Synchronization

and
Digital Receivers

Marie-Laure BOUCHERET
IRIT/ENSEEIHT
E-mail : [email protected]

Synchronization (SC, Gaussian) 1

 Synchronization algorithms
(Single carrier systems, Gaussian channels)

Synchronization (SC, Gaussian) 2

 Contents

¾ Impact of synchronization errors

¾ Analog vs digital demodulators
¾ Baseband signal generation
¾ Likelihood functions
¾ Carrier phase recovery
¾ Timing recovery
¾ Carrier frequency recovery
¾ Digital demodulators examples
¾ Advanced topics
¾ References

Synchronization (SC, Gaussian) 3

 Impact of synchronization errors (1)

¾ Carrier phase error:

BPSK, « NRZ »filter

¾ Maximum phase jitter is

determined by the
implementation loss in the
link budget.

Synchronization (SC, Gaussian) 4

 Impact of synchronization errors (2)

¾ Timing error
BPSK, « NRZ »filter

¾ Maximum timing jitter is

determined by the
implementation loss in the
link budget.

Synchronization (SC, Gaussian) 5

 Demodulation

¾ Functions to be implemented
‰Baseband conversion
‰I,Q generation
‰Carrier recovery
‰Timing recovery
‰Matched filtering
‰Demodulation/decoding

Synchronization (SC, Gaussian) 6

 Analog demodulators

• Typical analog demodulator architecture

PLL : baseband conversion + carrier frequency/phase correction

IF input Timing data

PLL Decoder
correction

PLL : baseband conversion + carrier frequency/phase

correction
Timing correction : FF/FB structure AFTER PLL

Synchronization (SC, Gaussian) 7

 Digital demodulators

A digital demodulator is NOT the sampled version of the

equivalent analog demodulator.

⇒ Specific algorithms suited to digital implementation have

been developped.

Main differences between digital and analog demodulators:

¾ Down conversion is INDEPENDENT from phase recovery
¾ Timing recovery is performed BEFORE phase recovery

Synchronization (SC, Gaussian) 8

 Receiver input signal

( ) (
y (t ) = Re x(t )e 2 jπ f0t + Re n(t )e2 jπ f0t )
x(t ) = e jϕ (t ) ∑ d k h(t − kT − τ )
k
ϕ (t ) = 2π∆ft + ϕ0

f0 : carrier frequency, Df :carrier frequency uncertainty

φ0 : phase offset, τ : timing offset
dk : emitted symbols
h(t): emission filter (wideband channel assumed)

Synchronization (SC, Gaussian) 9

 Baseband signal generation (1)

¾Analog implementation

LPF Real part

f0
π/2

LPF Imag part

This process can be digitally implemented

(DAF : digital anti aliasing filter)
Synchronization (SC, Gaussian) 10
 Baseband signal generation (2)

¾ Digital implementation (1)

s(t) is the real received passband signal (allocated bandwidth :
FI, centred at f0=FI)
s(t) sr sd x
HPB 2
Fe=4FI

-FI | HPB(f)|

0 Fe/8 Fe/4
f

Synchronization (SC, Gaussian) 11

 Baseband signal generation (3)

¾ Digital implementation (2)

hO(n)
Re(x(n))
s(t)

z-n Im(x(n))
+/-1
Fe=4Fi

Synchronization (SC, Gaussian) 12

 Likelihood functions (1)

r (t ) = x(t ) + n(t )
x(t ) = e jϕ (t ) ∑ d k h(t − kT − τ )
k
ϕ (t ) = 2π∆ft + ϕ0
ρ(T0) : signal observed during a period of duration T0

Φ = {ϕ0 , ∆f ,τ , {d k }} Vector of unknown parameters

{ { }}
ˆ = ϕˆ0 , ∆fˆ ,τˆ, dˆk
Φ Vector of parameters estimates

Synchronization (SC, Gaussian) 13

 Likelihood functions (1)

 ) = Pr( ρ (T0 ) / Φ
Λ (Φ )
In Gaussian channel:
 1 2

Λ (Φ  ) = exp  − ∫  ) dt 
r (t ) − s (t , Φ
 N0 
 T0 
2 jπ∆ft + jϕ0

s (t , Φ ) = Ae ∑ dk h(t − kT − τ )
k
 ) : signal replica
s (t , Φ

Synchronization (SC, Gaussian) 14

 Likelihood functions (2)

Sub-optimal likelihood functions :

- DD : Decision Directed
- NDA : Non-data aided (depends on modulation)

These sub-optimal likelihood functions are derived for

timing, phase and frequency.

Synchronization (SC, Gaussian) 15

 Likelihood functions (3)

Timing :
LNDA (τ ) = ∑ p (k ,τ )
2

r(t)
h*(-t) p (k ,τ )

kT + τ

Timing recovery is performed prior to phase recovery

Synchronization (SC, Gaussian) 16

 Likelihood functions (4)

Carrier phase:
¾ DD likelihood function

( ) (
LDD (ϕ ) = ∑ aˆk Re p (k ,τˆ)e − jϕ + ∑ bˆk Im p (k ,τˆ)e − jϕ
k k
)
¾NDA lokelihood function for general rotationnaly
symetric signal constellation (2p/N symetry)


*N
( )
LNDA (ϕ ) = Re  E d k ∑ p (k ,τˆ)e
k
N − jNϕ 



Synchronization (SC, Gaussian) 17

 Likelihood functions (5)

Examples of general rotationnaly symetric signal constellation

QPSK N=4

16QAM N=4

Synchronization (SC, Gaussian) 18

 Likelihood functions (6)

¾ Carrier frequency recovery

QAM
 
− j 2π∆fkT +ϕ  
 
2 
L ({ak } , ∆f ,ϕ ) = ∑ dk + 2∑ Re p(k ,τˆ)dk*e


  

 
k k  
 
PSK
 
− j 2π∆fkT +ϕ  
 

L ({ak } , ∆f ,ϕ ) = ∑ Re p(k ,τˆ)dk*e


  

 
k  
 

Synchronization (SC, Gaussian) 19

 Carrier phase recovery : DDMLFB (1)

Derivation of detector expression from Likelihood function

d
LDD (ϕ ) = 0 for ϕ = ϕˆ
dϕ
( )
⇒ ∑ Im d k* p (k ,τˆ)e − jϕ = 0 for ϕ = ϕˆ
k

( )
⇒ u ( k ) = Im d k* p (k ,τˆ)e − jϕ is a phase detector

Synchronization (SC, Gaussian) 20

 Carrier phase recovery : DDMLFB (2)
S curve (example for QPSK)
=> Phase ambiguity (solved by using differential encoding/decoding)

Synchronization (SC, Gaussian) 21

 Carrier phase recovery : DDMLFB (3)

DPLL

p(k ,tˆ)
decision

phase detector

1/(z-1) F(z)

Synchronization (SC, Gaussian) 22

 Carrier phase recovery : DDMLFB (4)

Other possible detectors

p(k ,τˆ) = w(k )

   
 * ˆ
u1(k ) = Im  w (k ).sgn w(k ) − dk  
   
  
u2 (k ) = Im  dˆk*   w(k ) − dˆk 
   
   
u3(k ) = Im  dk*.c sgn w(k ) − dˆk  
   
   
u4 (k ) = Im  dˆk*  sgn  w(k ) − dˆk 
   

Synchronization (SC, Gaussian) 23

 Carrier phase recovery : DDMLFB (5)

Phase equivalent scheme

j ϕˆ( z )
H ( z) =
F(z) ϕ ( z)
j 1 * −1 dz
2 BLT = v∫ H ( z ) H (z )
2 jπ γ z
1/(z-1)
2BLT
σ ∝
Es / N 0

Synchronization (SC, Gaussian) 24

 Carrier phase recovery : NDAMLFB (1)

Example for QPSK

d
LNDA (ϕ ) = 0 for ϕ = ϕˆ
dϕ

{
⇒ ∑ Im  p (k ,τˆ)e − jϕ }
4
 = 0 for ϕ = ϕˆ
k  

{
⇒ u ( k ) = Im  p (k ,τˆ)e − jϕ }
4
 is a phase detector
 

Synchronization (SC, Gaussian) 25

Carrier phase recovery : NDAMLFB (2)

Synchronization (SC, Gaussian) 26

 Carrier phase recovery : NDAMLFB (3)

p(k ,tˆ)
decision

phase detector

1/(z-1) F(z)

Synchronization (SC, Gaussian) 27

 Carrier phase recovery : NDAMLFF (1)

¾ Suited for burst transmission

¾ Two types of structures : block window, sliding window
¾ Example for QPSK

{
∑ Im  p(k ,τˆ)e− jϕ }
4
 = 0 for ϕ = ϕˆ
k 
1   π
⇒ ϕˆ = Arg  ∑ p 4 ( k ,τˆ )  + k
4 k  2
⇒ Phase ambiguity (kπ /2)

Synchronization (SC, Gaussian) 28

 Carrier phase recovery : NDAMLFF (2)

« Sliding window » estimator

p(k ,tˆ)
Delay (L samples) decision

Exp(-j .)
Phase estimator (*)

(*): averaging over 2L+1 samples

Synchronization (SC, Gaussian) 29

 Carrier phase recovery : NDAMLFF (3)

¾ « Block » estimator

p(k ,tˆ)
Delay (L samples) decision

Exp(-j .)

Phase estimator (*) L hold

Synchronization (SC, Gaussian) 30

 FF structures vs FB structures

¾ Advantage
ƒ No acquisition time

¾ Drawbacks
ƒ Smaller BLT => higher jitter, higher cycle slip
probability
ƒ Sensitivity to frequency deviation

Synchronization (SC, Gaussian) 31

 Timing recovery (1)

LNDA (τ) = ∑ p(k ,τ) = ∑ Re2 ( p(k ,τ) ) + ∑ Im2 ( p(k ,τ) )
2
k k k
d L τ τ d Re( p(k ,τ )) + 2 Im( p(k ,τ )) d Im( p(k ,τ ))
dτ NDA
(  ) = 2∑ Re( p ( k ,  ))
dτ
 ∑ 
dτ

k k

⇒Derivative vs timing is approximated by a difference

Re( p(k ,τ)) ∝ Re( p(k + λ ,τ)) − Re( p(k − λ ,τ))

Im( p(k ,τ)) ∝ Im( p(k + λ ,τ)) − Im( p(k − λ ,τ))

Synchronization (SC, Gaussian) 32

 Timing recovery (2)

Gardner:
l=1/2 => detector output is independent from carrier phase error.
 
GA(k ) = Re( p(k +1/2,τ)) Re( p(k,τ)) − Re( p(k +1,τ))






 
 
+ Im( p(k +1/2,τ)) Im( p(k,τ)) − Im( p(k +1,τ))






 

Synchronization (SC, Gaussian) 33

 Timing recovery (3)

Matched Fractionnal
2 to DPLL
filter delay

TED

τˆ

Loop is implented by
updating filter polyphasing the
matched filter

Synchronization (SC, Gaussian) 34

 Timing recovery (4)

¾ S curve (Gardner, quantized)

Synchronization (SC, Gaussian) 35

 Timing recovery (5)

¾ Timing estimator (Oerder and Meyr)

τˆ = − 1 Arg  L−1 N −1 p(k , n) 2 e2 jπNn 
 
∑ ∑
T 2π  k =0 n=0


 
p(k , n)  p(kT + nT / N )

where N is the number of samples per second

Example : N=4
 
τˆ = − 1 Arg  L−1 3 p(k , n) 2 j n 
∑ ∑
T 2π  k =0 n=0


 

Synchronization (SC, Gaussian) 36

 Timing recovery (6)
Implementation of Oerder and Meyr
 
τˆ = − 1 Arg  L−1 p(k , 0) L−1
 ∑{ } + j ∑ { p(k ,1) }
2 2 2 2 
− p (k , 2) − p (k ,3)
T 2π  k =0 k =0


 

D * Rs
4 Rs
E - Σ
M a 2 π τ / Ts
t
U *
Rs /K
a
X n

(1/4) - Σ
Rs
*

Synchronization (SC, Gaussian) 37

 Frequency recovery : general

¾ Feedback structures
ƒ « Frequency » detectors
ƒ « Time » detectors

¾ Feedforward structures
ƒ Type 1
ƒ Type 2

Synchronization (SC, Gaussian) 38

 Frequency recovery:FB structures (1)

¾ « Frequency » detector (1)

x(n)
SMF
2Rs FED
e(n)
FMF
y(n)
VCO F(z)

Synchronization (SC, Gaussian) 39

 Frequency recovery:FB structures (2)

¾ « Frequency » detector (2)

SMF : signal matched filter : g(t)

FMF : frequency matched filter : -2jptg(-t)

e(n)=Im(x(n)y*(n))

A simpler filter (SFMF) derived from FMF can be used

(g(t)=-j sgn(t) g(-t)

Acquisition range : +/-(1+a)Rs

No prior timing correction required

Synchronization (SC, Gaussian) 40
Frequency recovery:FB structures (3)

Synchronization (SC, Gaussian) 41

 Frequency recovery:FB structures (3)

¾ « Time » detectors

Any estimator can be used as a time detector.

Frequency offset range is +/- Rs/M

Timing has to be corrected prior to frequency detection

1 sample/symbol is sufficient.

Synchronization (SC, Gaussian) 42

 Frequency recovery:FF structures (1)

¾ Bellini
 N   N  => Cycle slip
ˆ =
∑ iα / 8π T ∑ i
   2 
∆fT 
i  

 −N



 −N 
 ai: unwrapped phase

¾ RCFE (reduced complexity frequency estimator)

 M 
ˆ = 1 Arg ∑ d  r r* 
2π∆fT 

Large D leads to better
MD  k k  k k −D 
 



performances but to
rk = p(k ,τˆ) lower frequency range.

Synchronization (SC, Gaussian) 43

Digital demodulators (1)

Synchronization (SC, Gaussian) 44

 Digital demodulators (2)

Choice of algorithms depends on specifications such as:

¾ Acquisition time (=> FF/FB structures)
¾ Maximum frequency deviation (=> frequency circuitry
needed)
¾ Eb/No (=> use of TD if low)
¾ .....

Synchronization (SC, Gaussian) 45

 Digital demodulators (3)

FED
Example: Receiver for TCM (in cooperation with CNES)
Synchronization (SC, Gaussian) 46
 Advanced topics
¾Evolutions of input specifications (for satellite
communications)
•Low Eb/No (use of efficient coding schemes such as
Turbo-Codes and LDPC)
•Bursty transmission
• Large frequency deviation (low-cost terminals, non
GEO sat.)
¾Critical function : phase recovery (classical algorithms
fail)
¾There is a need to develop new synchronisation schemes
Synchronization (SC, Gaussian) 47
 References (1)
Reports, books on synchronization
¾ F.M Gardner « Demodulator reference recovery techniques suited for digital
implementation » ESTEC contract n° 6847/86/NL/GG,1988
¾ F.M Gardner « Frequency detectors for digital demodulators via ML
derivation », ESTEC contract n° 8022/88/NL/DG,part 2, june 1990
¾ T Jesuprret, M Moeneclaey, G Asheid « Digital demodulator synchronization :
performance analysis », ESTEC contract n° 8437/89/NL/RE, June 1991
¾ H. Meyr, M Moeneclaey, S.A Fechtel « Digital communications receivers :
synchronization, channel estimation and signal processing », J. Wiley and
Sons, 1998
PhD Dissaertations
¾ D Mottier « Association des fonctions d’égalisation, de synchronisation et de
décodage canal pour les transmissions numériques à grande efficacité
spectrale », PhD disertation (in French), INSA de Rennes
¾ Ivar Mortensen "Traitement en bande de base pour charges utiles à
régénération bord" ,PhD dissertation (in French), ENST, 1997

Synchronization (SC, Gaussian) 48

 References (2)

¾ Catherine Morlet " Démodulateur embarqué multiporteuses pour applications

multimédia par satellites », PhD dissertation (in French), ENST, 2000
Phase estimation
¾ A.J Viterbi and A.M Viterbi « Non-linear Estimation of PSK-modulated carrier phase
with applications to burst digital transmission », IEEE on IT, 1983
¾ M Moeneclaey, G de Jonghe, « ML oriented NDA carrier Synchronization for General
Rotationnally Symmetric Signal Constellations », IEEE on COM, August 1994
¾ C.N Georghiades « Blind carrier Phase Acquisition for QAM constellations »,IEEE on
COM, November 1997
Frequency estimation
¾ S Bellini and al « Digital frequency estimation in burst mode QPSK transmission »,
IEEE on COM, vol COM 38, July 1990
¾ A.N D’Andrea, U mengali « Performance of a frequency detector based on the Maximu
Likelihood principle »
¾ F Classen and al. « Maximum Likelihood Open Loop Carrier Synchronizer for Digital
Radio», ICC 93
¾ J Zhang and al « Data-Aided estimation of carrier frequency for burst detection of
QAM », Electronics Lettters, October 200
Synchronization (SC, Gaussian) 49
 References (3)

Timing estimation
¾ F.M Gardner « A BPSK/QPSK timing error detector for sampled reciver »,
IEEE on COM, may 1986
¾ Oerder, Meyr « Digital filter and square timing recovery », IEEE on COM,
May 1988
¾ M.K Nezami, R Sudhakar « New schemes for improving Non-Data-Aided
symbol Timing Recovery for QAM receivers in flat fading channels », 2000

Synchronization (SC, Gaussian) 50