1. Aug 2004
Project: IEEE P Working Group for Wireless Lacal Area Networks (WLANs)
Submission Title: Inprocomm PHY Proposal for IEEE n: MASSDIC-OFDM
Date Submitted: Aug 13th, 2004
Source: Kim Wu et al., Inprocomm, Inc.
Address 9F, No. 93, Shuei-Yuan St. Hsinchu, 300 Taiwan, R.O.C
Voice: , FAX: ,
Re: [Response to Call for Proposals]
Abstract: We present a partial proposal for a high-data-rate physical layer of a Local Area Network, using MASSDIC-OFDM transmission. The air interface is based on 2x2 MIMO-OFDM, using up to 256-QAM for the modulation. Advanced FEC coding using modern LDPC code is proposed with signal-space diversity coding to significantly improve the performance. The proposal is extendable to 3x3 MIMO and 4x4 MIMO.
Purpose: [Proposing a PHY-layer interface for standardization by n]
Notice: This document has been prepared to assist the IEEE P It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein.
Release: The contributor acknowledges and accepts that this contribution becomes the property of IEEE and may be made publicly available by P
Kim Wu et al., MASSDIC-OFDM

2. Inprocomm PHY Proposal for IEEE 802.11n: MASSDIC-OFDM
Aug 2004
Inprocomm PHY Proposal for IEEE n: MASSDIC-OFDM
Kim Wu, Chao-Yu Chen, Tsung-Yu Wu,
Racy Cheng, Chi-chao Chao, Mao-Ching Chiu
Inprocomm, Inc.
Kim Wu et al., MASSDIC-OFDM

3. Assumptions in the proposal Main features of the proposal
Aug 2004
Content
Assumptions in the proposal
Main features of the proposal
Proposed Multiple-Antenna Signal Space DIversity Coded OFDM (MASSDIC-OFDM) PHY System architecture
Modulation, precoding, FEC, proposed receiver structure
Frame format
Preamble
FEC Coding
Compatibility to a
Simulation Results
Summary
Kim Wu et al., MASSDIC-OFDM

4. Assumptions in the proposal
Aug 2004
Assumptions in the proposal
The proposal is targeted at the physical layer
A MAC efficiency of 60% is assumed
To reach the 100 Mbps MAC Goodput, a minimum of 167 Mbps is required.
Kim Wu et al., MASSDIC-OFDM

5. Main features of the proposal 1/2
Aug 2004
Main features of the proposal 1/2
This proposal inherits the good features of a OFDM standard
Spectrally efficient, robust against narrowband interference
Low complexity in channel equalization
No ISI and intercarrier interference (ICI) if channel max delay is less than the guard interval
Powerful FEC code by bit-interleaved convolutional coded modulation
This proposal uses MIMO (2x2 Mandatory) architecture to double the capacity. And the MIMO is extendable to 4x4
This proposal uses variable guard interval to optimize the data rate against different channel delay spread
The operation bandwith is targeted at 20 MHz, the same as that in a.
Kim Wu et al., MASSDIC-OFDM

6. Main features of the proposal 2/2
Aug 2004
Main features of the proposal 2/2
This proposal uses 256-QAM to boost bandwidth efficiency
This proposal explores the signal and space diversity without sacrificing BW via linear constellation precoding technique (LCP) (optional)
With low rate (e.g. 3/4) FEC, OFDM is shown to be inferior to single-carrier transmission due to loss of multipath diversity
This problem is resolved by artificially making ICI among independent subcarriers
The LCP is a kind of signal-space diversity code
This proposal use modern powerful error control code: extended irregular repeat-accumulated (eIRA) low density parity check (LDPC) code to improve performance
Kim Wu et al., MASSDIC-OFDM

7. Aug 2004
Proposed PHY: Multiple-Antenna Signal Space DIversity Coded OFDM (MASSDIC-OFDM)
Kim Wu et al., MASSDIC-OFDM

8. MASSDIC-OFDM Tx Architecture
Aug 2004
MASSDIC-OFDM Tx Architecture
Linear constellation Precoding Q
QAM mapping
Scrambler
FEC p1 RF
Subcarrier grouping
Nc-IFFT
L-CP
Nc-IFFT
L-CP RF
Nt=2 ( 3, 4 optional), # of Tx antennas
Nc=128, the number of subcarriers per antenna
p1: bit level interleaver (optional)
Q : a 4x4 unitary matrix
L-CP: 1200 or 800 ms cyclic prefix
FEC: eIRA LDPC code (2676, 2007), (2676, 1784), (2676,1338)
Kim Wu et al., MASSDIC-OFDM

9. Linear constellation precoding matrix Q
Aug 2004
Linear constellation precoding matrix Q
Kim Wu et al., MASSDIC-OFDM

10. Performance improvement with LCP (ML-soft output sphere decoding)
Aug 2004
Performance improvement with LCP (ML-soft output sphere decoding)
Kim Wu et al., MASSDIC-OFDM

11. Subcarrier index Nc=128, K=4 and the subcarrier grouping for LP
Aug 2004
Subcarrier index Nc=128, K=4 and the subcarrier grouping for LP
Antenna 1 y0 y2
95 96
127
0 1 2
Antenna 2 y1 y3
225
128
224
Kim Wu et al., MASSDIC-OFDM

12. System Parameters Nt=2 (lower bit rates are the same as 802
System Parameters Nt=2 (lower bit rates are the same as a Nt=1 case)
Aug 2004
MAC Goodput Mbps
--- 32 42 47 63 75 89
100
Info.Data Rate Mbps *X 53 70 79
105
125
148
167
QAM Constellation 16 64
256
FFT Size
128
Coding Rate R
1/2 (CC)
1/2
2/3
3/4
Pilot Tones
Data Tones
Info. Length ms
6.4
Cyclic Prefix ms
1200
800
Null Tones 12
Symbol Length ms
7.6
7.2
Channel Bit Rate Mbps
158
222
* Rate X is dedicated for the header. The K=7, CC encoded data is 16-QAM modulated and inserted into the pilot tones of the first OFDM symbol of the first 2 antennas while those of the other antennas are nulled.
Kim Wu et al., MASSDIC-OFDM

13. System Parameters Nt=3 (Optional)
Aug 2004
System Parameters Nt=3 (Optional) 48 63 71 95
113
133
150
Info.Data Rate Mbps 80
105
118
158
188
222
250
QAM Constellation 16 64
256
FFT Size
128
Coding Rate R
1/2
2/3
3/4
Pilot Tones
Data Tones
100
Info. Length ms
6.4
Cyclic Prefix ms
1200
800
Null Tones 12
Symbol Length ms
7.6
7.2
Channel Bit Rate Mbps
159
237
333
Kim Wu et al., MASSDIC-OFDM

14. System Parameters Nt=4 (Optional)
Aug 2004
System Parameters Nt=4 (Optional)
MAC Goodput 64 84 95
126
150
178
200
Info.Data Rate Mbps
106
140
158
211
250
296
333
QAM Constellation 16
256
FFT Size
128
Coding Rate R
1/2
2/3
3/4
Pilot Tones
Data Tones
100
Info. Length ms
6.4
Cyclic Prefix ms
1200
800
Null Tones 12
Symbol Length ms
7.6
7.2
Channel Bit Rate Mbps
159
316
444
Kim Wu et al., MASSDIC-OFDM

15. Proposed Receiver Structure for 2x2 case
Aug 2004
Proposed Receiver Structure for 2x2 case RF
DU1
MIMO Equalization (MMSE, DEF, ML-S)
LCP-decoding (ML-S) RF
DU2
Information
bits
De-Scrambler
FEC-decoder
QAM Demapping
De-int.
Demodulation Unit ( DU )
Channel
Estimation,
Equalization
Sym.
Detection CP
Remove
FFT
Synch.
Kim Wu et al., MASSDIC-OFDM

16. Aug 2004
802.11n Frame format
If Pad bits > 64, then a repeated PHY header is substituted for the first 64 bits of the pad bits. The FCS only checks for payload.
First OFDM symbol
PLCP preamble
PHY
Header
Repeated PHY Header
Pad
bits
Payload: bytes
FCS
CRC-16
Data field
K=7, R=1/2 CC encoded and inserted into the pilot tones of the first OFDM Symbol of the fist two antennas while those of other antennas are nulled.
Rate
6 bits
Reserved
2 bits
LCP
1 bit
Interleaver
1 bit
Length
16 bits
Service
16 bits
HCS
16 bits
Tail
6 bits
Kim Wu et al., MASSDIC-OFDM

17. Operation for lower transmission rates
Aug 2004
Operation for lower transmission rates
For n Tx-Rx in 2x2 case operating in the lower transmission rate mode (6~54Mbps), the mapped information is encoded as complex 2x2 Alamouti space-time code in the 2 Tx antenna
For n Tx-Rx in 4x4 case, the mapped information is encoded as complex 4x4 Alamouti space-time code in the 4 Tx antenna
For n Tx-Rx in 3x3 case, the third Tx antenna is turned off while the others are the same as the case in 2x2
Kim Wu et al., MASSDIC-OFDM

18. Preamble structure 2X2 case
Aug 2004
Preamble structure 2X2 case
5.6 ms
2.4 ms X X X X X X -X -X -X -X Y Y Y Y Y Y -Y -Y -Y -Y
TX 1 S1 CP L1
Data 1
TX 2 S2 CP L2
Data 2
1.6 ms
6.4 ms
Si’s: Short Training Symbols, X is the same as a, Y is orthogonal to X.
Li’s: Long Training Symbols
CP: Cyclic Prefix
Kim Wu et al., MASSDIC-OFDM

19. Preamble structure 3X3 case
Aug 2004
Preamble structure 3X3 case
5.6 ms
2.4 ms X X X X X X -X -X -X -X Y Y Y Y Y Y -Y -Y -Y -Y Z Z Z Z Z Z -Z -Z -Z -Z
TX 1 S1 CP L1
Data 1
TX 2 S2 CP L2
Data 2
TX 3 S3 CP L3
Data 3
1.6 ms
6.4 ms
X ,Y, Z orthogonal.
Kim Wu et al., MASSDIC-OFDM

20. Preamble structure 4X4 case
Aug 2004
Preamble structure 4X4 case
5.6 ms
2.4 ms X X X X X X -X -X -X -X Y Y Y Y Y Y -Y -Y -Y -Y Z Z Z Z Z Z -Z -Z -Z -Z W W W W W W -X -W -W -W
TX 1 S1 CP L1
Data 1 S2 CP L2
Data 2 S3 CP L3
Data 3
TX 2 S4 CP L4
Data 4
1.6 ms
6.4 ms
X ,Y, Z W orthogonal.
Kim Wu et al., MASSDIC-OFDM

21. More on the Long Training Symbols in the frequency domain
Aug 2004
More on the Long Training Symbols in the frequency domain
with
The long training symbols are designed such that the Mean squared error (MSE) is minimized when ML-estimation is used
The long training symbols Li’s have the following properties:
The Li’s are orthogonal
The Li’s are nearly circular-shift orthogonal
Low PAPR
Kim Wu et al., MASSDIC-OFDM

22. Use the same scrambler as 802.11a
Aug 2004
Data Scrambler
Use the same scrambler as a
Kim Wu et al., MASSDIC-OFDM

23. Bit-level Interleaver
Aug 2004
Bit-level Interleaver
Kim Wu et al., MASSDIC-OFDM

24. Details of the Interleaver
Aug 2004
Details of the Interleaver
Denote by k the index before permutation and j the index after permutation. ( : the number of coded bits per OFDM symbol)
First permutation:
Second permutation:
where .
is obtained by concatenating two OFDM symbols that have been bit-interleaved by the above interleaver.
Kim Wu et al., MASSDIC-OFDM

25. Aug 2004
Pad bits
The number of OFDM symbols, ; the number of bits in the DATA field, ; and the number of pad bits, , are computed in terms of the length of the PSDU (LENGTH) by:
Kim Wu et al., MASSDIC-OFDM

26. Aug 2004
FEC Coding
Using extended irregular repeat-accumulate (eIRA) code (a kind of LDPC code)
For R=3/4: (2676, 2007)
For R=2/3: (2676, 1784)
For R=1/2: (2676,1338)
For information length other than 2007 (1784 or 1338) use code shortening
Header is CC encoded with R=1/2 and K=7 (the same as a)
Kim Wu et al., MASSDIC-OFDM

27. Aug 2004
FEC encoder structure
Kim Wu et al., MASSDIC-OFDM

28. Packet size accommodation with concatenation and shortening
Aug 2004
Packet size accommodation with concatenation and shortening
2007 (1784, 1338) bit data field
669 (892, 1338) bit parity
2676 bit codeword
For long packets, codewords are concatenated
669 (892, 1338)
bit parity
2676-N bit zero pad
N bit data field
For short blocks, codeword shortening is adapted.
The zero pad will not be transmitted.
Kim Wu et al., MASSDIC-OFDM

29. Details of the proposed FEC code
Aug 2004
Details of the proposed FEC code
In general, the parity check matrix H of an (n,k) eIRA code could be written as the following form :
, where is a random spare matrix,
is the inverse of a lower-triangular
matrix.
Kim Wu et al., MASSDIC-OFDM

30. Then, we put 1’s on randomly and avoid the length-4 cycle.
Aug 2004
We decide the column and row weight distribution of through Gaussian Approximation.
Then, we put 1’s on randomly and avoid the length-4 cycle.
Kim Wu et al., MASSDIC-OFDM

31. The generator matrix is in the format of
Aug 2004
The generator matrix is in the format of
Usually, it costs large complexity to calculate
. However, the which is a full upper-triangle matrix can be implemented by a differential encoder in the above figure.
Kim Wu et al., MASSDIC-OFDM

32. Short block size ( K bit)
Aug 2004
Packet Encoding
For a short block, it is not efficient if we encode it in the original code rate. Hence, we choose the other code rate to decrease the number of parity check bits.
If the block size is on the interval of the table shown below, we change the code rate.
Code Rate
(former)
Short block size ( K bit)
(later)
R = 1 / 2
850~480
R=2/3
<480
R=3/4
R = 2 / 3
<1000
Kim Wu et al., MASSDIC-OFDM

33. Aug 2004
Decoder Structure
For a LDPC code, it can be described as a Tanner graph with variable nodes , check nodes and edges.
Kim Wu et al., MASSDIC-OFDM

34. Each check node represents a parity check equation.
Aug 2004
Each check node represents a parity check equation.
Check node j is connected to a variable node i if and only if the element hji in the parity check matrix H is a 1
Through the Tanner graph, belief propagation algorithm is used to decode the eIRA code.
Kim Wu et al., MASSDIC-OFDM

35. Codec Complexity Analysis
Aug 2004
Codec Complexity Analysis
There are two kinds of operations in the decoding process: addition and check node operation (*).
The complexity of the two operations are similar.
The degree of a node means the number of edges connected to the node.
We define and respectively as the number of variable and check nodes of degree ; and as the maximum degree of variable and check nodes. *:
Kim Wu et al., MASSDIC-OFDM

36. check node operations in one iteration.
Aug 2004
There are
additions and
check node operations in one iteration.
Kim Wu et al., MASSDIC-OFDM

37. The weight distributions of column and row are:
Aug 2004
The weight distributions of column and row are:
Code Rate
Nv(i)
Nc(i)
R = 1 / 2
i: 1 2 4 5 6 7 1
1337
656
682
657
681
R = 2 / 3 3 9 10
891
453
1331
454
438
R = 3 / 4 11 12
668
1562
445
225
444
Kim Wu et al., MASSDIC-OFDM

38. Aug 2004
In our eIRA code, the numbers of operations per iteration are as follows:
Code Rate
addition
chk_op
Total
R = 1 / 2
15478
39603
55081
R = 2 / 3
12007
63642
75649
R = 3 / 4
8246
75555
83801
Kim Wu et al., MASSDIC-OFDM

39. Ratio to eIRA per iteration
Aug 2004
The Viterbi decoding complexity for Convolutional code (802.11a) with code word length 2676
Code rate
Addition
Ratio to eIRA per iteration
R = 1 / 2
256896
4.6643
R = 2 / 3
342528
4.5276
R = 3 / 4
385344
4.5984
(*): Here, we ignore the calculation of branch metric and the memory trace-back process.
And, we assume that the complexity of comparing is equal to that of addition.
Kim Wu et al., MASSDIC-OFDM

40. Aug 2004
Compatibility to a
The proposed n is compatible to a by defining the same PHY and MAC as that of a in low data rate mode (6Mbps~54Mbps mode).
A n device can distinguish between a and n packets by detecting different format of packet preambles.
Upon detection of a packets, the n device turns to operate in the a mode.
Kim Wu et al., MASSDIC-OFDM

41. Aug 2004
Simulation Results
PER vs. Eb/N0 performance for fading channel model B, C, D, E and AWGN channel.
PER vs. SNR performance for fading channel model channel B, C, D, E and AWGN channel.
Throughput vs. SNR performance for channel B, C, D, E and AWGN channel.
Kim Wu et al., MASSDIC-OFDM

42. 2X2 MIMO AWGN Channel PER vs. Eb/No for Different Data Rates
Aug 2004
2X2 MIMO AWGN Channel PER vs. Eb/No for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

43. 2X2 MIMO Channel B PER vs. Eb/No for Different Data Rates
Aug 2004
2X2 MIMO Channel B PER vs. Eb/No for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

44. 2X2 MIMO Channel C PER vs. Eb/No for Different Data Rates
Aug 2004
2X2 MIMO Channel C PER vs. Eb/No for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

45. 2X2 MIMO Channel D PER vs. Eb/No for Different Data Rates
Aug 2004
2X2 MIMO Channel D PER vs. Eb/No for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

46. 2X2 MIMO Channel E PER vs. Eb/No for Different Data Rates
Aug 2004
2X2 MIMO Channel E PER vs. Eb/No for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

47. 2X2 MIMO AWGN Channel PER vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO AWGN Channel PER vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

48. 2X2 MIMO Channel B PER vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel B PER vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

49. 2X2 MIMO Channel C PER vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel C PER vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

50. 2X2 MIMO Channel D PER vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel D PER vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

51. 2X2 MIMO Channel E PER vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel E PER vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

52. 2X2 MIMO AWGN Channel : throughput vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO AWGN Channel : throughput vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

53. 2X2 MIMO Channel B : Throughput vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel B : Throughput vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

54. 2X2 MIMO Channel C : throughput vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel C : throughput vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

55. 2X2 MIMO Channel D : throughput vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel D : throughput vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

56. 2X2 MIMO Channel E : throughput vs. SNR for Different Data Rates
Aug 2004
2X2 MIMO Channel E : throughput vs. SNR for Different Data Rates
Kim Wu et al., MASSDIC-OFDM

57. Complexity relative to 802.11a
Aug 2004
Complexity relative to a
802.11n 167Mbps
Ratio to a
802.11n 54Mbps
802.11a 54Mbps
#multipliers for Equalization
12.37
4 (MMSE receiver for Nt=2) 1
#multipliers for Equalization with LP
(Sphere decoding for LP)
791.68
4*4^3=256(the size of LP is 4x4; the complexity of SD is 4^3)
#addition for FEC(R=1/2)
6.65
2.15 ( average 10 iterations when per<0.1)
1(viterbi decoder)
#addition for
FEC(R=2/3)
6.83
2.21 (average 10 iterations when per<0.1)
FEC(R=3/4)
6.71
2.17 (average 10 iterations when per<0.1)
#multipliers for FFT
7.22
128log(128)/64log(64)
=2.33
Kim Wu et al., MASSDIC-OFDM

58. Summary A PHY proposal MASSDIC-OFDM is presented
Aug 2004
Summary
A PHY proposal MASSDIC-OFDM is presented
A MAC throughput of 100 Mbps is achieved with only 2X2 MIMO
Further extension to 3x3 MIMO and 4x4 is reserved as an option
The performance curves of the proposed PHY are shown with MMSE detection, improvement of 2~4 dB is possible with more advanced detection like MMSE-DFE, ML-sphere decoding
The complexity is estimated and compared to that of a 54Mbps
Kim Wu et al., MASSDIC-OFDM