3092
5, MAY 2017
Fundamental Limits of Cache-Aided
Interference Management
Navid Naderializadeh, Mohammad Ali Maddah-Ali, and Amir Salman Avestimehr
Abstract— We consider a system, comprising a library of
N files (e.g., movies) and a wireless network with a K T trans-
mitters, each equipped with a local cache of size of MT files
and a K R receivers, each equipped with a local cache of size of
M R files. Each receiver will ask for one of the N files in the
library, which needs to be delivered. The objective is to design
the cache placement (without prior knowledge of receivers’
future requests) and the communication scheme to maximize
the throughput of the delivery. In this setting, we show that the
 
sum degrees-of-freedom (sum-DoF) of min K T MT +K N
R MR
, K R
is achievable, and this is within a factor of 2 of the optimum,
under uncoded prefetching and one-shot linear delivery schemes.
This result shows that (i) the one-shot sum-DoF scales linearly
with the aggregate cache size in the network (i.e., the cumulative
memory available at all nodes), (ii) the transmitters’ caches
and receivers’ caches contribute equally in the one-shot sum-
DoF, and (iii) caching can offer a throughput gain that scales
linearly with the size of the network. To prove the result, we
propose an achievable scheme that exploits the redundancy of
the content at transmitter’s caches to cooperatively zero-force
some outgoing interference, and availability of the unintended
content at the receiver’s caches to cancel (subtract) some of
the incoming interference. We develop a particular pattern for
cache placement that maximizes the overall gains of cache-aided
transmit and receive interference cancellations. For the converse,
we present an integer optimization problem which minimizes the
number of communication blocks needed to deliver any set of
requested files to the receivers. We then provide a lower bound
on the value of this optimization problem, hence leading to an
upper bound on the linear one-shot sum-DoF of the network,
which is within a factor of 2 of the achievable sum-DoF.
Index Terms— Interference management, caching, wireless net-
works, degrees of freedom.
I. I NTRODUCTION
O VER the last decade, video delivery has emerged as
the main driving factor of the wireless traffic. In this
context, there is often a large library of pre-recorded content
(e.g. movies), out of which, users may request to receive a
specific file. One way to reduce the burden of this traffic is to
Manuscript received April 20, 2016; revised November 28, 2016; accepted
January 30, 2017. Date of publication February 16, 2017; date of current
version April 19, 2017. This work was supported in part by the NSF CAREER
under Grant 1408639, in part by the NSF NETS under Grant 1419632, in part
by the NSF EARS under Grant 1411244, and in part by the ONR Award under
Grant N000141612189. This paper was presented in part at the 2016 IEEE
International Symposium on Information Theory [1].
N. Naderializadeh and A. S. Avestimehr are with the Department of
Electrical Engineering, University of Southern California, Los Angeles,
CA 90007 USA (e-mail: naderial@usc.edu; avestimehr@ee.usc.edu).
M. A. Maddah-Ali is with Nokia Bell Labs, Holmdel, NJ 07733 USA
(e-mail: mohammad.maddah-ali@nokia.com).
Communicated by O. Simeone, Associate Editor for Communications.
Digital Object Identifier 10.1109/TIT.2017.2669942
0018-9448 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO.
employ memories distributed across the networks and closer
to the end users to prefetch some of the popular content. This
can help system to deliver the content with higher throughput
and less delay.
As a result, there have been significant interests in
both academia and industry in characterizing the impact of
caching on the performance of communication networks (see,
e.g. [2]–[18]). In particular, in a network with only one trans-
mitter broadcasting to several receivers, it was shown in [3]
that local delivery attains only a small fraction of the gain
that caching can offer, and by designing a particular pattern in
cache placement at the users and exploiting coding in delivery,
a significantly larger global throughput gain can be achieved,
which is a function of the entire cache throughout the network.
This also demonstrates that the gain of caching scales with
the size of the network. As a follow-up, this work has been
extended to the case of multiple transmitters in [4], where it
was shown that the gain of caching can be improved if several
transmitters have access to the entire library of files. Caching at
the transmitters was also considered in [5] and used to induce
collaboration between transmitters in the network. It is also
shown in [6], [7] that caches at the transmitters can improve
load balancing and increase the opportunities for interference
alignment. In particular, in [7], the authors consider wireless
networks with caches only at the transmitters’ side and in
this setting, they provide a lower bound on the normalized
delivery time and show the tightness of their bound for
several system parameters. More recently, the authors in [8]
evaluated the performance of cellular networks with edge
caching via a hypergraph coloring problem. In [9], the authors
studied the problem of maximizing the delivery rate of a
fog radio access network for arbitrary prefetching strategies.
In [10], the authors study the multiplexing gain of receiver
caching in Wyner’s networks. Furthermore, the authors in [11]
consider cloud radio access networks with caches at the
base stations and demonstrate the opportunities that coded
caching can provide in terms of reducing the backhaul
cost.
In this paper, we consider a general network setting with
caches at both transmitters and receivers, and demonstrate how
one can utilize caches at both transmitters and receivers to
manage the interference and enhance the system performance
in the physical layer. In particular, we consider a library of
N files and a wireless network with K T transmitters and
K R receivers, in which each transmitter and each receiver is
equipped with a cache memory of a certain size. In particular,
each transmitter and each receiver can cache up to MT and
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT 3093

M R files, respectively. The system operates in two phases.

The first phase is called the prefetching phase, where each
cache is populated up to its limited size from the content of the
library. This phase is followed by a delivery phase, where each
user reveals its request for a file in the library. The transmitters
then need to deliver the requested files to the receivers. Note
that in the prefetching phase, the system is still unaware of the
files that the receivers will request in the delivery phase. The
goal is to design the cache contents in the prefetching phase
and communication scheme in the delivery phase to achieve
the maximum throughput for arbitrary set of requested files.
Due to their practical appeal, in this work we focus on one-shot
linear delivery strategies. Interestingly, many of the previous
works on caching have relied on one-shot schemes for content
delivery (see, e.g. [5], [8]).
Our main result in this paper is the characterization of Fig. 1. Wireless network with K T transmitters and K R receivers, where each
transmitter and each receiver caches up to MT F packets and M R F packets,
the one-shot linear sum degrees-of-freedom (sum-DoF) of the respectively, from a library of N files, each composed of F packets.
network, i.e., number of the receivers that can be served
interference-free simultaneously, within a factor of 2 for all
system parameters.  In fact, we show  that the one-shot linear an integer optimization problem for the minimum number
sum-DoF of min K T MT +K N
R MR
, K R is achievable, and this is of communication blocks needed to deliver a fixed set of
requests for a given caching realization. We then show how
within a factor of 2 of the optimum. This result shows that the
we can focus on average demands instead of the worst-case
one-shot linear sum-DoF of the network grows linearly with
demands to derive an outer optimization problem on the
the aggregate cache size in the network (i.e., the cumulative
number of communication blocks optimized over the caching
memory available at all nodes). It also implies that caches
realizations. Finally, we present a lower bound on the value of
at the transmitters’ side are equally valuable as the caches
the aforementioned optimization problem, which leads to the
on the receivers’ side in the one-shot linear sum-DoF of the
desired upper bound on the one-shot linear sum-DoF of the
network. Our result, therefore, establishes a fundamental limit
network. This result illustrates that in this setting, caches at
on the performance of one-shot delivery schemes for cache-
transmitters’ side are equally valuable as caches at receivers’
aided interference management.
side. It also shows that caching offers a throughput gain that
To achieve the aforementioned sum-DoF, we propose a
scales linearly with the size of the network.
particular pattern in cache placement so that each piece of each
The rest of the paper is organized as follows. We present
file in the library is available in the caches of K TNMT transmit-
the problem formulation in Section II. We state the main
ters and K RNM R receivers. Once caching is done this way, we result in Section III. We prove the achievability of our main
can show that for delivering any set of requested contents to result in Section IV and the converse in Section V. Finally, in
the receivers, min K T MT +K N
R MR
, K R of the receivers can be Section VI, we conclude the paper and discuss some of the
served at each time, interference-free. This gain is achieved concurrent, independent works on topics similar to this paper.
by simultaneously exploiting the opportunity of collaborative
interference cancellation (i.e., zero-forcing) at the transmitters’ II. P ROBLEM F ORMULATION
side and opportunity of eliminating known interference contri- In this section, we first provide a high-level description of
butions at the receivers’ side. The first opportunity is created the problem setting and the main parameters in the system
by caching the pieces of each file at several transmitters. The model, and then we present a detailed description of the
second opportunity is available since pieces of a file requested problem formulation.
by one receiver have been cached at some other receivers and
thus do not impose interference at those receivers effectively. A. Problem Overview
Our proposed cache placement pattern maximizes the overall Consider a wireless network, as illustrated in Figure 1, with
gain achieved by these opportunities for any arbitrary set of K T transmitters and K R receivers, and also a library of N files,
receiver requests and this gain can be achieved even with a each of which contains F packets, where each packet is a
simple one-shot linear delivery scheme. vector of B bits. Each node in the network is equipped with
Moreover, we demonstrate that our achievable sum-DoF a local cache memory of a certain size that can be used to
is within a factor of 2 of the optimal sum-DoF for one- cache contents arbitrarily from the library before the receivers
shot linear schemes. To prove the outer bound, we take a reveal their requests and communication begins. In particular,
four-step approach in order to lower bound the number of each transmitter and each receiver is equipped with a cache
communication blocks needed to deliver any set of requested of size MT F and M R F packets, respectively.
files to the receivers. First, we show that the network can We assume that the system operates in two phases, namely
be converted to a virtual MISO interference channel in each the prefetching phase and the delivery phase. In the prefetch-
block of communication. Using this conversion, we next write ing phase, each node can cache contents arbitrarily from the

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3094
library subject to its cache size constraint. In particular, each
transmitter selects up to MT F packets out of the entire library
to store in its cache, and each receiver selects up to M R F
packets out of the entire library to store in its cache. In the
delivery phase, each receiver requests an arbitrary file from
the library. Since each receiver may have cached parts of its
desired file in the prefetching phase, the transmitters need to
deliver the rest of the requested packets to the receivers over
the wireless channel.
We assume that at each time, the transmitters employ a one-
shot linear scheme, where a subset of requested packets are
selected to be delivered interference-free to a corresponding
subset of receivers. Each transmitter transmits a linear combi-
nation of the subset of the selected packets which it has cached
in the prefetching phase. The interference is cancelled with the
aid of cached contents as follows. Since each requested packet
may be cached at multiple transmitters, the transmitters can
collaborate in order to zero-force the outgoing interference of
that packet at some of the unintended receivers. Moreover, the
receivers can also use their cached packets as side information
to eliminate the remaining incoming interference from to unde-
sired packets. Our objective is to design a cache placement
scheme and a delivery scheme which maximize the number
of packets that can be delivered at each time interference-free.
In this setting, we define the one-shot linear sum-degrees
of freedom as the ratio of the number of delivered packets
over the number of blocks needed for communicating those
packets for any set of receiver demands. Finally, we define
the one-shot linear sum-DoF of the network, denoted by
DoF∗L,sum (N, MT , M R ), as the maximum achievable one-shot
linear sum-DoF over all caching realizations.
B. Detailed Problem Description
We consider a discrete-time additive white Gaussian noise
channel, as illustrated in Figure 1, with K T transmitters
KT KR
denoted by {Txi }i=1 and K R receivers denoted by {Rxi }i=1 .
The communication at time t over this channel is modeled by

KT
Y j (t) = h j i X i (t) + Z j (t),
i=1
where X i (t) ∈ C denotes the signal transmitted by Txi , i ∈
[K T ]  {1, . . . , K T } and Y j (t) denotes the receive signal by
Rx j , j ∈ [K R ]. Moreover, h j i ∈ C denotes the channel gain
from Txi to Rx j , assumed to stay fixed over the course of
communication, and Z j (t) denotes the additive white Gaussian
noise at Rx j at time slot t, distributed as CN (0, 1). The
 at Tx i , i ∈ [K T ] is subject to the power
transmit signal
constraint E |X i (t)|2 ≤ P.
We assume that each receiver will request an arbitrary file
out of a library of N files {Wn }n=1 N , which should be delivered
by the transmitters. Each file Wn in the library contains F
packets {wn, f } Ff=1 , where each packet is a vector of B bits;
i.e., wn, f ∈ F2B . Furthermore, we assume that each node in the
network is equipped with a cache memory of a certain size that
can be used to cache arbitrary contents from the library before
the receivers reveal their requests and communication begins.
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017
In particular, each transmitter and each receiver is equipped
with a cache of size MT F and M R F packets, respectively.
We assume that the network operates in two phases, namely
the prefetching phase and the delivery phase, which are
described in more detail as follows.
Prefetching Phase: In this phase, each node can store an
arbitrary subset of the packets from the files in the library
up to its cache size. In particular, each transmitter Txi
chooses a subset Pi of the N F packets in the library, where
|Pi | ≤ MT F, to store in its cache. Likewise, each receiver
Rx j stores a subset Q j of the packets in the library, where
|Q j | ≤ M R F. Caching is done at the level of whole packets
and we do not allow breaking the packets into smaller sub-
packets. Also, we assume that all the transmitters and receivers
use uncoded prefetching in the sense that they are not allowed
to cache combinations of multiple packets in the library as a
single packet. Note that this phase takes place unaware of the
receivers’ future requests.
Delivery Phase: In this phase, each receiver Rx j , j ∈ [K R ],
reveals its request for an arbitrary file Wd j from the library
for some d j ∈ [N]. We let d = [d1 . . . d K R ]T denote the
demand vector. Depending on the demand vector d and the
cache contents, each receiver has already cached some packets
of its desired file and there is no need to deliver them. The
transmitters will be responsible for delivering the rest of the
requested packets to the receivers. In order to make sure that
any piece of content in the library is stored at the cache of
at least one transmitter in the network, we assume that the
transmitter cache size satisfies K T MT ≥ N.
Each transmitter first employs a random Gaussian coding
scheme ψ : F2B → C B̃ of rate log P + o(log P) to encode
each of its cached packets into a coded packet composed of
B̃ complex symbols, so that each coded packet carries one
degree-of-freedom (DoF). We denote the coded version of each
packet wn, f in the library by w̃n, f  ψ(wn, f ). Afterwards,
the communication takes place over H blocks, each of length
B̃ time slots. In each block m ∈ [H ], the goal is to deliver a
subset of the requested packets, denoted by Dm , to a subset
of receivers, denoted by Rm , such that each packet in Dm is
(1) intended to exactly one receiver in Rm . In addition, the set of
transmitted packets in all blocks and the cache contents of the
receivers should satisfy
 H

{wd j , f } f =1 ⊂
F
Dm ∪ Q j , ∀ j ∈ [K R ], (2)
m=1
which implies that for any receiver Rx j , j ∈ [K R ], each of its
requested packets should be either transmitted in one of the
blocks or already stored in its own cache.
In each block m ∈ [H ], we assume a one-shot linear
scheme where each transmitter transmits an arbitrary linear
combination of a subset of the coded packets in Dm that it
has cached. Particularly, Txi , i ∈ [K T ] transmits xi (m) ∈ C B̃ ,
where

xi (m) = v i,n, f (m) w̃n, f , (3)
(n, f ):
wn, f ∈Pi ∩Dm
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT 3095

and v i,n, f (m)’s denote the complex beamforming coefficients III. M AIN R ESULT AND I TS I MPLICATIONS
that Txi uses to linearly combine its coded packets in block m. In this section, we present our main result on the one-shot
On the receivers’ side, the received signal of each receiver linear sum-DoF of the network and its implications.
Rx j ∈ Rm in block m, denoted by y j (m) ∈ C B̃ , can be written Theorem 1: For a network with a library of N files, each
as containing F packets, and cache size of MT and M R files at

KT each transmitter and each receiver, respectively, the one-shot
y j (m) = h j i xi (m) + z j (m), (4) linear sum-DoF of the network, as defined in Definition 1,
i=1 satisfies
where z j (m) ∈ C B̃ denotes the noise vector at Rx j in block m.  
min K T MT +K
N
R MR
, K R ≤ DoF∗L,sum (N, MT , M R )
Then, receiver Rx j will use the contents of its cache to cancel  
(subtract out) the interference of some of undesired packets ≤ min 2 K T MT +K
N
R MR
, K R , (8)
in Dm , if they exist in its cache. In particular, each receiver
Rx j ∈ Rm , forms a linear combination L j,m , as for sufficiently large F.
In the following, we highlight the implications of Theorem 1
L j,m (y j (m), Q̃ j ) (5)
and its connections to some prior works:
to recover w̃d j , f ∈ Dm , where Q̃ j denotes the set of coded 1) (Within a factor of 2 characterization) The upper bound
packets cached at receiver Rx j . in (8) is within a factor of 2 of the lower bound in (8).
The communication in block m ∈ H to transmit the packets Therefore, Theorem 1 characterizes the one-shot linear
in Dm is successful, if there exist linear combinations (3) at sum-DoF of a cache-aided wireless network to within a
the transmitters’ side and (5) at receivers’ side, such that for factor of 2, for all system parameters.
all Rx j ∈ Rm , 2) (Aggregate cache size matters) The one-shot linear sum-
L j,m (y j (m), Q̃ j ) = w̃d j , f + z j (m). (6) DoF characterized in Theorem 1 is proportional to the
aggregate cache size that is available throughout the
The channel created in (6) is a point-to-point channel, whose network, even-though these caches are isolated.
capacity is log P + o(log P). Hence, since each coded packet 3) (Equal contribution of transmitter and receiver caches)
w̃d j , f is coded with rate log P + o(log P), it can be decoded Perhaps interestingly, the caches at both sides of the
with vanishing error probability as B increases. We assume network, i.e., the transmitters’ side and the receivers’
that the communication continues for H blocks until all the side, are equally valuable in the achievable one-shot linear
desired packets are successfully delivered to all receivers. sum-DoF of the network. Note that in practice, size of
Since each packet carries one degree-of-freedom, the one- each transmitter’s cache, MT , could be large. However,
shot linear sum-degrees-of-freedom (sum-DoF) of |Dm | is the number of transmitters (e.g., base stations) K T is
achievable in each block m ∈ [H ]. This implies that through- often small. On the other hand, size of the cache M R
out the H blocks of communication, the one-shot linear sum- at the receivers (e.g., cellphones) is small, whereas the
m=1 Dm
H
DoF of is achievable. Therefore, for a given caching number of receivers K R is large. Therefore K T MT could
H
realization, we define the one-shot linear sum-DoF to be be comparable with K R M R . Our result in Theorem 1
maximum achievable one-shot linear sum-DoF for the worst shows that neither caches at the transmitters nor caches
case demands; i.e., at the receivers should be ignored.
H
4) (Linear scaling of DoF with network size) Letting K T =
KT KR Dm K R = K , we observe that the one-shot linear sum-
{Pi }i=1 ,{Q j } j =1
DoFL,sum = inf sup
m=1
. (7) DoF scales linearly with the number of users in a
d H,{Dm }m=1
H H fully-connected interference channel. Note that without
caches, the one-shot linear sum-DoF of a fully-connected
This leads us to the definition of the one-shot linear sum- interference channel is bounded by 2, as shown in [19].
DoF of the network as follows. Hence, caching enables linear growth of the DoF without
Definition 1: For a network with a library N files, each the need for more complex physical layer schemes.
containing F packets, and cache size of MT and M R files at 5) (Role of transmitter and receiver caches) As we will show
each transmitter and receiver, respectively, we define the one- in Section IV, in (8), K TNMT represents the contribution
shot linear sum-DoF of the network as the maximum achiev- of collaborative zero-forcing at the transmitters’ side,
able one-shot linear sum-DoF over all caching realizations; and K RNM R represents the gain of canceling the known
i.e., interference at the receivers’ side.
K K
{Pi }i=1
T ,{Q } R
j j =1 6) (Connection to single-server coded caching [3]) A special
DoF∗L,sum (N, MT , M R ) = sup DoFL,sum case of our network model is the case with a single
K K
{Pi }i=1
T ,{Q } R
j j =1 transmitter, which was previously considered in [3].
s.t. |Pi | ≤ MT F, ∀i ∈ [K T ]  case, it can
In this  be shown that a sum-DoF of
|Q j | ≤ M R F, ∀ j ∈ [K R ], min 1 + K RNM R , K R is achievable, which is equivalent
K K to the global caching gain introduced in [3], indicating
{Pi }i=1
T ,{Q } R
j j =1
where DoFL,sum is defined in (7). the number of receivers in the network that can be served

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3096
simultaneously, interference-free. Hence, our result sub-
sumes the result of [3] by generalizing it for the case of
multiple transmitters.
7) (Connection to multi-server coded caching [4]) Another
special case of our network model is the case where
each transmitter has space to cache the entire library;
i.e., MT = N. This case was previously considered
in [4] and it can be verified that in this case, a sum-
DoF of min K T + K RNM R , K R is achievable. Hence, our
result can also be viewed as a generalization of the result
in [4] where the cache size of each transmitter may be
arbitrarily smaller than the entire library size.
Remark 1: In practice, the files in the library have nonuni-
form demands and some of them are more popular than the
rest. In this case, our algorithm can be used to cache and
deliver the N most popular files. If a user requests one of
the remaining less popular files, it can be directly served by
a central base station. The parameter N can be tuned, based
on the popularity pattern of the contents, in order to attain the
best average performance.
Example 1: As an illustrative example, consider a cellu-
lar network with 5 base stations as transmitters, each with
a 10 TB memory and 100 cellphones as receivers, each
with a 32 GB memory. Moreover, consider a library of
the 1000 most popular movie titles on Netflix, each with size
of 5 GB. Then, Theorem 1 implies that at each time, around 11
cellphones can be served simultaneously interference-free, no
matter what their demands are, in contrast to the naive time-
sharing scheme, where at each time only 1 cellphone can be
served. 
The rest of the paper is devoted to the proof of Theorem 1.
In particular, we illustrate the achievable scheme in Section IV
and we present the converse argument in Section V.
IV. ACHIEVABLE S CHEME
In this section, we prove the achievability of Theorem 1
by presenting an achievable scheme which utilizes the caches
at the transmitters and receivers efficiently to exploit the
zero-forcing and interference cancellation opportunities at the
transmitters’ and receivers’ sides, respectively. In particular,
we introduce a prefetching strategy which maximizes the gains
attained by the aforementioned opportunities in the delivery
phase, no matter what the receiver demands are.
We first explain our achievable scheme through a simple,
illustrative example and then proceed to mention our general
achievable scheme.
A. Description of the Achievable Scheme via an Example
Consider a system with K T = 3 transmitters and K R = 3
receivers, where each transmitter has space to cache MT = 2
files and each receiver has space to cache M R = 1 file. The
library has N = 3 files W1 = A, W2 = B, and W3 = C, each
consisting of F packets.
In the following, we will describe the prefetching and
delivery phases in detail.
Prefetching Phase: In this phase, each file Wn , n ∈ [3] in
  
the library is broken into 32 31 = 9 disjoint subfiles Wn,T ,R
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017
for any T ⊆ [K T ] = [3] and R ⊆ [K R ] = [3] such that
|T | = 2 and |R| = 1, where each subfile consists of F/9
packets. Each subfile Wn,T ,R is then stored at the caches of
the two transmitters in T and the single receiver in R. For
example, file A is broken into 9 subfiles as follows:
A12,1 , A12,2 , A12,3 , A13,1 , A13,2 , A13,3, A23,1 , A23,2 , A23,3,
where A12,1 is stored at transmitters Tx1 and Tx2 as well as
receiver Rx1 , A12,2 is stored at transmitters Tx1 and Tx2 as
well as receiver Rx2 , etc. We do the same partitioning for
files B and C, as well.
It is easy to verify that each transmitter caches 6 sub-
files of each file, hence the total size of its cached content
is 3 ∗ (6 ∗ F/9) = 2F packets which satisfies its memory
constraint. Also, each receiver caches 3 subfiles of each file
and its total cached content has size 3∗(3∗ F/9) = F packets,
hence satisfying its memory constraint. Note that in this phase,
we are unaware of receivers’ future requests.
Delivery Phase: In this phase, each receiver reveals its
request for a file in the library. Without loss of general-
ity, assume that receivers Rx1 , Rx2 and Rx3 request files
Wd1 = A, Wd2 = B and Wd3 = C, respectively. Note that
each receiver has already stored 3 subfiles of its desired file in
its own cache, and therefore the transmitters need to deliver the
6 remaining subfiles of each requested file. In particular, the
following 18 subfiles need to be delivered by the transmitters
to the requesting receivers:
A12,2 , A12,3 , A13,2 , A13,3, A23,2 , A23,3 to receiver Rx1 ,
B23,3, B13,1 , B12,3, B23,1 , B13,3, B12,1 to receiver Rx2 ,
C13,1 , C23,2 , C23,1 , C12,2 , C12,1 , C13,2 to receiver Rx3 . (9)
We now show that we can break the 18 subfiles in (9)
into 6 sets, each containing 3 subfiles, such that the subfiles
in each set can be delivered simultaneously to the receivers,
interference-free. Such a partitioning is illustrated through
the 6 steps in Figure 2, where each step takes F9 blocks.
In each step, 3 subfiles are delivered to all the receivers
simultaneously, while all the inter-user interference can be
eliminated. For example, in the first step, as in Figure 2–(a),
subfiles A12,2 , B23,3, C13,1 are respectively delivered to
receivers Rx1 , Rx2 , and Rx3 at the same time. In Figure 3, we
show in detail how the interference is cancelled in this step.
The transmit signals of transmitters Tx1 , Tx2 and Tx3 can be
respectively written as
X 1 = − h 32 Ã12,2 + h 23 C̃13,1 ,
X 2 = h 31 Ã12,2 − h 13 B̃23,3,
X 3 = − h 21 C̃13,1 + h 12 B̃23,3,
where for any subfile Wn,T ,R , W̃n,T ,R denotes its coded
version. For simplicity, in this example, we ignore the power
constraint at the transmitters. On the other hand, the received
signals by receivers Rx1 , Rx2 and Rx3 can be respectively
written as
Y1 = (h 12 h 31 − h 11 h 32 ) Ã12,2 + (h 11 h 23 − h 13 h 21 )C̃13,1 + Z 1 ,
Y2 = (h 23 h 12 − h 22 h 13 ) B̃23,3 + (h 22 h 31 − h 21 h 32 ) Ã12,2 + Z 2 ,
Y3 = (h 31 h 23 − h 33 h 21 )C̃13,1 + (h 33 h 12 − h 32 h 13 ) B̃23,3 + Z 3.
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT 3097

Fig. 2. Delivery phase for the example in Section IV-A for respective requests of files A, B and C by receivers Rx1 , Rx2 and Rx3 , where L(α, β)
denotes some linear combination of α and β. In every step, each pair of transmitters collaborate to zero-force the interference due to a specific subfile at a
certain undesired receiver. Moreover, each receiver also uses its cache contents to cancel the interference due to the other interfering packet. Therefore, the
communication is interference-free in all 6 steps.

Fig. 3. More detailed description of the linear encoding and decoding schemes used in the delivery phase step in Figure 2–(a). In this step, Tx1 and Tx2
zero-force A12,2 at Rx3 , Tx1 and Tx3 zero-force C13,1 at Rx2 , and Tx2 and Tx3 zero-force B23,3 at Rx1 . Moreover, Rx1 , Rx2 and Rx3 can cancel the
interference due to C13,1 , A12,2 , and B23,3 , respectively, since they already have each respective subfile in their own cache.

Now, note that receivers Rx1 , Rx2 and Rx3 can cancel B. Description of the General Achievable Scheme
the interference due to C13,1 , A12,2 , and B23,3, respectively, Our general achievable scheme is given in Algorithm 1.
since they already have each respective subfile in their own In this algorithm, we use the notation
cache. Therefore, all the interference in the network can be
K T MT K R MR
effectively eliminated and the receivers will be able to decode tT  , tR  , (10)
their desired subfiles. Likewise, one can verify that all the N N
receivers can receive their desired subfiles interference-free in and for now, we assume that tT and t R are integers. Recall
all the 6 steps of communication depicted in Figure 2. that in the example in Section IV-A, tT = 2 and t R = 1.
Consequently, the 18 subfiles in (9), each of which consists In the following, we will describe the prefetching and
of F/9 packets, are delivered to the receivers in 6 steps, each delivery phases in more detail.
consisting of F/9 blocks. Note that our particular file splitting 1) Prefetching Phase: For any file Wn in the library,
  
pattern in the prefetching phase and the particular scheduling n ∈ [N], we partition it into KtTT Kt RR disjoint subfiles of
pattern in the delivery phase allows us to maximally exploit equal sizes,1 denoted by
the two gains of zero-forcing the outgoing interference on the  
transmitters’ side and canceling the known interference on the Wn = Wn,T ,R T ⊆[K T ]:|T |=tT . (11)
receivers’ side, no matter what the receiver demands are in R⊆[K R ]:|R|=t R
the delivery phase. Therefore, the sum-DoF of 18∗F/9 6∗F/9 = 3 =
  1 Due to the assumption that F is sufficiently large, we can assume that it
  
K T MT +K R M R
min N , K R is achievable in this network. is an integer multiple of Kt T Kt R .
T R

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3098 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017

Algorithm 1 Achievable Scheme for Theorem 1

Prefetching Phase:
1: for n = 1, . . . , N  K  K   
2: Partition Wn into t T t R disjoint subfiles Wn,T ,R T ⊆[K ],|T |=t ,R⊆[K ],|R|=t of equal sizes.
T R T T R R
3: end
4: for i = 1, . . . , K T
5: Txi caches all Wn,T ,R for which i ∈ T .
6: end
7: for j = 1, . . . , K R
8: Rx j caches all Wn,T ,R for which j ∈ R.
9: end

Delivery Phase:
10: for j ∈ [K R ]
11: for T ⊆ [K T ]s.t. |T | = tT
12: for R ⊆ [K R ] \ { j }s.t. |R| = t R  
13: partition Wd j ,T ,R to t R[K![K R −(t R +1)]!
R −(t R +tT )]!
disjoint subfiles Wd j , T ,π,π π∈R of equal sizes.
π ∈[K R ]\(R∪{ j }),tT −1
14: end
15: end
16: end
17: for T ⊆ [K T ]s.t. |T | = tT
18: for R ⊆ [K R ]s.t. |R| = tT + t R
19: for π ∈ circ
R
20: Each transmitter Txi transmits a linear combination of the coded subfiles as in 
X i = Li,T ,π W̃dπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] : l ∈ [tT + t R ], i ∈ T ⊕ K T (l − 1)
using
 the linear combinations shown in Lemma 2 such that the  subfiles
Wdπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] : l ∈ [tT + t R ]
are simultaneously delivered to the receivers in R interference-free.
21: end
22: end
23: end

Based on the above partitioning, in the prefetching phase, Likewise, in the prefetching phase, each receiver Rx j stores
each transmitter Txi stores a subset Pi of the packets in the a subset Q j of the packets in the library as described below.
library as described below.  
  Q j = Wn,T ,R : j ∈ R . (13)
Pi = Wn,T ,R : i ∈ T . (12)
Illustration: For instance, in the example network consid- Illustration: For instance, in the example network consid-
ered in Section IV-A, transmitter Tx3 stores the following ered in Section IV-A, receiver Rx2 stores the following subset
subset of packets in its cache. of packets in its cache.
P3 = {W1,13,1, W1,13,2 , W1,13,3, W1,23,1 , W1,23,2, W1,23,3 , Q2 = {W1,12,2 , W1,13,2, W1,23,2 ,
W2,13,1 , W2,13,2 , W2,13,3 , W2,23,1, W2,23,2 , W2,23,3 ,
W2,12,2 , W2,13,2 , W2,23,2 ,
W3,13,1, W3,13,2 , W3,13,3 , W3,23,1, W3,23,2 , W3,23,3}
= {A13,1, A13,2 , A13,3 , A23,1, A23,2 , A23,3 , W3,12,2 , W3,13,2 , W3,23,2}
B13,1, B13,2 , B13,3 , B23,1, B23,2 , B23,3, = {A12,2 , A13,2 , A23,2 ,
C13,1 , C13,2 , C13,3 , C23,1 , C23,2 , C23,3 }.  B12,2 , B13,2 , B23,2,
C12,2 , C13,2 , C23,2 }. 
Based on the above caching strategy, we can verify that the
total number of packets cached by transmitter Txi equals
   This suggests that the total number of packets cached by
KT − 1 K R F tT receiver Rx j is equal to
N  K T  K R  = N F = MT F packets,
tT − 1 tR KT   
tT tR
 −1 KT KR − 1 F tR
hence satisfying its memory size constraint, where KtTT−1 N  K T  K R  = N F = M R F packets,
is tT tR − 1 KR
tT tR
the number of subsets T ⊆ [K T ] of size tT which include the
transmitter index i . which also satisfies its memory size constraint.

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT 3099

2) Delivery Phase: In this section, we first describe the
delivery phase for the case where tT + t R ≤ K R , so that the
first term in the lower bound in (8) is dominant. We will later
show how to deal with the case where tT + t R > K R .
In the delivery phase, the receiver requests are revealed,
and in particular, each receiver Rx j , j ∈ [K R ] requests a file
Wd j from the library and the transmitters need to deliver the
subfiles in
Finally, for a set T and an integer j , we let T ⊕m j denote
entry-wise addition of elements of T with j modulo m, as
defined in (15).
Lemma 1: Given the prefetching phase in Section IV-B1,
for any receivers’ demand vector d, the set of subfiles which
need to be delivered to the receivers can be partitioned into
disjoint subsets of size tT + t R as


{Wd j ,T ,R : j ∈
/ R} T ⊆[K T ]:|T |=tT
R⊆[K R ]:|R|=tT +t R
to receiver Rx j ; i.e., the subfiles of file Wd j which have not π∈circ
R

been already stored in the cache of receiver Rx j . × Wdπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] :
In the following, our goal is to show that the set of packets 
which need to be delivered to the receivers can be partitioned l ∈ [tT + t R ] . (16)
into subsets of size tT + t R such that the packets in each Proof: See Appendix A. 
subset can be scheduled together. To this end, we need to Illustration: For the example network mentioned in
further break each subfile to smaller subfiles. In particular, for Section IV-A, the set of 18 subfiles which need to be delivered
any j ∈ [K R ], T ⊆ [K T ] s.t. |T | = tT , R ⊆ [K R ] \ { j } s.t. to the receivers, as in (9), can be partitioned to the following
|R| = t R , we partition Wd j ,T ,R to t R[K![K R −(t R +1)]!
R −(t R +tT )]!
smaller 6 sets.
disjoint subfiles of equal sizes denoted by
  {A12,2, B23,3 , C13,1 } ∪ {A12,3, B13,1 , C23,2 }
Wd j ,T ,R = Wd j ,T ,π,π π∈R , (14) ∪{A13,2 , B12,3, C23,1 } ∪ {A13,3, B23,1, C12,2 }
π ∈[K R ]\(R∪{ j }),tT −1
∪{A23,2 , B13,3, C12,1 } ∪ {A23,3, B12,1, C13,2 }. (17)
where for a set S, S denotes the set of permutations of
Based on the partitioning of the small subfiles that need
S, and for any t ∈ {1, . . . , |S|}, S ,t denotes the set of all to
 K Tbe
 delivered
KR 
to the receivers in Lemma 1, we will have
permutations of all subsets of S of size t; i.e.,
tT tT +t R (tT + t R − 1)! steps of communication, where at
 each step, specific sets T and R and a permutation π are
S ,t = A . fixed as in (16), and each transmitter Txi will transmit a linear
A⊆S ,|A|=t combination of the coded subfiles for which i ∈ T ⊕ K T (l −1);
Remark 2: Note that in the example setting discussed in i.e.,
Section IV-A, t R[K![K R −(t R +1)]!
= 1, which implies that further 
R −(t R +tT )]!
X i = Li,T ,π W̃dπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] :
partitioning of the subfiles is not needed.
The advantage of further breakdown of the subfiles in (14) is 
that we can now partition the set of the subfiles which need to l ∈ [tT + t R ], i ∈ T ⊕ K T (l − 1) , (18)
be delivered to the receivers into certain subsets of size tT +t R
such that each subfile Wd j ,T ,π,π intended for receiver Rx j is where for any subfile Wd j ,T ,π,π , W̃d j ,T ,π,π denotes the
zero-forced at the receivers with indices in π . Moreover, since corresponding coded subfile containing PHY coded symbols,
this subfile is also already cached at the receivers with indices and Li,T ,π (.) represents the linear combination that transmitter
in π, the communication will be interference-free for each set Txi chooses for sending the subfiles in (18).
of the tT + t R subfiles. We will next show that under such a delivery scheme,
We show how to do such a partitioning in Lemma 1. there always exists a choice of linear combinations at the
In this lemma, we use the following notation: For a set R, transmitters so that at each step, the communication will be
R denote the set of (|R| − 1)! circular permutations
we let circ interference-free and all the tT + t R receivers in R can decode
of R.2 Moreover, for a set S, a permutation π ∈ S and two their desired packets, as we also showed in the example setting
integers i, j satisfying j ≥ i , we define π[i : j ] as in Section IV-A.
Lemma 2: For any subset of tT transmitters T ⊆ [K T ],
π[i : j ] = [π(i ⊕|S | 0) π(i ⊕|S | 1) π(i ⊕|S | 2) . . .
any subset of tT + t R receivers R ⊆ [K R ], and any circular
π(i ⊕|S | ( j − i ))], permutation π ∈ circ R , there exists a choice of the linear
KT
where for an integer m, i ⊕m j is defined as combinations {Li,T ,π (.)}i=1 in (18) such that the set of tT +t R
subfiles in
i ⊕m j = 1 + (i + j − 1 mod m). (15)  
Wdπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] :l ∈ [tT +t R ] ,
2 A circular permutation of a set R is a way of arranging the elements of
R around a fixed circle. The number of distinct circular permutations of a can be delivered simultaneously
 and interference-free
 by the
set R is equal to (|R| − 1)!. For example, if R = {1, 2, 3}, then circ
R = transmitters in T ⊕ K T (l − 1) to the receivers in R.
{[1, 2, 3], [1, 3, 2]}. l∈[tT +t R ]

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3100 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017

⎧ i
⎪ 
⎪
⎪ v i,l W̃dl ,{l,...,tT +l},{l+1,...,l+t R },{l+t R +1,...,l+t R +tT −1} , if i ∈ {1, . . . , tT − 1}
⎪
⎪
⎪
⎪l=1
⎨  i
Xi = v i,l W̃dl ,{l,...,tT +l},{l+1,...,l+t R },{l+t R +1,...,l+t R +tT −1} , if i ∈ {tT , . . . , tT + t R } (19)
⎪
⎪l=i−tT +1
⎪
⎪
⎪
⎪
+t R
tT
⎪
⎩ v i,l W̃dl ,{l,...,tT +l},{l+1,...,l+t R },{l+t R +1,...,l+t R +tT −1} , if i ∈ {tT + t R + 1, . . . , 2tT + t R − 1}
l=i−tT +1
+t R −1
2tT 
Yj = h ji Xi + Z j
i=1
T+j
t
= h j i v i, j W̃d j ,{ j,...,tT + j },{ j +1,..., j +t R },{ j +t R +1,..., j +t R +tT −1}
i= j
j +t
 T −1 t
T +l
+ h j i v i,l W̃dl ,{l,...,tT +l},{l+1,...,l+t R },{l+t R +1,...,l+t R +tT −1}
l= j +1 i=l
j −1 t
 T +l
+ h j i v i,l W̃dl ,{l,...,tT +l},{l+1,...,l+t R },{l+t R +1,...,l+t R +tT −1} + Z j (20)
l= j −t R i=l

Proof: For ease of notation and without loss of generality,
assume
T = {1, . . . , tT }
T ⊕ K T (l − 1) = { l, . . . , tT + l}
R = {1, . . . , tT + t R }
π = [1, . . . , tT + t R ].
First, we need to determine the subset of the subfiles which
is available at each transmitter. It is easy to verify that
• If i ∈ {1, . . . , tT − 1}, then transmitter Txi has subfiles
 i=l
W̃dπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] :

l ∈ {1, . . . , i } ;
• If i ∈ {tT , . . . , tT + t R }, then transmitter Txi has subfiles

W̃dπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] :
 exists a choice of linear combination coefficients {v i,l } such
l ∈ {i − tT + 1, . . . , i } ;
• and if i ∈ {tT + t R + 1, . . . , 2tT + t R − 1}, then transmitter
Txi has subfiles
 Remark 3: As mentioned in Section II, we assume that the
W̃dπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1] :
 channel gains remain constant over the course of communica-
l ∈ {i − tT + 1, . . . , tT + t R } .
Since each transmitter sends a linear combination of
the subfiles that it has, the transmit signal of transmit-
ter Txi can be written as in (19) at the top of this
page. This implies that the received signal at receiver Rx j ,
j ∈ {1, . . . , tT + t R } can be written as in (20) at the top of
this page.
Now, note that in (20), the first term corresponds to the
desired subfile of receiver Rx j , while the second and third
terms correspond to the undesired subfiles whose interference
needs to be canceled at this receiver. However, note that the
subfiles in the third term are already cached at receiver Rx j
and hence it is able to cancel their incoming interference.
Hence, in order for all receivers Rx j , j ∈ {1, . . . , tT + t R }
to receive their subfiles interference-free, there should exist a
choice of linear combination coefficients {v i,l } such that
T+j
t
h j i v i, j = 1, ∀ j ∈ {1, . . . , tT + t R } (21)
i= j
T +l
t
h j i v i,l = 0, ∀ j ∈ {1, . . . , tT + t R },
∀l ∈ { j + 1, . . . , j + tT − 1}. (22)
Equations (21)-(22) introduce a system of tT (tT + t R ) linear
equations. On the other hand, the number of variables {v i,l }
is also equal to tT (tT + t R ). This indicates that there always
that (21)-(22) are satisfied. Finally, note that by scaling all the
transmit signals by a large enough factor, the power constraint
at all the transmitters can also be satisfied. Hence the proof is
complete. 
tion. However, for the delivery scheme presented in the proof
of Lemma 2, this assumption can be relaxed, since we only
need the channel gains to remain unchanged for each block
of communication and they can be allowed to vary among
different blocks.
Remark 4: In the delivery scheme presented in the proof
of Lemma 2, we only used zero-forcing at the transmitters
in order to cancel their outgoing interference, which is DoF-
optimal. However, in general one can use any scheme that
exploits the collaboration among the transmitters in order to
optimize the actual rates in the finite-SNR regime (such as the
schemes suited for the MIMO broadcast channels [20]).

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT
C. Analysis of the Sum-DoF of the Proposed
Achievable Scheme
As a result of Lemmas 1 and 2, it is clear that for any set
of receiver demands in the delivery phase, we can schedule
all the requested subfiles in groups of size tT + t R . Now,
if tT and/or t R are not integers, we can split the memories
and the files proportionally so that for each new partition,
the aforementioned scheme can be applied for updated tT and
t R which are integers. Hence, combining the schemes over
different partitions allows us to serve tT + t R simultaneously,
interference-free, for any values of tT and t R such that
t T + t R ≤ K R .3
Finally, if tT + t R > K R , then since we cannot serve more
than K R receivers, we can neglect some of the caches at either
the transmitters’ side or the receivers’ side and use a fraction
of the caches with new sizes KNT ≤ MT ≤ MT and M R ≤ M R
K M +K M
so that T T N R R = K R . We can then use Algorithm 1 to
serve all the K R receivers simultaneously without interference.
As we showed in Section IV-B1, our prefetching phase
respects the cache size constraint of all the transmitters and
receivers. Moreover,
  given our prefetching phase, each receiver
−1
Rx j caches KtTT Kt RR−1 F
= MNR F packets of each
( tT )( Kt RR )
KT
file in the library. Hence, for each set of requested files by
the receivers, a total of K R 1 − MNR F packets need to be
delivered by the transmitters to the receivers.
Therefore, based on the delivery phase mentioned in
Section IV-B, the number of blocks required to deliver all
the K R 1 − MNR F packets to the receivers is equal to
MR
K R 1− F
N the optimal trade-off between DoF gain and complexity
min{tT +t R ,K R } . This suggests that for any set of receiver
demands, sum-DoF of
MR
KR 1 − N F
= min{tT + t R , K R }
M
K R 1− NR F In this section, we prove the converse of Theorem 1.
min{tT +t R ,K R }
 
K T MT + K R M R
= min , KR
N one-shot linear sum-DoF. In order to prove the converse, we
is achievable, hence completing the proof of achievability of
Theorem 1.
D. Complexity Analysis of the Proposed Achievable Scheme
In this section, we outline the complexity of the main
ingredients of the achievable scheme. First note that in the
prefetching phase,  each of the N files in the library is
partitioned into KtTT Kt RR disjoint subfiles of equal sizes. After-
   optimized over the caching realizations. Finally, we present
wards, the delivery phase consists of KtTT tTK+tR R (tT + t R − 1)!
steps of communication, in each of which tT + t R subfiles
are delivered to tT + t R receivers. Finally, for the decoding,
each of the
 KR receivers
 in the network receives a subfile
in (tT K+tRR )! KtTT tTK+tR R steps of communication, and at each
step it applies interference cancellation and then point-to-point
decoding in order to decode its desired subfiles.
3 In [3], this method is referred to as memory-sharing, which resembles
time-sharing in network information theory.
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3101
We will now point out the following two remarks regarding
the complexity of our achievable scheme.
1) For the scenarios inspired by cellular networks where
there are a few transmitters with large cache sizes and
many receivers with small cache sizes, the complexity
is not very high. For example, consider a network with
a library of 500 files, 5 base stations each capable of
caching 200 files and 50 receivers each capable of caching
10 files. In this case, it is easy to verify that each file is
partitioned to 500 subfiles in the prefetching phase and
then each subfile is subsequently partitioned to 48 subfiles
in the delivery phase. Such a partitioning can be easily
handled considering the fact that the size of each file can
be of the order of Gigabytes. Moreover, each receiver is
involved in 23520 communication steps in each of which
it decodes one of its desired subfiles.
2) For the network configurations in which the complexity
of file partitioning in the prefetching phase and the
number of communication steps in the delivery phase is
higher, one can break the network into smaller network
configurations (by partitioning the sets of transmitters
and receivers and/or the files and the memories) so that
for each subnetwork, the complexity is low enough.
We note that in each subnetwork, we can still achieve a
considerable DoF gain, which implies that our proposed
achievable scheme can achieve reasonable gains by taking
the feasible complexity into account. This provides us
with with the flexibility in terms of trading DoF gain
with complexity, so that we can achieve reasonable
gains within feasible ranges of complexity. Characterizing
requires new techniques which are beyond the scope of
this paper and we leave it as future work.
V. C ONVERSE
In particular, we show that the lower bound on the one-shot
linear sum-DoF in (8) is within a factor of 2 of the optimal
take four steps as detailed in the following sections. First, we
demonstrate how in each block of communication, the network
can be converted into a virtual MISO interference channel.
Second, we use this conversion to write an integer optimization
problem for the minimum number of communication blocks
needed to deliver a set of receiver demands for a given caching
realization. Third, we show how we can focus on average
demands instead of the worst-case demands to derive an outer
optimization problem on the number of communication blocks
a lower bound on the value of the aforementioned outer
optimization problem, which leads to the desired upper bound
on the one-shot linear sum-DoF of the network.
A. Conversion to a Virtual MISO Interference Channel
KT
Consider any caching realization {Pi }i=1 , {Q j } Kj =1
R
and
any demand vector d. As discussed in Section II, in each com-
munication blocks a subset of requested packets are selected
3102 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017

to be sent to a corresponding subset of distinct receivers. Now, since each of the vectors vl , l ∈ [L] is non-
Now, we can state the following lemma, which bounds the zero, it has at least one non-zero element at some position
number of packets that can be scheduled together in a single nl ∈ [|Tl |] which we denote by ql . Hence, using appropriate
communication block using a one-shot linear scheme. linear transformations on the elements of vl , it can be writ-
Lemma 3: Consider a single communication block where a ten as
set {wnl , fl }l=1
L of L packets are scheduled to be transmitted  
1
together to L distinct receivers. In order for each receiver to vl = ql Pl , (30)
v̄l
successfully decode its desired packet, the number of these
concurrently-scheduled packets should be bounded by where Pl is a |Tl | × |Tl | permutation matrix (which is derived
by swapping the first and the nlt h rows of the |Tl |×|Tl | identity
L ≤ min |Tl | + |Rl |, (23) matrix) and v̄l is a vector of size (|Tl | − 1) × 1. Also, for any
l∈[L]
two distinct pairs l = j , we can apply the inverse permutation
where for any l ∈ [L], Tl and Rl denote the set of trans-
Pl−1 on the channel gain vector h j l to get h̄ j l = Pl−1 h j l , and
mitters and receivers which have cached the packet wnl , fl ,
we can write h̄ j l as
respectively.  (1) 
Proof: For ease of notation and without loss of gen- h̄ j l
erality, suppose that in the considered block, L packets h̄ j l = (2) , (31)
h̄ j l
{w1,1 , . . . , w L ,1 } are scheduled to be sent to L receivers
{Rx1 , . . . , Rx L }, respectively. Each transmitter Txi , i ∈ [K T ] where h̄ (1) (2)
j l is a scalar and h̄ j l is of size (|Tl | − 1) × 1.
will transmit Therefore, the nulling condition in (28) can be rewritten as

xi = v i,l,1 w̃l,1 , (24)   
1
l: i∈Tl hTjl vl = (Pl h̄ j l )T ql Pl (32)
v̄l
where we have dropped the dependency on the block index,  
1
since we are focusing on a single block. On the other hand, = ql h̄Tjl PlT Pl (33)
v̄l
the received signal of receiver Rx j , j ∈ [L] can be written as 
! 1 
= ql h̄ (1) h̄ (2)T
(34)

KT jl jl v̄l
yj = h j i xi + z j (25) (1) (2)T
i=1
= ql h̄ j l + h̄ j l v̄l = 0, (35)

KT  where we have used the identity PlT Pl = I since the permu-
= h ji v i,l,1 w̃l,1 + z j (26)
tation matrix Pl is an orthogonal matrix.
i=1 l: i∈Tl
Now, since the packet sent by the virtual transmitter Tx l is

L 
available in the caches of at most |Rl | receivers in the network,
= h j i v i,l,1 w̃l,1 + z j . (27)
the interference of each transmitter should be nulled at least
l=1 i∈Tl
at L − |Rl | − 1 unintended receivers. This implies that the free
Therefore, (27) implies that we can effectively convert beamforming variables at transmitter l, i.e., v̄l , should satisfy
the network into a new MISO interference channel with L at least L − |Rl | − 1 linear equations in the form of (35). This
virtual transmitters {Txl } L , where Txl is equipped with is not possible unless the number of equations is no greater
l=1
|Tl | antennas, and L single-antenna receivers {Rx j } Lj=1, in than the number of variables, or
which each virtual transmitter Txl intends to send the coded
packet w̃l,1 to receiver Rxl . Each antenna in the new network L − |Rl | − 1 ≤ |Tl | − 1 ⇒ L ≤ |Tl | + |Rl |. (36)
corresponds to a transmitter in the original network. Hence, Since the above inequality holds for all l ∈ [L], the proof
the channel vectors are correlated in the new network. In fact, is complete. 
as (27) suggests, all the antennas corresponding to the same
transmitter in the original network have the same channel gain
vectors to the receivers in the new network. B. Integer Program Formulation
In the constructed MISO interference channel, we take a Equipped with Lemma 3, we define a set of packets Dm
similar approach as in [19] in order to bound the one-shot selected to be transmitted at block m to be feasible if its
linear sum-DoF of the network. Each virtual transmitter T xl size satisfies condition (23) in Lemma 3. We can then write
in the constructed MISO network will select a beamforming the following integer program (P1) to minimize the number
vector vl ∈ C|Tl |×1 (which consists of the coefficients chosen of required communication blocks for any given caching
by the original transmitters corresponding to its antennas) to realization and set of receiver demands:
transmit its desired symbol. Denoting the channel gain vector
between transmitter T xl and receiver Rx j as h j l ∈ C|Tl |×1 , the min H (P1-1)
decodability conditions can be written as 
H 
KR
 
s.t. Dm = Wd j \ Q j (P1-2)
hTjl vl = 0, ∀l = j s.t. j ∈
/ Rl (28) m=1 j =1
hTjj v j = 0, ∀ j ∈ [L]. (29) Dm is feasible, ∀m ∈ [H ], (P1-3)

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT 3103

where (P1-2) states that all the demanded packets that are not 
N  
an,T ,R ≤ M R F, ∀ j ∈ [K R ]
cached at the requesting receivers need to be delivered by the
n=1 T ⊆∅ [K T ] R⊆[K R ]:
transmitters over the H blocks of communication. j ∈R
(P3-4)
C. Relaxing Worst-Case Demands to Average Demands an,T ,R ≥ 0, ∀n ∈ [N], ∀T ⊆∅ [K T ], ∀R ⊆ [K R ].
and Optimizing Over Caching Realizations (P3-5)
We can now write an optimization problem to minimize the Similar techniques to the above relaxation from worst-case
number of communication blocks required for delivering the demands to average demands have also been used in [3],
worst-case demands optimized over the caching realizations. [17], [18], [21]. Such a relaxation is helpful since for the
However, before that, we need to introduce some notation. case of average distinct demands, there is a clear and precise
KT
Given any caching realization {Pi }i=1 , {Q j } Kj =1
R
, we can characterization for the number of packets that need to be
break each file Wn , n ∈ [N], in the library into (2 K T −1)(2 K R ) delivered to all the receivers. This is in contrast to the case
subfiles {Wn,T ,R }T ⊆∅ [K T ],R⊆[K R ] , where Wn,T ,R denotes of worst-case demands, since in this case, the upper bound on
the subfile of Wn exclusively stored in the caches of the the number of delivered packets might be arbitrarily loose.
transmitters in T and receivers in R, and we use the shorthand
notation T ⊆∅ [K T ] to denote T ⊆ [K T ], T = ∅. We define D. Lower Bound on the Number of Communication Blocks
an,T ,R as the number of packets in Wn,T ,R . Having the optimization problem in (P3), we now present
Denoting the answer to the optimization problem (P1) by the following lemma which provides a lower bound on the
H ∗ {Pi }i=1
KT
, {Q j } Kj =1
R
, d , the below optimization problem value of (P3).
yields the number of communication blocks required for Lemma 4: The value of the optimization problem (P3) is
MR 2
delivering the worst-case demands, minimized over all caching K R N F 1− N
realizations: bounded from below by K T MT +K R M R .
Proof: See Appendix B. 
min max H ∗ {Pi }i=1
KT
, {Q j } Kj =1
R
,d (P2-1) Since the total number of packets delivered over the channel
K K d
{Pi }i=1
T ,{Q } R
j j =1 is K R 1 − MNR F in the optimization problem (P3), Lemma 4
  immediately yields the following upper bound on the one-shot
s.t. an,T ,R = F, ∀n ∈ [N] (P2-2)
linear sum-DoF:
T ⊆∅ [K T ] R⊆[K R ]
K R 1− MNR F K T MT + K R M R

N  
an,T ,R ≤ MT F, ∀i ∈ [K T ] DoF∗L,sum(N, MT , M R ) ≤ = .
KR NF
M
1− NR
2
N − MR
n=1 R⊆[K R ] T ⊆[K T ]:
i∈T K T MT+K R M R
(P2-3) Combining the above bound with the trivial bound on the

N   one-shot linear sum-DoF which is the number of receivers,
an,T ,R ≤ M R F, ∀ j ∈ [K R ] K R , we have
n=1 T ⊆∅ [K T ] R⊆[K R ]:
 
T +K R M R
j ∈R DoF∗L,sum (N, MT , M R ) ≤ min K T MN−M R
, K R . (37)
(P2-4)
Now, consider the following two cases:
an,T ,R ≥ 0, ∀n ∈ [N], ∀T ⊆∅ [K T ], ∀R ⊆ [K R ]. • M R ≤ 2 : In this case, (37) implies that
N

(P2-5) " #
∗ K T MT + K R M R
DoFL,sum(N, MT , M R ) ≤ min , KR
To lower bound the value of the above optimization prob- N − N2
lem, we can write the following optimization problem, which  
K T MT + K R M R
yields the number of communication blocks averaged over ≤ min 2 , KR .
N
all the π(N, K R ) = (N−K
N!
R )!
permutations of distinct receiver
demands, denoted by P N,K R : • M R > N2 : In this case, (8) implies that one-shot linear
sum-DoF of
1  " #
min H ∗ {Pi }i=1
KT
, {Q j } Kj =1
R
,d N
K T MT +K R 2
KT
{Pi }i=1
KR
,{Q j } j =1 π(N, K R ) DoFL,sum(N, MT , M R ) > min , KR
d∈P N,K R N

(P3-1) KR
  > 2
s.t. an,T ,R = F, ∀n ∈ [N] (P3-2)
T ⊆∅ [K T ] R⊆[K R ] can be achieved, while the upper bound in (37) implies
that DoF∗L,sum (N, MT , M R ) ≤ K R .
N  
an,T ,R ≤ MT F, ∀i ∈ [K T ] (P3-3) Therefore, in both cases, the inner bound in (8) is within
n=1 R⊆[K R ] T ⊆[K T ]: a factor of 2 of the outer bound in (8), which completes the
i∈T proof of the converse of Theorem 1.
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3104
VI. D ISCUSSION AND C ONCLUDING R EMARKS
In this work, we considered a wireless network setting
with arbitrary numbers of transmitters and receivers, where
all transmitters and receivers in the network are equipped with
cache memories of specific sizes. We characterized the one-
shot linear sum-DoF of the network to within a gap of 2.
In particular, we showed that the one-shot linear sum-DoF
of the network is proportional to the aggregate cache size in
the network, even though the cache of each node is isolated
from all the other nodes. We presented an achievable scheme
which loads the caches carefully in order to maximize the
opportunity for zero-forcing the outgoing interference from the
transmitters and interference cancellation due to previously-
cached content at the receivers. We also demonstrated that the
achievable one-shot linear sum-DoF of our scheme is within
a multiplicative factor of 2 of the optimal one-shot linear
sum-DoF by bounding the number of communication blocks
required to deliver any set of requested files to the receivers
using an integer programming approach.
Since the initial submission of this work and its presentation
in [1], quite a few works have appeared on the problem of
cache-aided interference management. For example, in [22],
the authors proposed a linear programming approach and
showed its approximate optimality for cache-aided wireless
networks with at most three receivers. Furthermore, in [23],
the authors considered cache-aided wireless networks with
arbitrary number of receivers but with caches only at the trans-
mitters’ side and with the addition of fronthaul links, through
which the transmitters can communicate with a central cloud
processor, and characterized the optimal normalized delivery
time to within a factor of 2. Finally, in a more recent work,
the authors in [24] considered cache-aided wireless networks
with delivery strategies beyond one-shot linear schemes, and
characterized the fundamental limits of cache-aided wireless
networks to within a constant factor of 13.5.
Based on these developments, we can now use the results
in [24] to understand how much gain beyond one-shot linear
schemes is theoretically possible in cache-aided wireless net-
works. To that end, let us focus on the following two regimes.
• Large number of transmitters and receivers (K T = K R =
K ): For an equal and large number of transmitters and
receivers K T = K R = K , the sum-DoF in [24], which we
denote by DoF∗sum (N, MT , M R ), approximately scales as
DoF∗sum (N, MT , M R ) ≈ K ,
while the one-shot linear sum-DoF approximately scales
as
MT + M R
DoF∗L,sum (N, MT , M R ) ≈ K .
N
We note that in this regime, both of the above sum-DoF’s
scale with the number of nodes in the network. However,
in this regime, interference alignment already achieves a
sum-DoF of K2 and therefore, caching can only provide a
factor of 2 gain, whereas using a one-shot linear delivery
scheme, we can also achieve the same scaling. Note that
according to [19], in this regime, the one-shot linear sum-
DoF of the network without caching is upper bounded
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017
by 2. Hence, our result demonstrates that while restricting
the network to a simple PHY delivery scheme, caching
can provide a gain which scales with the number of users.
• Limited number of transmitters (K T  K R ): Motivated
by cellular networks, for a constant number of transmit-
ters K T = C, the sum-DoF in [24] approximately scales
as
DoF∗sum (N, MT , M R ) ≈ C K R MR
N ,
while the one-shot linear sum-DoF scales as
C MT +K R M R K R ↑ K R M R
DoF∗L,sum (N, MT , M R ) = N ≈ N .
This indicates that in this case, interference alignment
can only be helpful to within a constant factor compared
to the one-shot linear schemes. However, this multiplica-
tive gain comes at the cost of the high complexity of
interference alignment, and moreover, such a gain is not
materialized within any practical ranges of signal-to-noise
ratio (SNR).
There are several interesting directions following this work.
First, in this work we assumed all the links in the network
to be present in the network topology. However, due to
fading effects, some links between certain transmitter-receiver
pairs might be absent from the network topology. It would
be interesting to study what type of caching strategies are
optimal in this case and to explore its connections to the
index coding problem [25]–[27]. Another direction would
be to combine caching with more sophisticated interference
management schemes. Some initial results have been reported
in [28], in which the authors used the replication in the cache
contents at the transmitters in order to improve the system
performance using the ITLinQ scheme [29]–[31]. It would be
interesting to study the role of transmitter and receiver caches
illustrated in this work in improving the achievable system
throughput that more sophisticated delivery schemes such as
ITLinQ can provide.
A PPENDIX A
P ROOF OF L EMMA 1
For any T ⊆ [K T ] s.t. |T | = tT and for any l ∈ [tT + t R ],
it is clear that the set T ⊕ K T (l − 1) is of size tT . Also, for
any R ⊆ [K R ] s.t. |R| = tT + t R , and for any permutation
R , the vector π[l + 1 : l + t R ] is of size t R and the
π ∈ circ
vector π[l + t R + 1 : l + t R + tT − 1] is of size tT − 1.
Furthermore, note that
Wdπ(l) ,T ⊕ K T (l−1),π[l+1:l+t R ],π[l+t R +1:l+t R +tT −1]
is a subfile of the file Wdπ(l) requested by receiver Rxπ(l) .
However, since π(l) ∈ / π[l + 1 : l + t R ], receiver Rxπ(l) has
not stored the packets in this subfile in its cache and therefore,
this subfile needs to be delivered to this receiver.
Finally, each set inside the union in (16) is composed of
tT + t R subfiles. The number of such sets is equal to
  
KT KR
(tT + t R − 1)!. (38)
tT tT + t R
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT 3105

+K R
K T
1   
KR N
Hence, the total number of subfiles in (16) is equal to 1
   = an,T ,R
KT KR N s
(tT + t R − 1)!(tT + t R ) s=1 j =1 T ⊆[K T ]: R⊆[K R ]: n=1
tT t +t |T |∈[s] |R|=s−|T |
 T  R  /R
j∈
KT KR
= (tT + t R )!. (39) 1 
KT 
KR
1 
KR   
N
tT tT + t R = an,T ,R
N r +r
 K On Kthe
R −1
other
 hand, each receiver Rx j has already cached r=1 r =0 j =1 T ⊆[K T ]: R⊆[K R ]: n=1
T
subfiles as in (13) in its cache, and needs the rest |T |=r |R|=r
tT t R −1    /R
j∈
of the subfiles of its requested file, i.e., KtTT K Rt R−1 subfiles,
1 
KT 
KR   
N
t R ![K R −(t R +1)]!
KR − r
where each subfile is further partitioned into = an,T ,R (41)
[K R −(t R +tT )]! N r +r
r=1 r =0 T ⊆[K T ]: R⊆[K R ]: n=1
smaller subfiles. Hence, the total number of small subfiles that |T |=r |R|=r
need to be delivered to all the receivers is equal to R −1
1 
K T K
     br,r
KT KR − 1 t R ![K R − (t R + 1)]! = , (42)
KR N r +r
t t [K R − (t R + tT )]! r=1 r =0
 T  R 
KT KR
= (tT + t R )!, where for any r ∈ [K T ] and r ∈ [K R − 1] ∪ {0}, we
tT tT + t R
which equals the total number of small subfiles in (16), define
calculated in (39). Consequently, the set of requested subfiles
which are not cached at the corresponding receivers can be 
KR   
N
br,r  an,T ,R
partitioned as in (16), hence the proof is complete. 
j =1 T ⊆[K T ]: R⊆[K R ]: n=1
A PPENDIX B |T |=r |R|=r
/R
j∈
P ROOF OF L EMMA 4
  
N
According to the constraint (23), each of the packets of = (K R − r ) an,T ,R .
order s, which are available at s nodes, either on the trans- T ⊆[K T ]: R⊆[K R ]: n=1
mitter side or the receiver side, can be scheduled with at most |T |=r |R|=r
s − 1 packets of the same order. Therefore, for any given
caching realization and set of demands, we have the lower Moreover, adding the constraint in (P3-3) over all transmit-
bound ters yields
 
∗ KT KR
H {Pi }i=1 , {Q j } j =1, d

KT 
N  
+K R 
K T KR   K T MT F ≥ an,T ,R (43)
a d j ,T ,R
≥ i=1 n=1 R⊆[K R ] T ⊆[K T ]:
KR i∈T
s=K R j =1 T ⊆[K T ]: R⊆[K R ]:
|T |∈[s] |R|=s−|T |
/R
j∈
N   KT 
R −1 
K = an,T ,R (44)
KR   a d j ,T ,R
+ n=1 R⊆[K R ] i=1 T ⊆[K T ]:
s i∈T
s=1 j =1 T ⊆[K T ]: R⊆[K R ]:
|T |∈[s] |R|=s−|T | N   KT 
/R
j∈ = r an,T ,R . (45)
+K R 
K T KR   n=1 R⊆[K R ] r=1 T ⊆[K T ]:
a d j ,T ,R
≥ . (40) |T |=r
s
s=1 j =1 T ⊆[K T ]: R⊆[K R ]:
|T |∈[s] |R|=s−|T | Likewise, adding the constraint in (P3-4) over all receivers
/R
j∈
yields
Now, denoting the objective function in (P3-1) by
KT
H̄ {Pi }i=1 , {Q j } Kj =1
R
, we have

KR 
N  
  K R MR F ≥ an,T ,R (46)
KT
H̄ {Pi }i=1 , {Q j } Kj =1
R
j =1 n=1 T ⊆∅ [K T ] R⊆[K R ]:
j ∈R
T RK +K
1 1 
N  
KR 
≥ = an,T ,R (47)
π(N, K R ) s
s=1 n=1 T ⊆∅ [K T ] j =1 R⊆[K R ]:
⎡ ⎤
j ∈R
⎢   
KR N
⎥ 
N  
KR 
×⎣ π(N − 1, K R − 1) an,T ,R ⎦
j =1 T ⊆[K T ]: R⊆[K R ]: n=1
= r an,T ,R , (48)
|T |∈[s] |R|=s−|T | n=1 T ⊆∅ [K T ] r =0 R⊆[K R ]:
/R
j∈ |R|=r

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3106 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 63, NO. 5, MAY 2017

⎛ ⎞
and from (45) and (48), we have 
N  
= KR ⎝ an,T ,R ⎠
(K T MT +⎡K R M R )F n=1 T ⊆∅ [K T ] R⊆[K R ]

N
⎢  
KT  
KR 
N  
≥ ⎣ r an,T ,R − an,T ,R
n=1 R⊆[K R ] r=1 T ⊆[K T ]: j =1 n=1 T ⊆∅ [K T ] R⊆[K R ]:
|T |=r ⎤ j ∈R
 
KR  ≥ K R (N − M R )F, (53)
⎥
+ r an,T ,R ⎦
T ⊆[K T ] r =0 R⊆[K R ]:
where the inequality is due to (P3-2) and (46). Therefore, we
|R|=r can continue (42) to bound the objective function in (P3-1) as

KT 
KR   
N
KT
= (r + r ) an,T ,R H̄ {Pi }i=1 , {Q j } Kj =1
R

r=1 r =0 T ⊆[K T ]: R⊆[K R ]: n=1

R −1
1 
K T K
|T |=r |R|=r br,r
 R −1
K T K ≥
r +r N r +r
≥ br,r . (49) r=1 r =0
KR − r R −1
r=1 r =0 1 K T K
(K R − r )br,r
≥
Now, using the Cauchy-Schwarz inequality, we can write KR N r +r
* * r=1 r =0
R −1
+ K R −1 + K R −1  K K −1 2
K +  r +r +  KR − r 1  T 
R
br,r ≤ , br,r , br,r . ≥ br,r (54)
KR − r r +r K R N F(K T MT + K R M R )
r =0 r =0 r =0 r=1 r =0
(50)  2
1
≥ K R (N − M R )F (55)
Summing the above inequality over r yields K R N F(K T MT + K R M R )
2
MR
 R −1
K T K KR N F 1 − N
br,r = ,
K T MT + K R M R
r=1 r =0⎡* * ⎤
+ K R −1 + K R −1 where (54) and (55) follow from (52) and (53), respectively.
KT +  r +r +  KR − r
≤ ⎣, br,r , br,r ⎦ This completes the proof. 
KR − r r +r
*
r=1 r =0 * r =0
+ K T K R −1 + K T K R −1
+  r + r +  K R − r R EFERENCES
≤ , br,r , br,r (51)
KR − r r +r [1] N. Naderializadeh, M. A. Maddah-Ali, and A. S. Avestimehr, “Funda-
r=1 r =0 * r=1 r =0 mental limits of cache-aided interference management,” in Proc. IEEE
+ K T K R −1
- +  K R − r Int. Symp. Inf. Theory (ISIT), Jul. 2016, pp. 2044–2048.
≤ (K T MT + K R M R )F , br,r , (52) [2] N. Golrezaei, K. Shanmugam, A. G. Dimakis, A. F. Molisch,
r +r and G. Caire, “Femtocaching: Wireless video content delivery
r=1 r =0 through distributed caching helpers,” in Proc. INFOCOM, Mar. 2012,
where in (51) we have invoked the Cauchy-Schwarz inequality pp. 1107–1115.
[3] M. A. Maddah-Ali and U. Niesen, “Fundamental limits of
again and (52) follows from (49). On the other hand, we have caching,” IEEE Trans. Inf. Theory, vol. 60, no. 5, pp. 2856–2867,
May 2014.
 R −1
K T K
[4] S. P. Shariatpanahi, S. A. Motahari, and B. H. Khalaj,
br,r “Multi-server coded caching,” IEEE Trans. Inf. Theory, vol. 62, no. 12,
r=1 r =0 pp. 7253–7271, Dec. 2016.
[5] A. Liu and V. K. N. Lau, “Exploiting base station caching in MIMO

K T KR −1 
KR   
N
cellular networks: Opportunistic cooperation for video streaming,” IEEE
= an,T ,R Trans. Signal Process., vol. 63, no. 1, pp. 57–69, Jan. 2015.
r=1 r =0 j =1 T ⊆[K T ]: R⊆[K R ]: n=1 [6] M. A. Maddah-Ali and U. Niesen, “Cache-aided interference channels,”
|T |=r |R|=r in Proc. IEEE Int. Symp. Inf. Theory (ISIT), Jun. 2015, pp. 809–813.
/R
j∈ [7] A. Sengupta, R. Tandon, and O. Simeone, “Cache aided wireless
⎡ networks: Tradeoffs between storage and latency,” in Proc. Annu. Conf.
⎛ ⎞ Inf. Sci. Syst. (CISS), Mar. 2016, pp. 320–325.
⎢
KT KR KR
⎢⎜    N
⎟ [8] B. Azari, O. Simeone, U. Spagnolini, and A. M. Tulino, “Hypergraph-
= ⎢⎜ an,T ,R ⎟ based analysis of clustered co-operative beamforming with application
⎢⎝ ⎠ to edge caching,” IEEE Wireless Commun. Lett., vol. 5, no. 1, pp. 84–87,
r=1 r =0 j =1 ⎣ T ⊆[K T ]: R⊆[K R ]: n=1 Jan. 2016.
|T |=r |R|=r [9] S.-H. Park, O. Simeone, and S. Shamai (Shitz), “Joint optimization of
⎛ ⎞⎤ cloud and edge processing for fog radio access networks,” IEEE Trans.
Wireless Commun., vol. 15, no. 11, pp. 7621–7632, Nov. 2016.
⎜ ⎟⎥ [10] M. Wigger, R. Timo, and S. Shamai (Shitz), “Complete interference
⎜    N ⎟⎥
⎜ ⎟ mitigation through receiver-caching in Wyner’s networks,” in Proc. IEEE
−⎜ an,T ,R ⎟ ⎥
⎜ ⎟⎥ Inf. Theory Workshop (ITW), Sep. 2016, pp. 335–339.
⎝T ⊆[K T ]: R⊆[K R ]: n=1 ⎠⎦ [11] Y. Ugur, Z. H. Awan, and A. Sezgin, “Cloud radio access networks with
|T |=r |R|=r coded caching,” in Proc. 20th Int. ITG Workshop Smart Antennas (WSA),
j ∈R Mar. 2016, pp. 1–5.

Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
NADERIALIZADEH et al.: FUNDAMENTAL LIMITS OF CACHE-AIDED INTERFERENCE MANAGEMENT
[12] M. Ji, G. Caire, and A. F. Molisch, “Wireless device-to-device caching
networks: Basic principles and system performance,” IEEE J. Sel. Areas
Commun., vol. 34, no. 1, pp. 176–189, Jan. 2016.
[13] M. Afshang, H. S. Dhillon, and P. H. J. Chong, “Fundamentals of cluster-
centric content placement in cache-enabled device-to-device networks,”
IEEE Trans. Commun., vol. 64, no. 6, pp. 2511–2526, Jun. 2016.
[14] M. Tao, E. Chen, H. Zhou, and W. Yu, “Content-centric sparse multicast
beamforming for cache-enabled cloud RAN,” IEEE Trans. Wireless
Commun., vol. 15, no. 9, pp. 6118–6131, Sep. 2016.
[15] X. Peng, J. C. Shen, J. Zhang, and K. B. Letaief, “Backhaul-aware
caching placement for wireless networks,” in Proc. IEEE Global Com-
mun. Conf. (GLOBECOM), Dec. 2015, pp. 1–6.
[16] S. S. Bidokhti, M. Wigger, and R. Timo, “Erasure broadcast networks
with receiver caching,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT),
Barcelona, Spain, Jul. 2016, pp. 1819–1823.
[17] S. S. Bidokhti, M. Wigger, and R. Timo. (May 2016). “Noisy
broadcast networks with receiver caching.” [Online]. Available:
https://arxiv.org/abs/1605.02317
[18] S. S. Bidokhti, M. Wigger, and R. Timo, “An upper bound on the
capacity-memory tradeoff of degraded broadcast channels,” in Proc. Int.
Symp. Turbo Codes Iterative Inf. Process., Brest, France, Sep. 2016,
pp. 350–354.
[19] M. Razaviyayn, G. Lyubeznik, and Z.-Q. Luo, “On the degrees of free-
dom achievable through interference alignment in a MIMO interference
channel,” IEEE Trans. Signal Process., vol. 60, no. 2, pp. 812–821,
Feb. 2012.
[20] H. Weingarten, Y. Steinberg, and S. Shamai (Shitz), “The capacity region
of the Gaussian multiple-input multiple-output broadcast channel,” IEEE
Trans. Inf. Theory, vol. 52, no. 9, pp. 3936–3964, Sep. 2006.
[21] C.-Y. Wang, S. H. Lim, and M. Gastpar. (Jan. 2016). “A new
converse bound for coded caching.” [Online]. Available:
https://arxiv.org/abs/1601.05690
[22] F. Xu, M. Tao, and K. Liu. (May 2016). “Fundamental tradeoff between
storage and latency in cache-aided wireless interference networks.”
[Online]. Available: https://arxiv.org/abs/1605.00203
[23] A. Sengupta, R. Tandon, and O. Simeone. (May 2016). “Cloud
and cache-aided wireless networks: Fundamental latency trade-offs.”
[Online]. Available: https://arxiv.org/abs/1605.01690
[24] J. Hachem, U. Niesen, and S. Diggavi. (Nov. 2016). “Degrees of freedom
of cache-aided wireless interference networks.” [Online]. Available:
https://arxiv.org/abs/1606.03175
[25] Y. Birk and T. Kol, “Informed-source coding-on-demand (ISCOD) over
broadcast channels,” in Proc. 17th Annu. Joint Conf. IEEE Comput.
Commun. Soc. (INFOCOM), vol. 3. Mar. 1998, pp. 1257–1264.
[26] Z. Bar-Yossef, Y. Birk, T. Jayram, and T. Kol, “Index coding with
side information,” in Proc. IEEE 47th Annu. Symp. Found. Comput.
Sci. (FOCS), Oct. 2006, pp. 197–206.
[27] S. El Rouayheb, A. Sprintson, and C. Georghiades, “On the index coding
problem and its relation to network coding and matroid theory,” IEEE
Trans. Inf. Theory, vol. 56, no. 7, pp. 3187–3195, Jul. 2010.
[28] N. Naderializadeh, D. T. Kao, and A. S. Avestimehr, “How to utilize
caching to improve spectral efficiency in device-to-device wireless
networks,” in Proc. 52nd Annu. Allerton Conf. Commun., Control,
Comput. (Allerton), Oct. 2014, pp. 415–422.
[29] N. Naderializadeh and A. S. Avestimehr, “ITLinQ: A new approach for
spectrum sharing in device-to-device communication systems,” IEEE J.
Sel. Areas Commun., vol. 32, no. 6, pp. 1139–1151, Jun. 2014.
[30] N. Naderializadeh and A. S. Avestimehr, “ITLinQ: A new approach
for spectrum sharing,” in Proc. IEEE Int. Symp. Dyn. Spectr. Access
Netw. (DYSPAN), Apr. 2014, pp. 327–333.
[31] N. Naderializadeh and A. S. Avestimehr, “ITLinQ: A new approach for
spectrum sharing in device-to-device communication systems,” in Proc.
IEEE Int. Symp. Inf. Theory (ISIT), Jun. 2014, pp. 1573–1577.
Authorized licensed use limited to: Technische Universitaet Berlin. Downloaded on July 27,2023 at 09:28:28 UTC from IEEE Xplore. Restrictions apply.
3107
Navid Naderializadeh received the B.S. degree in electrical engineering
from Sharif University of Technology, Tehran, Iran, in 2011, the M.S.
degree in electrical and computer engineering from Cornell University,
Ithaca, NY, USA, in 2014, and the Ph.D. degree in electrical engineering
from University of Southern California, Los Angeles, CA, USA in 2016.
He is currently a research scientist at Intel Labs. His research interests
include using information-theoretic concepts to develop spectrum sharing
mechanisms in wireless networks, particularly in device-to-device (D2D) and
cacheaided communication systems. Dr. Naderializadeh ranked first in the
Iranian nationwide university entrance exam in 2007. He was a recipient of
Jacobs scholarship in 2011. He was selected as a 2015-16 Ming Hsieh Institute
Ph.D. Scholar. He has also been a finalist in the Shannon Centennial student
competition at Nokia Bell Labs in 2016.
Mohammad Ali Maddah-Ali received the B.Sc. degree from Isfahan Univer-
sity of Technology, Isfahan, Iran, and the M.A.Sc. degree from the University
of Tehran, Tehran, Iran, both in electrical engineering. From 2002 to 2007,
he was with the Coding and Signal Transmission Laboratory (CST Lab),
Department of Electrical and Computer Engineering, University of Waterloo,
Waterloo, ON, Canada, working toward the Ph.D. degree. From March 2007
to December 2007, he was working at the Wireless Technology Laboratories,
Nortel Networks, Ottawa, ON, Canada, in joint program between CST Lab and
Nortel Networks. From January 2008 to August 2010, he was a Postdoctoral
Fellow at the Department of Electrical Engineering and Computer Sciences in
the University of California at Berkeley. Since September 2010, he has been at
Bell Laboratories, Alcatel-Lucent, Holmdel, NJ, as a communication network
research scientist. His research interests include wireless communications,
content delivery networks, and multiuser information theory. Dr. Maddah-Ali
received several awards including the 2016 Information Theory Society Paper
Award, the 2015 IEEE Communications Society and Information Theory
Society Joint Paper Award, and the Natural Science and Engineering Research
Council of Canada (NSERC) Postdoctoral Fellowship.
Amir Salman Avestimehr received the B.S. degree in electrical engineering
from Sharif University of Technology, Tehran, Iran, in 2003 and the M.S.
degree and Ph.D. degree in electrical engineering and computer science, both
from the University of California, Berkeley, in 2005 and 2008, respectively.
He is currently an Associate Professor at the EE department of University of
Southern California, Los Angeles, CA. He was also a postdoctoral scholar at
the Center for the Mathematics of Information (CMI) at the California Institute
of Technology, Pasadena, in 2008. His research interests include information
theory, the theory of communications, and their applications.
Dr. Avestimehr has received a number of awards for research and teaching,
including the Communications Society and Information Theory Society Joint
Paper Award in 2013, the Presidential Early Career Award for Scientists
and Engineers (PECASE) in 2011, the Michael Tien 72 Excellence in
Teaching Award in 2012, the Young Investigator Program (YIP) award from
the U. S. Air Force Office of Scientific Research in 2011, the National
Science Foundation CAREER award in 2010, and the David J. Sakrison
Memorial Prize in 2008. He has been a Guest Associate Editor for the IEEE
T RANSACTIONS ON I NFORMATION T HEORY Special Issue on Interference
Networks and General Co-Chair of the 2012 North America Information
Theory Summer School and the 2012 Workshop on Interference Networks.