2014 IEEE International Symposium on Information Theory

Degrees of Freedom for Multiple-Multicast Traffic

Shaileshh Bojja Venkatakrishnan

Pramod Viswanath

Sreeram Kannan

University of Illinois, Urbana-Champaign University of Illinois, Urbana-Champaign University of California, Berkeley

Email: bjjvnkt2@illinois.edu
Email: pramodv@illinois.edu
Email: ksreeram@eecs.berkeley.edu

AbstractWe propose a new coding scheme for interference

alignment in a single hop fast fading wireless network with
general message demands. For the X-Channel, the Degrees of
Freedom (DoF) region achievable by the scheme is shown to touch
a known outer-bound at several points. For multiple-multicast
demands we show that the achievable region is at least half
of the cut-set bound region. The key innovation in our scheme
is the reduction of the vector space alignment problem to a
combinatorial arrangement problem. Finally, we use the scheme
to give a poly-logarithmic bound for the flow-cut gap in fast
fading Gaussian wireless networks with multiple multicasts.

I. I NTRODUCTION
Interference is a first order issue in wireless networks. Traditional methods to handle interference such as power control
and scheduling can be suboptimal and limit the maximum
throughput in the network. Interference alignment (IA) is a
relatively recent technique to mitigate this problem in which
the transmitted signals are beam-formed in such a way that the
interference is restricted to only a subspace at each receiver.
The DoF of the K-user interference channel with ergodic
fading has been analyzed in [1] where they show that a
DoF per message of up to 21 is achievable. [2] looks at an
interference network in which every transmitter has a single
multicast message. The DoF region for networks with more
general message demands has not yet been studied. In this
paper, we present an IA scheme that generalizes the scheme
in [2] for (i) multiple-unicasts (also referred to as the XChannel), in which each transmitter can have an independent
message to each receiver and (ii) multiple-multicasts, in which
each transmitter can have an independent multicast message to
each subset of receivers. For the case of multiple-unicasts, we
show that the DoF outer bound in [3] can be achieved under
a symmetry condition. We also show that our scheme can
achieve a DoF of within 12 of the cut-set bound in both cases.
Finally we present an approximate DoF region of general fast
fading Gaussian networks under a conjecture on the flow-cut
gap of node-capacitated undirected graphs.
II. N ETWORK M ODEL
Following the model presented in [3], we consider a single
hop wireless network with K transmitters and K receivers,
each having a single antenna. Let Hji be a generic channel
matrix between transmitter i and receiver j. Assume that the
causal channel state information is known globally. Then, the
input-output relation is given by:
X
Yj (t) =
Hji (t)Xi (t) + Zj (t), j = 1, 2, . . . , K
i{1,...,K}

978-1-4799-5186-4/14/$31.00 2014 IEEE

where t represents the channel use index, which can be

interpreted as time or frequency index. Xi (t) is the signal
transmitted by the transmitter i, Yj (t) is the signal received
by receiver j and Zj (t) represents the additive white Gaussian
noise at receiver j at time k. [3] has given an outer bound for
the DoF region of the X-network as:

out
D
= [dij ]KK : (m, n) {1, 2, . . . , K}{1, 2, . . . , K},
K
X

dm
q +

q=1

K
X

dpn dm
n 1


(1)

p=1

where dij refers to the DoF of the message from transmitter

i to receiver j. Throughout this paper we restrict ourselves
to the single antenna case. Notation: For any k N, let [k]
denote the set {1, . . . , k}. For any two matrices U and V we
let U V mean the (column) subspace spanned by U is
contained in the subspace spanned by V. Similarly U V
means the subspace spanned by both the matrices coincide.
III. X N ETWORK : I NTERFERENCE A LIGNMENT S CHEME
In the alignment scheme discussed in [1] for the K-user
interference channel, a key problem is to construct a matrix
V such that, for generic matrices T1 , . . . , Tn , the spaces
T1 V, . . . , Tn V are aligned, i.e., T1 V . . . Tn V. One
approximate method that guarantees asymptotic alignment is
done as follows [4, section 4.6.2]:
V =[(T1 )1 (T2 )2 . . . (Tn )n w,
n
X
s.t.
i 1, 1 , . . . , n Z+ ],
i=1

where w is a generic vector used to evolve the space and

a parameter. Let us call such a vector w used to evolve
subspaces in the above fashion as a base-vector for the
subspace.
A. Relative arrangement problem
In the case of a K-user X-channel each transmitter can have
[1]
[N ]
a message to each receiver. Let {Xi , . . . , Xi } denote a set
of N message vectors (of equal dimension) from transmitter i.
Each message is meant for a particular receiver in [K] denoted
[l]
by label(Xi ). Suppose we use N base-vectors w1 , . . . , wN to
[1]
[N ]
evolve the beamforming matrices Vi , . . . , Vi respectively
[1]
,...,X
[N ] ) denote
for each transmitter i [K]. Now, let (X
i
i

836

2014 IEEE International Symposium on Information Theory

(a) DoF matrix d

(b) Message demand matrix n

(c) Arrangement of the messages as an array B

Fig. 1. An example showing the optimal arrangement of messages for the X channel with the given message demand. The red dotted lines show the interfering
messages at receiver 1. Notice that all the interfering messages are contained under (in the same columns as) the largest interference (at all receivers)

an ordering of the set of messages of transmitter i, such that

[l] is beamformed using V[l] , i.e.,
the message X
i
i
Xi (t) =

N
X

[l] [l]
Vi X
i (t)

l=1

is the signal transmitted by the ith user. A beam-forming ma[l]

[l]
[l]
trix Vi undergoes the linear transformation Vi Hji Vi
when subjected to the channel matrix Hji between transmitter
[l]
i and receiver j. Now, suppose further that Vi s have the
following property:
Property 1. At each receiver j, the interference subspaces in
[l]
[l] ) 6= j} align
the set {Hji Vi : i {1, . . . , K}, label(X
i
with each other l [N ], while all the message subspaces in
[l]
[l] ) = j} become linearly
{Hji Vi : i {1, . . . , K}, label(X
i
independent to the interference spaces and to each other.
With this property the ordering of the messages become
important. For example, consider a 2-user X-channel where
[1]
[2]
transmitter 1 has the messages {X1 , X1 } for receiver 1,
[3]
message {X1 } for receiver 2, while transmitter 2 has the
[1]
[2]
[3]
message {X2 } for receiver 1 and messages {X2 , X2 } for
receiver 2. Suppose user 1 transmits its messages as:
[1]

[1]

[2]

[2]

[3]

[3]

X1 (t) = V1 X1 (t) + V1 X1 (t) + V1 X1 (t).

Now, consider the following two ways by which user 2 can
transmit:
[1]

[1]

[2]

[2]

[3]

[3]

[1]

[3]

[2]

[2]

[3]

[1]

(i) X2 (t) =V2 X2 (t) + V2 X2 (t) + V2 X2 (t)

(ii) X2 (t) =V2 X2 (t) + V2 X2 (t) + V2 X2 (t).
[l]

If the Vi s satisfy Property 1, then the interference subspace

takes a total of 3 signalling dimensions in case (i), but only 2
in case (ii) at the receivers. The messages themselves occupy
3 dimensions. Therefore, the messages have a DoF of 3/6
in case (i) but only 3/5 in case (ii), at the receivers. Clearly
the way we start out with message arrangements dictates the
efficiency of the communication scheme. This is precisely the
problem we address. In the sections that follow we describe
how best to permute the message tuples at each transmitter in
order to achieve the optimal DoF.

B. Alignment scheme
X networks represent the most general class of nonmulticast communication scenario possible in a single-hop
wireless network. Let d = [dij ]KK [0, 1]KK denote the
DoF matrix, where the (i, j)th entry dij refers to the DoF of
the message from transmitter i to receiver j. Since the vertices
of the DoF region given in (1) are rational, we consider only
the achievability of points with rational coordinates. Let us
first
those points d on the outer bound for which
PK consider
i
d
=
D
i [K] for some D R, i.e., the sumj=1 j
DoF of the messages from each transmitter are equal. We later
consider the scheme for a general DoF point. For any rational
point d, let Z+ be such that n , d has all integral
entries, i.e., nij Z i, j [K] where nij = dij . We interpret
nij as the number of messages from transmitter i to receiver j.
Therefore, every transmitter has a total of N = D messages.
The following sections III-C and III-D are the two key steps
involved in our scheme.
C. Step 1: Combinatorial message alignment
Since each transmitter has N messages in total, let us use a
set of N base-vectors W = {wi , 1 i N } (where wi s are
generic) in order to evolve the beamforming matrices. Notice
that the same set of base-vectors is used in all the transmitters.
As discussed previously in section III-A, the ultimate aim here
is to optimally assign each message to one of the beamforming
matrices.
We view such an assignment of messages to beamforming
matrices or equivalently, to base-vectors, as an array B where
B(i, j) denotes the message from transmitter i listed under
base vector j. Fig. 1(c) shows an example assignment for K =
3, N = 10. We have listed only the message labels and not
the actual messages itself in the assignment B because for any
assignment of messages, exchanging of messages having the
same label does not change the performance.
Let us call a column a j-block (for j [K]) if all the entries
of that column have the label j. A column which is not a jblock for any j is called a -block. Let Nj denote the number
of j-blocks in B for j [K] and Ij , N Nj .
Proposition 1. For any point d in the outer bound region (1)
such that the sum-DoF of messages from all transmitters are

837

2014 IEEE International Symposium on Information Theory

(a) Gn (V, E)

(b) Time layered version

(c) Steady state

Fig. 2. An example showing a node capacitated graph Gn (V, E) with a single source, node 1, having a multicast message to nodes 2 and 3. In Fig. 2(a),
two possible Steiner trees are shown in red and blue dotted lines. Fig. 2(b) shows the passing of messages with time. At each time instant, the interior nodes
forward the message received in the previous time slot while the source generates a new message and in Fig. 2(c), we have shown the steady state traffic in
the network

equal, there exists an arrangement B such that:

K
X

nij + Ij

j [K]

(2)

i0 =1

where n, , Ij are as was defined above.

Proof: Since d is within the outer bound region (1), we
must have:
K
X
i0 =1

nij +

nij 0

j [K].

(3)

j 0 [K]:j 0 6=j

Let 1 = mini {ni1 : 1 i K} be the smallest message

to receiver 1 from a transmitter. We first arrange 1 of the ni1
messages from each transmitter i together
in 1 columns. This
P

implies I1 = N 1 . In fact, I1 = j 0 [K]:j 0 6=1 nij 0 where i is

a transmitter having the largest number of interfering messages
or equivalently the smallest number
P of non-interfering messages for 1. As such, we have I1 j 0 [K]:j 0 6=1 nij 0 i [K].
Therefore, from equation (3) we get:
K
X

ni1 + I1 .

(4)

i0 =1

Now, we sequentially perform this operation for each receiver

(2) holds. This is possible because,
PK so that
PKequation
i

N
since j nij i [K] and hence
j
j=1
j=1 j
the above operations require no more than N = |W | columns.
The remaining messages can be arbitrarily assigned to any
unassigned columns. The proposition follows.
Fig. 1 shows an illustration of a K = 3 case with = 20.
D. Step 2: Evolution of beamforming matrices
The next step is to generate the beamforming matrices
from the base-vectors. The subspaces are created such that
they satisfy Property 1, i.e., an interfering message from a
subspace remains within the subspace while a non-interfering
message becomes linearly independent to the subspace at the
receiver. This property, together with the previous combinatorial arrangement of the messages allows us to achieve
interference alignment. Let us use a symbol expansion of
2
[l]
= t ( + 1)K K where t = . Let Vi denote

the beamforming matrix associated with base-vector wl for

l [N ] at transmitter i. Now, the evolution is done as:

[l]
nm
Vi =
H
nm wl :

(m,n)[K]2

B(m,l)6=n

nm {0, . . . , } if m 6= i,
nm {0, . . . , 1} if m = i


,

(5)

[l]

i.e., for each Vi we use all the K 2 channel matrices except

those associated with the messages listed under the base vector
wl in B. In the following we show that contructing the
beamforming matrices this way guarantees alignment. The
arguments we use in this section are based on the results in [3]
and [2].
Proposition 2. The set of beamforming matrices generated
according to Equation (5) satisfies Property 1.
Proof: The proof is as follows:
1) Alignment of interfering messages: We first show that
at every receiver all the interfering vectors are aligned within
a subspace. Let:

[l]
nm

V ,
Hnm wl : nm {0, . . . , } ,

(m,n)[K]

B(m,l)6=n

(6)

[l] : l [N ], i s.t. B(i, l) 6= j] where for

and let V(j)
, [V
matrices A, B, [A B] stands for the augmented matrix. Now
[l]
for any interference message coded via the matrix Vi it is
easy to see that the received signal space:
[l]
[l] V(j)

Hji Vi V

since left multiplication by the channel matrix only increases

[l]
the exponent ji in the vectors of Vi by 1, and such vectors
[l] . In other words, our construction
are already included in V
ensures that all the interfering messages in column l of B
are aligned at the receivers for each l. Such a columnwise
alignment implies a global alignment of all the interfering
messages within a subspace. Hence we conclude that for any

838

2014 IEEE International Symposium on Information Theory

receiver j [K], the interference component is contained

within the subspace V(j).

We now proceed to show linear
independence between the signal and interference spaces at
the receiver.
2) Linear Independence of Signal and Interference: Let

S(j)
denote the subspace formed by the beamforming matrices
corresponding to the message signals at receiver j:
[l]

S(j)
= [Hji Vi : B(i, l) = j, i [K], l [N ]]

(7)

so that if S(j)
and V(j)
are linearly independent, then V(j)
can be zero-forced to retrieve the information of the messages

in S(j).
Let (j) denote the combined space:

(j) = [S(j)
V(j)].

(8)
2

[l]

Note that Vi is of dimension ( + 1)K K by K1 ( +

2
[l] is ( + 1)K 2 K by ( + 1)K 2 K .
1)K 2K+1 while V
Using Proposition 1 we see that (j) is a tall matrix. Since all
channel matrices are assumed to be diagonal, each element of
(j) is a monomial term of several random variables. Hence
by the argument used in [2] and [3, Lemma 1] we conclude
that (j) has full-rank almost surely. This proves the linear
independence of the message subspaces with the interference
subspace and with each other, as claimed in Property 1.
With Property 1 being true, looking back at the message
assignment algorithm of section III-C, we see that a j-block
is an interference block for all receivers except j, a -block
is an interference block for all the receivers and Ij counts the
number of interference blocks for receiver j. What we have
ensured is that all the interfering messages at receiver j are
aligned under the largest interfering message for receiver j.
Since the message blocks in (j), j [K] with a dimension
2
2
of ( + 1)K K by ()K1 ( + 1)K 2K+1 have fullcolumn rank, in the asymptotic case [4] we can expect a DoF
of:
2

[l]

|Hji Vi |
1
()K1 ( + 1)K 2K+1
=
= lim

(l + 1)K 2 K
[l]

(t) or equivalently a sumcorresponding to each message X

i
i
i
DoF of nj / = dj for each source-destination pair. It is
crucial to note that we are able to use a time expansion
with a scaling of t = because our arrangement satisfied
PropositionP
1. For any other arrangement of B, it is possible
0
K
that maxj { i0 =1 nij +Ij } is larger than in which case (j)
would no longer be full-rank for all j. Therefore any DoF point
in the outer-bound 1 with equal sum-DoF is achievable. We
conclude this section by stating the achievable DoF region for
the scheme.
Theorem 1. The achievable DoF region of the scheme for the
X-network is given by the set: d [0, 1]KK such that,

X
X
dij |s| (9)
(|s| 1)D + max
dij +
i

c
j:js

i,j
js,i[K]

s {1, . . . , K}, s 6= {}, where D = maxi

j[K]

dij .

The details are left to a longer version of this paper. Note

that the achievable region (9) touches the outer bound (1) at
points where the sum-DoF of messages from the transmitters
are equal.
IV. D O F I NNER B OUND
In the previous section we presented a scheme for a multiple
unicast traffic model. For a general multicast model we can
essentially use the same two step procedure for alignment as
outlined in sections III-C and III-D. A technical difficulty
in this model is that the message labels can now denote
sets of receivers instead of just single receivers. One natural
implication of our alignment framework is:
Proposition 3. In a K-user multiple multicast network, any
d such that:
K
X X

dis

s[K] i=1
js

1
2

and

dis

s[K]

1
2

i, j

(10)

is achievable using our alignment scheme,

i.e., we can show that our scheme comes within 12 of the
cut-set bound. Since the X-channel and multiple multicast
channels are only a special case of the above generalized
multiple multicast channel, this inner bound holds even in
those cases.
V. A PPROXIMATE D O F OF GENERAL G AUSSIAN N ETWORK
We now characterize flow on a wireless network with
multicast demands. In [5] the authors propose a novel layering
technique for a network with multiple unicast message demands. Such a scheme comes to within a logarithmic factor of
the cut-set bound. In the following, we show that this method
can be generalized to networks with multicast demands using
our IA scheme in the physical layer. Consider a wireless
network G with vertex set V and an undirected edge set E.
Assume flat-fading between each pair of nodes connected by
an edge, with channel coefficients of unit variance. Let us
suppose there are k multicast demands with source node i
and destination set Si corresponding to demand i [k]. Let
Gi = i Si and t = maxi |Gi |. Without loss of generality
we assume the i s are all distinct, i.e., each source node has
only one multicast message.
A. Achievability
In order to obtain the flow-cut gap for the Gaussian network, we first consider an undirected node-capacitated graph
Gn (V, E), on the same vertex and edge sets, with an incom)
ing and outgoing node capacity of C(P
for all the nodes.
2
Here the node capacity bounds the total multicast message
rate forwarded (resp. received) by a node, with the rate of
each multicast message counted only once. Each Steiner tree
spanning Gi provides a dissemination path for the messages
from the source i to the destination set Si .

839

2014 IEEE International Symposium on Information Theory

Our main observation is that any multicasting over the

node-capacitated graph Gn can be emulated on the Gaussian
network, or in other words:
g
n
Rach
(P ) Rach
(P )

(11)

g
n
where Rach
(P ) and Rach
(P ) stand for the achievable rate
regions in the Gaussian and the node-capacitated network
respectively. Consider the steady-state bipartite graph corresponding to the node-capacitated network as illustrated in
Fig. 2. This steady state network also has a node capacity
of 12 C(P ). From our interference alignment scheme we know
that half of the cut-set bound is achievable at high SNR for
a complete bipartite topology (10). It is easy to see that the
achievability of 12 C(P ) does not change for any other bipartite
topology (for feasible multicasts). Now, by routing information
in the Gaussian network via any set of Steiner trees that lead
to that steady state rate, we can conclude that any feasible flow
achievable in the node-capacitated network Gn (V, E) can be
achieved on the Gaussian network G(V, E). This proves (11).

Further, since the cut-set bound region for the node capacitated
graph is exactly the same as the polymatroidal cut-set bound
region, using the result in [5] we have:


g
n
Rcut
(P/bd3 )
1 g
P
Rcut
(P )
n

Rcut Rcut

(16)
3
2
bd
2f (k, t)
f (k, t)
where d is the maximum degree of the graph and b =
2
eE(log |h| )
. Now, it is shown in [6] that a flow-cut gap of
2
O(log3 (kt)) is achievable for the setting of edge-capacited
undirected graphs. We conjecture that a similar result holds
true even in our case:
Conjecture 1. In a node-capacitated undirected graph a
flow-cut gap of f (k, t) = O(logc kt) is achievable for some
constant c > 0.
Under this conjecture, combining the results so far we get:
Theorem 2. The capacity of a Gaussian wireless network with
k multicast demands as described above can be approximated
to within a poly-logarithmic factor by the cut-set bound:
g
Rcut
(P/bd3 )
g
Rach
(P ).
2O(logc kt)

B. Cut-set bounds
Let us now relate the corresponding cut-set bounds.
g
n
(P ) and Rcut
(P ) denote the Gaussian and nodeLet Rcut
capacitated cut-set regions respectively for power P . For any
cut V in the Gaussian network G(V, E), let D() =
{1 i k : i , Si c 6= } denote the set of
commodities that are separated by cut . We then have:
X
Ri cutg () = I(X ; Yc |Xc ).
(12)
iD()

The cut value above can be upper bounded by evaluating the

capacity of a colored Gaussian network as explained in [5].
Now, in the node-capacitated network case, for any cut
V , let F = {(uv) E : u , v c } denote the edges
crossing the cut in Gn . We define the value of the cut as
in the polymatroidal network case [5], i.e., for an assignment
function g that maps each edge in F to either of its end nodes,
we have:
cutn () , min |g(F )|
g

C(P )
2

(13)

where g(F ) is the range of g. For the concurrent flow

problem [6], where the message demand rates are given by
Ri = Di i [k] for fixed D1 , . . . , Dk , we have:
min P

cutn ()

i:iD()

Di

(14)

where is called the sparsest cut. For the optimal we

define the flow-cut gap as a function f (k, t), such that
1

f (k,t) . If the flow-cut gap is also independent of the demand

rates D1 , . . . , Dk , we have
n
Rach

n
Rcut
.
f (k, t)

(17)

VI. C ONCLUSION
We have presented a novel interference alignment scheme
that simplifies the problem of interference alignment to a combinatorial problem in networks with general message demands.
This framework allows us to easily see the DoF region for
simpler networks like the K user interference network. For
the X network we have shown that the achievable region
of our scheme touches a previously known outer bound. We
have also extended the result of achievability of half of the
cut-set bound to general multicast networks. And finally, we
have demonstrated a poly-logarithmic flow-cut gap in general
Gaussian wireless networks with multicast demands.
R EFERENCES
[1] V. R. Cadambe and S. A. Jafar, Interference alignment and the degrees
of freedom of the k-user interference channel, IEEE transactions on
Information Theory, vol. 54, no. 8, pp. 34253441, August 2008.
[2] L. Ke, A. Ramamoorthy, Z. Wang, and H. Yin, Degrees of freedom
region for an interference network with general message demands, CoRR,
vol. abs/1101.3068, 2011.
[3] V. R. Cadambe and S. A. Jafar, Interference alignment and the degrees
of freedom of wireless x networks, IEEE transactions on information
theory, vol. 55, no. 9, pp. 38933908, September 2009.
[4] S. A. Jafar, Interference Alignment - A New Look at Signal Dimensions
in a Communication Network, ser. Foundations and Trends in Communications and Information Theory. now, 2010, vol. 7, no. 1, dOI:
10.1561/0100000047.
[5] S. Kannan and P. Viswanath, Capacity of multiple unicast in wireless
networks: A polymatroidal approach, IEEE Transactions on Information
Theory, accepted March 2014.
[6] P. N. Klein, S. A. Plotkin, S. Rao, and E. Tardos, Approximation
algorithms for steiner and directed multicuts, Journal of Algorithms,
vol. 22, no. 2, pp. 241269, 1997.

(15)

840