1. False L-STF Detection Issue
Month Year
doc.: IEEE yy/xxxxr0
July 2019
False L-STF Detection Issue
Date:
Authors:
Name
Affiliations
Address
Phone
Steve Shellhammer
Qualcomm
Geert Awater
Richard van Nee
Bin Tian
Youhan Kim
Steve Shellhammer, Qualcomm
John Doe, Some Company

2. July 2019
Introduction
OFDM-Based PPDUs begin with a legacy short training field (L- STF) consisting of ten repetitions of a short training sequence
Non-HT (11a), HT (11n), VHT (11ac), HE (11ax)
Receivers use the STF for packet detection, as well as for other functions like AGC, coarse time and frequency offset estimation
Here we will show that it is highly likely that the narrowband portion of a WUR PPDU may falsely trigger the L-STF detector of a STA not operating in WUR mode
This means that the STA will confuse a WUR PPDU with an L-STF, making the STA think it has detected a non-WUR PPDU
This can occur in the OBSS case, where the STA does not detect the WUR preamble
Steve Shellhammer, Qualcomm

3. Autocorrelation Metric
July 2019
Autocorrelation Metric
It is common for STAs to calculate the autocorrelation of the receive signal, with a time offset equal to the short training sequence duration (800 ns)
One possible detection metric is based on the normalized autocorrelation of one or more pairs of short training sequences
The normalized autocorrelation over 𝑁 short training sequence is given below
𝐶 𝑥𝑥,𝑁 (𝑡)= 1 𝑁−1 𝑘=0 (𝑁−1)𝑇−1 𝑥 𝑡+𝑘 𝑥 𝑡+𝑘+𝑇 ∗ 1 𝑁 𝑘=0 𝑁𝑇−1 𝑥 𝑡+𝑘 2
Where 𝑇 is the duration of a short training sequence, in samples
Steve Shellhammer, Qualcomm

4. L-STF Detection Simulations
July 2019
L-STF Detection Simulations
We simulated a typical L-STF detector on both the Sync and Data Fields, using the 2 and 4 µs MC-OOK symbols in Annex AC
For the Sync Field and the HDR Data Field (where the 2 µs MC-OOK symbols are used) Examples 1 and 2 in Annex AC we see a low normalized autocorrelation metric value as well as low false L-STF detection rates
For the LDR Data Field (where the 4 µs MC-OOK symbols are used) we see high normalized autocorrelation metric values as well as high false L-STF detection rates, for all three Examples
Steve Shellhammer, Qualcomm

5. Evaluation – 2 µs On Symbols
July 2019
Evaluation – 2 µs On Symbols
We evaluated the three examples of 2 µs MC-OOK On Symbols in Annex AC
We considered the case of six STFs (𝑁=6) used in the L-STF detector, which we consider a typical case (each implementation may be different)
As part of the evaluation we included the cyclic shift in the Symbol Randomizer
We then selected the highest normalized autocorrelation metric (over the possible cyclic shifts)
2 µs MC-OOK On Symbol
Maximum Normalized Autocorrelation Metric
Example 1
0.30
Example 2
0.26
Example 3
0.58
While the detector threshold is implementation specific, in our simulations we have found the false alarms rate is highly dependent on the autocorrelation metric
Examples 1 and 2 had negligible L-STF false alarm rates
Examples 3 had high L-STF false alarm rates, which is indicated by its high correlation metric
Steve Shellhammer, Qualcomm

6. Evaluation – 4 µs On Symbols
July 2019
Evaluation – 4 µs On Symbols
We evaluated the three examples of 4 µs MC-OOK On Symbols in Annex AC
We considered the case of six STFs (𝑁=6) used in the L-STF detector, which we consider a typical case (each implementation may be different)
As part of the evaluation we included the cyclic shift in the Symbol Randomizer
We then selected the highest normalized autocorrelation metric (over the possible cyclic shifts)
4 µs MC-OOK On Symbol
Maximum Normalized Autocorrelation Metric
Example 1
0.57
Example 2
0.58
Example 3
0.52
The detector threshold is implementation specific, but these high correlation metrics generally leads to high false alarms in L-STF detection as shown in our simulations
Steve Shellhammer, Qualcomm

7. July 2019
System Impact
STAs not operating in WUR mode will often falsely detect L-STF during the data portion of the WUR PPDU
Once the STA determines that this is not a valid PPDU it will go back to L-STF detection mode again and likely false detect again
This will likely continue throughout the entire WUR PPDU
This will cause these STAs to consume power for the entire WUR PPDU
These STAs may miss actual non-WUR PPDUs since the receiver is occupied processing, due to these false L-STF detections
Depending on the implementation this will cause many PHY errors which will lead to a negative impact on performance and channel access
Steve Shellhammer, Qualcomm

8. Maximum Normalized Autocorrelation Metric
July 2019
New 4 µs Symbol
Through a global optimal search we identified a new 4 µs Symbol that has a significantly lower normalized autocorrelation metric while also having good WUR PER performance
New Symbol = [-1, 1, 1, 1, -1, 1, 0, -1, -1, -1, 1, -1, -1]
Here we add the New Symbol Maximum Normalized Autocorrelation Metric to the Table
4 µs MC-OOK On Symbol
Maximum Normalized Autocorrelation Metric
Example 1
0.57
Example 2
0.58
Example 3
0.52
New Symbol
0.22
A Maximum Normalized Autocorrelation metric of 0.22 is significantly lower, and in our simulations the L-STF false alarm rate is insignificant
Steve Shellhammer, Qualcomm

9. WUR Performance using the New Symbol
July 2019
WUR Performance using the New Symbol
We simulated the performance of the new symbol in the LDR Data Field
We compare the PER performance to the Example 1 in Annex AC
In both cases, for the 2 µs symbol in the Sync Field, we used Example 1 from Annex AC
It has the lowest overall normalized autocorrelation metric
We simulated for AGWN and Channel Model D for both Single TX Antenna and Four TX Antennas cases
Steve Shellhammer, Qualcomm

10. AWGN – Single TX Antenna
July 2019
AWGN – Single TX Antenna
Small Loss in performance with New Symbol
Note, that Example 1 has the best AWGN performance of any of the three examples in Annex AC, due to its low PAPR
Steve Shellhammer, Qualcomm

11. Channel Model D – Single TX Antenna
July 2019
Channel Model D – Single TX Antenna
Small Loss in performance with New Symbol
Steve Shellhammer, Qualcomm

12. AWGN – Four TX Antennas Small Loss in performance with New Symbol
July 2019
AWGN – Four TX Antennas
Small Loss in performance with New Symbol
Steve Shellhammer, Qualcomm

13. Channel Model D – Four TX Antennas
July 2019
Channel Model D – Four TX Antennas
Negligible Loss in performance with New Symbol
Steve Shellhammer, Qualcomm

14. July 2019
Conclusions
Examples 1 and 2 of the 2 µs On Symbols from Annex AC both have low L-STF false alarm rates, while Example 3 has a high false alarm rate
All Three Examples of the 4 µs On Symbols from Annex AC have high L-STF false alarm rates
These high L-STF false alarm rates will cause excessive power consumption and may result in missed packet detection of non- WUR PPDUs
The New 4 µs Symbol provided in this presentation this issue can be resolved with only a very small loss in WUR performance
There are two possible approaches to addressing this issue
Details on Next Slide
Steve Shellhammer, Qualcomm

15. Options for Resolving the Issue
July 2019
Options for Resolving the Issue
Mandate Use of MC-OOK On Symbols in Annex AC
In Table AC-1 (2 µs Symbols) keep Examples 1 and 2
In Table AC-2 (4 µs Symbols) specify the New Proposed Symbol and possibly other similar sequences with proven good autocorrelation metric and demodulation metrics
Add text in the Draft stating that the MC-OOK On Symbol shall be based on the Examples in Tables AC-1 and AC2
Provide a list of Recommended MC-OOK On Symbols
For 2 µs Symbols keep Examples 1 and 2 in Table AC-1
For 4 µs Symbols specify the New Proposed Symbol in Table AC-2. We can also and another similar sequence with proven good autocorrelation metric and demodulation metrics
If an implementation chooses a different sequence other than those specified in Annex AC, additional transmit testing shall be added in the transmit specification clause to verify it autocorrelation metric less than TBD (e.g. 0.35) threshold
Steve Shellhammer, Qualcomm