1. September 2009
Project: IEEE P Working Group for Wireless Personal Area Networks (WPANs)
Submission Title: [Suggested Improvements to SUN FPP Proposal]
Date Submitted: [September 2009]
Source: [T. Schmidl, A. Batra, S. Hosur, K. Le] Company [Texas Instruments]
Address [12500 TI Blvd, Dallas, TX USA]
Voice:[ ], FAX: [ ],
Re: [In response to SUN FPP proposal]
Abstract: [This presentation describes several suggested improvements to the SUN FPP proposal]
Purpose: [For information]
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
Tim Schmidl, Texas Instruments Inc.

2. Suggested Improvements to SUN FPP Proposal
September 2009
Suggested Improvements to SUN FPP Proposal
Tim Schmidl, Texas Instruments Inc.

3. Problems with Current Numerology
September 2009
Problems with Current Numerology
In the current proposal, a few of the modes result in a non-integer number of data bits per OFDM symbol
Examples:
NDSC = 100, BPSK, R = 1/2, Spreading = 4  NDBPS = 12.5
NDSC = 22, BPSK, R = 1/2, Spreading = 2  NDBPS = 5.5
NDSC = 22, BPSK, R = 3/4, Spreading = 1  NDBPS = 16.5
where NDSC = number of data sub-carriers and NDBPS = number of data bits per symbol
Having to support a non-integer number of data bits per symbol is awkward and leads to a more complicated design and different data path for these modes
Tim Schmidl, Texas Instruments Inc.

4. Proposed OFDM Numerology
September 2009
Proposed OFDM Numerology
Supported NDSC should be 96, 48, 24, 12, and 4  guarantees an integer number of data bits per symbol and common data rates across all FFT sizes
MCS
SNR (dB)
Number of Data Sub-Carrier (NDSC)
96 (8 pilots)
48 (4 pilots)
24 (2 pilots)
12 (2 pilots)
4 (2 pilots)
NCBPS
NDBPS
Rate
-5.4 24 12
93.75 1
-2.4 48
187.5 6
46.88 2
0.6 96
375 3
3.6
192
750 4
6.2 72
562.5 36
281.25 18
140.63 8 5
9.0 16
62.5
12.4
NCBPS = number of coded bits per symbol, NDBPS = number of data bits per symbol, Rate is in kbps, SNR measured at PER = 10%, 1000 bytes
MCS table was modified to achieve ~3dB step size in SNR
Tim Schmidl, Texas Instruments Inc.

5. Proposed Modulation, Coding Set (MCS) Values
September 2009
Proposed Modulation, Coding Set (MCS) Values
By slightly modifying the MCS values, we can add more robustness to the system and ensure roughly uniform step sizes in required SNR values for a fixed packet error rate
MCS table was modified to achieve ~3 dB step sizes in SNR
MCS
Constellation
Rate
FDS
Real Output
BPSK
1/2 4
YES 1 2
QPSK 3
DCM-QPSK NO
3/4 5
16 QAM 6
Tim Schmidl, Texas Instruments Inc.

6. September 2009
Puncturing Patterns
Define correct puncturing pattern for R = 3/4 (specification shows a puncturing pattern for R = 3/4, but it is actually for R = 2/3)
Source Data
Encoded Data
Bit Stolen Data
sent/received
Bit Inserted Data
Decoded Data
X0 X1 X2 X X4 X5
A0 A1 A2 A A4 A5
B0 B1 B2 B B4 B5
Stolen Bit
Inserted
Dummy Bit
A0 B0 A1 B A3 B3 A4 B5
y y y y y y5
Punctured Coding (r = ¾)
Tim Schmidl, Texas Instruments Inc.

7. Dual-Carrier Modulation
September 2009
Dual-Carrier Modulation
QPSK: 2 bits mapped to symbol, then on to IFFT tone
MCS 4-6: there is no spreading – system has to rely solely on FEC
As the code strength decreases, start to experience diversity loss
DCM (Dual-carrier modulation): info from two QPSK symbols (4 bits) are spread onto 2 tones separated by half of signal BW
A new form of spreading which has been shown to provide a gain in frequency diversity over QPSK
Probability that there is a deep fade on both tones is quite small
Even if there is a deep fade on one of the two tones, the 4 bits of information can be recovered using a simple joint MAP decoder
Tim Schmidl, Texas Instruments Inc.

8. Frequency Domain Spreading
September 2009
Frequency Domain Spreading
SUN is expected to generally operate in low Doppler spread environments, so time diversity will be limited  better to exploit the frequency diversity offered by frequency selective channels
Frequency domain spreading of 2 can be obtained by generating data symbols for the negative frequencies and copying the complex conjugates to the positive frequencies.
Frequency domain spreading of 4 can be obtained by generating data symbols for the left half of the negative frequencies. The entire block can be copied to the right half of the negative frequencies, and then complex conjugates can be used for the positive frequencies.
Tim Schmidl, Texas Instruments Inc.

9. September 2009
STF Structure
Because of the low SNR requirements, two repetitions of STF may not be enough. We propose to have 4 repetitions to ensure packets can be detected in the presence of low SNR:
To improve boundary detection performance, we propose to negate the last repetition of the last STF (“-s”)
Tim Schmidl, Texas Instruments Inc.

10. Repetition Period for STF
September 2009
Repetition Period for STF
Cyclic prefix is always set to 1/4 OFDM symbol, so multipath less than 1/4 symbol is fully contained within the cyclic prefix
If repetition period for STF is shorter than the length of cyclic prefix, then long channels will degrade the detection of the STF and multipath protection is determined by repetition length of the STF rather than cyclic prefix.
Propose to set repetition length of STF ≥ length of cyclic prefix
Option #
Current repetition length
Proposed repetition length 1
1/8 symbol
1/4 symbol 2 3 4
1/2 symbol 5
Tim Schmidl, Texas Instruments Inc.

11. Design of STF and LTF Sequences (1)
September 2009
Design of STF and LTF Sequences (1)
The lowest MCS levels produce real IFFT outputs, so the transmitter for these MCSs can be simplified (e.g. only 1 DAC is needed).
If we design the STF and LTF to be real sequences in the time domain (complex conjugate symmetry in the frequency domain), then the power used in the transmitter can scale with the data rate (lower real MCS’s use 1 DAC, while higher MCS’s use 2 DAC’s)
Real sequences tend to have higher peak to average power ratios than complex sequences, so the sequences should be designed to have low PAR
Tim Schmidl, Texas Instruments Inc.

12. Design of STF and LTF Sequences (2)
September 2009
Design of STF and LTF Sequences (2)
PAR (dB) for STF PAR (dB) for LTF
Original Complex Sequence
QPSK real sequence
Option 1
2.2496
3.7261
Option 2
4.2346
Option 3
3.4006
4.3983
Option 4
Option 5
3.0103
Original Complex Sequence
BPSK real sequence
Option 1
5.6003
5.2695
Option 2
3.1794
4.4236
Option 3
3.8258
4.1538
Option 4
5.8345
4.1017
Option 5
4.2597
4.3983
Tim Schmidl, Texas Instruments Inc.

13. Design of STF and LTF Sequences (3)
September 2009
Design of STF and LTF Sequences (3)
STF Sequences
STF_freq(Option-1) = sqrt(104/24)*[0,zeros(1,3),j,zeros(1,3),-j,zeros(1,3),-j,zeros(1,3),j,zeros(1,3),j,zeros(1,3),-j ,zeros(1,3),j,zeros(1,3),-j,zeros(1,3),j,zeros(1,3),j,zeros(1,3),j,zeros(1,3),j,zeros(1,31),-j,zeros(1,3),-j,zeros(1,3),-j,zeros(1,3),-j ,zeros(1,3),j,zeros(1,3),-j,zeros(1,3),j,zeros(1,3),-j,zeros(1,3),-j,zeros(1,3),j,zeros(1,3),j,zeros(1,3),-j,zeros(1,3)]
STF_freq(Option-2) = sqrt(52/12)*[0,zeros(1,3),-1,zeros(1,3),-j,zeros(1,3),1,zeros(1,3),j,zeros(1,3),1,zeros(1,3),-j ,zeros(1,15),j,zeros(1,3),1,zeros(1,3),-j ,zeros(1,3),1,zeros(1,3),j,zeros(1,3),-1,zeros(1,3)]
STF_freq(Option-3) = sqrt(26/6)*[0,zeros(1,3),-j,zeros(1,3),-1,zeros(1,3),-j,zeros(1,3),0,zeros(1,3),j,zeros(1,3),-1,zeros(1,3), j,zeros(1,3)]
STF_freq(Option-4) = sqrt(14/6)*[0,0,-j,0,-1,0,-j,0,0,0,j,0,-1,0,j,0]
STF_freq(Option-5) = sqrt(6/2)*[0,0,-j,0,0,0,j,0]
* change Option-1 multiplier to sqrt(108/24) if 100 data sub-carriers are kept.
Tim Schmidl, Texas Instruments Inc.

14. Design of STF and LTF Sequences (4)
September 2009
Design of STF and LTF Sequences (4)
LTF Sequences
LTF_freq(Option-1)= [0, 1,-1,-1, 1, 1,-1,-1, 1, 1,-1, 1, 1, 1, 1,-1, 1,-1,-1,-1,-1,-1,-1, 1, 1, 1, 1, 1,-1,-1, 1,-1, 1, 1, 1,-1,-1,-1,-1, 1, 1,-1,-1,-1
,-1, 1,-1,-1,-1,-1, 1,-1,-1 ,zeros(1,23), -1,-1, 1,-1,-1,-1,-1, 1,-1,-1,-1,-1, 1, 1,-1,-1,-1,-1, 1, 1, 1,-1, 1,-1,-1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1,-1, 1,-1
, 1, 1, 1, 1,-1, 1, 1,-1,-1, 1, 1,-1,-1, 1]
LTF_freq(Option-2)= [0,-1,-1,-1,-1, 1,-1, 1,-1, 1, 1,-1, 1, 1,-1,-1, 1,-1,-1,-1, 1, 1, 1,-1, ,-1, zeros(1,11),-1,-1,-1,-1, 1, 1, 1,-1,-1,-1, 1,
-1,-1, 1, 1,-1, 1, 1,-1, 1,-1, 1,-1,-1,-1,-1]
LTF_freq(Option-3)= [0,-1,-1,-1,-1,-1, 1,-1, 1, 1, 1,-1,-1, 1, zeros(1,5), 1,-1,-1, 1, 1, 1,-1, 1,-1,-1,-1,-1,-1]
LTF_freq(Option-4)= [0,-1,-1,-1,1,1,-1,1,zeros(1,1),1,-1,1,1,-1,-1,-1]
LTF_freq(Option-5)= [0,-1,-1,1, zeros(1,1),1,-1,-1]
If 100 data sub-carriers are kept then use
LTF_freq(Option-1 alternative)=[0, 1, 1,-1,-1, 1, 1,-1, 1, 1, 1,-1, 1, 1,-1, 1,-1, 1,-1,-1,-1,-1,-1, 1, 1, 1, 1,-1,-1,-1, 1,-1, 1,-1,-1, 1, 1, 1, 1,-1
,-1,-1, 1,-1,-1,-1, 1,-1,-1,-1,-1, 1, 1,-1, 1 ,zeros(1,19), 1,-1, 1, 1,-1,-1,-1,-1, 1,-1,-1,-1, 1,-1,-1,-1, 1, 1, 1, 1,-1,-1, 1,-1, 1,-1,-1,-1, 1, 1, 1, 1,
-1,-1,-1,-1,-1, 1,-1, 1,-1, 1, 1,-1, 1, 1, 1,-1, 1, 1,-1,-1, 1, 1]
Tim Schmidl, Texas Instruments Inc.

15. September 2009
Proposed PHY Header
Useful to have reserved bits in the PHY header to allow for evolution of the standard
Currently, there are only 7 MCS values, therefore, at most 5 bits are needed to describe RATE
Typical systems will need at most 4 retransmissions to achieve application level PER, so only need 2 bits to define scrambler seed
Proposed PHY Header structure: 36 bits
Tim Schmidl, Texas Instruments Inc.

16. September 2009
Bit Interleaver
Current draft: Options 1 and 2 use a two-stage interleaver, while Options 3 and 4 use a one-stage interleaver.
Proposal: Make implementation more consistent by using IEEE a two-stage interleaver for all options:
Tim Schmidl, Texas Instruments Inc.

17. Harmonized Symbol Rate (1)
September 2009
Harmonized Symbol Rate (1)
It would be beneficial to harmonize symbol rate with the other modulation options in TG4g, this implementation, i.e., all symbol clocks can be derived from same source
Let’s look at two very common crystals – 20 and 24 MHz (just examples)
Review of FSK symbol rates:
For FSK, 24 MHz crystal can support all data rates with integer dividers
Modulation
Data Rate (kbps)
Over-sampling Factor
Required Sample Rate (ksps)
Divider assuming 20 MHz crystal
Divider assuming 24 MHz crystal
FSK 40 2 80
250
300 50
100
200
240
120
160
320
62.5 75
400 60
Tim Schmidl, Texas Instruments Inc.

18. Harmonized Symbol Rate (2)
September 2009
Harmonized Symbol Rate (2)
How does 24 MHz crystal work for current OFDM sample rate?
Results in non-integer dividers which complicates implementation
Better choice is to use a sample rate = 1.2 Msps for 128-point FFT mode:
Recommendation change: Df = kHz  Tsym = ms, Tcp = 26.7 ms
Modulation
FFT Size
Over-sampling Factor
Required Sample Rate (ksps)
Divider assuming 24 MHz crystal
OFDM with 1.25 Msps 64 4
2500
9.6
128
5000
4.8
Modulation
FFT Size
Over-sampling Factor
Required Sample Rate (ksps)
Divider assuming 24 MHz crystal
OFDM with 1.2 Msps 64 4
2400 10
128
4800 5
Tim Schmidl, Texas Instruments Inc.