Presentation is loading. Please wait.

Presentation is loading. Please wait.

Hanks H. Zeng Ye Li Jack H. Winters AT&T Labs - Research Red Bank, NJ

Similar presentations


Presentation on theme: "Hanks H. Zeng Ye Li Jack H. Winters AT&T Labs - Research Red Bank, NJ "— Presentation transcript:

1 A FAST SELECTIVE-DIRECTION MMSE TIMING RECOVERY ALGORITHM FOR SPATIAL-TEMPORAL EQUALIZATION IN EDGE
Hanks H. Zeng Ye Li Jack H. Winters AT&T Labs - Research Red Bank, NJ September 27, 2000

2 OUTLINE EDGE Spatial-Temporal equalization
Selective-direction timing recovery Results Conclusions

3 THIRD GENERATION TDMA SYSTEM: EDGE
EDGE: Enhanced Data Rates for Global Evolution High data rate (384 kbps) service based on GSM, evolution of IS-136 and GSM 8PSK at ksps 26 symbol training sequence 1/3, 3/9 or 4/12 reuse Both ISI and CCI 58 3 8.25 3 26 58 s

4 SMART ANTENNAS IN TDMA SYSTEMS
Currently, 2 receive antennas at base stations with IS-136 Combining changed from maximal ratio to MMSE combining in 1999 => 3-4 dB SINR gain Planned upgrade to 4 antennas for further improvement Spatial processing (track channel across time slot) Want to use smart antenna techniques in EDGE Spatial-Temporal processing (constant weights across time slot)

5 SMART ANTENNAS IN EDGE Spatial-Temporal processing using DDFSE for interference suppression 5 feedforward taps, 5 feedback taps, 8-state Viterbi Equalizer Degradation due to shortened equalizer => timing critical

6 WEIGHT ESTIMATION AND TIMING RECOVERY
Previous technique: Weight estimation: Train as MMSE-DFE Timing: MPE - Minimize ratio of precursor to cursor energy Issue: Much poorer performance with strong precursor versus postcursor ISI

7 WEIGHT ESTIMATION AND TIMING RECOVERY
Solution: Selective-direction equalization Since time slot is processed as a block, select either the forward or reverse direction for processing, with timing that minimizes the MMSE (trained as MMSE- DFE).

8 WEIGHT ESTIMATION AND TIMING RECOVERY
Issue: Computational complexity of training is doubled (with two directions) - important since inverse of large matrix is required, which is computationally intensive Solution: As shown in paper, the training algorithm can be modified so that the same matrix inverse is used for both directions => minimal computational-complexity increase with our technique

9 BER with 2 antennas, HT profile, and single interferer

10 BLER with single interferer (MCS-5)

11 CONCLUSIONS Proposed a selective-direction MMSE timing recovery algorithm for ST processing: Improves performance with minimal increase in computational complexity Applied algorithm to EDGE: 3 dB lower SIR for 1% raw BER with 2 antennas as compared to previous technique Up to 25 dB interference suppression at 10% BLER with two antenna versus single antenna receiver


Download ppt "Hanks H. Zeng Ye Li Jack H. Winters AT&T Labs - Research Red Bank, NJ "

Similar presentations


Ads by Google