Cliquez pour modifierConference le style deson sous-titres du masque IEEE International Communications (ICC) 23-27 May, 2010, Cape Town, South Africa

A Combined Time and Frequency Algorithm for Improved Channel Estimation in TDS-OFDM Liu Ming, Matthieu Crussière, Jean-François Hélard Institute of Electronics and Telecommunications of Rennes (IETR) European University of Brittany (UEB) Rennes, France May 24, 2010

1

Outline • Introduction to TDS-OFDM signal

• PN-based channel estimation in time domain • Proposed data-aided channel estimation in frequency domain • Simulation

• Conclusion

Ming LIU

2

TDS-OFDM Signal • Cyclic Prefix (CP) - OFDM

GI

Symbol K-1

GI

Symbol K

• Time Domain Synchronous (TDS) – OFDM 1 PN

PN

PN Symbol K-1

Symbol K

− Guard interval (GI) − Training sequence for channel estimation & synchronization − Adopted by the Chinese digital terrestrial TV broadcasting (DTMB) 1. TDS-OFDM is also known as Known Symbol Padding-OFDM and Pseudo Random Postfix-OFDM.

Ming LIU

3

PN removal & OverLap and Add •

Received TDS-OFDM PN

Symbol K

PN

Symbol K+1

PN

hˆ PN

PN

Symbol K

Symbol K+1

OLA

•

Orthogonality rebuilt signal Symbol K

Symbol K+1

Ming LIU

4

TDS-OFDM System Model

Time domain data symbols:

Transmitted signal: Received signal: After Overlap and Add (OLA)[1]:

linear convolution circular convolution

Received data symbols in freq. domain: Equalization: [1] M. Liu, M. Crussiere, J.-F. Helard, O. Pasquero, “Analysis and Performance Comparison of DVB-T and DTMB Systems for Terrestrial Digital TV,” in Proc. of the IEEE International Conference on Communication Systems, 2008.

Ming LIU

5

Outline • Introduction to TDS-OFDM signal

• PN-based channel estimation in time domain • Proposed data-aided channel estimation in frequency domain • Simulation

• Conclusion

Ming LIU

6

Time Domain PN-based Channel Estimation (1) •

Challenge: – ISI degrades the estimation.

•

OFDM Data

GI

OFDM Data

Solutions: – Iteratively remove ISI from training sequence, e.g. method in [5]. – Isolate ISI from training sequence, – e.g. CP or Zero Padding PN 1

•

GI structure in DTMB system: – Define:

PN + Prefix & Postfix

– Equivalent: PN + CP

CP G

PN 2 D

– Circular convolution of PN & channel

[5] S. Tang, K. Peng, K. Gong, J. Song, C. Pan, Z. Yang, “Novel Decision-Aided Channel Estimation for

TDS-OFDM Systems,” in Proc. of the IEEE ICC’08, 2008, pp. 946-950.

Ming LIU

7

Time Domain PN-based Channel Estimation (2) • Auto-correlation property of PN sequence

C(n) D

• Circular cross-correlation of received PN sequence and perfect one D-1

n

-1

Imperfect correlation

noise

• Further improvement can be done in the paper. • Mean square error (MSE) of time domain channel estimation method

• Interference on the OFDM data caused by imperfect PN removal

Ming LIU

8

Outline • Introduction to TDS-OFDM signal

• PN-based channel estimation in time domain • Proposed data-aided channel estimation in frequency domain • Simulation

• Conclusion

Ming LIU

9

Turbo-like Channel Estimation •

No frequency pilot symbol for Pilot-Symbol-Assisted (PSA) channel estimation

•

State-of-the-art turbo-like channel estimation PN

Received data

Channel Estimation

Data-aided channel estimation

Combination

Interleaver

Mapper

Local PN

S/P data

PN subtraction

OLA

FFT

Equalization

Deinterleaver

Demapper

Decoder

– Rebuild data symbols based on the soft information output from the decoder. – Use rebuilt data symbols as “known training symbols”. – High complexity & time delay, e.g. LDPC decoder & extremely deep interleaver (170 or 510 OFDM symbols! ) in the DTMB system.

Ming LIU

10

Proposed Data-aided Channel Estimation Time domain estimation PN

Received data

Correlation

Frequency domain data-aided estimation Combination

Wiener filtering

Average over Bc

Estimation Rebuild data

PN generation

S/P

data

PN subtraction

OLA

FFT

Equalization

Soft demapper

– Exclude channel decoder & deinterleaver/interleaver from feedback loop.

– Rebuild data symbols based on the soft information output from demapper. – Use averaging over coherence bandwidth to refine the channel estimate.

– Use Wiener filtering to obtain an improved channel estimate. – Combine time and frequency domain channel estimates. – Low complexity & time delay. Ming LIU

11

Rebuild Soft Data Symbols • Use the Log-Likelihood Ratio (LLR) from the soft-output demapper:

• Compute probabilities for each constellation points:

• Estimated soft data symbols:

for QPSK:

Ming LIU

12

Freq. Domain Data-aided Channel Est. • Instantaneous data-aided channel estimate

• Select virtual pilot positions • Average over coherence bandwidth • Repeat averaging over all virtual pilot positions • Wiener filtering based interpolation

Interpolated estimate

CFR ˆ (k ) CFR samples in the pilots H' p

† Instantaneous estimate Hˆ ( k )

Real channel response

M virtual pilot position kp

Frequency index (k)

Ming LIU

13

MMSE Based Combination • Weighted combination: • MSE of the combined channel estimate:

• Minimize combination MSE:

• Minimum MSE (MMSE) combination factor:

•

Combined estimate is used for PN removal & equalization in next iteration. Ming LIU

14

Outline • Introduction to TDS-OFDM signal

• PN-based channel estimation in time domain • Proposed data-aided channel estimation in frequency domain • Simulation

• Conclusion

Ming LIU

15

Simulation Parameters •

Simulation parameters are chosen according to DTMB standard

[1].

Signal bandwidth

7.56 MHz

FFT size

3780

Constellation

QPSK

Guard Interval length

420 symbols (1/9, 55.6 ms)

Interleaving depth

B=52, M=240

Channel coding

LDPC (R=0.8) + BCH (762, 752)

Channel model

COST 207 typical urban 6 paths (TU-6) and single frequency network (SFN) channels

Averaging length

9 subcarriers for TU-6, 3 subcarriers for SFN

[1] Framing structure, channel coding and modulation for digital television terrestrial broadcasting system,

Chinese National Standard GB 20600-2006.

Ming LIU

16

Mean Square Error (MSE) Performance 0

-1

10

10

-1

-2

10

MSE of CFR estimation

MSE of CFR estimation

10

6.8 dB method in [5]

-3

10

PN based 1st iteration 2nd iteration 3rd iteration 1st iteration [5] 2nd iteration [5] 3rd iteration [5]

-4

10

-5

10

10 times -2

10

-3

10

Proposed method

-4

10 5

10

15

20 SNR (dB)

TU-6 channel

PN based estimation 1st iteration 2nd iteration 3rd iteration

25

30

5

10

15

20

25

30

SNR (dB)

SFN channel

• The proposed channel estimation method outperforms the one proposed in [5]. • 6.8 dB gain over the PN-based one in terms of required SNR to achieve a MSE level of 1×10-3 in TU-6 channel. • MSE reduced about ten times in SFN channel.

Ming LIU

17

Bit Error Rate (BER) Performance -1

-1

10

10

-2

-2

10

10

1.7 dB

BER

BER

Close to perfect case

1dB

1.7 dB

-3

-3

10

10

0.4 dB proposed data-aided method PN-based channel estimation perfect channel estimation channel estimation in [5]

-4

10

7

8

9

10

11

SNR (dB)

TU-6 channel

proposed data-aided method initial time domain channel estimation perfect channel estimation

-4

10

12

13

14

15

5

6

7

8

9

10

11

12

13

14

15

SNR (dB)

SFN channel

• 1.7 dB and 0.4 dB gain over method in [5] and the PN-based one, respectively, in terms of required SNR to achieve BER of 5×10-5 after LDPC and BCH decoder in TU-6. • Very close to performance of perfect channel estimation • 1.7 dB gain in SFN channel and the gap between the proposed method and perfect channel estimation is 1 dB.

Ming LIU

18

Computational Complexity Additional complexity from data-aided channel estimation

Steps rebuild data instantaneous estimation averaging Wiener filtering compute MSE combine Total real multiplications real additions

Basic operations O(1) 4N, 2N 2K, 2N 2KN, 2N(K-1) 4N, 2N 4N, 2N (12+2K)N+2K, (6+2K) N → O(N) ~ O(N2) K: number of virtual pilots N: FFT size Ming LIU

19

Outline • Introduction to TDS-OFDM signal

• PN-based channel estimation in time domain • Proposed data-aided channel estimation in frequency domain • Simulation

• Conclusion

Ming LIU

20

Conclusion • Propose a combined time and frequency channel estimation algorithm. – time domain channel estimation based on circular correlation of received & local PN sequences. – a low-complexity data-aided channel estimation method excluding the channel decoder from the feedback loop.

– averaging over coherence bandwidth and Wiener filtering based interpolation to improve the data-aided channel estimation. – MMSE based combination of time and frequency domain estimates.

• The proposed channel estimation method outperforms the PN-based one as well as the typical one in the literature.

Ming LIU

21

Thank you!

LIU Ming [email protected] Institute of Electronics and Telecommunications of Rennes (IETR) 20, Av. des Buttes des Coesmes, 35708 Rennes, France

Ming LIU

22