Insufficiency of linear coding in network information flow
Summary (3 min read)
I. INTRODUCTION
- Associated with the sources are messages and associated with the sinks are demands.
- The goal is for each sink to deduce its demanded messages from its in-edges by having information propagate from the sources through the network.
- The following result was recently shown in [2] .
- Also, the authors compute the exact network coding capacity and the linear network coding capacity of this network for any finite-field alphabet (Corollary IV.5).
II. INSUFFICIENCY OF NETWORK LINEAR CODES OVER FINITE FIELDS
- The authors establish the existence of a solvable network that has no linear solution over any finite field and any vector dimension.
- Suppose there exists a fractional coding solution over alphabet with, also known as Proof.
- If all messages other than are fixed, then each edge of the path from to can take on at most different values.
- Suppose the authors impose on a network code the constraint that for every node with in-degree one, the out-edges must carry the same symbol as the lone in-edge, and for every source with exactly one message, the out-edges must carry the source's lone message.
Lemma II.2:
- The network has a scalar linear solution over any ring with characteristic two, but has no linear solution for any vector dimension over a finite field with odd characteristic.
- Proof: A scalar linear solution (as illustrated in Fig. 1 ) to the network over any ring of characteristic two is given by the following edge functions and sink decoding functions: (any edge function not shown is assumed to be an identity mapping).
- Note that the fact that the alphabet is a ring with characteristic two is used only in decoding the message at node where .
- Let be the identity matrix and for each and , let be the vector carried on the edge from to .
- Since by (14) , the matrices and are invertible.
Lemma II.3:
- The network has a scalar linear solution over any ring where is a unit, but has no linear solution for any vector dimension over a finite field with characteristic two.
- Proof: A scalar linear solution (as illustrated in Fig. 2 ) to the network over any ring where is invertible is given by the following edge functions and sink decoding functions (any edge function not shown is assumed to be an identity mapping):.
- This contradicts the assumption that the demand at node can be recovered, since is not uniquely determined by the node's inputs.
- Since is the only node that produces message and node demands message , and since there is a unique directed path from to , the coding capacity of is at most by Lemma II.1.
Theorem II.4:
- There exists a solvable network that has no linear solution over any finite field and any vector dimension.
- The proof is achieved with , which combines networks and .
- The symbols and indicate addition and subtraction in the ring of integers modulo , the symbol indicates addition in the ring (i.e., bitwise XOR), and indicates the result of exchanging the order of the bits in a 2-bit binary word .
- Note that the functions and are linear over but not over GF or , the function is linear over and GF but not over , and the function is not linear over any of these.
- In decoding at node , the authors used the fact that if the 2-bit binary representation of is , then the following binary representations also hold: Corollary II.5: Proof: By Theorem II.4, the network is solvable and therefore the coding capacity is at least (independent of the alphabet size by Lemma I.1).
III. INSUFFICIENCY OF NETWORK LINEAR CODES OVER RINGS AND MODULES
- The authors start by showing how to extend nonsolvability over finite fields to nonsolvability over finite commutative rings with identity.
- If a network does not have a linear solution over any finite field in dimensions, then it does not have a linear solution over any finite commutative ring with identity in dimensions, also known as Theorem III.1.
- The authors need only show that the demands are met.
- It follows by induction that along every edge, if is the vector carried by the new coding and is the vector carried by the old coding, then (component-wise).
- The corollary establishes that linear network codes are insufficient over a class of rings that includes finite fields.
Corollary III.2:
- There exists a solvable network such that for every vector dimension there is no linear solution over any finite commutative ring with identity.
- In [11] , solvable networks were given, whose minimum alphabet size required for a solution could be made arbitrarily large.
- By combining such a network with the network used in the proof of Theorem II.4 (i.e., taking the disjoint union of them) one obtains a solvable network with no linear solution for any vector dimension and an arbitrarily large minimum alphabet size for a solution.
- From this fact and Corollary III.2, the authors immediately obtain the following corollary.
Corollary III.3: For each
- , there exists a solvable network which has no scalar solution for any alphabet of cardinality smaller than , and such that for every vector dimension there is no linear solution over any finite commutative ring with identity.
- The authors can talk about linearity in even more generality than the above, if they are willing to separate the set of coefficients allowed in linear functions from the set of inputs to the linear functions (the set of messages).
- Or the authors can let the set of coefficients be any field and let the message set be any vector space over .
- This action must satisfy the analogues of the usual vector space laws: for any and , the authors have (the first here is the ring zero and the second is the group zero).
- The notions of scalar linear solution and linear solution for a network now easily generalize to the context of an -module .
Theorem III.4:
- There exists a solvable network that does not have an -linear solution over for any ring , any finite -module with more than one element, and any vector dimension.
- If the authors have a scalar -linear solution over , then for each demand at a sink node they get an equation of the form where are the source messages and is the composition of decoding and edge functions given by the specified solution; this equation must hold for all choices of from .
- (The authors thank Daniel Goldstein for this clean proof of the result and Lance Small for the reference.).
- The division into cases will be on whether in the final ring (or, equivalently, whether for all in ).
- If , then the proof of Lemma II.2 shows that does not have a scalar -linear solution over .
IV. ASYMPTOTIC INSUFFICIENCY OF NETWORK LINEAR CODES OVER FINITE FIELDS
- Throughout this section, is a finite field, all matrices have entries in , denotes the identity matrix for each , and denotes the vector carried on the edge from a node to a node , where and are two adjacent nodes in some given network.
- The following notation will be used in proofs in this section.
- Therefore, using (40) and the identity gives which, in turn, implies Since the right-hand side of (41) has terms and must be written as a linear combination of the terms on the left-hand side of (41), the term on the left-hand side of (41) must have a zero matrix coefficient in the linear combination.
- Therefore, the number of possible values for the right-hand side is no larger than the number of possible values for the left-hand side.
- The next corollary follows immediately from IV.3 and IV.4, and together with Corollary II.5 shows that the coding capacity of (i.e., ) is exactly 10% greater than the maximum linear coding capacity (i.e., ) over any finite field.
Did you find this useful? Give us your feedback
Citations
2,442 citations
2,190 citations
Cites background from "Insufficiency of linear coding in n..."
...It is known that linear codes are insufficient for optimal inter-session network coding [ 18 ], and even if we limit ourselves to linear codes, determining how to perform the coding is NP-hard [6]....
[...]
890 citations
Cites background from "Insufficiency of linear coding in n..."
...It is known that linear codes are insufficient for optimal inter-session network coding [18], and even if we limit ourselves to linear codes, determining how to perform the coding is NP-hard [6]....
[...]
738 citations
451 citations
Cites background from "Insufficiency of linear coding in n..."
...One rea son for this is that coding for multiple connections is a very difficult problem—one that, in fact, remains currently open with only cumbersome bounds on the asymptotic capability of coding [16] and examples that demonstrate the insufficiency of various classes of linear codes [17], [18], [19], [20]....
[...]
References
8,533 citations
"Insufficiency of linear coding in n..." refers background in this paper
...The authors thank Kris Popat for providing a helpful suggestion....
[...]
3,660 citations
2,628 citations
1,525 citations