ISSL Project Review DGM72Y11

1
Analysis of DSRC Multi-Channel MAC Performance
Gavin Holland, HRL Laboratories, LLC Fan Bai,
Hariharan Krishnan, GM RD
August 26, 2008
2
Purpose and Objectives

• Purpose
• To evaluate and characterize the performance of
  DSRCs multi-channel MAC coordination protocol
  (IEEE 1609.4 standard) for applications that
  depend on periodic, broadcast-based messages,
  such as cooperative collision warning (CCW)
  safety applications
• Objectives
• Analyze the impact of several design tradeoffs
  and parameter settings that have not yet been
  studied in-depth or on a large-scale, or are
  currently unspecified, for a plausible network
  configuration and scenario
• CCH/SCH dwell-time ratio
• CCH/SCH channel switching frequency
• Channel switching guard interval length
• Local clock synchronization error
• Methodology
• Realistic, large-scale simulation based on the
  HRL/GM WAVES simulation toolset
• Plausible network densities and traffic loading
  based on recommended defaults

V2V Non-Safety
V2V Safety
Algorithms Protocols
DSRC
Expert Analysis
V2I Financial
Orig. V2I FT App
New V2I FT App
RSU
HRL WAVES Large-Scale Simulator
3
Conclusions and Recommendations

• Summary of Conclusions
• The CCW packet success probability appears to
  decrease linearly with a decrease in CCH/SCH
  dwell-time ratio for a reasonably dense,
  plausible network configuration
• However, queue occupancy suggests that a sharp
  decrease in performance could occur at ratios
  higher than expected due to problems with
  broadcast scheduling
• A channel switching frequency of 20Hz (50ms equal
  dwell) appears to provide reasonable performance
  for CCW, due, in part, to reduced queue
  occupancies
• The channel switching guard interval seems overly
  conservative in comparison to other radios, but
  the impact on performance does not seem to be as
  severe as presumed
• Local clock synchronization error must be severe
  to be significantly detrimental
• Overall, 1609.4 multi-channel switching appears
  to provide a feasible solution for simultaneous
  CCW and non-safety applications for the scenarios
  considered
• Suggested Changes to IEEE 1609.4 Standard
• Provide a MAC layer management API/interface to
  enable upper-layer protocols and applications to
  register and receive MAC state changing events
• Enables performance optimizing intelligent
  adaptation mechanisms based on explicit MAC
  monitoring and feedback
• For example, CCW could adapt its transmissions to
  start/stop packet generation in (loose?)
  synchronization with the MACs switching on/off
  the CCH
• 2. Tighten the requirements on the
  channel-switching guard interval (i.e. 4ms) to
  improve system efficiency -- a detailed study
  should be conducted to see if system efficiency
  could be improved without significantly
  compromising synchronization

4
Presentation Roadmap

• Executive Summary
• Preliminaries
• Analysis
• Conclusions

v
5
DSRC Multi-Channel MACCoordination Protocol
OBU 1
OBU 2
OBU 1
OBU 2
RSU
RSU
Fig 1. Channel Diversity for Increased Capacity
Fig 2. DSRC Channel Allocation
Fig 3. DSRC Multi-Channel Coordination

• Purpose of Multi-Channel Switching
• Support different classes of applications
  (safety/non-safety) with different latency,
  range, and bandwidth requirements
• Support different priority levels across and
  within applications, take advantage of channel
  diversity (Figs. 1 2)
• Minimize contention between high-priority,
  safety-oriented control traffic (on Control
  Channel (CCH)) and non-safety application traffic
  (on Service Channels (SCH))
• Operation
• Nodes are tightly synchronized, where a CCH
  interval MUST be scheduled at the beginning of
  every second (Fig. 3)
• Time is divided into control channel (CCH)
  intervals and service channel (SCH) intervals (a
  split-phase approach)
• OBU MUST periodically switch to CCH for safety,
  high-priority, and WAVE Service Announcements
  (WSA) packets
• OBU MUST only use WSM packet format to transmit
  data packets on CCH (as a hedge against CCH
  congestion)
• OBU MAY select service from WSAs and switch to
  specified SCH provided by OBU/RSU
• Guard Intervals are scheduled on entry to every
  interval to allow for differences in
  synchronization errors and switching delays

Efficient multi-channel coordination is important
for support of safety/non-safety application
coexistence
6
Assumptions Regarding Unspecified (or Loosely
Defined) Protocols
SCH Queue
Fig 1. Packet Scheduling
Fig 2. State Management

• Packet Scheduling
• Standard requires cancellation of packet
  transmissions that do not complete at or before
  the end of the guard interval
• Annex B suggests alternative approach, whereby
  packet is only transmitted if it is predicted to
  complete before end of interval
• For our analysis, we use a similar approach,
  letting 802.11p decide if a packet can be
  transmitted in time
• State Management
• Standard does not define how MAC state (e.g.
  Backoff, NAV) should be handled across channel
  switches
• For our analysis, we implement a state freezing
  protocol that stores the context of selected
  state parameters prior to switching away from the
  channel, and restores the context upon channel
  re-entry
• Note, this assumes that the state of the channel
  is approximately the same across channel
  switches, which most likely is not the case for
  channels with small coherence time or very
  dynamic mobility, so care is taken to ensure that
  the context is relevant prior to restoration to
  limit the impact of stale state

Plausible packet scheduling and state management
protocols were used in our analysis
7
Application Model and Performance Metrics

• Cooperative Collision Warning (CCW)
• OBU periodically broadcasts vehicle status to
  neighbors (e.g. position, velocity, sensor
  status) on CCH
• Neighbors use this information to infer collision
  likelihood and react accordingly
• Assumptions
• All vehicles have continuously backlogged SCH
  traffic and continuous access to service provider
• Packet payload is 100 bytes, start of
  transmission staggered randomly between vehicles
• Broadcast interval default 100ms (but varied in
  some cases to observe impact of changes in load)
• Performance Metrics (Network Metrics)
• Packet Success Probability (PSP) Prob(packet
  transmitted received)
• Per-Packet Latency Time received Time
  transmitted
• Performance Metrics (Application Metrics)
• Inter-Reception Time (IRT) Time received Time
  last received
• Application-Level Reliability (ALR) Prob(IRT lt
  tolerable time window T)

Application model, assumptions, and metrics
selected to represent and characterize a
challenging scenario
8
Vehicular, Radio, and Channel Models

• Vehicular Models
• Large Scale
• 1920 vehicles on 8 lane 1.6km stretch of freeway,
  360 neighbors within 300m range
• Eastbound lanes stationary (5m apart), Westbound
  lanes moving at 25 mph (10m apart)
• Small Scale
• 50 vehicles placed randomly in 2 x 0.5 km area,
  with no mobility
• Used for illustrating phenomena and
  characteristics that dont require large-scale
  scenario
• Radio Model
• Configured to match settings of DSRC-compatible
  radios
• Receiver sensitivity -95dBm
• Noise factor 5.0
• Antenna was 0dB omni at 1.2m high with 1.2dB of
  cable loss
• Data rate 6Mbps
• Transmission power 9dBm (300m effective range)
• Channel Model
• Fading model empirically derived from field test
  data
• RSSI samples binned by distance and fit to
  Nakagami ltm,Ogt

Fig 1. Large Scale Vehicular Model
Fig 2. Measured and Simulated Packet Delivery
Ratio
Vehicular, radio, and channel models selected to
approximate a plausible deployment scenario
9
Presentation Roadmap

• Executive Summary
• Preliminaries
• Analysis
• Conclusions

v
10
Impact of Varying CCH Dwell Time Percentage
Sync Interval
Hard Constraints
.....
ICCH ISCH ISync
CCH Interval
SCH Interval
Guard Interval
nxISync 1 sec, n n??
Start of UTC second
Start of next UTC second
CCH Dwell Time Percentage (ICCH / Isync)x100
Impact on Applications?
25 CCH Dwell
Performance Tradeoffs?
Some Valid Configurations
75 CCH Dwell
Reasonable Values?

• Background
• The standard allows the CCH Interval (ICCH) and
  the SCH Interval (ISCH) to be different, just as
  long as the length of the Sync Interval (ISync
  ICCHISCH) is a divisor of 1 second
• Allowance has even been given for them to be
  dynamically adaptable in the future
• Issue
• What impact does varying the ratio of CCH/SCH
  have on CCW performance?
• Approach
• Set ISync 1sec and measure impact of different
  ICCH and ISCH within specified constraints
• CCW broadcast interval 100ms, and queues are
  large enough to hold 375 packets
• Simulation scenario is congested freeway (1920
  vehicles), confidence intervals are 95

11
CCW Packet Success Probability
Fig 1. Average Packet Success Probability
Fig 2. Peak and Average Queue Occupancy

• Observations
• Baseline result for 100 CCH dwell time (no
  switching) is shown for comparison
• Intuitively, continuous SCH backlog results in
  proportional change in available CCH capacity
  with change in CCH dwell time percentage (Fig. 1)
• Fig. 2 shows sharp increase in queue occupancy
  between 60-40, suggesting saturation of network
• Interestingly, queue occupancy does not exceed
  maximum so decrease in PSP is not due to queue
  overflow

What causes packet success probability to
decrease with decreasing CCH dwell time
percentage?
12
MAC Collision Avoidance and Error Recovery
Fig 2. CW Growth During Error-Recovery
Fig 1. Illustration of Basic Access Method for
Scheduling Packet Transmissions

• Collision Avoidance
• Before transmitting a frame, the node senses the
  busy status of the medium
• If the medium is idle for at least a DIFS, the
  node may transmit its frame immediately
• However, if the medium is busy, then the node
  invokes the backoff procedure to avoid a
  collision
• Select a random integer BO over 0,CW, where CW
  is an integer in the range CWmin CW CWmax
  (initially, CW CWmin)
• Compute a backoff time BT BO x SlotTime, and
  set a timer that decrements BT for every idle
  slot following a DIFS/EIFS
• When BT 0, transmit the frame immediately
• Error Recovery
• If it is detected that transmission of the frame
  failed, then the node invokes error recovery
• Retransmit the frame after another backoff, but
  draw BT from the interval 0, CW min(2 x CW
  1, CWmax)
• Repeat with increasing CW until the transmission
  succeeds, or the maximum number attempts is
  reached
• There is no error recovery for broadcast or
  multicast frames because there is no mechanism to
  detect failure

The lack of error recovery for MAC-level
broadcast and multicast packets can cause
significant problems
13
Issues with MAC-level Broadcast and Multicast
EIFS
Frame
Defer
Frame
DIFS
BTD BTE
Collision!
Fig 1. Timeline Showing how Collision Occurs
with the Current MAC Backoff Procedure

• Problem
• Since there is no error recovery for broadcast or
  multicast frames, then CW does not increase (CW
  CWmin)
• Thus, the probability of collision is high when
  more than CWmin 1 backlogged nodes attempt
  transmission
• Since CWmin is small (7 or 15, depending on the
  version of the standard), then scalability is a
  serious problem
• To illustrate, Fig. 1 shows a timeline for 5
  nodes, where B-E queue broadcast frames when A is
  transmitting
• Here, nodes C picks the smallest BT, followed by
  D and E, who select identical BTs, followed by B
• After A is done, C transmits next, then, since D
  and E have the same BTs, they transmit
  simultaneously
• Since D and E have no way of knowing that their
  transmissions collided, then their CW is
  unchanged
• For A-C, even though they detected the collision
  and defer by an EIFS, their CW also does not
  increase

Without error recovery, senders of broadcast and
multicast frames cannot react to
congestion-related losses
14
Channel Collisions due to Channel-Switching-Induce
d Broadcast Synchronization (CSIBS)
SCHn Interval
CCH Interval
Packets
Packets
A
A
Time
Time
IP
IP
WSMP
WSMP
LLC
LLC
v red points show application-level packet
transmissions v green points show when packet
was transmitted by MAC v yellow points how when
packets were dropped by MAC v blue points show
RSSI of received packets
MAC
MAC
Channel Switch
...
...
SCH1
SCHn
CCH
SCH1
SCHn
CCH
PHY
PHY
Guard Interval
CCH
SCHn
CCH Interval
Fig 2. Illustration of source of collisions
Fig 1. Profile showing collisions due to
channel-switching-induced broadcast
synchronization

• Problem
• In DSRC, many nodes may have broadcast frames
  queued for one channel while tuned to another
• Thus, upon entry to the channel all of the nodes
  attempt to transmit using BT computed using CW
  CWmin
• CWmin 15 slots in 802.11 R2003, Section 15.3.2,
  and 7 slots in earlier revisions
• Thus, there are large packet losses due to
  collisions at the beginning of the interval until
  queues are emptied
• This is illustrated in Figs 1 2, which
  illustrate the problem for SCH/CCH switching and
  CCW applications
• During SCH Interval, packets continue to be
  queued for the CCH by the CCW application
• After switch to CCH, queued packets are
  transmitted in a burst with high probability of
  collision

Collisions due to channel-switching-induced
broadcast synchronization are the main cause of
PSP decrease
15
Measured CCW Packet Inter-Reception Time and
Application-Level Reliability
Fig 1. Peak and Average Inter-Reception Time
Fig 2. Measured Application-Level Reliability

• Observations
• Peak and average IRT (Fig. 1) also increase with
  decrease in dwell , as expected, with CSIBS
  causing sharp increase for CCH dwell lt 40
• Note even for the baseline case of CCH dwell
  100, IRT gt 100ms (the CCW broadcast interval)
  because the network load is sufficiently high
  that there are congestion losses even without
  channel switching
• Fig. 2 shows the measured ALR for varying values
  of the application tolerance window T, computed
  using the average measured PSP and t 100ms
• Note over all measured dwell times a tolerance
  window of T 2sec is sufficient for an ALR gt
  90, and T 1sec is sufficient for an ALR of gt
  90 for dwell times 40

IRT and ALR values are reasonable for CCH dwell gt
40 for the simulated CCW scenario
16
Impact of Varying Channel Switching Frequency
Varying Channel Switching Frequency vary ISync
ISync 50ms (20 Hz)
Some Valid Configurations
ISync 100ms (10 Hz)

• Background
• The standard allows the Sync Interval (ISync) to
  be varied (within aforementioned constraints)
• Issue
• What impact does varying the channel switching
  frequency have on CCW performance?
• Approach
• Set ICCH ISCH and measure impact of different
  ISync within specified constraints
• Two CCW broadcast intervals (t 100ms, and t
  200ms), which indicates the different levels of
  network load, are used to observe impact of load
• Simulation scenario is large-scale freeway (1920
  vehicles), confidence intervals are 95

17
CCW Packet Success Probability
Fig 1. Average Packet Success Probability

• Observations
• Baseline result of 0Hz (no switching) is provided
  as a basis for comparison, as before
• For both broadcast intervals, there is 40
  decrease in PSP between 0Hz and 5Hz, which is
  caused by CSIBS
• For t 100ms, the PSP continues downward with
  increasing frequency, but only slightly so up to
  50Hz
• For t 200ms, the PSP first trends upwards with
  increasing frequency before it falls, owing to
  the fact that an increase in switching frequency
  decreases CSIBS by decreasing the chance that,
  within the same spatial region, there are
  sufficient vehicles with backlogged packets to
  cause collisions upon switching to the CCH

Increasing channel switching frequency up to 50Hz
does not appear to have a significant impact on
PSP
18
CCW Packet Inter-Reception Time
Fig 1. Peak and Average Inter-Reception Time for
t 100ms
Fig 2. Peak and Average Inter-Reception Time for
t 200ms

• Observations
• For both broadcast intervals, peak and average
  IRT is largely unaffected as the channel
  switching frequency is varied over the range of
  5Hz to 50Hz.
• For Fig. 1, this is seemingly counterintuitive
  since an increase in the switching frequency
  results in higher switching overhead and higher
  end-to-end packet latency. In fact, this is the
  case, but the IRT metric only considers the time
  between successive packets, which are relatively
  unchanged due to the high percentage of queued
  packets being transmitted close together at the
  beginning of a CCH.
• The sharp trend upwards at 100Hz for both
  broadcast intervals is due to the inability of
  the MAC to schedule packets before the end of the
  interval, resulting in higher switching overhead
  due to channel idle time.

Increasing channel switching frequency up to 50Hz
also does not appear to have a significant impact
on IRT
19
Impact of Guard Interval Overhead

• Background
• An unavoidable overhead of channel switching in
  DSRC is the guard interval
• The standard defines the guard interval to be the
  sum of the maximum synchronization tolerance and
  switching delay (4ms by default)
• The synchronization tolerance is a function of
  the stability of the local clock, the
  availability of an external synchronization
  source, and the use of the TSF timestamp field
  protocol
• The switching delay is the maximum time it takes
  a radio to switch between channels
• Issue
• What impact does the size of the guard interval
  have on channel switching performance?
• Approach
• Set ICCH ISCH and vary ISync and guard interval
  to measure impact of guard interval overhead
• CCW broadcast interval t 50ms, and CCH dwell
  time was varied from 5ms to 90ms
• Simulation scenario is small-scale (50 vehicles),
  confidence intervals are 95

20
Impact of Guard Interval Overhead on CCW
Fig 1. Maximum available CCH capacity for two
guard intervals
Fig 2. Relative PSP for 4ms interval vs. no
guard interval

• Observations
• Fig. 1 shows the maximum available CCH capacity
  versus dwell time for a 4ms and 224us guard
  interval, where the latter is the guard interval
  for 802.11 FHSS radios (which we consider here to
  be a practical lower-bound).
• Also shown in Fig. 1 is an estimate of the dwell
  time Ld at which the simulated load will saturate
  the network, which is based on the simple
  formula Ld Lg / (1 l / c), where Lg is the
  guard interval, l is the simulated load, and c is
  the estimated CCH capacity (both in packets per
  ms)
• Note that, for our simulated scenario, Ld
  11.3ms, for a guard interval of 4ms, versus Ld
  0.633ms for 224us
• Fig. 2. shows the impact on the PSP, plotted as
  the ratio of with and without a 4ms guard
  interval. Notice that the relative PSP drops
  significantly near to the predicted saturation
  point.

While the default guard interval is conservative,
its impact on the PSP is minimal for reasonable
dwell times
21
Impact of Synchronization Drift

• Background
• The standard allows for the coexistence of
  networks with and without external time sources
• Although availability of an external time source
  (e.g. GPS) is assumed, it is not required
• Issue
• How bad does synchronization drift have to be
  before performance is severely degraded?
• Approach
• Set ICCH ISCH and randomly drift local clocks
  in t us increments every ms to measure impact
  of synchronization error
• CCW broadcast interval t 100ms, and ISync 100ms
  (50ms CCH dwell time)
• Simulation scenario is small-scale (50 vehicles),
  confidence intervals are 95

22
Impact of Synchronization Drift on CCW
Fig 2. Peak and Average Inter-Reception Time
Fig 1. Average Packet Success Probability

• Observations
• Here, we seek to only estimate how badly
  performance is impacted with varying degrees of
  synchronization drift.
• Fig. 1 shows the PSP as a function of average
  clock drift, in us of drift per ms. Intuitively,
  PSP decreases with increasing drift as the
  probability that the sender and receiver are on
  the CCH at the same time decreases.
• Interestingly, the trend appears to slow with
  larger intervals, which is due to the fact that
  as nodes become unsynchronized their CCH and SCH
  intervals overlap, reducing the impact of
  collisions due to CSIBS
• However, the peak IRT does not share the same
  trend, and is worse than in the synchronized case
  shown earlier

Synchronization error is a cause of concern, but
it takes large drift to impact performance
significantly
23
Presentation Roadmap

• Executive Summary
• Preliminaries
• Analysis
• Conclusions

v
24
Recap of Conclusions and Recommendations

• Summary of Conclusions
• The CCW packet success probability appears to
  decrease linearly with a decrease in CCH/SCH
  dwell-time ratio for a reasonably dense,
  plausible network configuration
• However, queue occupancy suggests that a sharp
  decrease in performance could occur at ratios
  higher than expected due to problems with
  broadcast scheduling
• A channel switching frequency of 20Hz (50ms equal
  dwell) appears to provide reasonable performance
  for CCW, due, in part, to reduced queue
  occupancies
• The channel switching guard interval seems overly
  conservative in comparison to other radios, but
  the impact on performance does not seem to be as
  severe as presumed
• Local clock synchronization error must be severe
  to be significantly detrimental
• Overall, 1609.4 multi-channel switching appears
  to provide a feasible solution for simultaneous
  CCW and non-safety applications for the scenarios
  considered
• Suggested Changes to IEEE 1609.4 Standard
• Provide a MAC layer management API/interface to
  enable upper-layer protocols and applications to
  register and receive MAC state changing events
• Enables performance optimizing intelligent
  adaptation mechanisms based on explicit MAC
  monitoring and feedback
• For example, CCW could adapt its transmissions to
  start/stop packet generation in (loose?)
  synchronization with the MACs switching on/off
  the CCH
• 2. Tighten the requirements on the
  channel-switching guard interval (i.e. 4ms) to
  improve system efficiency -- a detailed study
  should be conducted to see if system efficiency
  could be improved without significantly
  compromising synchronization

25

• Questions?
• Thanks!

26
BACKUPS
27
Wireless Access in Vehicular Environments
Simulation (WAVES) Toolset
Vehicular Application Scenario Generation
WAVE Simulator
Output Processing
DSRC/Vehicular Models
Vehicular Mobility Scenario Generation
Engine
QualNet
Scenario Repository
Flowing Hwy
CORSIM Fwy
CORSIM Urban Inters.
28
CCW Packet Inter-Reception Time
CCH Dwell Time 100
CCH Dwell Time 20
Fig 1. Peak and Average Inter-Reception Time

• Observations
• Peak and average IRT (Fig. 1) also increase with
  decrease in dwell , as expected, with CSIBS
  causing sharp increase for CCH dwell lt 40
• Note even for baseline case of CCH dwell 100,
  IRT gt 100ms (the CCW broadcast interval) because
  the network load is sufficiently high that there
  are congestion losses even without channel
  switching
• This is illustrated in Fig. 2, which shows the
  samples sorted by IRT for 100 and 20 dwell
  times, and the characteristic stair-step
  pattern of the IRT with steps falling near
  multiples of the broadcast interval

Fig 2. Samples Sorted According to IRT
IRT is reasonable for CCH dwell gt 40 for the
simulated CCW scenario
29
CCW Application-Level Reliability
Fig 2. Measured Application-Level Reliability
Fig 1. Predicted Application-Level Reliability

• Observations
• Fig. 1 shows the predicted ALR for varying values
  of the application tolerance window T, computed
  using the average measured PSP and t 100ms
• Fig. 2 shows the ALR computed directly from the
  simulation measurements
• Overall, the predicted and measured ALR are in
  good agreement for T gt 500ms, since the IRT is
  seldom larger than those values, and dependence
  on the dwell time is clearly evident
• Also, note that over all measured dwell times a
  tolerance window of T 2sec is sufficient for an
  ALR gt 90, and T 1sec is sufficient for an ALR
  of gt 90 for dwell times 40

Reasonable ALR values can be achieved for CCH
dwell times 40 for the simulated CCW scenario
30
CCW Packet Success Probability
(b) t 200ms
(a) t 100ms
Fig 2. Peak and Average Queue Occupancies
Fig 1. Average Packet Success Probability

• Observations
• Baseline result of 0Hz (no switching) is provided
  as a basis for comparison, as before
• For both broadcast intervals, there is 40
  decrease in PSP between 0Hz and 5Hz, which is
  caused by CSIBS
• For t 100ms, the PSP continues downward with
  increasing frequency, but only slightly so up to
  50Hz
• For t 200ms, the PSP first trends upwards with
  increasing frequency before it falls, owing to
  the fact that an increase in switching frequency
  decreases CSIBS by decreasing the chance that,
  within the same spatial region, there are
  sufficient vehicles with backlogged packets to
  cause collisions upon switching to the CCH
• The difference in backlogged packets is shown in
  the peak and average queue occupancies of Fig. 2.

Increasing channel switching frequency up to 50Hz
does not appear to have a significant impact on
PSP
31
CCW Per-Packet Latency
Fig 1. Average Per-Packet Latency for t 100ms
Fig 2. Average Per-Packet Latency for t 200ms

• Observations
• For the heavier load (t 100ms), the per-packet
  latency increases steadily with channel switching
  frequency
• For the lighter load (t 200ms), the per-packet
  latency actually decreases as the channel
  switching frequency is varied over the range of
  5Hz to 20Hz, and then is effectively unchanged
  until the sharp increase at 100Hz
• This decrease mirrors the drop in the queue
  occupancies, resulting in a drop in the impact of
  CSIBS
• The sharp trend upwards at 100Hz for both
  broadcast intervals is due to the inability of
  the MAC to schedule packets before the end of the
  interval, resulting in higher switching overhead
  due to channel idle time

Increasing the channel switching frequency
decreases the impact of CSIBS for the lightly
loaded network
32
Demonstration of Decreased CSIBS with Increase in
Channel Switching Frequency
v red points show application-level packet
transmissions v green points show when packet
was transmitted by MAC v yellow points how when
packets were dropped by MAC v blue points show
RSSI of received packets
Guard Interval
CCH Interval
Fig 1. Profile showing reduced number of
collisions due to CSIBS when channel switching
frequency is increased (100Hz)

• Observations
• Increasing channel switching frequency results in
  decreased CSIBS because of lower queue
  occupancies
• During SCH, fewer packets are queued for CCH
  resulting in smaller burst of packets and,
  subsequently, fewer collisions

Impact of CSIBS can decrease with increase in
channel switching frequency due to lower queue
occupancy
33
Application Reliability as a Function of Distance
for Varying Tolerance Time Windows
Impact on application reliability for all loads
is significant for all but the largest tolerance
window