The Effect of Delay in CFO Tracking Loop
The Effect of Delay in CFO Tracking Loop
in Multi-Carrier Systems: Analysis
in Multi-Carrier Systems: Analysis
Presenter: Pin-Hsun Lin
Advisor: Prof. Tzi-Dar Chiueh
Date: Mar. 1
st, 2004
2
NTU Confidential
Outline
Outline
• Introduction
– Synchronization and tracking loop
– Number of delay in an OFDM CFO tracking loop and some alternatives of the loop to reduce the delay
• Analysis
– Find the loop difference equation
– Fundamental knowledge for further analysis – The transition matrix and it’s function
Synchronization preliminaries
Synchronization preliminaries
• Synchronization: estimation + compensation • All synchronization algorithms can be coarsely
divided into
– Feedforward type
fast but high complexity, usually used in acquisition *. – Feedback type
slow but low complexity, usually used in tracking.
(several complicated algorithms are achieved by iteration)
4
NTU Confidential
Why there’s a tracking
Why there’s a tracking
“feedback loop” ?
“feedback loop” ?
• To make estimation more accurate, the data
(redundant) must be used as more sufficient as
possible.
• In CP-OFDM systems, there’re 2 redundancy can
be used:
– CP (time domain)
– Pilot (frequency domain)
• If we want to use pilots with time domain
compensation, a feedback loop is inevitable.
(Time domain compensation is intuitively the
easiest one
(not the only one)but
it’s sensitive to ICI
and modulation type.
)
• If we want to use CP, there’s no feedback loop
but
it’s sensitive to timing error.
Loop characteristics
Loop characteristics
• Performance indices of tracking loop:
– Convergence speed
– Steady state error variance – Steady state error distribution
• Several factors can affect the above:
– SNR
– Loop delay
– Constellation size
6
NTU Confidential
The effect of loop delay and SNR to
The effect of loop delay and SNR to
steady state phase error distribution: an
steady state phase error distribution: an
example *
example *
α =(K1+K2)/K1
D is number of delay
* T. Koizumi and H. Miyakawa, “Statistical analyses of digital phase-locked loops with time delay,” IEEE Trans. Commun., July 1977, p. 731-p.735
Number of Delay in the Loop and
Number of Delay in the Loop and
Some Alternatives of the loop
Some Alternatives of the loop
8
NTU Confidential
Alternatives of the tracking loop
Alternatives of the tracking loop
BUF WLS NCO + -5_1 4_1 3_1 2_1 FFT
Delay 1 Delay 1BR Delay 1
Delay 1 1 CFO∧ arg{ } LF BUF WLS NCO + -4_1 3_1 3_1 2_1 FFT Delay 1 Delay 1 Delay 1 1 CFO∧ arg{ } LF Original:
Remove the bit-reversal:
• Serial in/serial out deteriorates the convergence performance .
Alternatives of the tracking loop
Alternatives of the tracking loop
FFT BUF WLS LF + -3_k 3_1 2_1 Delay 1 Delay 1 Delay 1 1 CFO∧ arg{ } Circular conv
1 0
NTU Confidential
Alternatives of the tracking loop
Alternatives of the tracking loop
FFT BUF WLS LF NCO + -3_N/2 2_N/2 1_N/2 Delay 1 Delay 1 Delay 1/2 1 CFO∧ arg{ }
Find the
Find the
Loop Difference Equation
Loop Difference Equation
1 2
NTU Confidential
Mapping the tracking loop as a
Mapping the tracking loop as a
PLL problem
PLL problem
sm,p FFT CFO Estimator LF AWGN ej2πεp arg{ } LF Noise Delay 1 or 2 NCO NCO CFO Time domain CFO Tracking LoopApproximation of Time- domain CFO Tracking Loop
Xm,p
CFO
1/Xm,p
Sm,p
The distribution of the Phase detector
The distribution of the Phase detector
output (1)
output (1)
Sum of N-1 independent R.Vs given Φ
If can be approximated by Gaussian
1 4
NTU Confidential
Test of the Gaussianity (1)
Test of the Gaussianity (1)
If
where then
The accuracy can be examined by Berry-Esseen theorem [6][9]:
In my case the upper bound is: where
Test of the Gaussianity (2)
Test of the Gaussianity (2)
For good Gaussianity:
• FFT size is large • SNR is low
1 6
NTU Confidential
The distribution of the Phase detector
The distribution of the Phase detector
output (2)
output (2)
The stochastic difference equation of
The stochastic difference equation of
the loop
the loop
The difference equation of the loop filter:
NTU Confidential
Fundamental Knowledge for
Fundamental Knowledge for
Further Analysis
Further Analysis
Problem formulation
Problem formulation
t (sample) phase error φ (quantized by k-bit) acceptable phase error t=0 t=1 t=2 t=3 t=n t=n+1 1 2 3 2k 1 2 3 2k 1 2 3 2k 1 2 3 2k 1 2 3 2k 1 2 3 2k2 0
NTU Confidential
The Markov property and Markov
The Markov property and Markov
chain [8]
chain [8]
• Using only transition probability and initial condition
can describe the whole process
• A discrete time Markov chain is a Markov process whose state space is finite or countable set and whose time
index is 0,1,2…
• In digital systems, finite word length makes the loop’s behavior as a finite state Markov chain.
The transition probability and
The transition probability and
transition matrix
transition matrix
[8]
[8]
• A simple 1-step state transition diagram and transition matrix: p1,2 1 2 3 4 P2,3 p2,4 p3,3 p4,4
Absorbing state Transient state
Starting state destination state
p1,4 1 2 3 4 1 2 3 4
2 2
NTU Confidential
The existence of Markov property
The existence of Markov property
Although ω depends on Φ, given:
3 m 2 m 1 m m 2 m 1 m m 1 m m
φ
,
φ
,
φ
,
φ
φ
,
φ
,
φ
φ
,
φ
-the p.d.f. is of Φm with is known
No delay
1-symbol delay 2-symbol delay
The Transition Matrix and
The Transition Matrix and
It’s Function
It’s Function
2 4
NTU Confidential
Find the transition matrix
Find the transition matrix
The number of terms ω depends only on the number of accumulator (only 2-term in 2nd order loop)
combines these two terms as only one
(the resulting p.d.f. is the convolution of those of the 2 terms)
Using the compound p.d.f. we can get the transition p.d.f. integrating the transition p.d.f. to get the transition p.m.f. Here’s the transition matrix
The functions of the transition
The functions of the transition
matrix: steady state analysis
matrix: steady state analysis
transition Matrix P
Stationary
Phase error variance
BER and
BER degradation Stationary
p.m.f.
Apply Eigenvalue decomposition
If P is nonnegative, Irreducible, aperiodic
2 6
NTU Confidential
The functions of the transition
The functions of the transition
matrix: transient state analysis
matrix: transient state analysis
transition matrix
Mean time to absorption (convergence)
Complexity of higher order loop (1)
Complexity of higher order loop (1)
p12,24 1,2 p23,33 p24,44 p4,4 p14,44 2,3 2,4 3,3 4,4 1,4 p12,23 p33,33 p44,44
•A 2nd order state transition diagram and transition matrix:
1,1 1,2 1,3 1,4 2,1 2,2 2,3 2,4 3,1 3,2 3,3 3,4 4,1 4,2 4,3 4,4 1,1 1,2 1,3 1,4 2,1 2,2 2,3 2,4 3,1 3,2 3,3 3,4 4,1 4,2 4,3 4,4 Previous state
This state This state
next state
42x42 transition matrix
• This kind of transition matrix can describe 1st order loop
with 1 delay or 2nd order loop without delay
2 8
NTU Confidential
Complexity of higher order loop (2)
Complexity of higher order loop (2)
• Assume phase value is represented by k-bit, 2nd order loop, d-delay
The transition matrix is 2k(d+2)x2k(d+2)
(actually the independent parameters are only 2k(d+2)(2k-1), as the following shows)
The complexity is too large for computer to do matrix operation
We must use some approximation
Conclusion
Conclusion
• As the difference equation of the OFDM CFO tracking loop has Markov property, we can analyze it by
modeling it as a Markov chain (if finite word length effect is considered).
• After getting the transition matrix from the difference equation, we can find:
– Stationary p.m.f. – Stationary variance
– BER and it’s degradation – Mean time to converge
• For a higher order loop, the dimension of the transition matrix is too large for computer, it’s necessary to resort to some approximation of the transition matrix.
3 0
NTU Confidential