CFO Compensation in Frequency
CFO Compensation in Frequency
Domain
Domain
Presenter: Pin-Hsun Lin
Advisor: Prof. Tzi-Dar Chiueh
Date: Aug. 18
th2003
Outline
Outline
• Motivation
• Time-delay in a loop
– What are the impacts of delay in a loop?
– How the error performance degrades with the prolonged settling time?
– Under what condition the conventional method is improper? – Loop filter design for a stable system
• With/without consideration of phase error variance
• Preliminary remedies
– Frequency domain compensation
• Circular convolution, interpolator, rotation
Motivation
Motivation
• In the 802.11a project the time domain CFO
tracking is said to be unstable since there’s a
large delay (FFT block)
• Find out how the delay affects the burst
communication and how to solve the problem
caused by the effect efficiently.
Model of delay in a loop
Model of delay in a loop
FFT FFT CFO EstimatorCFO Estimator NCO NCO Up to 2 OFDM symbols delay • Pipeline registers
• Latency of signal processing blocks:
―FFT, CFO estimation, TFO estimation, etc. Some causes of the delay in feedback loop in a
communication system include:
ACC ACC 2 2 1 1 NCO NCO ACC ACC
If symbol based estimator is used
Impacts of delay in a loop:
Impacts of delay in a loop:
The optimal natural frequency is decreased [1] The optimal natural frequency is decreased [1] The error variance increases [1] The error variance increases [1] Delay in a loop increases Delay in a loop increases Trade-off
The settling time increases [3] [4].
The settling time increases [3] [4].
Impacts of delay in a loop:
Impacts of delay in a loop:
the model of loop with delay [1][2]
the model of loop with delay [1][2]
LO , R LO , T
φ
φ
-KD KD K0/S K0/S F(s)F(s) VCO , Rφ
)
t
(
n
∫
∫
-∞∞ ∞ ∞ - SPN f - H f df SWN f H f df WN PN 2 2 2 2 2 | ) ( | ) ( | ) ( 1 | ) ( 2
)
(
)
(
s
H
s
k
k
s
F
f p LO WN PN S kP f f S , 2 2 Close loop transfer function 0 L P f is the laser line-width
Delay
τ
Delay
τ
Impacts of delay in a loop:
Impacts of delay in a loop:
the increased error variance and the
the increased error variance and the
decreased optimal natural frequency [1][2]
decreased optimal natural frequency [1][2]
Bit rate=565Mbps
MHz
1
=
f
δ
Optimal loop design No modification
according to the loop delay
How the error rate performance degrades
How the error rate performance degrades
with the prolonged settling time?
with the prolonged settling time?
The length of settling time
The length of settling time
Accuracy of coarse synchronization Accuracy of coarse synchronization Length of training sequence Length of training sequence
Error rate performance
Error rate performance
Delay in a loop
Under what condition the
Under what condition the
conventional method is improper?
conventional method is improper?
If the previous relationship is valid, then
under the following conditions the
conventional methods are improper:
• In burst communication (not such long time
for convergence)
• When training symbol is very short like
802.11a
Loop filter design for a stable system:
Loop filter design for a stable system:
w/o consideration of phase error variance
w/o consideration of phase error variance
[3]
[3]
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 Kp K f )] 1 2 /( , 0 [ ) ) 2 / 1 cos(( ) 2 / tan( ) 2 / sin( 4 ) ( ) 2 / tan( ) ) 1 sin(( 2 ) ( 0 M M k M k k f p f Analytical method:
The stable region is enclosed by:
M is the samples of delay
M=0 M=1
M=2
Loop filter design for a stable system:
Loop filter design for a stable system:
w/o consideration of phase error variance
w/o consideration of phase error variance
[4]
[4]
• Replacing z=exp(u+jv)
into the denominator of the loop transfer function.
• scan u>0 and v=0~2*pi • The region doesn’t cover
by the spirals is the stable
region as the right figure 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 kf
Numerical method:
Loop filter design for a stable system:
Loop filter design for a stable system:
with consideration of phase error variance
with consideration of phase error variance
[5]
[5]
• Given delay want to find an F(s) that minimizes the phase error variance:
1
)
(
)
(
)
(
)
(
)
(
)
(
' `
s
Y
s
M
S
X
s
N
s
M
s
N
e
theorem s Cauchy ion approximat e Pad sThen F (S) stabilizes the loop iff:
NQ Y MQ X s F ) (
where Q is any stable proper and rational function. Then find Q by minimizing .
2(The solution is complicated so isn’t shown here.) 1.
2. 3.
Preliminary Remedies
Compensate the error in frequency
Compensate the error in frequency
domain
domain
FFT FFT CFO compensatorCFO compensator CFO EstimatorCFO Estimator• The latency of the frequency domain compensator must be smaller than the time domain one.
• The additional complexity must be moderate.
Frequency domain compensation:
Frequency domain compensation:
Circular convolution
Circular convolution
• Time domain rotation is equivalent to
frequency domain circular convolution.
)
n
+
Y
HW
(
E
=
r
n
+
Y
HW
=
r
H⇒
d H)
n
+
Y
HW
(
AWE
)
n
+
Y
HW
(
E
Eˆ
W
H H 1-Time domain compensation:
Frequency domain compensation:
H 1 1
W
Eˆ
W
N
1
=
A
⇒
Eˆ
W
=
AW
--⇒
Frequency domain compensation:
Frequency domain compensation:
Circular convolution (cont’d)
Circular convolution (cont’d)
Trunca
t ion
Frequency domain compensation:
Frequency domain compensation:
Circular convolution (cont’d)
Circular convolution (cont’d)
• The length to be truncated can be determined by:
• The computational complexity can be further optimized by the Chinese remainder theory
(CRT) and the latency can be further improved. • Low latency architecture is under researched.
<
σ
,
|
h
h
|
=
Interpolator [6]
Interpolator [6]
FFT FFT N NP NP interpolator interpolator N CFO Estimator CFO Estimator Zero padding Zero padding P P 1st stage 2nd stageInterpolator (cont’d)
Interpolator (cont’d)
• The constant BER degradation between no CFO and the cubic interpolator may be because not enough information is
included to do the compensation. • The sufficient and necessary
conditions for the usage of interpolator is needed be investigated.
~0.001
CFO normalized to the sub-carrier spacing
Rotation [7]
Rotation [7]
• Rotation is the easiest method with the lowest
latency and the worst error performance.
• When CFO is small, the effect of CFO can be considered as a phase rotation.
the residual CFO can be compensated by frequency
domain rotation.
• The ICI can’t be removed by the rotation.
• The resulted SNR degradation is related to how accuracy the coarse synchronization can achieve.
Rotation (cont’d)
Rotation (cont’d)
0 s 2 N E ) R f Δ N π ( 10 ln 3 10OFDM
BER
M
log
SER
2≤
•
Given BER, we can get SER by the following approximation for M-ary modulation :•
Using the SER we can get the corresponding SNR. With the SNR and the following approximation we can get the SNR degradation (SINRnon-ideal-SNRideal in dB).Conclusion
Conclusion
• The impacts of delay in a loop were introduced.
• 3 Loop filter design methods for a stabilize a time-delay system were introduce.
• 3 frequency domain compensation methods were introduced • Research the relationship between the error rate
performance degradation and the prolonged settling time.
• Validate the sufficient and necessary conditions for the usage of interpolator.
• Merge the circular convolution and the interpolator and find a low latency architecture.
Reference
Reference
• [1] M. A. Grant, W. C. Michie and M. J. Fletcher, “The performance of optical
phase-locked loops in the presence of nonnegligible loop propagation delay,”
IEEE Journal of Lightwave Technology, Vol. 5, No.4, April, 1987, pp. 592-597. • [2] S. Norimatsu and K. Iwashita, “PLL propagation Delay-time influence on
linewidth requirements of optical PSK homodyne detection,” IEEE Journal of
Lightwave Technology, Vol. 9, No.10, Oct, 1991, pp. 1367-1375.
• [3] J.W.M. Bergmans, “Effect of loop delay on stability of discrete-time PLL, “Circuits and Systems I: Fundamental Theory and Applications. IEEE
Transactions on, Volume: 42 Issue: 4, April 1995, pp. 229 -231
• [4] A. D. Gloria, D. Grosso and M. Olivieri and G. Restani, “A novel stability
analysis of a PLL for timing recovery in hard disk drives,” Circuits and Systems
I: Fundamental Theory and Applications, IEEE Transactions on , Volume: 46 Issue: 8 , Aug. 1999 pp. 1026 -1031
• [5] O. Yaniv and D. Raphaeli, “Near-optimal PLL design for decision-feedback
carrier and timing Recovery,” IEEE Trans, Commu. Vol. 49, No. 9, Sept 2001, pp.
1669-1678
• [6] M. Luise, M. Marselli and R. Reggiannini, “Low-complexity blind carrier
frequency recovery for OFDM signals over frequency-selective radio channels,”
Communications, IEEE Transactions on, Vol. 50, No. 7, July 2002 pp. 1182 -1188 • [7] T. Pollet, M. V. Bladel and M. Moeneclaey, “BER sensitivity of OFDM systems
to carrier frequency offset and Wiener phase noise,” IEEE Trans, Commu. Vol.