown in Figure 9.
PRBS CHANNEL OSCILLOSCOPE
The symbol clock is used as the external trig- CLOCK DATA FILTER
IN GENERATOR UNDER TEST
ger, while the horizontal time base is adjust- EXTERNAL
VERTICAL
ed so that one symbol period is displayed. INPUT TRIGGER

The eye pattern of a band-limited channel CLOCK

OUT
will have the form of that shown in Figure
10. The amount of peak ISI distortion is cal-
culated as Figure 9. ISI degradation measurement test setup.
X
20 log10 (dB)
Y PERFORMANCE TESTING
The magnitude of this value is the increase
in E/No ratio required to compensate for the One standard of demodulator performance is bit error rate, or BER, which is defined as
band-limiting effects of the filter. BER = NE/NT

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com
 The Communications Edge ™
Tech-note
Author: J. Mark Steber

the complementary error function

No = the noise spectral density or total
noise power in a one-hertz band-
width
This relationship is plotted in Figure 11.
Demodulator performance can he deter-
mined by measuring the required E/No for a
X Y
desired error rate, and comparing this value
to theoretical. The difference in E/No is the
penalty incurred due to carrier recovery and
STR circuitry losses and intersymbol inter-
ference.
A typical test setup for BER measurement is
shown in Figure 12.
SAMPLING INSTANT
For BPSK system testing, the pseudo-ran-
dom pattern generator supplies a known
Figure 10. Measured eye diagram of an unequalized filter.
sequence of ‘ones” and zeros” to the modula-
tor. The BPSK output is then combined
1.0
with noise via the white noise generator, and
the noise floor is increased until a specific
0.1 E/No, is measured. Finally, the noisy BPSK
signal is applied to the demodulator under
0.01 test, with the demodulated data routed to
ERROR PROBABILITY

the error detector. Phase ambiguity resolu-

10-3
tion (±180°) is usually not a problem, as
most commercial error detectors can process
10-4
both the recovered PRBS DATA sequence,
10-5 and its complement, DATA. The error
detector then counts the bits which are in
10-6 error, relative to the transmitted sequence,
and provides an indication of BER.
-10 -5 0 5 10 15 Db QPSK system testing requires additional
E/No
phase ambiguity resolution circuitry.
Typically, a unique word preamble is
Figure 11. Ideal P(e) curve (BPSK-QPSK)
appended to the PRBS data frame. The pre-
amble is then decoded at the output of the
where, the result
demodulator, and those data lines which are
P(e) = 1 erfc E
NE is the number of detected bits which are
in error and NT is the total number of bits
where,
2 √ No
in error are inverted. Finally, the two data
lines are multiplexed together, and the com-
in the measurement gating period. posite bit stream is routed to the error detec-
P(e) is the system probability of error,
tor. The measured demodulator BER can
Under the common assumption of an addi- (BPSK-QPSK)
then be compared to the theoretical perfor-
tive white gaussian-noise channel (AWGN), [E is the energy/bit = signal power / bit
mance curves.
very definitive statements can be made con- rate. -Ed. ]
cerning theoretical error rates for PSK sys- ∞ E/No can be measured indirectly at the
2
tems [1, 20]. Of fundamental importance is
erfc(x) ∆
∫
√π x
e-y2 dy
demodulator input by using the relationship

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com
 The Communications Edge ™
Tech-note
Author: J. Mark Steber

E/No = STb/No = S/NoRb

PSEUDO DIGITAL CONTROL
RANDOM ERROR
E/No = 10 log10 S - 10 log10 Rb
dB
PATTERN DETECTOR - 10 log10 No
GENERATOR
where, Rb = the bit rate

UNIQUE AMBIGUITY
Signal power can be measured directly in
WORD RESOLUTION dBm with an RF voltmeter. The spectral
PREAMBLE CIRCUITRY
density of the noise can be measured in dBm
Hz with a wave analyzer.
RF
PSK CONCLUSION
SIGNAL
MODULATOR
GENERATOR
This two-part article has presented an
overview of the various practical techniques
used for the demodulation of PSK encoded
DEMOD-
BPF UNDER signals. Many topics have only been touched
TEST on, and the reader is referred to the selected
PSK bibliography for a sampling of the per-
tinent literature.
WHITE
RF WAVE
NOISE
VOLTMENTER ANALYZER
The design of a PSK demodulator requires
GENERATOR
the blending of RF, analog, and digital cir-
cuitry in proportions dictated by system
constraints, such as the required baud rate,
Figure 12. Typical test setup for BER measurement.
channel bandwidth, and maximum allow-
able acquisition time. As noted throughout
this article, these constraints suggest certain
circuit realizations, some of which have
become de facto standards. It is up to the
designer, however, to select the techniques
that will yield superior demodulator perfor-
mance at a reasonable cost.

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com
 The Communications Edge ™
Tech-note
Author: J. Mark Steber

SELECTED PSK BIBLIOGRAPHY

1. Schwartz, M. Information Transmission, 8. Simon, M. K., “Tracking Performance of White Gaussian Noise,” IEEE Transactions
Modulation, and Noise, McGraw Hill, Costas Loops with Hard-Limited In on Communications, VOL COM-20,
New York, 1980. Phase Channel,” IEEE Transactions on No. 2, April 1972, pp. 94-104.
Communications, VOL COM-26, No. 4,
2. Feher, K., Digital Communications with 15. Texas Instrument Applications Report,
April 1978, pp. 420-432.
Microwave Applications, Prentice Hall, Bulletin SCA-206, “Digital Phase-Locked
New Jersey, 1981. 9. Weber, C. L., “Demod-Remod Coherent Loop Design using 5N54/74L5297.”
Tracking Receiver for QPSK and SQPSK,”
3. Lindsey. W. C. and Simon, M. K., 16. Mueller, K, H., “A New Approach to
IEEE Transactions on Communications,
Telecommunication Systems Engineering, Optimum Pulse Shaping in Sampled
VOL COM-28, No. 12, Dec. 1980,
Prentice Hall, New Jersey, 1973. Systems Using Time-Domain Filters,
pp. 1945-1953.
“BSTJ, VOL 52, No. 5, May-June 1973,
4. Blanchard, A., Phase Locked Loops,
10. Wolejsza, C. J., and Chakraborty, D., pp. 723-729.
Wiley, New York, 1976,
“TDMA Modem Design Criteria,”
17. Feher, K., and DeCristofaro, R.,
5. Simon, M. K., and Lindsey, W. C., Comsat Technical Review, Volume 9,
“Transversal Filter Design and
“Optimum Performance of Suppressed Number 2A, Fall 1979.
Application in Satellite Communications,”
Carrier Receivers with Costas Loop
11. Holmes, J. K., Coherent Spread Spectrum IEEE Transactions on Communications,
Tracking,” IEEE Transactions on
Systems, Wiley, New York, 1981. VOL COM-24, No. 11, Nov. 1976,
Communications, VOL COM-25, No. 2
pp. 262-1268.
(Feb. 1977) pp. 215-227. 12. Le-Ngoc, T., and Feher, K., “A Digital
Approach to Symbol Timing Recovery 18. Oppenheim, A. V., and Schafer, R. W.,
6. Simon, M. K., “On the Calculation in
Systems,” IEEE Transactions on Digital Signal Processing, Prentice Hall,
Costas Loops with Arbitrary Arm Filters,”
Communications, VOL COM-28, No. New Jersey, 1975.
IEEE Transactions on Communications,
12, Dec. 1980, pp. 1993-1999.
VOL COM-26, No. 1 (Jan. 1978) 19. Feher, K., Digital Communications -
pp. 179-183. 13. Lindsey, W. C., “A Survey of Digital Satellite Earth Station Engineering,
Phase Locked Loops,” Proceedings of the Prentice Hall, New Jersey, 1983.
7. Ritter, S., “An Optimum Phase
IEEE, VOL 69, No. 4, April 1981,
Reference Detector for Fully Modulated 20. Wozencraft, J, M. and Jacobs, I. M.,
pp. 410-430.
Phase-Shift Keyed Signals,” IEEE Principles of Communication
Transactions on Aerospace and 14. Cessna, J. R., and Levy, D. M., “Phase Engineering, Wiley, New York, 1965.
Electronic Systems, VOL AES-5, No. 4, Noise and Transient Times for a Binary
July 1969, pp. 627-631. Quantized Digital Phase-Locked Loop in

Copyright © 1984 Watkins-Johnson Company

Vol. 11 No. 2 March/April 1984
Revised and reprinted © 2001 WJ Communications, Inc.

WJ Communications, Inc. • 401 River Oaks Parkway • San Jose, CA 95134-1918 • Phone: 1-800-WJ1-4401 • Fax: 408-577-6620 • e-mail: [email protected] • Web site: www.wj.com