We present and discuss several novel applications of deep learning for the physical layer. By interpreting a communications system as an autoencoder, we develop fundamental new way to think about design end-to-end reconstruction task that seeks jointly optimize transmitter receiver components in single process. show how this idea can be extended networks multiple transmitters receivers concept radio transformer means incorporate expert domain knowledge machine model. Lastly, demonstrate...

We conduct an in depth study on the performance of deep learning based radio signal classification for communications signals. consider a rigorous baseline method using higher order moments and strong boosted gradient tree compare between two approaches across range configurations channel impairments. effects carrier frequency offset, symbol rate, multi-path fading simulation over-the-air measurement lab software radios training strategies both. Finally we conclude with discussion remaining...

We address the problem of learning an efficient and adaptive physical layer encoding to communicate binary information over impaired channel. In contrast traditional work, we treat unsupervised machine focusing on optimizing reconstruction loss through artificial impairment layers in autoencoder (we term this a channel autoencoder) introduce several new regularizing which emulate common wireless impairments. also discuss role attention models form radio transformer network for helping...

We introduce a novel physical layer scheme for single user Multiple-Input Multiple-Output (MIMO) communications based on unsupervised deep learning using an autoencoder. This method extends prior work the joint optimization of representation and encoding decoding processes as end-to-end task by expanding transmitter receivers to multi-antenna case. widely used domain appropriate wireless channel impairment model (Rayleigh fading channel), into autoencoder problem in order directly learn...

Channel modeling is a critical topic when considering accurately designing or evaluating the performance of communications system. Most prior work in learning new modulation schemes has focused on using simplified analytic channel models such as additive white Gaussian noise (AWGN), Rayleigh fading channels other similar compact parametric models. In this paper, we extend recent training generative adversarial networks (GANs) to approximate wireless responses more reflect probability...

This paper presents a novel method for synthesizing new physical layer modulation and coding schemes communications systems using learning-based approach which does not require an analytic model of the impairments in channel. It extends prior work published on channel autoencoderto consider case where stochastic response is known or can be easily modeled closed form expression. By adopting adversarial learning approximation information encoding, we jointly learn solution to both tasks...

We explore unsupervised representation learning of radio communication signals in raw sampled time series representation. demonstrate that we can learn modulation basis functions using convolutional autoencoders and visually recognize their relationship to the analytic bases used digital communications. also propose evaluate quantitative metrics for quality encoding domain relevant performance metrics.

We introduce a novel physical layer scheme for Multiple Input Output (MIMO) communications based on unsupervised deep learning using an autoencoder. This method extends prior work the joint optimization of representation and encoding decoding processes as single end-to-end task by expanding transmitter receiver to multi-antenna case. domain appropriate wireless channel impairment model (the multi-input multi-output Rayleigh fading channel), into autoencoder problem in order directly learn...

We introduce learned attention models into the radio machine learning domain for task of modulation recognition by leveraging spatial transformer networks and introducing new appropriate transformations. This model allows network to learn a localization capable synchronizing normalizing signal blindly with zero knowledge signal's structure based on optimization classification accuracy, sparse representation, regularization. Using this architecture we are able outperform our prior results in...

Radio emitter recognition in dense multi-user environments is an important tool for optimizing spectrum utilization, identifying and minimizing interference, enforcing policy. data readily available easy to obtain from antenna, but labeled curated often scarce making supervised learning strategies difficult time consuming practice. We demonstrate that semi-supervised techniques can be used scale beyond datasets, allowing discerning recalling new radio signals by using sparse signal...

This paper presents a novel physical layer scheme for multiple-input multiple-output (MIMO) communication systems based on unsupervised deep learning (DL) using an autoencoder in interference channel (IC) environment. Moreover, it extends the single-input single-output (SISO) to consider fading conditions. In both schemes, two system encoders and decoders are jointly optimized presence of minimize their symbol error rate (SER). We analyze resulting SER performance varying...

We introduce a powerful recurrent neural network based method for novelty detection to the application of detecting radio anomalies. This approach holds promise in significantly increasing ability naive anomaly detect small anomalies highly complex complexity multi-user bands. demonstrate efficacy this on number common real over air communications bands interest and quantify performance terms probability an false alarm rates across range interference band power ratios compare baseline methods.

We investigate sequence machine learning techniques on raw radio signal time-series data. By applying deep recurrent neural networks we learn to discriminate between several application layer traffic types top of a constant envelope modulation without using an expert demodulation algorithm. show that complex protocol sequences can be learned and used for both classification generation tasks this approach.

This paper investigates the significance of designing a reliable, intelligent, and true physical environment-aware precoding scheme by leveraging an accurately designed channel twin model to obtain realistic state information (CSI) for cellular communication systems. Specifically, we propose fine-tuned multi-step design process that can render CSI very close actual environment. After generating precise CSI, execute using obtained at transmitter end. We demonstrate two-step parameters' tuning...

We introduce a method for detecting, localizing and identifying radio transmissions within wide-band time-frequency power spectrograms using feature learning convolutional neural networks on their 2D image representation. By doing so we build foundation higher level contextual spectrum event understanding, labeling, reasoning in complex shared many-user environments by developing tools which can rapidly understand label sequences of events based experience labeled data rather than...

End-to-end autoencoder (AE) learning has the potential of exceeding performance human-engineered transceivers and encoding schemes, without a priori knowledge communication-theoretic principles. In this work, we aim to understand what extent for which scenarios claim holds true when comparing with fair benchmarks. Our particular focus is on memoryless multiple-input multiple-output (MIMO) multi-user (MU) systems. Four case studies are considered: two point-to-point (closed-loop open-loop...