Daniela Tuninetti

On the optimality of uncoded cache placement

Caching is an effective way to reduce peak-hour network traffic congestion by storing some contents at users local cache. Maddah-Ali and Niesen (MAN) initiated a fundamental study of caching systems by proposing a scheme (with uncoded cache placement and linear network coding delivery) that is provably optimal to within a factor 4.7. In this paper, when the cache contents and the user demands are fixed, we connect the caching problem to an index coding problem and show the optimality of the MAN scheme under the conditions that (i) the cache placement phase is restricted to be uncoded (i.e, pieces of the files can only copied into the users cache), and (ii) the number of users is no more than the number of files. As a consequence, further improvements to the MAN scheme are only possible through the use of coded cache placement.

On caching with more users than files

Caching is an efficient way to reduce peak hour network traffic congestion by storing some content at the users cache without knowledge of later demands. Recently, Maddah-Ali and Niesen proposed a two-phase, placement and delivery phase, coded caching strategy for centralized systems (where coordination among users is possible in the placement phase), and for decentralized systems. This paper investigates the same setup under the assumption that the number of users is larger than the number of files. By using the same uncoded placement strategy of Maddah-Ali and Niesen, a novel coded delivery strategy is proposed to profit from the multicasting opportunities that arise because a file may be demanded by multiple users. The proposed delivery method is proved to be optimal under the constraint of uncoded cache placement for centralized systems with two files. Moreover it is shown to outperform known caching strategies for both centralized and decentralized systems.

Gaussian Half-Duplex Relay Networks: Improved Constant Gap and Connections With the Assignment Problem

This paper considers a Gaussian relay network where a source transmits a message to a destination with the help of N half-duplex relays. The information theoretic cut-set upper bound to the capacity is shown to be achieved to within 1.96(N+2) bits by noisy network coding, thereby reducing the previously known gap. This gap is obtained as a special case of a more general constant gap result for Gaussian half-duplex multicast networks. It is then shown that the generalized degrees-of-freedom of this network is the solution of a linear program, where the coefficients of the linear inequality constraints are proved to be the solution of several linear programs referred as the assignment problem in graph theory, for which efficient numerical algorithms exist. The optimal schedule, that is, the optimal value of the 2N possible transmit-receive configuration states for the relays, is investigated and known results for diamond networks are extended to general relay networks. It is shown, for the case of N=2 relays, that only N+1=3 out of the 2N=4 possible states have a strictly positive probability and suffice to characterize the capacity to within a constant gap. Extensive experimental results show that, for a general N -relay network with N≤8 , the optimal schedule has at most N+1 states with a strictly positive probability. As an extension of a conjecture presented for diamond networks, it is conjectured that this result holds for any half-duplex relay network and any number of relays. Finally, a network with N=2 relays is studied in detail to illustrate the channel conditions under which selecting the best relay is not optimal, and to highlight the nature of the rate gain due to multiple relays.

On the Gaussian Half-Duplex Relay Channel

This paper considers the Gaussian half-duplex relay channel (G-HD-RC): a channel model where a source transmits a message to a destination with the help of a relay that cannot transmit and receive at the same time. It is shown that the cut-set upper bound on the capacity can be achieved to within a constant gap, regardless of the actual value of the channel parameters, by either partial-decode-and-forward or compress-and-forward. The performance of these coding strategies is evaluated with both random and deterministic switch at the relay. Numerical evaluations show that the actual gap is less than what analytically obtained, and that random switch achieves higher rates than deterministic switch. As a result of this analysis, the generalized degrees-of-freedom of the G-HD-RC is exactly characterized for this channel. In order to get insights into practical schemes for the G-HD-RC that are less complex than partial-decode-and-forward or compress-and-forward, the exact capacity of the linear deterministic approximation (LDA) of the G-HD-RC at high signal-to-noise-ratio is determined. It is shown that random switch and correlated nonuniform inputs bits are optimal for the LDA. It is then demonstrated that deterministic switch is to within one bit from the capacity. This latter scheme is translated into a coding strategy for the original G-HD-RC and its optimality to within a constant gap is proved. The gap attained by this scheme is larger than that of partial-decode-and-forward, thereby pointing to an interesting practical tradeoff between gap to capacity and complexity.

On the Capacity of the Two-User Gaussian Causal Cognitive Interference Channel

This paper considers the two-user Gaussian causal cognitive interference channel (GCCIC), which consists of two source-destination pairs that share the same channel and where one full-duplex cognitive source can causally learn the message of the primary source through a noisy link. The GCCIC is an interference channel with unilateral source cooperation that better models practical cognitive radio networks than the commonly used model which assumes that one source has perfect noncausal knowledge of the other sources message. First, the sum-capacity of the symmetric GCCIC is determined to within a constant gap. Then, the insights gained from the study of the symmetric GCCIC are extended to more general cases. In particular, the whole capacity region of the Gaussian Z-channel, i.e., when there is no interference from the primary user, and of the Gaussian S-channel, i.e., when there is no interference from the secondary user, are both characterized to within 2 bits. The fully connected general, i.e., no-symmetric, GCCIC is also considered and its capacity region is characterized to within 2 bits when, roughly speaking, the interference is not weak at both receivers. The parameter regimes where the GCCIC is equivalent, in terms of generalized degrees-of-freedom, to the noncooperative interference channel (i.e., unilateral causal cooperation is not useful), to the non-causal cognitive interference channel (i.e., causal cooperation attains the ultimate limit of cognitive radio technology), and to bilateral source cooperation are identified. These comparisons shed light into the parameter regimes and network topologies that in practice might provide an unbounded throughput gain compared to currently available (non cognitive) technologies.

New outer bounds for the interference channel with unilateral source cooperation

This paper studies the two-user interference channel with unilateral source cooperation, which consists of two source-destination pairs that share the same channel and where one full-duplex source can overhear the other source through a noisy in-band link. Novel outer bounds of the type 2R1 + R2 and R1 + 2R2 are developed for the class of injective semi-deterministic channels with independent noises at the different source-destination pairs. The bounds are then specialized to the Gaussian noise case. Interesting insights are provided about when these types of bounds are active, or in other words, when unilateral cooperation is too weak and leaves some system resources underutilized.

Gaussian half-duplex relay channels: Generalized degrees of freedom and constant gap result

This paper considers the Gaussian relay channel where the relay node operates in half-duplex mode. The exact capacity of the linear deterministic approximation of the Gaussian channel at high SNR is derived first. This result is then used to inspire an achievable scheme valid for any SNR in the original channel. The scheme is quite simple: it uses successive decoding and does not incur in the typical delay of backward decoding. The achievable rate is then showed to be at most 3 bits away from the cut-set upper bound, which allows to analytically determine the generalized Degrees-of-Freedom of the channel. A closed form expression for the gDoF-optimal fraction of time the relay node transmits is found as well.

Gaussian half-duplex relay networks: Improved gap and a connection with the assignment problem

This paper studies a Gaussian relay network, where the relays can either transmit or receive at any given time, but not both. Known upper (cut-set) and lower (noisy network coding) bounds on the capacity of a memoryless full-duplex relay network are specialized to the half-duplex case and shown to be to within a constant gap of one another. For fairly broad range of relay network sizes, the derived gap is smaller than what is known in the literature, and it can be further reduced for more structured networks such as diamond networks. It is shown that the asymptotically optimal duration of the listen and transmit phases for the relays can be obtained by solving a linear program; the coefficients of the linear constraints of this linear program are the solution of certain `assignment problems for which efficient numerical routines are available; this gives a general interesting connection between the high SNR approximation of the capacity of a MIMO channel and the `assignment problem in graph theory. Finally, some results available for diamond networks are extended to general networks. For a general relay network with 2 relays, it is proved that, out of the 4 possible listen/transmit states, at most 3 have a strictly positive probability. Numerical results for a network with K - 2 <; 9 relays show that at most K-1 states have a strictly positive probability, which is conjectured to be true for any number of relays.

The capacity to within a constant gap of the Gaussian half-duplex relay channel

This paper studies the Gaussian half duplex relay channel, where the relay node can not transmit and receive at the same time. The main contribution lies in showing that both Partial-Decode-Forward and Compress-Forward achieve the CutSet upper bound to within a constant gap regardless of the channel parameters. This provides a closed form characterization of the Generalized Degrees-of-Freedom (gDoF) of the channel, which for certain channel parameters is strictly smaller than the gDoF of the full duplex channel. Half duplex channels can convey information through the random switch between the receive and retransmit phases; this work shows numerically that random switch achieves larger rates compared to deterministic switch, which is usually considered in the literature.

On the capacity of full-duplex causal cognitive interference channels to within a constant gap

This paper considers the two-user Gaussian Causal Cognitive Interference Channel (GCCIC), which consists of two source-destination pairs that share the same channel and where one full-duplex cognitive source can causally learn the message of the primary source through a noisy link. The GCCIC is an interference channel with unilateral source cooperation that models practical cognitive radio networks. Different achievable strategies are shown to be at most a finite number of bits away from an outer bound for a set of the channel parameters that, roughly speaking, excludes the case of weak interference at both receivers.