September, 2003 Erlich, Infineon Technologies Slide 1 doc.: IEEE 802.15- 03/0350 Submiss ion Project: IEEE P802.15 Working Group for Wireless Personal Area Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs) Networks (WPANs) Submission Title: [Synchronous Simultaneously Operating Pico nets] Date Submitted: [September 2003] Source: [Yossi Erlich] Company [Infineon Technologies] Address [P.O.Box 8631, Poleg Industrial Area, Netanya 42504, Israel ] Voice:[+972-9-8924100], FAX: [+972-9-8658756], E-Mail: [[email protected]] Re: [] Abstract: [A synchronization mechanism is presented. This synchronization improves the performance under simultaneously operation pico-nets scenarios (SOP). The proposed solution has some drawbacks. However we think that it is worth considering the approach since the currently inspected methods do not sufficiently treat the SOP problem ] Purpose: [Suggest a solution to the SOP problem] 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
24
Embed
Doc.: IEEE 802.15-03/0350 Submission September, 2003 Erlich, Infineon TechnologiesSlide 1 Project: IEEE P802.15 Working Group for Wireless Personal Area.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
September, 2003
Erlich, Infineon TechnologiesSlide 1
doc.: IEEE 802.15-03/0350
Submission
Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs)Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANs)
Abstract: [A synchronization mechanism is presented. This synchronization improves the performance under simultaneously operation pico-nets scenarios (SOP). The proposed solution has some drawbacks. However we think that it is worth considering the approach since the currently inspected methods do not sufficiently treat the SOP problem]
Purpose: [Suggest a solution to the SOP problem]
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.
September, 2003
Erlich, Infineon TechnologiesSlide 2
doc.: IEEE 802.15-03/0350
Submission
Synchronous Simultaneously OperatingPico nets
A MB-OFDM Extension
September, 2003
Erlich, Infineon TechnologiesSlide 3
doc.: IEEE 802.15-03/0350
Submission
Do We Give Up Dense Utilization?
• SOP, Specifically the “near-far” scenario, is a major unsolved 802.15.3a issue
TV
TV
DV
D
DVD LapTop
dref=7m
dint=0.5m
dint=2m
PDA
Neighbor’s Apartment
Victim Pico netWithin a 3 SOP scenario
September, 2003
Erlich, Infineon TechnologiesSlide 4
doc.: IEEE 802.15-03/0350
Submission
Principles
• We suggest a MB-OFDM extension that enables 3 or 4 SOP, with very low dint/dref without performance degradation
• The OFDM symbols, transmitted by SOP, do not overlap in the T-F space
• Rough time synchronization among neighbor UWB• Low-level synchronization mechanism that requires -
no inter pico-nets management communication and is (almost) independent on the MAC layer
• No substantial additional complexity (cost)
September, 2003
Erlich, Infineon TechnologiesSlide 5
doc.: IEEE 802.15-03/0350
Submission
T-F OFDM Allocation3-SOP Alternative
• OFDM symbol duration T=312.5nSec• 242.4 nSec - OFDM info length• 60.6 nSec - Cyclic prefix (or zero pad) • 9.5 nSec - Fast-Hopping time
• Transmit N consecutive OFDM symbols within each band (N=4)• For 3-bands devices – Slow hopping - Half PRF• For 6-bands devices (advanced modems) – Fast hopping
NT=1.25uSec
Tg=NT=1250nSec
Tp =6NT=1.9uSec
1
1
1
2
2
2
1
1
1
2
2
2
1
1
1
2
2
2
1
1
1
2
2
2
1
1
1
2
2
2
1
1
1
2
2
2
1
1
1
2
2
2
1
1
1
2
2
2
• Same hopping order for all channels• Inter channel guard time Tg=N·T = 1.25uSec (for N=4)
– The actual guard time is Tg + 9.5nSec = 1.26uSec
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
time
frequency
3 ba
nds6
band
s
NT=1.25uSec Tp =6NT=1.9uSec
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
September, 2003
Erlich, Infineon TechnologiesSlide 6
doc.: IEEE 802.15-03/0350
Submission
T-F OFDM Allocation4-SOP Alternative
• Maintain the same guard time Tg=N·T = 1.25uSec for (N=4)• Double the transmission time within each band (transmit 2·N consecutive OFDM
symbols)• Same duty cycle per band• Same line code
time
frequency
3 b
an
ds6 b
an
ds
2NT=2.5uSec
Tg=NT=1.25uSec
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
1 1 2 2 33 4 4
Tp = 12NT = 15uSec
September, 2003
Erlich, Infineon TechnologiesSlide 7
doc.: IEEE 802.15-03/0350
Submission
Synchronization Signals
• Every Tsync≈50uSec (an integer number of the transmission period Tp) the communication halts (~4uSec+Tg) for a Sync signal
• The Sync signal is composed of 6 transmissions for the 3-bands case, and 12 for the 6-bands case
• Each symbol is a sequence (303nSec) with good autocorrelation properties
• The drawn example is plotted for 3-SOP time
frequency3
ba
nd
s6 b
an
ds
T = 312.5nSec
13T = 4 uSec
NT
1
1
1
2
2
2
3
3
3
1
1
1
2
2
2
3
3
3
2 3
3
2 3
3
1
1
1
2
2
3
1
1
1
2
2
3
1
1
1
2
2
2
3
3
3
1
1
1
2
2
2
3
3
3
Tg=NTTg
T
Tg
Tsync = 50uSec 50uSec
time
Sync Signal Sync SignalSync Signal
September, 2003
Erlich, Infineon TechnologiesSlide 8
doc.: IEEE 802.15-03/0350
Submission
Synchronization Mechanism
• Every device selects (per Tsync=50uSec) whether it transmits or receives the Sync signal (selection policy will be presented)
• If the earliest detected Sync was received before the receiver’s local timer, the device advances the local timer accordingly
• The timing correction is never done within burst reception/transmission
time
Earliest Sync signal
Power
ReceiverNominal Time
correction
Received Sync signals
Result:– Every device periodically
advances its local timer according to the fastest device in the neighborhood
– The fastest clock dictates synchronous transmission time scale
September, 2003
Erlich, Infineon TechnologiesSlide 9
doc.: IEEE 802.15-03/0350
Submission
PNCPNC
Channel #2S2=[1 0]
Channel #3S3=[1 1 0]
Sync Transmission Policy
• Inter-device successful synchronization happens when the faster transmits and the slower receives
• A PNC at channel ‘n’ transmits at Sync #m if Sn[modulus(m,n)]=1• A non-PNC device at channel ‘n’ transmits at Sync #m if Sn[modulus(m,n)]=0• Within a pico-net, the Sync indices are synchronized Whenever the PNC transmits the
non-PNC receive (and vice versa)
• Every pair of nodes at neighbor channels, synchronize at least every 7 Sync intervals
• Every pair of nodes within the same pico-net, synchronize at least every 7 Sync intervals
• This method was designed to limit synchronization error due to clock drift effects
September, 2003
Erlich, Infineon TechnologiesSlide 10
doc.: IEEE 802.15-03/0350
Submission
Synchronization Error
• The synchronization error is a result of 3 effects: 1. Drift (Skewed clock of all transmitters ‘seen’ by a single receiver)
• In steady state < 13*40PPM*Tsync = 26nSec• Under acceleration (to be explained) < 13*240PPM*Tsync = 156nSec
2. Propagation delay (Sync signals and data signals)• Worst case < 4*15m/c = 200nSec
3. Near-Far (Distant Sync signals masked by near transmitters)• Detection error < 3*303nSec = 909nSec
• Total effect (under acceleration) < 1.265uSec, which is by 5.5nSec longer than 4T+9.5nSec
• Devices position scenarios where this upper bound is reached are very rare
September, 2003
Erlich, Infineon TechnologiesSlide 11
doc.: IEEE 802.15-03/0350
Submission
Devices Clusters
• Define a ‘Cluster’ by the set of devices sharing the same synchronized time scale– All devices that ‘see’ each other – are
considered as a part of the same cluster– Particularly all devices within a pico net
belong to the same cluster• Initializing device
– Search Sync signals– No Sync signal No Cluster Be the first– Otherwise (Detected Sync signal)
• Join the cluster’s Sync signals (Denote this cluster “Primary Cluster”)
• Use Sync Tx. policy Snew = [1, 0, 0, 0, 0, 0, 0]; (transmit Sync every 7 Tsync intervals)
• Potentially, merge the cluster with one or more other clusters (“Secondary Clusters”)
September, 2003
Erlich, Infineon TechnologiesSlide 12
doc.: IEEE 802.15-03/0350
Submission
Clusters Merging (by a new device)
time
time
Tsync=50uSec
SuccessfulMerge
Accelerating Sync (0.31 Sec)
JoinedAccelerating
Tsync=50uSec
Primary Cluster
Secondary Cluster
• For the Sync signal transmission, the merging device uses a timer which is 200 PPM faster then its free running clock (advance by 10nSec every Sync signal)
• The slowest drift with respect to any other cluster would be 160PPM, which means that after at most Tmerge= 0.31Sec (50uSec/160PPM) the whole Tsync interval was swept
• Therefore it is guaranteed that all “Secondary Clusters” joined the Primary cluster
• The “Drift Effect” on the synchronization within the acceleration period was increased from 26nSec (13*240PPM*50uSec) to 156nSec (13*240PPM*50uSec)
September, 2003
Erlich, Infineon TechnologiesSlide 13
doc.: IEEE 802.15-03/0350
Submission
Periodic Clusters Merging Attempts (1)
• Consider a case where, due to a device movement, inter-cluster interference shows up
• Devices within a well-covered area (e.g. plugged repeaters) are protected from such merging requirements
• Periodic merging attempts are done by randomizing clock acceleration incidents
• When one cluster is ‘accelerating’ and the other cluster is not, for a complete Tmerge=0.31Sec period, then successful merging is guaranteed
• Within each cluster, there is a single device (dynamically selected) that randomizes acceleration incidents
– After M(n)·Tmerge time since the end of each acceleration, the accelerating device initiates another acceleration
– {M(n)} are randomized IID with probability ½ between {1.5, 5} (given as a distribution example)
• An accelerating device that senses Sync acceleration that was not initiated by itself, leaves duty
• All other devices (potentially only PNC), monitor Sync accelerations• A device that senses no acceleration for a certain time (e.g. 6·Tmerge) :
– Initiates an immediate acceleration– Becomes an accelerator (starts the random process)
• It is guaranteed that a set of devices that simultaneously become accelerators is decimated exponentially in time, and finally a single accelerator remains (each acceleration is expected to half the set population)
(*) Tmerge = 0.31Sec
September, 2003
Erlich, Infineon TechnologiesSlide 15
doc.: IEEE 802.15-03/0350
Submission
Periodic Clusters Merging Attempts (3)
• The figure shows the waiting-time statistics from clusters interaction until they successfully merge– At 80% of cases, within less
then 6∙Tmerge=1.9Sec– At 90% of cases, within less
then 10∙Tmerge=3.1Sec– At 95% of cases, within less
then 15.4∙Tmerge=4.8Sec– At 99% of cases, within less
then 26∙Tmerge=8.1Sec• The duration is short with
respect to device movements
September, 2003
Erlich, Infineon TechnologiesSlide 16
doc.: IEEE 802.15-03/0350
Submission
Work to be Done
• The presented work is incomplete• Full simulations should be carried out• We should design parameters such as:
– Sync signal length• Extend the Sync signal Increase sensitivity Merge before interruption
– Sync signal interval (Tsync=50uSec?)• Increase Sync rate Accelerate merging
– Per-band transmission duration (N=4?)– Merging procedure– And more…
• We should explore many issues such as:– Effects on other networks– Multiple merging devices– Big clusters– And more…
September, 2003
Erlich, Infineon TechnologiesSlide 17
doc.: IEEE 802.15-03/0350
Submission
Disadvantages
1. Long transmission time within each band: 1.25uSec for 3-SOP instead of 303uSec - 615nSec (currently proposed)
– Longer effect on narrowband systems– UWB: Requires longer interleaving for maintaining frequency diversity The N=4 duration is designed for rare worst case positioning. Some
shortening may be done by simulations analysis2. System complication
We think that the complication is small, judging against the solved SOP problem. It’s more appealing to take the asynchronous solution – Is this the right solution?
3. The 1/6 duty cycle– The achievable rate with QPSK is limited (the 480Mbps requires higher
constellation) Complexity penalty– 3dB higher effect on narrowband system
4. Co-channel Interference– Alien pico-net devices which re-use the same channel interfere more
severely
September, 2003
Erlich, Infineon TechnologiesSlide 18
doc.: IEEE 802.15-03/0350
Submission
Suggestion
• We encourage the WG members to join us exploring this approach
• We support the MB-OFDM and consider it as the best asynchronous solution
• We suggest that the WG considers synchronous solution AFTER the MB-OFDM proposal is accepted
• A PNC at channel ‘n’ transmits at Sync #m if Sn[modulus(m,n)]=1• A non-PNC device at channel ‘n’ transmits at Sync #m if Sn[modulus(m,n)]=0• Within a pico-net, the Sync indices are synchronized Whenever the PNC
transmits the non-PNC receive (and vice versa)• Consider two devices from different channels, n1 and n2.
Define D(n1,n2) as the maximal number of synchronization intervals* between two successful incidents {a device from channel #n1 transmits and a device from channel #n2 receives}.
• For the selected sequences:– ’Neighbor’ channels satisfy:
• D(0,1)≤7 D(1,2)≤4 D(2,3)≤6 D(3,0)≤6– Other channel pairs satisfy:
• D(0,2)≤6 D(1,3)≤9
(*) The synchronization interval is Tsync=~50uSec
September, 2003
Erlich, Infineon TechnologiesSlide 21
doc.: IEEE 802.15-03/0350
Submission
The Drift Effect (2)(Back Slide)
• For 40 PPM maximal relative drift and for Tsync=50uSec:– The maximal relative drift between two devices from different channels:
– For the 3-channels case the maximal relative drift between two devices from different channels:• D(n1,n2)≤7 Drift effect ≤ 14nSec
– For two devices within the same pico-net (channel #n),Define D(n,n) as the maximal number of elapsing Sync intervals from transmission of Sync signal by the faster device, until the next transmitted Sync massage is received (directly/indirectly) by the slower device
• D(n,n)≤7 Drift effect ≤ 14nSecNote: The worst case is when the two devices are non-PNC devices within channel #0
• For any device, consider two devices that are within detection range, one within the same pico net of the receiving device, and the other uses a neighbor channel (interfering).Claim: The drift effect between the two nodes is upper bounded by 13 synchronization intervals:
Drift effect ≤ 40PPM•13•Tsync =26nSecThis claim can be proven by simply showing that the synchronization time is upper bounded by D(n1,n1)+D(n1,n2)-1, where ‘n1’ is the index of the receiver’s channel, and n2 is the other channel.
• For robustness under Sync signal miss-detection - double drift effect could be assumed
September, 2003
Erlich, Infineon TechnologiesSlide 22
doc.: IEEE 802.15-03/0350
Submission
The Propagation Delay Effect(Back Slide)
• Consider the receiver at device “A”• Consider any interfering transmitter from other channels (other pico nets)
– Directly communicates Sync signals with “A”– “Int”’s local time is between –(dint-A/c) to +(din-At/c) with respect to “A” ’s local time– “Int”’s signal is received at “A” between 0 to +2d int-A/c with respect to “A”’s local time
• Consider any co-cannel transmitter (named “B”) from the same pico net– “B” is synchronized with the PNC, and the PNC is synchronized with “A”– “B”’s time is between –(dPNC-B+dA-PNC)/c to +(dPNC-B+dA-PNC)/c with respect to “A”– “B”’s signal is received at “A” between –(dPNC-B+dA-PNC-dA-B)/c to +(dPNC-B+dA-PNC-dA-B)
with respect to “A”’s local time• The total difference between the two signals as seen at device “A”:
– E = 2(dPNC-B+dA-PNC-dA-B+dint-A)/c– For dPNC-B, dA-PNC, dint-A ≤ d = 15m E ≤ 4d/c = 200nSec
AB
Int
PNCSync path
Communication path
Interference path
September, 2003
Erlich, Infineon TechnologiesSlide 23
doc.: IEEE 802.15-03/0350
Submission
The Near-Far Effect(Back Slide)
• Sync signals from near transmitters masks Sync signals from far transmitters
• The worst case is a delayed detection error of 303nSec
• A synchronization between two transmitters seen by a third device may involve up to 3 such detection errors
Near-Far effect < 3•303nSec=909nSec
September, 2003
Erlich, Infineon TechnologiesSlide 24
doc.: IEEE 802.15-03/0350
Submission
Clusters Merging (by a new device)(Back Slide)
Sync signals’ interference to nodes in secondary clusters
• Within the merging process some devices at the “Secondary Clusters” experience interferences
• Less then once per 7*Tsync=350uSec• For 3-bands devices -
– A merging Sync signal either affects 2 non-consecutive OFDM symbols, or a single OFDM symbol, or none. Such a Sync signal is transmitted at a 1/7 rate
– For N=4, the noise effect is at 2/3 rate once every 525uSec– For N=5, the noise effect is at 4/5 rate once every 437uSec