Road to High Speed WLAN. Xiaowen Wang

Transcription

1 Road to High Speed WLAN Xiaowen Wang

2 Introduction n standardization process. Technologies enhanced throughput Raw data rate enhancement Overhead management Final remarks LSI Confidential 2

3 Background Jan. 2004, 11n working group started with the goal of reaching network throughput of over 200Mb/s. Since then, several industry consortiums began to work on proposals for 11n, namely, TGnSync and WWiSE. Jan complete proposals were submitted to 11n by TGnSync, WWiSE and MITMOT. Till July 2005, no one proposal can pass the confirmation vote. The three parties began the work of merging to a joint proposal. EWC (enhanced wireless connection) formed in Oct by members of TGnSync and WWiSE working on the joint proposal. Jan proposal from EWC was approved as confirmed proposal. First letter ballot stated in March Nov. 2006, draft 1.06 was approved. Draft 2.0 was approved in March LSI Confidential 3

4 Starting Point Highest raw link data rate in a/g is 54Mb/s Coding rate: 3/4 Modulation: 64QAM Symbol duration: 3.2µs GI duration: 0.8µs With network efficiency of 90%, the network throughput would be 48Mb/s. 16µs training and 4µs signal field IFS ACK and other control packages LSI Confidential 4

5 Solutions Increase raw physical layer link data rate Increase coding rate Increase constellation density Increase bandwidth MIMO: spatial multiplexing Reduce overhead Physical layer signaling, training, MAC layer header, control and signaling package. Increase data package length. LSI Confidential 5

6 Data Rate Enhancement

7 Advanced Coding Goal: achieve higher data rate and enhance performance Convolutional code: New code rate: 5/6 LDPC code: optional Code rates: ½ 2/3 3/4 5/6. Turbo code had been considered, but does not make it to the final draft. STBC: 1 spatial stream to 2 Tx antennas: optional Data rate increase: x10/9 Overhead associated: signaling and training LSI Confidential 7

8 High Density Constellation 256QAM was considered but not accepted in the current draft The SNR requirement is too high, over 30dB required for PER of 10%. Possible data increase: x2 Overhead associated: signaling LSI Confidential 8

9 Bandwidth Increase Bandwidth is doubled from 20MHz to 40MHz: optional One user can have as many as 108 subcarriers. Data rate increase: x2.25 Overhead associated Signaling MAC layer probing LSI Confidential 9

10 MIMO Spatial multiplexing: map Nsts spatial streams (independent data streams) to Ntx transmitter antennas. k k k Y Q X N 1 N N N 1 STS TX STS STS Direct mapping: Nsts = Ntx, either identity or CSD matrices Spatial expansion (optional): Nsts < Ntx, simple CSD or selective diversity Beamforming (optional): decide the steering matrix on the run according to the channel information by the sounding packet Data rate increase: x4 Overhead: Signaling Training MAC support LSI Confidential 10

11 MAC Layer Procedure of Beamforming Implicit feedback Use the reciprocity of the channel Beamformer send training request, beamformee send training back, then beamformer starts beamforming. Calibration is needed to ensure reciprocity. Explicit feedback Null sounding packet is sent by the beamformer and then beamformee estimate the channel and quantize and feedback the channel information to the beamformer. LSI Confidential 11

12 Overhead Management

13 OFDM Symbol Level GI (guard interval) length 11a/g, GI length is ¼ of symbol length There are 3 different GI legnth ¼ of symbol length, 0.8μs 1/8 of symbol length, 0.4μs ½ of symbol length, 1.6μs FFT size of symbol length MIT/MOT proposed to double the symbol length to 128, but was not adopted by the current draft. To realize the efficiency benefit, the MAC layer package has to be certain length which is not the case in practice. LSI Confidential 13

14 PPDU Level Signal field has to increase to accommodate the more complex PHY. Current HT signal field has 2 OFDM symbols transmitted in BPSK. Legacy signal field is kept for backward compatibility. Training requirement is basically proportional to the number of spatial streams. A 4μs HT-STS is added to improve AGC accuracy Long training design criteria All tones of different channel corresponding to different antenna pairs should be independently trained. LTS LTS U NTX NLTF The column of U should be orthognal. LTS All transmitter chains should be excited all the time. No beam should be formed as different transmitter chains transmit correlated signals. CSD*Walsh LSI Confidential 14

15 Greenfield Operation In the HT mixed format, a legacy (11a/g) training and signal field are transmitted. Attempt to reuse the legacy training had been considered. AGC accuracy is the reason for adding HT-STF and HT-LTF Signal field is kept for backward compatibility. L-STF L-LTF L-signal HT-signal HT-STF HT-LFT1 HT-LFT1 Greenfield operation: no legacy transmission exists Greenfield condition is assured by MAC protection scheme. Save 16μs compared to HT-mixed format Only 2 pilots instead of 4 pilots are used. HT-STF HT-LTF HT-signal HT-LFT2 HT-LTF2 Each HT-LTF is only half the size (one symbol) of the L-LTF, more sophisticated channel estimation algorithm is required to ensure the performance. LSI Confidential 15

16 MAC Level Aggregation HT PHY increases the signal field and training HT PHY decreases the data portion MAC layer data packet size is limited by higher layer (such as TCP/IP) Aggregate several MAC layer packets and transmit them in one PPDU. Block ACK One ACK package is at least 24μs no matter how fast the PHY is. Block ACK is not only a QoS scheme, but also increase the throughput. RIFS LSI Confidential 16

17 Final Remarks n is the first wireless standard that MIMO takes the center stage. Spatial multiplexing STBC Transmitter beamforming WiFi announced last August, they are going to began certificate pre- 11n products this year. The feature list is still in working, but only baseline system would be certified. Convolutional code with rate ½, 2/3, ¾, 5/6. 2x2 20MHz spatial multiplexing Greenfield operation Packet aggregation and block ACK Lessons learned from this successful MIMO OFDM practice. System becomes more and more flexible. Control pane becomes more and more a challenge Cross layer optimization and tradeoff. LSI Confidential 17

18 LSI Confidential 18