January 2016 doc.: IEEE 802.15-16-0018-00-007a Project: IEEE … · 2016. 1. 10. · doc.: IEEE...
Transcript of January 2016 doc.: IEEE 802.15-16-0018-00-007a Project: IEEE … · 2016. 1. 10. · doc.: IEEE...
doc.: IEEE 802.15-16-0018-00-007a
Submission Slide 1
Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs)
Submission Title: National Taiwan University 802.15.7r1 OCC Proposal
Date Submitted: 10 January, 2016Source: Hsin-Mu Tsai, National Taiwan UniversityAddress: No.1, Sec. 4, Roosevelt Rd., Taipei 10617, TaiwanVoice: +886 2 33664888 ext. 316, E-Mail: [email protected]
Re: CFP Response
Abstract: CFP Response
Notice: This document has been prepared to assist the IEEE P802.15. 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 P802.15.
<Hsin-Mu Tsai>, <National Taiwan University>
January 2016
doc.: IEEE 802.15-16-0018-00-007a
Submission
January 2016
<Hsin-Mu Tsai>, <National Taiwan University>Slide 2
National Taiwan University802.15.7r1 OCC Proposal
Hsin-Mu (Michael) [email protected]
doc.: IEEE 802.15-16-0018-00-007a
Submission
Table of Content
• Requirements• Background• PHY proposal – RS-FSK• MAC proposal –
Frame Format for RS-FSK
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 3
doc.: IEEE 802.15-16-0018-00-007a
Submission
REQUIREMENTS
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 4
doc.: IEEE 802.15-16-0018-00-007a
Submission
Transmitter RequirementsOverall objective:Retain the transmitter’s original function • The transmission should not create flickers visible to
human eyes• The transmitting LED should still retain dimming
capability• Use only ON and OFF – binary levels to modulate data
(simple transmitter design, minimum modifications to existing systems, cost efficient)
• Color is not used to modulate data (color is determined by user preference)
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 5
doc.: IEEE 802.15-16-0018-00-007a
Submission
Receiver Requirements
Overall Objective: Use existing hardware
• Use off-the-shelf, unmodifiedrolling shutter camera as the receiver– Adding software (e.g. smartphone app) instantly gives
existing devices the reception capability– The start of the exposure duration is assumed NOT
controllable (unsynchronized channel)– Camera frame rate is NOT accurate and can vary over
time
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 6
doc.: IEEE 802.15-16-0018-00-007a
Submission
Receiver Requirements• Cameras using different image sensors should be
able to demodulate the same transmissionsimultaneously– Parameters that could be different include
• Resolution• Row read-out duration• Frame rate • Exposure duration• TX-RX distance
• Camera receiver should be able to receive even when the transmitter does not occupy the entire image (operate at longer distance & in line-of-sightscenarios)
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 7
doc.: IEEE 802.15-16-0018-00-007a
Submission
BACKGROUND
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 8
doc.: IEEE 802.15-16-0018-00-007a
Submission
Light-to-Rolling-Shutter-Camera Channel
• Main characteristics:1. Rolling exposure process
(rolling shutter sampling)2. Time gap 3. Low-pass filtering due to integration
during exposure
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 9
doc.: IEEE 802.15-16-0018-00-007a
Submission
Global versus Rolling Shutter
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 10
Global Shutter:All pixels exposedat the same time
Rolling Shutter:Pixels are exposed
row by row
doc.: IEEE 802.15-16-0018-00-007a
Submission
Comparison of Sampling Schemes<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 11
doc.: IEEE 802.15-16-0018-00-007a
Submission
Idle Period<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 12
Time
Idle
Exposure(1st frame)
Exposure(2nd frame)
s1
s2s3
s4s5
s6
s7
s8
s9s10
s11
s12
Signal Loss
Symbol Loss
Receiving CameraTransmitted Signal
Idle period results in significant signal & symbol loss
doc.: IEEE 802.15-16-0018-00-007a
Submission
Idle Period + Un-occupied Area<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 13
Time
Idle
Exposure(1st frame)
Exposure(2nd frame)
s1
s2s3
s4s5
s6
s7
s8
s9s10
s11
s12
Signal Loss
Symbol Loss
Receiving CameraTransmitted Signal
Smaller light in the image results in even more loss
No signal
No signal
doc.: IEEE 802.15-16-0018-00-007a
Submission
Idle Period + Un-occupied Area<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 14
Time
Idle
Exposure(1st frame)
Exposure(2nd frame)
s1
s2s3
s4s5
s6
s7
s8
s9s10
s11
s12
Signal Loss
Symbol Loss
Receiving CameraTransmitted Signal
Different light location results in different portion
of signal / symbol loss
No signal
doc.: IEEE 802.15-16-0018-00-007a
Submission
Idle Period + Un-occupied Area = Time gap = Lossy Channel
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 15
Transmitting Light
Phone 1Phone 2 Phone 3
Time
Exposure(1st frame)
Exposure(2nd frame)
Idle (17%)
Exposure(1st frame)
Exposure(2nd frame)
Idle (38%)
Exposure(1st frame)
Exposure(2nd frame)
Idle (33%)
2m 1m15cm
Loss 60%
Loss 40% Loss
50%
Different cameras have diverse loss ratio !
doc.: IEEE 802.15-16-0018-00-007a
Submission
Examples of Camera Parameters<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 16
ImageResolution
(X x Y)
Frame Rate (fps)
Measured Read-out
Duration (𝝁s)
Time Gap (ms)
(Percentage of Frame Duration)
AppleiPhone 6 Plus
1920x1080 30 21.42 10.20 (30.60%)
Apple iPhone 5s
1920x1080 29.98 20.65 11.03 (33.10%)
AppleiPhone 4s
1920x1080 29.87 24.48 7.04 (21.03%)
HTC New One
1920x1080 29.94 19.08 12.79(38.30%)
Samsung Galaxy S4
1920x1080 29.93 25.53 5.84 (17.48%)
Point Grey Flea3
2048x1080 30 14.73 17.42 (52.27%)
doc.: IEEE 802.15-16-0018-00-007a
Submission
Low-pass Filtering due to Exposure
• Pixel intensity of y-th row of pixel(s):
– 𝑟(𝑦, 𝑡): received signal of the y-th row at time t– 𝑇*: start of exposure for this image frame– 𝑇+: exposure duration– 𝑇,: row read-out duration
• Long 𝑇+ results in significant low-pass filtering!
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 17
I[y] =
Z T0+(y�1)Tr+Te
T0+(y�1)Tr
r(y, t)dt
doc.: IEEE 802.15-16-0018-00-007a
Submission
PHY PROPOSAL -ROLLING-SHUTTERFREQUENCY SHIFT KEYING
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 18
doc.: IEEE 802.15-16-0018-00-007a
Submission
Recap: Rolling Shutter
Sampling
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 19
doc.: IEEE 802.15-16-0018-00-007a
Submission
Rolling Shutter –Frequency Shift Keying (RS-FSK)
Slide 20Time
Low frequencysquare wave
High frequencysquare wave
Wide Strip Narrow Strip
TX: Use signal frequency to represent data.RX: Use strip width to demodulate data.
Time
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>
doc.: IEEE 802.15-16-0018-00-007a
Submission
to represent bits.
Relation between Frequency and Strip Width
Slide 21
Time
estimates withselects a frequency from :
Transmitter Receiver
Transmitted SignalImage Y axis
……
……
.: read-out duration.
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>
doc.: IEEE 802.15-16-0018-00-007a
Submission
RS-FSK Advantages1. Average intensity stays the same for different symbols
over one symbol period (avoid flickers) 2. The waveform can be demodulated by receivers with
different (rolling shutter) sampling rates (i.e., read-out duration)
3. Demodulation is possible even when partial symbol is lost(cope with lossy channel)
4. Low-pass filtering does not destroy the signal as long as the exposure duration is not an integer multiples of the signal period
5. Dimming can be realized by changing the duty cycle of the signal
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 22
doc.: IEEE 802.15-16-0018-00-007a
Submission
YIN - Accurate Frequency Estimation for RXAdvantages:1. Time-domain autocorrelation
handles variable signal length2. Parabolic interpolation
improves accuracy when is not an integer3. Measures to handle slow varying DC,
i.e., non-uniform illumination surface
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 23
S4(front) S4(back) One(back) iP5s(front) iP5s(back)0%
25%50%75%
100%
Prob
abilit
y
loss mixed redundancy
S4(front) S4(back) One(back) iP5s(front) iP5s(back)0
0.250.5
0.751
PRR
raw splitter splitter+parity
S4(front) S4(back) One(back) iP5s(front) iP5s(back)0369
1215
Smartphone−Camera
Thro
ughp
ut(B
ps)
raw splitter splitter+parity
Experimental resultsshow that all test camerascan all decode the sametransmission!
doc.: IEEE 802.15-16-0018-00-007a
Submission
RS-FSK Advantages1. Average intensity stays the same for different symbols
over one symbol period (avoid flickers) 2. The waveform can be demodulated by receivers with
different (rolling shutter) sampling rates (i.e., read-out duration)
3. Demodulation is possible even when partial symbol is lost(cope with lossy channel)
4. Low-pass filtering does not destroy the signal as long as the exposure duration is not an integer multiples of the signal period
5. Dimming can be realized by changing the duty cycle of the signal
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 24
doc.: IEEE 802.15-16-0018-00-007a
Submission Slide 25
Time
……
……
Exposureof
1st frame
Exposureof
2nd frame
Idle (not receiving)
s1
s2
s3
Low SymbolRate
s1s2s3s4
……
s8s9
High SymbolRate
For better throughput:High-order modulation is required!
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>
High-Order Modulation - Partial Symbol Loss
doc.: IEEE 802.15-16-0018-00-007a
Submission
RS-FSK Advantages1. Average intensity stays the same for different symbols
over one symbol period (avoid flickers) 2. The waveform can be demodulated by receivers with
different (rolling shutter) sampling rates (i.e., read-out duration)
3. Demodulation is possible even when partial symbol is lost(cope with lossy channel)
4. Low-pass filtering does not destroy the signal as long as the exposure duration is not an integer multiples of the signal period
5. Dimming can be realized by changing the duty cycle of the signal
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 26
doc.: IEEE 802.15-16-0018-00-007a
Submission
Z Te
r(i+ 1, t)dt
=
Ion
+ Io↵
2
· k · 1f+ I
on
(Te mod
1
f)
RS-FSK When Considering Exposure
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 27
1/ft
𝐼./
𝐼.00
𝑟𝑜𝑤3
𝑇+
1/ft
𝐼./
𝐼.00
𝑟𝑜𝑤345
𝑇+
Average intensityover k signal period bright!
Z Te
r(i, t)dt
=
Ion
+ Io↵
2
· k · 1f+ I
on
(Te mod
1
f)
“variation”:
Average intensityover k signal period bright!“variation”:
Tr
Actual captured image
doc.: IEEE 802.15-16-0018-00-007a
Submission
Z Te
r(i+ 3, t)dt
=
Ion
+ Io↵
2
· k · 1f+ I
o↵
(Te mod
1
f)
RS-FSK When Considering Exposure
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 28
1/ft
𝐼./
𝐼.00
𝑟𝑜𝑤346
𝑇+
Average intensityover k signal period
mixed“variation”:
1/ft
𝐼./
𝐼.00
𝑇+
𝑟𝑜𝑤347
Average intensityover k signal period
dark!“variation”:
Z Te
r(i+ 2, t)dt
=
Ion
+ Io↵
2
· k · 1f
+ {↵ · Ion
+ (1� ↵) · Io↵
} (Te mod
1
f)
doc.: IEEE 802.15-16-0018-00-007a
Submission
The Case Where RS-FSK is not Observed
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 29
1/ft
𝐼./
𝐼.00
𝑟𝑜𝑤3 𝑇+Average intensityover k signal period
No “variation”! No stripe pattern!
When exposure is an integer multiple of the signal period
Te = k · 1f
𝑇+
The result stays the same for all rows!
𝑟𝑜𝑤8
Z Te
r(i, t)dt
=
Ion
+ Io↵
2
· k · 1f+ I(Te mod
1
f)
) Te mod
1
f= 0
0
doc.: IEEE 802.15-16-0018-00-007a
Submission
RS-FSK with Different Exposure Durations
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 30
1/ft
𝐼./
𝐼.00
𝑟𝑜𝑤3
1/ft
𝐼./
𝐼.00
𝑟𝑜𝑤3
Te
Te = k · 1f+ (Te mod
1
f)
Z Te
r(i, t)dt
=
Ion
+ Io↵
2
· k · 1f+ I(Te mod
1
f)
Te
k = 1
k = 3 Average intensityover k signal period
”Variation” over
As the exposure duration increases, “variation” becomes small compared to the average intensity
Te
Te mod
1
f
doc.: IEEE 802.15-16-0018-00-007a
Submission
RS-FSK with Exposure / Low-pass Filtering1. In most cases, with arbitrary exposure duration
setting, RS-FSK signal remains detectable in the received images.
– The observed signal frequency in the images is the same as the transmitted signal frequency
2. When the exposure duration is (approximately) an integer multiple of the signal period, the signal is NOT detectable.
– This can be regarded as channel fading for that signal frequency3. As the exposure duration increases, the difference
between bright and dark strips (ON/OFF states) becomes small.
– When this difference becomes less than the intensity resolution of the image, the signal is no longer visible.
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 31
doc.: IEEE 802.15-16-0018-00-007a
Submission
Validation with Experimental Results
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 32
0.00 3.66 7.33 10.99 14.65 18.31 21.98 25.64 29.300%
20%
40%
60%
80%
100%
Exposure Duration (ms)
Det
ectio
n Er
ror P
roba
bilit
y
FFTDIPYIN
Reasonable error probability up to tens of milliseconds
• Tested with 3 flavors of algorithms
• Each demodulation is regarded as erroneous if the estimated signal period is off by more than one read-out duration
doc.: IEEE 802.15-16-0018-00-007a
Submission
RS-FSK Advantages1. Average intensity stays the same for different symbols
over one symbol period (avoid flickers) 2. The waveform can be demodulated by receivers with
different (rolling shutter) sampling rates (i.e., read-out duration)
3. Demodulation is possible even when partial symbol is lost(cope with lossy channel)
4. Low-pass filtering does not destroy the signal as long as the exposure duration is not an integer multiples of the signal period
5. Dimming can be realized by changing the duty cycle of the signal
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 33
doc.: IEEE 802.15-16-0018-00-007a
Submission
Dimming with Duty Cycle
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 34
t
𝐼./
𝐼.00
𝐼./
𝐼.001/f t1/f
50% duty cycle 25% duty cycle
𝐼./
𝐼.00t1/f
75% duty cycle
• Change square wave duty cycle to adjust observed brightness
• Duty cycle is independent of and does not affect signal frequency f
doc.: IEEE 802.15-16-0018-00-007a
Submission
MAC PROPOSAL –FRAME FOR RS-FSK
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 35
doc.: IEEE 802.15-16-0018-00-007a
Submission
MAC Proposal
Main Objective: Compensate for significant signal loss and unsynchronized channelProblems & Proposed Solutions:1. Mis-aligned symbol boundary
2. Lost symbol detection
3. Recover from loss
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 36
Embedded Sequence Number
Symbol Splitter
XOR-based Parity Code
doc.: IEEE 802.15-16-0018-00-007a
Submission
1. Problem: Mixed-Symbol Frames<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 37
TX & RX have different frame rates
TX & RX have the same frame rates
f1
f2
Time
TX RX
s1
s2
s3
s4
s5
s6Time
TX RX
Incorrect frequencyestimation!
firstsymbol
secondsymbol
doc.: IEEE 802.15-16-0018-00-007a
Submission
1. Solution: Symbol Splitter (SS)
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 38
Time
TX RX
s1
ss
s2
s3
s4
f1
f2ss
ss
ssf3
firstsymbol
secondsymbol
SS
Symbol Splitter:micro symbol with known signal frequency
• Insert SS between consecutive data symbols• Linear search in the image to locate SS
(the location with strong power at SS frequency)• Separate image areas for frequency estimation
doc.: IEEE 802.15-16-0018-00-007a
Submission
2. Problem: Lost Symbol Detection
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 39
��
�� �� ��
����
�� �� � �� ���
����
���
���� ���� ����
When TX frame duration < RX frame duration Ts,tx T
f,rx
Entire symbol could be lost in the gap!1. How does the receiver know a symbol is lost?2. How to recover the lost data symbol?
doc.: IEEE 802.15-16-0018-00-007a
Submission
2. Solution: Embedded Sequence Number<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 40
Time
s1
s2
s3
s4
s5
s6
s7
s8
Assume no consecutive symbol losses*:
How to embed sequence numbers into symbols?
1 , 2 , 3 , 1 , 2 —> 3 is lost1 , 2 , 3 , 1 , 2 —> 2 is lost1 , 2 , 3 , 1 , 2 —> 1 is lost
XX
X
s9
seq# 1
seq# 2
seq# 3
seq# 1
seq# 2
seq# 3
seq# 1
seq# 2
seq# 3
{ }…
….
……
.
……
.] ] ]*This is true when the TX & RX frame rates are not different by more than 2 times
3 sequencenumbers needed
3 sets of frequenciesto represent3 sequence
numbers
doc.: IEEE 802.15-16-0018-00-007a
Submission
3. Solution: XOR-Based Parity Code
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 41
Time
s1
s2
s3
s4
P
s5
f1
s6
f2
gap
TX RX
• Add parity symbol P every N-1data symbols
• Recover when 1 out of N symbol is lost
Determine N from loss probability:
Light sizeRead-out durationExposure duration
Transmitted symbol durationReceiving frame duration
doc.: IEEE 802.15-16-0018-00-007a
Submission
Derivation of Symbol Loss Probability
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 42
Light size (number of rows)Read-out durationExposure durationTransmitted symbol durationReceiving frame duration
H :Tr :
Te :Ts,tx :Tf,rx :
Ploss
=max
✓Tgap
� Ts,tx
Tf,rx
, 0
◆
=max
✓Tf,rx � (H � 1)Tr � Te � T
s,tx
Tf,rx
, 0
◆
doc.: IEEE 802.15-16-0018-00-007a
Submission
Validation with Experimental Results<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 43
Compatible with a wide range of devices!
doc.: IEEE 802.15-16-0018-00-007a
Submission
Validation with Experimental Results<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 44
Support longer distances /
smaller light!
doc.: IEEE 802.15-16-0018-00-007a
Submission
Validation with Experimental Results<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 45
S4(front) S4(back) One(back) iP5s(front) iP5s(back)0%
25%50%75%
100%
Prob
abilit
y
loss mixed redundancy
S4(front) S4(back) One(back) iP5s(front) iP5s(back)0
0.250.5
0.751
PRR
raw splitter splitter+parity
S4(front) S4(back) One(back) iP5s(front) iP5s(back)0369
1215
Smartphone−Camera
Thro
ughp
ut(B
ps)
raw splitter splitter+parity
doc.: IEEE 802.15-16-0018-00-007a
Submission
Preamble – Read-Out Duration Estimation
Slide 46
Time
Transmittedsignal
iPhone 5c iPhone 5s iPhone 6 Plus
= 108 px
250 Hz
= 164 px = 158 px
Packet format
Data ……
Preamble Symbol (known frequency 𝑓: )
Learn 𝑇, from the preamble:
Different cameras have different 𝑇, !
<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>
Channel Estimation - 𝑇,
doc.: IEEE 802.15-16-0018-00-007a
Submission
Proposed Frame Format<January 2016>
<Hsin-Mu Tsai>, <National Taiwan University>Slide 47
�������������
���� �����
��������������
�����������
�����������
�����������
�����������
������������
�����������
Additional frame header fields for open systems:1. Selection of PHY data and SS frequencies
(Number of frequencies used: 2 byte; Actual used frequency in Hz, 2 byte per frequency)
2. Parity density N (1 byte)3. Checksum (optional, 2-4 bytes)