CFDP Performance Over WeatherDependent KaBand Channel

1
CFDP Performance Over Weather-Dependent Ka-Band
Channel
Sung U / Jay Gao Jet Propulsion Laboratory

• SpaceOps 2006
• 22 June 2006
• Rome, Italy

2
Motivations

• Weather model and protocol have not been
  considered jointly and previous CFDP analysis has
  been based on single state channel model (e.g., a
  fixed bit error rate)
• This work is to provide the framework to analyze
  the CFDP latency in the correlated weather
  channel based on the real noise temperature
  measurements
• Our approach is to use Gilbert-Elliot (G-E)
  channel to model the bursty nature of weather
  events but assign different BER to each weather
  state

3
CCSDS File Delivery Protocol (CFDP)

• File Delivery Protocol for the space environment
• No connection establishment
• Deferred NAK
• File PDUs are transmitted
• at least once before receiver sends NAK (hence
  deferred NAK)

Metadata PDU
File Data PDUs
Initial Tx
EOF PDU
ACK EOF
NAK PDUs
Receiving Entity
Sending Entity
File Data PDUs
Multiple re-tx spurts
NAK PDUs
File Data PDUs
Finished PDU
ACK Finished
4
Pictorial View
Example of file transmissions from the Mars
Reconnaissance Orbiter (MRO) to the Deep Space
Network (DSN) under different weather conditions
5
Correlated Weather Channel Model
, where
is the threshold and T is the noise temperature
at the antenna
6
Ka-band CFDP Modeling

• Ka-band channel has two levels of dynamics
• 1) In one pass, the channel is bursty, i.e,.
  good/bad weather states are correlated (we
  assume DSN has always a pass to MRO)
  
• 2) Within a weather state, each packet
  experiences random errors based on the current
  weather state

7
Temperature data from Deep Space Network (DSN)
at Madrid site to obtain the Markov parameter

• Data is re-sampled to every 40 minutes (round
  trip) time scale to characterize weather state on
  each ARQ spurt

Raw Noise Sample Plot (points 1.44 min apart)
Availability
, where
is the threshold.
8
Data Sampling
Temperature data point
time
Within the sampling window1)Calculate the
average AVG (mean) sampling 2)Choose the
maximum data point MAX sampling (most
conservative approach)3)Choose the minimum data
point MIN sampling (most optimistic approach)
9
Bad weather
Weather Distributions collected from Average
Sampling
PDF
CDF
Good weather

CDF of geometric random variable with
mean 1/lambda
10
Assumptions

• 1) It is assumed that each good and bad weather
  data follows geometric r.v. distribution for
  simplicity
  
• 2) Other weather factors are not considered,
  besides temperature at the DSN antenna
  
• 3) Margin and error control scheme is not
  considered
  
• 4) CFDP delay performance is focused on the
  protocol level interaction
  
• 5) In reality, time scale of weather change could
  occur on minute level that impacts frame error
  process within each spurt
  
• 6) No link outages are assumed

11
Analysis of CFDP Latency Cumulative
Distribution Function (CDF) of the number of
transmissions required (Ns) to complete the file
transmission
P(i,k) probability that i spurts are required
to successfully transmit one PDU given weather
wk
Above equations can be solved recursively
12
Evaluation Parameters
BERs and corresponding PDU error rates

4000 repetitive weather realizations, each with
1,000,000 points are generated (according to the
two-state Markov chain) with a random starting
point to evaluate the average CDF of CFDP Latency
using the mathematical analysis.
13
CDF of CFDP Latency
File Size 10MB BER at bad weather 10-3 and
10-4
File Size 1MB BER at bad weather 10-3 and
10-4

x-axis is the maximum number of CFDP
transmissions required to achieve certain
percentile file transfer completion (y-axis) the
average required transmissions are shown in the
next slide.
14
Average CFDP Latency
average number of transmissions required
File size 1MB
File size 10MB
Difference in the average number of
transmissions for 10MB and 1 MB file size
15
Interpreting the latency distribution

• The CDF of the number of transmissions required
  for completing 99 file transfer, shows the
  maximum number of transmissions (therefore,
  number of round-trip times), when desiring
  certain percentile of file transfer completion
• ?It provides the statistical upper bound on the
  delay that could be experienced by a file
  
• However, the average number of transmissions
  required is usually much smaller
  
• File size has a moderate effect on the delay
  performance but less significant than the effect
  of different BER values
  

16
CDF comparisons different sampling methods

• MIN
• sampling
• ?lower
• bound

• MAX
• sampling
• upper bound

17
Average Latency of CFDP with different sampling
methods
MAX sampling File Size 10MB
BER at bad weather 10-3 and 10-4
MIN sampling File Size 10MB
BER at bad weather 10-3 and 10-4
-MIN sampling provides a lower bound on latency
-MAX sampling provides an upper bound of the CFDP
delay performance on the given weather
availability
18
Latency performance with different parameters

• BER
• Changing good weather BER from 10-5 to 10-8
• - Up to 5 transmissions reductions for
  maximum number of transmissions required and
  average 2.5 transmissions reductions
  
• Changing bad weather BER from 10-3 to 10-4
• - Up to 5 transmissions reductions for maximum
  number of transmissions required and average
  2.5 transmissions reductions
  

BER at bad weather significantly affects the
maximum latency, since 10-3 BER ? PDU error
rate 0.9997 10-4 BER ? PDU error rate
0.5507

• File size 10MB ? 1MB
• Up to 2 transmissions reductions for the maximum
  number of transmissions required to achieve 99
  percentile file completion
  
• On average, approximately 1 transmission
  reduction
  
• Sampling errors
• Sampling errors are not significant for
  determining the average number of transmissions
  required for the 99 file completion
  

19
Conclusions and Future Work

• Analyzed the CFDP performance with the correlated
  weather model with different file size, and BERs
  on sampled DSN weather data
  
• ?Correlated channel model provides more
  realistic CFDP performance under Ka-Band channel
  than previous work
  
• Improvements are needed
• -Markov Assumption two-state model provides a
  coarse fitting to the burst length statistics of
  the Good and Bad weather states
  
• (1) Multi-state Markov model, or (2) Statistics
  based on the season or months
  
• -Further, evaluation with different PDU sizes
  and weather availabilities
  
• -Examining effect of starting on good weather
  state (i.e., the effect of foreknowledge of
  weather)
  
• -Better weather mitigation strategy using
  forecasting and high margin

20
References

• 1. CCSDS File Delivery Protocol, CCSDS 727.0-B-2,
  URL http//ccsds.org/
  
• 2. Gao, J.L., and Segui, J.S., Performance
  Evaluation of the CCSDS File Delivery Protocol
  Latency and Storage Requirement, IEEE Aerospace
  Conference. 2004.
• 3. W. Li,   C. L. Law,    V.K. Dubey,  and   J.T.
  Ong,   Ka-band land mobile satellite channel
  model incorporating weather effects, IEEE
  Communications Letters, 5. 2001.
• 4. V.Y.Y.Chu, P.Sweeney, J.Paffett,
  M.N.Sweeting, Characterizing error sequences of
  the low earth orbit satellite channel and
  optimization with hybrid-ARQ schemes, Proc. IEEE
  GLOBECOM, Sydney, pp.3405-3410, 1998.
• 5. Sun, J., Gao J., Shambayatti, S., and
  Modiano, E., Ka-Band Link Optimization with Rate
  Adaptation, IEEE Aerospace Conference,  March,
  2006.
• 6. Sklar, Digital Communications Fundamentals
  and Applications, 2nd ed., Prentice Hall, 2nd,
  2001.
  
• 7. Gilbert, E. N., Capacity of a Burst-Noise
  Channel, Bell Sys. Tech. J., 39,pp. 1253-1266,
  Monograph 3683, September 1960.
  
• 8. Elbert, B., Introduction to Satellite
  Communications, 2nd ed., Artech House, 1998.
  
• 9. Garcia, L., Introduction to probability for
  electrical and computer engineers, 2nd ed.,
  Addison Wesley, 2nd, 1993.
  
• 10. Hoeffding W., Probability inequalities for
  sums of bounded random variables, American
  Statistical Association Journal, pages 13-30,
  1960.

21

• Back up

22
Overall comparison 10MB file size
weather-dependent (correlated) channel vs.
single weather state channel
CDF of 10MB file transmission latency with
weather-dependent channel vs. Independent channel
model