On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.
-
Upload
blaise-knight -
Category
Documents
-
view
229 -
download
3
Transcript of On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.
![Page 1: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/1.jpg)
On coding for networks with errors
Tracey HoCaltech
BIRS, August 2011
![Page 2: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/2.jpg)
Network error correction problem
2
s2
z unknown adversarial links
t1
t2
• Problem of reliable communication under arbitrary/adversarial errors on z links (or packets)
• The set of erroneous links is fixed but unknown to the network user → cannot do error correction on a link by link basis
• Combinatorial error model, algebraic coding…
network
s1
![Page 3: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/3.jpg)
Coherent & non-coherent models• Coherent: network topology known, centralized code
design• Non-coherent (multicast): network topology unknown,
use distributed random linear network coding (RLNC)• each packet comprises a vector of symbols from
a finite field Fq
• each node sends linear combinations of its received packets with coefficients chosen uniformly at random from Fq
• record network transformation in packet headers, or encode information in subspaces
(a1,…, an) (b1,…,
bn)
(γa1 +λb1,…,γan+λbn)
![Page 4: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/4.jpg)
Uniform & non-uniform error models
• Uniform errors− Coherent case: equal capacity network links, errors on
any z links− Non-coherent case: equal capacity packets, errors on
any z packets• Non-uniform errors
− Unequal capacity network links− Restrictions on which links can be erroneous− Erroneous nodes
![Page 5: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/5.jpg)
Outline•Single-source multicast, uniform errors (background)•Multiple-source multicast, uniform errors
T. Dikaliotis, T. Ho, S. Jaggi, S. Vyetrenko, H. Yao, M. Effros and E. Erez, "Multiple-access Network Information-flow and Correction Codes," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.
•Non-uniform errors: unequal link capacitiesS. Kim, T. Ho, M. Effros and S. Avestimehr, "Network error correction with unequal link capacities," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.T. Ho, S. Kim, Y. Yang, M. Effros and A. S. Avestimehr, "On network error correction with limited feedback capacity," ITA 2011.
• Non-multicast nested networks, uniform & non-uniform errorsO. Tekin, S. Vyetrenko, T. Ho and H. Yao, “Erasure correction for nested receivers,” Allerton 2011.
![Page 6: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/6.jpg)
Outline•Single-source multicast, uniform errors (background)•Multiple-source multicast, uniform errors
T. Dikaliotis, T. Ho, S. Jaggi, S. Vyetrenko, H. Yao, M. Effros and E. Erez, "Multiple-access Network Information-flow and Correction Codes," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.
•Non-uniform errors: unequal link capacitiesS. Kim, T. Ho, M. Effros and S. Avestimehr, "Network error correction with unequal link capacities," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.T. Ho, S. Kim, Y. Yang, M. Effros and A. S. Avestimehr, "On network error correction with limited feedback capacity," ITA 2011.
• Non-multicast nested networks, uniform & non-uniform errorsO. Tekin, S. Vyetrenko, T. Ho and H. Yao, “Erasure correction for nested receivers,” Allerton 2011.
![Page 7: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/7.jpg)
Background –single-source multicast,uniform errors
• Min cut = C• Error correction capacity = C – 2z, achievable with linear
network codes (Cai & Yeung 03)• Capacity-achieving (asymptotically in packet length) non-
coherent codes with polynomial-time decoding:− Probabilistic construction (Jaggi et al. 07)− Lifted Gabidulin codes (Koetter & Kschischang 08, Silva,
Kschischang & Koetter 08)
7
![Page 8: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/8.jpg)
Gabidulin codes
M = qF
n
rM = nqF
1
r
![Page 9: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/9.jpg)
M = rnqF
1
* nqF
1
r+tqF
n
GM =
GMChannel GM +E
=
Gabidulin codes
• Corrects any error E of rank < rank distance/2• Gabidulin codes are max rank distance codes (d = t+1, n≥r+t)
![Page 10: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/10.jpg)
Lifted Gabidulin codes
• Transmit [I x], where x is a codeword of a Gabidulin code• Achieves error correction capacity = C – 2z for
noncoherent single-source multicast• Can be decoded with low complexity by generalized rank
metric decoding (Silva , Kschischang & Koetter 08)
![Page 11: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/11.jpg)
Outline•Single-source multicast, uniform errors (background)•Multiple-source multicast, uniform errors
T. Dikaliotis, T. Ho, S. Jaggi, S. Vyetrenko, H. Yao, M. Effros and E. Erez, "Multiple-access Network Information-flow and Correction Codes," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.
•Non-uniform errors: unequal link capacitiesS. Kim, T. Ho, M. Effros and S. Avestimehr, "Network error correction with unequal link capacities," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.T. Ho, S. Kim, Y. Yang, M. Effros and A. S. Avestimehr, "On network error correction with limited feedback capacity," ITA 2011.
• Non-multicast nested networks, uniform & non-uniform errorsO. Tekin, S. Vyetrenko, T. Ho and H. Yao, “Erasure correction for nested receivers,” Allerton 2011.
![Page 12: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/12.jpg)
Multiple-source multicast,uniform errors
Challenges:• Sources do not know each other’s information
− Independent coding at sourcesCoding across different sources needed within
network in order for them to share network capacityMultiple access property
Straightforward extensions of single-source codes are suboptimal
Related work: code construction in (Siavoshani, Fragouli & Diggavi
08) achieves capacity for C1+C2=C
Sink
Source 2Source 1
![Page 13: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/13.jpg)
Multiple-source multicast,uniform errors
Challenges:• Sources do not know each other’s information
− Independent coding at sources• Coding across different sources needed within
network in order for them to share network capacity− Multiple access property
Straightforward extensions of single-source codes are suboptimalE.g. code construction in (Siavoshani, Fragouli &
Diggavi 08) achieves capacity only for C1+C2=C
Sink
Source 2Source 1
![Page 14: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/14.jpg)
Multiple-source multicast,uniform errors
Challenges:• Sources do not know each other’s information
− Independent coding at sources• Coding across different sources needed within
network in order for them to share network capacity− Multiple access property
• Straightforward extensions of single-source codes are suboptimal
Related work: code construction in (Siavoshani, Fragouli & Diggavi
08) achieves capacity for C1+C2=C
Sink
Source 2Source 1
![Page 15: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/15.jpg)
Multiple-source multicast,uniform errors
Challenges:• Sources do not know each other’s information
− Independent coding at sources• Coding across different sources needed within
network in order for them to share network capacity− Multiple access property
• Straightforward extensions of single-source codes are suboptimal
Related work: • code construction in (Siavoshani, Fragouli & Diggavi
09) achieves capacity for m1 +m2 = m
Sink
Source 2Source 1
![Page 16: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/16.jpg)
Capacity region• U = set of source nodes• mS = min cut capacity between sources in subset S of U
and each sink• ri = rate from the ith source• Theorem: The capacity region (coherent & non-coherent)
under any z link errors is given by the cut set bounds
• Example: z = 1r1 ≤ m1 -2z =1r2 ≤ m2 -2z = 1(r1 + r2 ≤ m -2z = 2)
USzmrSi
Si
,2
Sink
Source 2
Source 1
![Page 17: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/17.jpg)
Multiple Access Network Information-flow And Correction (MANIAC) codes• Low encoding and decoding complexity• Coherent & non-coherent (by lifting)• Main ideas:
− Interior nodes do normal random linear network coding in base field
− Each source encodes with a Gabidulin code over a different extension field
− Enables successive decoding at the sinks
![Page 18: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/18.jpg)
Field Extensions: example
4221011
0001
F 22
413
02
F 12
1613
8
F
![Page 19: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/19.jpg)
Field Extensions
Fq
Fq
Fqn1
n1n2
![Page 20: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/20.jpg)
Encoding
M1x X1
r 1+2z
r1
=
nn
r1 +2z
M2G2x X2
r2
=
nn
r2 +2z
G1 can correct rank z errors over Fq
G2 can correct rank z errors over Fq n1
r 2+2z
n = kn1n2 where ni≥ri +2z
G1
![Page 21: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/21.jpg)
Network transmission
Y=T1G1M1+T2G2M2+E
![Page 22: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/22.jpg)
Network transmission
EMG
MGTTY
22
1121
1G
2G
Y=T1G1M1+T2G2M2+E (1)
![Page 23: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/23.jpg)
Network transmission
EMG
MGTTY
22
1121
EMG
MTGTY
22
1211
invertible whp for RLNC q
FTGTD 211
1G
2G
1G
2G
D1
G
Y=T1G1M1+T2G2M2+E
1n
(1)
(2)
mm
![Page 24: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/24.jpg)
Network transmission
EMG
MGTTY
22
1121
EMG
MTGTY
22
1211
1G
2G
1G
2G
Y=T1G1M1+T2G2M2+E
(D-1Y) = EDMG
M 1
22
1
2G
D-1E
invertible whp for RLNC
(2)
(1)
(3)
q
FTGTD 211D1
G1nmm
![Page 25: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/25.jpg)
Network transmission
EMG
MGTTY
22
1121
EMG
MTGTY
22
1211
1G
2G
1G
2G
(D-1Y) )( 1
22 EDMG 2
G
Y=T1G1M1+T2G2M2+E
lowest r2+2zrows
(D-1Y) = EDMG
M 1
22
1
2G
D-1E
(D-1E)
invertible whp for RLNC
(2)
(1)
(4)
q
FTGTD 211D1
G1nmm
![Page 26: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/26.jpg)
(D-1Y) = G2M2+(D-1E)
Decoding
Matrix of rank z over Fq
Matrix of rank z over Fq
n1
• M2 can be decoded (same as single-source decoding problem)
• Decoder then subtracts effect of M2 and decodes M1
(This is why we need the extension field)
![Page 27: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/27.jpg)
Outline•Single-source multicast, uniform errors (background)•Multiple-source multicast, uniform errors
T. Dikaliotis, T. Ho, S. Jaggi, S. Vyetrenko, H. Yao, M. Effros and E. Erez, "Multiple-access Network Information-flow and Correction Codes," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.
•Non-uniform errors: unequal link capacitiesS. Kim, T. Ho, M. Effros and S. Avestimehr, "Network error correction with unequal link capacities," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.T. Ho, S. Kim, Y. Yang, M. Effros and A. S. Avestimehr, "On network error correction with limited feedback capacity," ITA 2011.
• Non-multicast nested networks, uniform & non-uniform errorsO. Tekin, S. Vyetrenko, T. Ho and H. Yao, “Erasure correction for nested receivers,” Allerton 2011.
![Page 28: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/28.jpg)
Uniform and non-uniform errors• Uniform errors:
− Multicast error correction capacity = min cut – 2z− Worst-case errors occur on the min cut
• Non-uniform errors:− Model: network with unequal link capacities, adversarial
errors on any z fixed but unknown links− Not obvious what are worst-case errors
• Cut size versus link capacities• Feedback across cuts matters
− Related work: Adversarial nodes (Kosut, Tong & Tse 09)
![Page 29: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/29.jpg)
Network cuts
29
S
U
• Both forward and feedback links matter in the unequal link capacity case
• Only forward link capacities matter in the equal link capacity case
• Feedback can provide information about errors on upstream links
![Page 30: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/30.jpg)
Two-node cut set bounding approach
30
Two-node network with n forward links and m feedback links.
S
U
![Page 31: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/31.jpg)
Tighter cut set bounding approach
• The two-node cut set bounding approach is equivalent to adding reliable, infinite-capacity bidirectional links between each pair of nodes on each side of the cut
• Tighter bounds can be obtained by taking into account which forward links affect or are affected by which feedback links
• Equivalent to adding a link (i,j) only if there is a directed path from node i to node j on that does not cross the cut
31
Zigzag network
![Page 32: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/32.jpg)
z=1
Capacity=5
z=1
Achieve rate 3 using new code construction
Zigzag network example
y1
y2
f
y3
Converse:• If there are more than q5
codewords, there exist two codewords of the form
x={y1, y2, f, y3, y4} x’= {y1’,y2, f, y3, y4’}
• When x is transmitted, adversary changes y1 to y1’. Value on feedback link is unchanged.
• When x’ is transmitted, adversary changes y4 ’ to y4.
• In both cases, sink observes{y1 ’, y2, f, y3, y4} → confusion
y4
∞
∞
![Page 33: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/33.jpg)
z=1
Capacity=5
z=1
Achieve rate 3 using new code construction
Zigzag network example
y1
y2
f
y3
• Without feedback link, capacity = 2• How to use feedback link to increase
capacity?y4
∞
∞
![Page 34: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/34.jpg)
Zigzag network example - achievability
• z adversarial links• n>z links of capacity a between source
and A• one feedback link of capacity b ≤ a from
A to B (no advantage for larger b)• m>z links of capacity c from node B to
sink
![Page 35: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/35.jpg)
Zigzag network example - achievability • (x1,x2,…,xn): an (n,n-z) MDS code in GF(qa-c)
(error detection)
• (y1,y2,…,yn+m): an (n+m,n+m-2z) MDS code in GF(qc) (error correction)
• View (xi, yi) and feedback value f as vectors representing elements of GF(qa), in terms of powers of a primitive element α
• Links i=1,…,z transmit (xi, yi) +αc-(i-1)(a-c)f (codebook size is multiplied by qb)
• Links i=z+1,…,n transmit (xi, yi)
• Links i=n+1,…,n+m transmit yi
![Page 36: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/36.jpg)
Zigzag network example - achievability • If A’s incoming links are not consistent with
any codeword, A sends error message to B• If B receives an error message or a
feedback link value that doesn’t match the correct value received from S, • B sends an error message with all the
information received from s and A to sink u using a repetition code.
• Upon receiving an error message, sink compares against information received from A to find the adversarial link(s).
• Error messages are sent a finite number of times, so amortized overhead goes to zero.
![Page 37: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/37.jpg)
Outline•Single-source multicast, uniform errors (background)•Multiple-source multicast, uniform errors
T. Dikaliotis, T. Ho, S. Jaggi, S. Vyetrenko, H. Yao, M. Effros and E. Erez, "Multiple-access Network Information-flow and Correction Codes," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.
•Non-uniform errors: unequal link capacitiesS. Kim, T. Ho, M. Effros and S. Avestimehr, "Network error correction with unequal link capacities," Special issue of the IT Transactions dedicated to the scientific legacy of Ralf Koetter, Feb 2011.T. Ho, S. Kim, Y. Yang, M. Effros and A. S. Avestimehr, "On network error correction with limited feedback capacity," ITA 2011.
• Non-multicast nested networks, uniform & non-uniform errorsO. Tekin, S. Vyetrenko, T. Ho and H. Yao, “Erasure correction for nested receivers,” Allerton 2011.
![Page 38: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/38.jpg)
Background - non-multicast
• Finding the capacity region of a general non-multicast network even in the erasure-free case is an open problem.− Coding across different receivers’ packets (inter-
session coding) is sometimes required.− We don’t know in general when intra-session coding is
sufficient.• We can derive cut set bounds by studying three-layer
networks
![Page 39: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/39.jpg)
Three-layer nested networks• A model for temporal demands
(decoding deadlines)• Links ↔ Packets• Receivers ↔ Deadlines by
which certain information must be decoded
• Each receiver receives a subset of the information received by its successor (Nested structure)
• Packet erasures can occur• Non-multicast problem I1 I1, I2 I1, I2, I3
t1 t2 t3
![Page 40: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/40.jpg)
Erasure models
We consider 2 different erasure models:• Arbitrary z-erasures
− At most z links can be erased, locations are unknown a priori.
• Sliding window erasure model
![Page 41: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/41.jpg)
Arbitrary z-erasures - Upper bound
• We can write cutset bounds for each erasure event from the perspective of each sink
• Need to combine these individual bounds from the perspective of multiple erasure events for multiple sinks
• A set of random processes Yis is a decoding set for Mi if |Yi
s|=mi-z and Yi
s is upstream of ti
• For each ti there are decoding sets Yis
zm
m
i
i
![Page 42: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/42.jpg)
Cutset-combining procedure• Xmi – set of links upstream of ti
• Repeat rules: extend: H(Xmi-1|M1, …, Mi-1) ≤H(Yi
s|M1, …, Mi-1)+ mi-z-mi-1
extract: H(Yis|M1, …, Mi-1) =ui+ H(Yi
s|M1, …, Mi)
combine: (mi-z) H(Xmi|M1, …, Mi)≤mi∑s H(Yi
s|M1, …, Mi)
• H(Yis|M1, …, Mi-1) – “residual capacity” after taking care of M1..Mi-1
• Same upper-bounding procedure applies when there are z/2 errors
zm
m
i
i
![Page 43: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/43.jpg)
Examplem1=3,m2=5, m3=7, m4=11, z=1
u2
u1
u3
(2,1,0)(2,0,0)
(0,4,0)(2,1,1)
(2,0,3)
(0,0,6)(0,4,1)
3D:
u1≤23u1+2u2 ≤83u1+2u2+u3 ≤96u1+5u2+4u3
≤24
u2
u1
(2,0)
(0,4)
(2,1)
2D: u1≤23u1+2u2
≤8
u1
0 21D:
u1≤2
![Page 44: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/44.jpg)
Example
• Exponentially large number of bounds that can be derived• Only some are tight – how to find them?
m1=3,m2=5, m3=7, m4=11, z=1
u1≤23u1+2u2 ≤83u1+2u2+u3 ≤96u1+5u2+4u3 ≤246u1+4u2+2u3+u4≤209u1+6u2+4u3+3u4 ≤366u1+5u2+4u3+2u4 ≤286u1+4.5u2+4u3+3u4 ≤309u1+7.5u2+7u3+6u4 ≤54
4D: 12
13
23
1245
1345
2345
12341235
124567
134567
234567
123467123567123456123457123456123457 1234567
1234567
1234567
1234567
1234567
![Page 45: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/45.jpg)
1,2,…,9,10 11,12,13,14
15,16,17,18
19,20,21,22
I1
I2
I3
I40.2 0.40.40.40
“As uniform as possible” coding
0.75
0.18750.1875
0.25
0.1875
0.2 0.2 0.2
0.50.50
0.5
0.25
0.50.250
m1 = 10, m2 = 14, m3 = 18, m4 = 22, u1 = 6, u2 = 3, u3 = 3, u4 = 4, z=2
75.0210
6
1
1
zm
u25.0
214
3
2
2
zm
u1875.0
218
3
3
3
zm
u5.0
21018
3
13
3
zmm
u 2.0222
4
4
4
zm
u4.0
21022
4
14
4
zmm
u5.0
21422
325.0*4
24
4
zmm
u
![Page 46: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/46.jpg)
Main Result
Theorem: The z-erasure correction capacity region of a 3-layer nested network is achieved by the “as uniform as possible” coding scheme.
![Page 47: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/47.jpg)
Proof Idea
• Let (u1,u2,…,un) be given. Fill up the table “as uniformly as possible”:
1,2,…,m1 m1 +1,…,m2 m2 +1,…,m3 … mn-1 +1,…,mn
I1 T1,1
I2 T2,1 T2,2
I3 T3,1 T3,2 T3,3
… … … … …
In Tn,1 Tn,2 Tn,3 … Tn,n
![Page 48: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/48.jpg)
Proof Idea-cont.
• Use Ti,j to decide which sequence of bounds to combine• Inductively prove that theoretical bound on the residual
information-carrying capacity of the first mk links after transmitting I1,I2,…,Ik matches the residual capacity in the table.
![Page 49: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/49.jpg)
Model: sliding window erasure
• Fix p and a threshold T. For y≥T, at most py out of y consecutive links can be erased.
• Erasures occur with rate p in long term.• Erasure bursts cannot be too long (Controlled by T ).
![Page 50: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/50.jpg)
Main Result
• Theorem: Let the rate vector be achievable. Then the rate vector
is achieved by “as uniform as possible” coding.
• Note: asymptotically optimal for m1 >> T >>1
u
Tp
p
mTm
1
1
1
log1
1
1
1
u
![Page 51: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/51.jpg)
Proof Idea
• Let be achieved in erasure-free case.• Let
• Use the “as uniform as possible” coding corresponding to • Turns out that 1-p times the original data is recovered
under the worst-case erasure
1
1log1
1
mTm
)v,...,v,(v v n21
v
![Page 52: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/52.jpg)
Proof Idea-cont.
u
Tp
p
mTm
1
1
1
log1
1
1
1
Coding Scheme
Worst-case erasure
Comparison between erasure-free case
We conclude that
is achieved by “as uniform as possible” coding.
![Page 53: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/53.jpg)
Conclusion
• Generalizing the network error correction problem →new types of approaches
• Achievability and converse results for non-uniform errors and general non-multicast problems still wide-open
![Page 54: On coding for networks with errors Tracey Ho Caltech BIRS, August 2011.](https://reader031.fdocuments.us/reader031/viewer/2022012918/56649e425503460f94b3559b/html5/thumbnails/54.jpg)
• Thank you!