Mehrdad Tahmasbi

On the capacity achieving probability measures for molecular receivers

In this paper, diffusion-based molecular communication with ligand receptor receivers is studied. Information messages are assumed to be encoded via variations of the concentration of molecules. The randomness in the ligand reception process induces uncertainty in the communication; limiting the rate of information decoding. We model the ligand receptor receiver by a set of finite-state Markov channels and study the general capacity of such a receiver. Furthermore, the i.i.d. capacity of the receiver is characterized as a lower bound for the general capacity. It is also proved that a finite support probability measure can achieve the i.i.d. capacity of the receiver. Moreover, a bound on the number of points in the support of the probability measure is obtained.

Second-order asymptotics of covert communications over noisy channels

We consider the problem of covert communication over noisy binary input Discrete Memoryless Channels (DMCs). Covertness is measured with respect to an adversary in terms of the divergence between the channel output distribution induced with and without communication. We characterize the exact second order asymptotics of the number of bits that can be reliably transmitted with a probability of error less than ∈ and a divergence less than δ. The main technical contribution of this paper is a detailed analysis of how to expurgate a random code while maintaining its channel resolvability properties.

Second order asymptotics for degraded wiretap channels: How good are existing codes?

We develop three new results regarding the second-order asymptotics of secure communication over wiretap channels. We first establish the optimal second-order asymptotics for a class of degraded wiretap channels without feedback under an effective secrecy criterion. We then derive the optimal second-order asymptotics for degraded wiretap channels with feedback. We finally develop a new converse bound for channel resolvability with non-capacity achieving distributions, which we use to develop useful converse bounds for asymmetric degraded wiretap channels. Our results are illustrated with several numerical examples, and suggest that known coding techniques already achieve their best performance.

Learning adversary's actions for secret communication

We analyze the problem of secure communication over a wiretap channel with an active adversary, in which the legitimate transmitter has the opportunity to sense and learn the adversarys actions. Specifically, the adversary has the ability to switch between two channels and to observe the corresponding output at every channel use; the encoder, however, has causal access to observations impacted by adversarys actions. We develop a joint learning/transmission scheme in which the legitimate users learn and adapt to the adversarys actions. For some channel models, we show that the achievable rates, which we define precisely, are arbitrarily close to those obtained with hindsight, had the transmitter known the actions ahead of time. This suggests that there is much to exploit and gain in physical-layer security by monitoring the environment.

Modulation recognition using side information and hybrid learning

Recent applications of machine learning to modulation recognition have demonstrated the potential of deep learning to achieve state-of-the-art performance. We propose to further extend this approach by using flexible time-space decompositions that are more in line with the actual learning task, as well as integrate side-information, such as higher order moments, directly into the training process. Our promising preliminary results suggest that there are many more benefits to be reaped from such approaches.