Calculating the capacity (with or without feedback) of channels with memory and continuous alphabets is a challenging task. It requires optimizing the directed information (DI) rate over all channel input distributions. The objective is a multi-letter expression, whose analytic solution is only known for a few specific cases. When no analytic solution is present or the channel model is unknown, there is no unified framework for calculating or even approximating capacity. This work proposes a novel capacity estimation algorithm that treats the channel as a `black-box', both when feedback is or is not present. The algorithm has two main ingredients: (i) a neural distribution transformer (NDT) model that shapes a noise variable into the channel input distribution, which we are able to sample, and (ii) the DI neural estimator (DINE) that estimates the communication rate of the current NDT model. These models are trained by an alternating maximization procedure to both estimate the channel capacity and obtain an NDT for the optimal input distribution. The method is demonstrated on the moving average additive Gaussian noise channel, where it is shown that both the capacity and feedback capacity are estimated without knowledge of the channel transition kernel. The proposed estimation framework opens the door to a myriad of capacity approximation results for continuous alphabet channels that were inaccessible until now.
When estimating the directed information between two jointly stationary Markov processes, it is typically assumed that the recipient of the directed information is itself Markov of the same order as the joint process. While this assumption is often made explicit in the presentation of such estimators, a characterization of when we can expect the assumption to hold is lacking. Using the concept of d-separation from Bayesian networks, we present sufficient conditions for which this assumption holds. We further show that the set of parameters for which the condition is not also necessary has Lebe…
It is known that the capacity region of a two user physically degraded discrete memoryless (DM) broadcast channel (BC) is not enlarged by feedback. An identical result holds true for a physically degraded Gaussian BC, established later using a variant of the Entropy Power Inequality (EPI). In this paper, we extend the latter result to a physically degraded Gaussian Vector BC (PD-GVBC). However, the extension is not EPI based, but employs a recent result on the factorization of concave envelopes. While the existing concave envelope factorization results do not hold in the presence of feedback, …
We consider the multiple unicast problem under network coding over directed acyclic networks with unit capacity edges. There is a set of n source-terminal (s_i - t_i) pairs that wish to communicate at unit rate over this network. The connectivity between the s_i - t_i pairs is quantified by means of a connectivity level vector, [k_1 k_2 ... k_n] such that there exist k_i edge-disjoint paths between s_i and t_i. Our main aim is to characterize the feasibility of achieving this for different values of n and [k_1 ... k_n]. For 3 unicast connections (n = 3), we characterize several achievable and …
Four estimators of the directed information rate between a pair of jointly stationary ergodic finite-alphabet processes are proposed, based on universal probability assignments. The first one is a Shannon--McMillan--Breiman type estimator, similar to those used by Verdú (2005) and Cai, Kulkarni, and Verdú (2006) for estimation of other information measures. We show the almost sure and $L_1$ convergence properties of the estimator for any underlying universal probability assignment. The other three estimators map universal probability assignments to different functionals, each exhibiting relati…
This report studies data-driven estimation of the directed information (DI) measure between two{em discrete-time and continuous-amplitude} random process, based on the $k$-nearest-neighbors ($k$-NN) estimation framework. Detailed derivations of two $k$-NN estimators are provided. The two estimators differ in the metric based on which the nearest-neighbors are found. To facilitate the estimation of the DI measure, it is assumed that the observed sequences are (jointly) Markovian of order $m$. As $m$ is generally not known, a data-driven method (that is also based on the $k$-NN principle) for es…
Recently, several methods have been proposed for estimating the mutual information from sample data using deep neural networks and without the knowing closed form distribution of the data. This class of estimators is referred to as neural mutual information estimators. Although very promising, such techniques have yet to be rigorously bench-marked so as to establish their efficacy, ease of implementation, and stability for capacity estimation which is joint maximization frame-work. In this paper, we compare the different techniques proposed in the literature for estimating capacity and provide…
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；University of Washington（华盛顿大学）
We conduct an axiomatic study of the problem of estimating the strength of a known causal relationship between a pair of variables. We propose that an estimate of causal strength should be based on the conditional distribution of the effect given the cause (and not on the driving distribution of the cause), and study dependence measures on conditional distributions. Shannon capacity, appropriately regularized, emerges as a natural measure under these axioms. We examine the problem of calculating Shannon capacity from the observed samples and propose a novel fixed-$k$ nearest neighbor estimator…
A methodology is developed to realized optimal channel input conditional distributions, which maximize the finite-time horizon directed information, for channels with memory and feedback, by information lossless randomized strategies. The methodology is applied to general Time-Varying Multiple Input Multiple Output (MIMO) Gaussian Linear Channel Models (G-LCMs) with memory, subject to average transmission cost constraints of quadratic form. The realizations of optimal distributions by randomized strategies are shown to exhibit a decomposion into a deterministic part and a random part. The deco…
Channel capacity plays a crucial role in the development of modern communication systems as it represents the maximum rate at which information can be reliably transmitted over a communication channel. Nevertheless, for the majority of channels, finding a closed-form capacity expression remains an open challenge. This is because it requires to carry out two formidable tasks a) the computation of the mutual information between the channel input and output, and b) its maximization with respect to the signal distribution at the channel input. In this paper, we address both tasks. Inspired by impl…
We extend several results of the third author and C. Malvenuto on graph-different permutations to the case of directed graphs and introduce new open problems. Permutation capacity is a natural extension of Sperner capacity from finite directed graphs to infinite digraphs. Our subject is combinatorial in nature, but can be equally regarded as zero-error information theory.
This work develops a new method for estimating and optimizing the directed information rate between two jointly stationary and ergodic stochastic processes. Building upon recent advances in machine learning, we propose a recurrent neural network (RNN)-based estimator which is optimized via gradient ascent over the RNN parameters. The estimator does not require prior knowledge of the underlying joint and marginal distributions. The estimator is also readily optimized over continuous input processes realized by a deep generative model. We prove consistency of the proposed estimation and optimiza…
In this paper, we consider some long-standing problems in communication systems with access to noisy feedback. We introduce a new notion, the residual directed information, to capture the effective information flow (i.e. mutual information between the message and the channel outputs) in the forward channel. In light of this new concept, we investigate discrete memoryless channels (DMC) with noisy feedback and prove that the noisy feedback capacity is not achievable by using any typical closed-loop encoder (non-trivially taking feedback information to produce channel inputs). We then show that …
Ben Gurion University of the Negev（内盖夫本·古里安大学）；Cornell University（康奈尔大学）
Directed information (DI) is a fundamental measure for the study and analysis of sequential stochastic models. In particular, when optimized over input distributions it characterizes the capacity of general communication channels. However, analytic computation of DI is typically intractable and existing optimization techniques over discrete input alphabets require knowledge of the channel model, which renders them inapplicable when only samples are available. To overcome these limitations, we propose a novel estimation-optimization framework for DI over discrete input spaces. We formulate DI o…
In many channel measurement applications, one needs to estimate some characteristics of the channels based on a limited set of measurements. This is mainly due to the highly time varying characteristics of the channel. In this contribution, it will be shown how free probability can be used for channel capacity estimation in MIMO systems. Free probability has already been applied in various application fields such as digital communications, nuclear physics and mathematical finance, and has been shown to be an invaluable tool for describing the asymptotic behaviour of many large-dimensional syst…
A complete set of practical estimators for the conditional, simple and joint algorihmic complexities is presented, from which a semi-metric is derived. Also, new directed information estimators are proposed that are applied to causality inference on Directed Acyclic Graphs. The performances of these estimators are investigated and shown to compare well with respect to the state-of-the-art Normalized Compression Distance (NCD).
The problem of estimating the directed information rate between two discrete processes $\{X_n\}$ and $\{Y_n\}$ via the plug-in (or maximum-likelihood) estimator is considered. When the joint process $\{(X_n,Y_n)\}$ is a Markov chain of a given memory length, the plug-in estimator is shown to be asymptotically Gaussian and to converge at the optimal rate $O(1/\sqrt{n})$ under appropriate conditions; this is the first estimator that has been shown to achieve this rate. An important connection is drawn between the problem of estimating the directed information rate and that of performing a hypoth…
We investigate the problem of continuous-time causal estimation under a minimax criterion. Let $X^T = \{X_t,0\leq t\leq T\}$ be governed by the probability law $P_θ$ from a class of possible laws indexed by $θ\in Λ$, and $Y^T$ be the noise corrupted observations of $X^T$ available to the estimator. We characterize the estimator minimizing the worst case regret, where regret is the difference between the causal estimation loss of the estimator and that of the optimum estimator. One of the main contributions of this paper is characterizing the minimax estimator, showing that it is in fact a Baye…
Recently, several methods have been proposed for estimating the mutual information from sample data using deep neural networks. These estimators ar referred to as neural mutual information estimation (NMIE)s. NMIEs differ from other approaches as they are data-driven estimators. As such, they have the potential to perform well on a large class of capacity problems. In order to test the performance across various NMIEs, it is desirable to establish a benchmark encompassing the different challenges of capacity estimation. This is the objective of this paper. In particular, we consider three scen…
In this paper, we investigate a Bayesian sparse reconstruction algorithm called compressive sensing via Bayesian support detection (CS-BSD). This algorithm is quite robust against measurement noise and achieves the performance of a minimum mean square error (MMSE) estimator that has support knowledge beyond a certain SNR threshold. The key idea behind CS-BSD is that reconstruction takes a detection-directed estimation structure consisting of two parts: support detection and signal value estimation. Belief propagation (BP) and a Bayesian hypothesis test perform support detection, and an MMSE es…
In this paper, we developed a model-based and a data-driven estimator for directed information (DI) to infer the causal connectivity graph between electrocorticographic (ECoG) signals recorded from brain and to identify the seizure onset zone (SOZ) in epileptic patients. Directed information, an information theoretic quantity, is a general metric to infer causal connectivity between time-series and is not restricted to a particular class of models unlike the popular metrics based on Granger causality or transfer entropy. The proposed estimators are shown to be almost surely convergent. Causal …
We consider the multiple unicast problem with three source-terminal pairs over directed acyclic networks with unit-capacity edges. The three $s_i-t_i$ pairs wish to communicate at unit-rate via network coding. The connectivity between the $s_i - t_i$ pairs is quantified by means of a connectivity level vector, $[k_1 k_2 k_3]$ such that there exist $k_i$ edge-disjoint paths between $s_i$ and $t_i$. In this work we attempt to classify networks based on the connectivity level. It can be observed that unit-rate transmission can be supported by routing if $k_i \geq 3$, for all $i = 1, \dots, 3$. In…
There is an increasing interest in studying the neural interaction mechanisms behind patterns of cognitive brain activity. This paper proposes a new approach to infer such interaction mechanisms from electroencephalographic (EEG) data using a new estimator of directed information (DI) called logit shrinkage optimized directed information assessment (L-SODA). Unlike previous directed information measures applied to neural decoding, L-SODA uses shrinkage regularization on multinomial logistic regression to deal with the high dimensionality of multi-channel EEG signals and the small sizes of many…
Directed Information: Estimation, Optimization and Applications in Communications and Causality
Ben-Gurion University of the Negev（内盖夫本-古里安大学）；The Hebrew University of Jerusalem（耶路撒冷希伯来大学）；Indian Institute of Science（印度科学研究所）；Technical University of Munich（慕尼黑工业大学）
We study the problem of efficiently broadcasting packets in multi-hop wireless networks. At each time slot the network controller activates a set of non-interfering links and forwards selected copies of packets on each activated link. A packet is considered jointly received only when all nodes in the network have obtained a copy of it. The maximum rate of jointly received packets is referred to as the broadcast capacity of the network. Existing policies achieve the broadcast capacity by balancing traffic over a set of spanning trees, which are difficult to maintain in a large and time-varying …
We propose a new estimator to measure directed dependencies in time series. The dimensionality of data is first reduced using a new non-uniform embedding technique, where the variables are ranked according to a weighted sum of the amount of new information and improvement of the prediction accuracy provided by the variables. Then, using a greedy approach, the most informative subsets are selected in an iterative way. The algorithm terminates, when the highest ranked variable is not able to significantly improve the accuracy of the prediction as compared to that obtained using the existing sele…
We introduce a general framework for treating channels with memory and feedback. First, we generalize Massey's concept of directed information and use it to characterize the feedback capacity of general channels. Second, we present coding results for Markov channels. This requires determining appropriate sufficient statistics at the encoder and decoder. Third, a dynamic programming framework for computing the capacity of Markov channels is presented. Fourth, it is shown that the average cost optimality equation (ACOE) can be viewed as an implicit single-letter characterization of the capacity.…
This paper describes a framework in which directed information is defined on abstract spaces. The framework is employed to derive properties of directed information such as convexity, concavity, lower semicontinuity, by using the topology of weak convergence of probability measures on Polish spaces. Two extremum problems of directed information related to capacity of channels with memory and feedback, and non-anticipative and sequential rate distortion are analyzed showing existence of maximizing and minimizing distributions, respectively.
We study the problem of computing the rate-distortion function for sources with feed-forward and the capacity for channels with feedback. The formulas (involving directed information) for the optimal rate-distortion function with feed-forward and channel capacity with feedback are multi-letter expressions and cannot be computed easily in general. In this work, we derive conditions under which these can be computed for a large class of sources/channels with memory and distortion/cost measures. Illustrative examples are also provided.
We extend the Blahut-Arimoto algorithm for maximizing Massey's directed information. The algorithm can be used for estimating the capacity of channels with delayed feedback, where the feedback is a deterministic function of the output. In order to do so, we apply the ideas from the regular Blahut-Arimoto algorithm, i.e., the alternating maximization procedure, onto our new problem. We provide both upper and lower bound sequences that converge to the optimum value. Our main insight in this paper is that in order to find the maximum of the directed information over causal conditioning probabilit…
A notion of directed information between two continuous-time processes is proposed. A key component in the definition is taking an infimum over all possible partitions of the time interval, which plays a role no less significant than the supremum over "space" partitions inherent in the definition of mutual information. Properties and operational interpretations in estimation and communication are then established for the proposed notion of directed information. For the continuous-time additive white Gaussian noise channel, it is shown that Duncan's classical relationship between causal estimat…
We consider finite state channels (FSCs) with feedback and state information known causally at the encoder. This setting is quite general and includes: a memoryless channel with i.i.d. state (the Shannon strategy), Markovian states that include look-ahead (LA) access to the state and energy harvesting. We characterize the feedback capacity of the general setting as the directed information between auxiliary random variables with memory to the channel outputs. We also propose two methods for computing the feedback capacity: (i) formulating an infinite-horizon average-reward dynamic program; and…
The index coding problem studies the fundamental limit on broadcasting multiple messages to their respective receivers with different sets of side information that are represented by a directed graph. The generalized lexicographic product structure in the side information graph is introduced as a natural condition under which the corresponding index coding problem can be decomposed into multiple interacting subproblems. The capacity region is characterized in terms of the subproblem capacity regions combined in the same product structure. The proof is based on dual uses of random coding---one …
The Finite Transmission Feedback Information (FTFI) capacity is characterized for any class of channel conditional distributions $\big\{{\bf P}_{B_i|B^{i-1}, A_i} :i=0, 1, \ldots, n\big\}$ and $\big\{ {\bf P}_{B_i|B_{i-M}^{i-1}, A_i} :i=0, 1, \ldots, n\big\}$, where $M$ is the memory of the channel, $B^n {\stackrel{\triangle}{=}} \{B_j: j=\ldots, 0,1, \ldots, n\}$ are the channel outputs and $A^n{\stackrel{\triangle}{=}} \{A_j: j=\ldots, 0,1, \ldots, n\}$ are the channel inputs. The characterizations of FTFI capacity, are obtained by first identifying the information structures of the optimal …
As a class of state-dependent channels, Markov channels have been long studied in information theory for characterizing the feedback capacity and error exponent. This paper studies a more general variant of such channels where the state evolves via a general stochastic process, not necessarily Markov or ergodic. The states are assumed to be unknown to the transmitter and the receiver, but the underlying probability distributions are known. For this setup, we derive an upper bound on the feedback error exponent and the feedback capacity with variable-length codes. The bounds are expressed in te…
Directed information or its variants are utilized extensively in the characterization of the capacity of channels with memory and feedback, nonanticipative lossy data compression, and their generalizations to networks. In this paper, we derive several functional and topological properties of directed information for general abstract alphabets (complete separable metric spaces) using the topology of weak convergence of probability measures. These include convexity of the set of consistent distributions, which uniquely define causally conditioned distributions, convexity and concavity of directe…
University of Michigan（密歇根大学）；Stanford University（斯坦福大学）
Directed information (DI) is a useful tool to explore time-directed interactions in multivariate data. However, as originally formulated DI is not well suited to interactions that change over time. In previous work, adaptive directed information was introduced to accommodate non-stationarity, while still preserving the utility of DI to discover complex dependencies between entities. There are many design decisions and parameters that are crucial to the effectiveness of ADI. Here, we apply ideas from ensemble learning in order to alleviate this issue, allowing for a more robust estimator for ex…
The multicast capacity of a directed network is closely related to the $s$-$t$ maximum flow, which is equal to the $s$-$t$ minimum cut capacity due to the max-flow min-cut theorem. If the topology of a network (or link capacities) is dynamically changing or have stochastic nature, it is not so trivial to predict statistical properties on the maximum flow. In this paper, we present a coding theoretic approach for evaluating the accumulate distribution of the minimum cut capacity of weighted random graphs. The main feature of our approach is to utilize the correspondence between the cut space of…
We consider finite state channels where the state of the channel is its previous output. We refer to these as POST (Previous Output is the STate) channels. We first focus on POST($α$) channels. These channels have binary inputs and outputs, where the state determines if the channel behaves as a $Z$ or an $S$ channel, both with parameter $α$. %with parameter $α.$ We show that the non feedback capacity of the POST($α$) channel equals its feedback capacity, despite the memory of the channel. The proof of this surprising result is based on showing that the induced output distribution, when maximiz…
In this paper, we consider the problem of evaluating capacity expression of a multiple access channel (MAC) with noiseless feedback. So far, the capacity expression for this channel is known through a multi letter directed information by Kramer [1]. Recently, it was shown in [2] that one can pose it as a dynamic optimization problem, however, no dynamic program was provided as the authors claimed there is no notion of state that is observed by both the senders. In this paper, we build upon [2] to show that there indeed exists a state and therefore a dynamic program (DP) that decomposes this dy…
A code design problem for memory devises with restricted state transitions is formulated as a combinatorial optimization problem that is called a subgraph domatic partition (subDP) problem. If any neighbor set of a given state transition graph contains all the colors, then the coloring is said to be valid. The goal of a subDP problem is to find a valid coloring with the largest number of colors for a subgraph of a given directed graph. The number of colors in an optimal valid coloring gives the writing capacity of a given state transition graph. The subDP problems are computationally hard; it …
The capacity region of the Finite-State Multiple Access Channel (FS-MAC) with feedback that may be an arbitrary time-invariant function of the channel output samples is considered. We characterize both an inner and an outer bound for this region, using Masseys's directed information. These bounds are shown to coincide, and hence yield the capacity region, of FS-MACs where the state process is stationary and ergodic and not affected by the inputs. Though `multi-letter' in general, our results yield explicit conclusions when applied to specific scenarios of interest. E.g., our results allow us t…
In multi user Gaussian relay networks, it is desirable to transmit private information to each user as well as common information to all of them. However, the capacity region of such networks with both kinds of information is not easy to characterize. The prior art used simple linear deterministic models in order to approximate the capacities of these Gaussian networks. This paper discusses the capacity region of the deterministic Y-channel with private and common messages. In this channel, each user aims at delivering two private messages to the other two users in addition to a common message…
For any class of channel conditional distributions, with finite memory dependence on channel input RVs $A^n {\stackrel{\triangle}{=}} \{A_i: i=0, \ldots, n\}$ or channel output RVs $B^n {\stackrel{\triangle}{=}} \{B_i: i=0, \ldots, n\}$ or both, we characterize the sets of channel input distributions, which maximize directed information defined by $ I(A^n \rightarrow B^n) {\stackrel{\triangle}{=}} \sum_{i=0}^n I(A^i;B_i|B^{i-1}) $ and we derive the corresponding expressions, called "characterizations of Finite Transmission Feedback Information (FTFI) capacity". The main theorems state that opt…
We derive sequential necessary and sufficient conditions for any channel input conditional distribution ${\cal P}_{0,n}\triangleq\{P_{X_t|X^{t-1},Y^{t-1}}:~t=0,\ldots,n\}$ to maximize the finite-time horizon directed information defined by $$C^{FB}_{X^n \rightarrow Y^n} \triangleq \sup_{{\cal P}_{0,n}} I(X^n\rightarrow{Y^n}),~~~ I(X^n \rightarrow Y^n) =\sum_{t=0}^n{I}(X^t;Y_t|Y^{t-1})$$ for channel distributions $\{P_{Y_t|Y^{t-1},X_t}:~t=0,\ldots,n\}$ and $\{P_{Y_t|Y_{t-M}^{t-1},X_t}:~t=0,\ldots,n\}$, where $Y^t\triangleq\{Y_0,\ldots,Y_t\}$ and $X^t\triangleq\{X_0,\ldots,X_t\}$ are the channel…
We consider capacity of discrete-time channels with feedback for the general case where the feedback is a time-invariant deterministic function of the output samples. Under the assumption that the channel states take values in a finite alphabet, we find an achievable rate and an upper bound on the capacity. We further show that when the channel is indecomposable, and has no intersymbol interference (ISI), its capacity is given by the limit of the maximum of the (normalized) directed information between the input $X^N$ and the output $Y^N$, i.e. $C = \lim_{N \to \infty} \frac{1}{N} \max I(X^N \…
In this paper, we prove the existence of fundamental relations between information theory and estimation theory for network-coded flows. When the network is represented by a directed graph G=(V, E) and under the assumption of uncorrelated noise over information flows between the directed links connecting transmitters, switches (relays), and receivers. We unveil that there yet exist closed-form relations for the gradient of the mutual information with respect to different components of the system matrix M. On the one hand, this result opens a new class of problems casting further insights into …
Information theoretic quantities play an important role in various settings in machine learning, including causality testing, structure inference in graphical models, time-series problems, feature selection as well as in providing privacy guarantees. A key quantity of interest is the mutual information and generalizations thereof, including conditional mutual information, multivariate mutual information, total correlation and directed information. While the aforementioned information quantities are well defined in arbitrary probability spaces, existing estimators add or subtract entropies (we …
We present novel data-processing inequalities relating the mutual information and the directed information in systems with feedback. The internal blocks within such systems are restricted only to be causal mappings, but are allowed to be non-linear, stochastic and time varying. These blocks can for example represent source encoders, decoders or even communication channels. Moreover, the involved signals can be arbitrarily distributed. Our first main result relates mutual and directed informations and can be interpreted as a law of conservation of information flow. Our second main result is a p…
The development of optimal and efficient machine learning-based communication systems is likely to be a key enabler of beyond 5G communication technologies. In this direction, physical layer design has been recently reformulated under a deep learning framework where the autoencoder paradigm foresees the full communication system as an end-to-end coding-decoding problem. Given the loss function, the autoencoder jointly learns the coding and decoding optimal blocks under a certain channel model. Because performance in communications typically refers to achievable rates and channel capacity, the …
