Training neural encoders requires a differentiable channel model for backpropagation. This can be bypassed by approximating the channel distribution using pilot signals. A common method for this is the use of generative adversarial networks (GANs). In this paper, we introduce diffusion models (DMs) for channel generation and propose an efficient training algorithm. Our DMs provide a solution that achieves near-optimal end-to-end symbol error rates (SERs). Importantly, DMs outperform GANs in high signal-to-noise ratio regions. Here, in particular, we explore the trade-off between sample quality and speed. We also show that the right noise scheduling can significantly reduce sampling time with a minor increase in SER.

Small neural networks (NNs) used for error correction were shown to improve on classic channel codes and to address channel model changes. We extend the code dimension of any such structure by using the same NN under one-hot encoding multiple times, then serially-concatenated with an outer classic code. We design NNs with the same network parameters, where each Reed-Solomon codeword symbol is an input to a different NN. Significant improvements in block error probabilities for an additive Gaussian noise channel as compared to the small neural code are illustrated, as well as robustness to channel model changes.

The use of deep learning-based techniques for approximating secure encoding functions has attracted considerable interest in wireless communications due to impressive results obtained for general coding and decoding tasks for wireless communication systems. Of particular importance is the development of model-free techniques that work without knowledge about the underlying channel. Such techniques utilize for example generative adversarial networks to estimate and model the conditional channel distribution, mutual information estimation as a reward function, or reinforcement learning. In this paper, the approach of reinforcement learning is studied and, in particular, the policy gradient method for a model-free approach of neural network-based secure encoding is investigated. Previously developed techniques for enforcing a certain co-set structure on the encoding process can be combined with recent reinforcement learning approaches. This new approach is evaluated by extensive simulations, and it is demonstrated that the resulting decoding performance of an eavesdropper is capped at a certain error level.

The Jensen inequality is a widely used tool in a multitude of fields, such as for example information theory and machine learning. It can be also used to derive other standard inequalities such as the inequality of arithmetic and geometric means or the Hölder inequality. In a probabilistic setting, the Jensen inequality describes the relationship between a convex function and the expected value. In this work, we want to look at the probabilistic setting from the reverse direction of the inequality. We show that under minimal constraints and with a proper scaling, the Jensen inequality can be reversed. We believe that the resulting tool can be helpful for many applications and provide a variational estimation of mutual information, where the reverse inequality leads to a new estimator with superior training behavior compared to current estimators.

End-to-end learning of communication systems with neural networks and particularly autoencoders is an emerging research direction which gained popularity in the last year. In this approach, neural networks learn to simultaneously optimize encoding and decoding functions to establish reliable message transmission. In this paper, this line of thinking is extended to communication scenarios in which an eavesdropper must further be kept ignorant about the communication. The secrecy of the transmission is achieved by utilizing a modified secure loss function based on cross-entropy which can be implemented with state-of-the-art machine-learning libraries. This secure loss function approach is applied in a Gaussian wiretap channel setup, for which it is shown that the neural network learns a trade-off between reliable communication and information secrecy by clustering learned constellations. As a result, an eavesdropper with higher noise cannot distinguish between the symbols anymore.

We study deterministic approximations of the Gaussian two-user multiple access wiretap channel (G-MAC-WT) and the Gaussian wiretap channel with a helper (G-WT-H). These approximations enable results beyond the recently shown 2/3 and 1/2 secure degrees of freedom (s.d.o.f.) for the G-MAC-WT and the G-WT-H, respectively. While the s.d.o.f. were obtained by real interference alignment, our approach uses signal-scale alignment. We show achievable schemes which are independent of the rationality of the channel gains. Moreover, our results can differentiate between channel strengths, in particular, between both users, and will establish secrecy rates dependent on this difference. We can show that the resulting achievable secrecy rates tend to the s.d.o.f. for vanishing channel gain differences. Moreover, we extend previous and develop new techniques to prove generalized s.d.o.f. bounds for varying channel strengths and show that our achievable schemes reach the bounds for certain channel gain parameters. We believe that our analysis is the next step toward a constant-gap analysis of the G-MAC-WT and the G-WT-H.

It is well-known that wireless channel reciprocity together with fading can be exploited to generate a common secret key between two legitimate communication partners. This can be achieved by exchanging known deterministic pilot signals between both partners from which the random fading gains can be estimated and processed. However, the entropy and thus quality of the generated key depends on the channel coherence time. This can result in poor key generation rates in a low mobility environment, where the fading gains are nearly constant. Therefore, wide-spread deployment of wireless channel-based secret key generation is limited. To overcome these issues, we follow up on a recent idea which uses unknown random pilots and enables “on-the-fly” key generation. In addition, the scheme is able to incorporate local sources of randomness but performance bounds are hard to obtain with standard methods. In this paper, we analyse such a scheme analytically and derive achievable key rates in the Alice-Bob-Eve setting. For this purpose, we develop a novel approximation model which is inspired by the linear deterministic and the lower triangular deterministic model. We claim that our novel approach provides an intuitive and clear framework to analyse similar key generation problems.

Recent investigations have shown that the sum secure degrees of freedom of the Gaussian wiretap channel with a helper is 1/2. The achievable scheme for this result is based on the real interference alignment approach. While providing a good way to show degrees of freedom results, this technique has the disadvantage of relying on the Khintchine-Groshev theorem and is therefore limited to almost all channel gains. This means that there are infinitely many channel gains, where the scheme fails. Furthermore, the real interference alignment approach cannot be used to yield stronger constant-gap results. We approach this topic from a signal-scale alignment perspective and use the linear deterministic model as a first approximation. Here we can show a constant-gap sum capacity for certain channel gain parameters. We transfer these results to the Gaussian model and discuss the results.

The use of wireless channel reciprocity properties for secret key generation has been an attractive method for many wireless applications. Most of the available methods rely on some sort of pilot signalling from the transceiver. This reduces the intrinsic security of the generated key. In this paper we have introduced a novel key generation and exchange method that requires no pilot signalling and have also included an algorithm to restore wireless channel reciprocity properties for varying transmission power. Finally we presented information theoretic analysis and the result of our implementation on off-the-shelf TelosB motes.

Recent investigations have shown that approximations of certain channels by the linear deterministic model (LDM) are sub-optimal in some cases. Replacing the LDM by a more sophisticated deterministic model possibly results in improvements of previous results. An example is use of the lower triangular deterministic model (LTDM) to replace the LDM. Due to the integrated dependence on the fine channel gains, results can be obtained which were not previously possible. In this paper, we investigate the IMAC under the approximation of this refined model. It was previously demonstrated, that multi-user gain is present in the conventional deterministic IMAC model. However, the capacity could be shown just for discrete points due to limitations of the approximation model. In this paper, we propose a scheme for the new approximation model, which reaches the upper bound completely and therefore provides the sum-capacity. Moreover, the scheme itself has a much more simplistic structure.

Recent investigations have shown that multi-user gain can be enabled in frequency-flat time-invariant single-antenna cellular networks, for example the interfering multiple access channel (IMAC). These investigations have shown gain beyond results through treating interference as noise (TIN) techniques in the weak interference regime, rendering TIN sub-optimal for these networks. However, it was shown previously that multi-user gain for the multiple access channel interfering with a point-to-point link (MAC-P2P) is limited to the regime equation, with a relapse of the rate to IC level above this regime. Previous results for the IMAC were limited to the very weak interference regime α ≤ 1 over 2. The question is if multi-user gain of the deterministic IMAC is also limited to this regime. We answer this question with no and show the sum-capacity approximated by the lower triangular deterministic model for the IMAC and that there is significant multi-user gain for α ≥ 2 over 3. We therefore explore the GDoF of the deterministic IMAC for the whole interference range.

The MAC-BC duality of information theory and wireless communications is an intriguing concept for efficient algorithm design. However, no concept is known so far for the important cellular channel. To make progress on this front, we consider in this paper the linear deterministic cellular channel. In particular, we prove duality of a network with two interfering MACs in each cell and a network with two interfering BCs in each cell. The operational region is confined to the weak interference regime. First, achievable schemes as well as upper bounds will be provided. These bounds are the same for both channels. We will show, that for specific cases the upper bound corresponds to the achievable scheme and hence establishing a duality relationship between them.

There has been major progress over the last decade in understanding the classical interference channel (IC). Recent key results show that constant bit gap capacity results can be obtained from linear deterministic models (LDMs). However, it is widely unrecognized that the time-invariant, frequency-flat cellular channel, which contains the IC as a special case, possesses some additional generalized degrees of freedom (GDoF) due to multi-user operation. This was proved for the LDM cellular channel very recently but is an open question for the corresponding Gaussian counterpart. In this paper, we close this gap and provide an achievable sum-rate for the Gaussian cellular channel which is within a constant bit gap of the LDM sum capacity. We show that the additional GDoFs from the LDM cellular channel carry over. This is enabled by signal scale alignment. In particular, the multi-user gain reduces the interference by half in the 2-user per cell case compared to the IC.