A Very Brief Introduction to Machine Learning With Applications to Communication Systems

Abstract

Given the unprecedented availability of data and computing resources, there is widespread renewed interest in applying data-driven machine learning methods to problems for which the development of conventional engineering solutions is challenged by modelling or algorithmic deficiencies. This tutorial-style paper starts by addressing the questions of why and when such techniques can be useful. It then provides a high-level introduction to the basics of supervised and unsupervised learning. For both supervised and unsupervised learning, exemplifying applications to communication networks are discussed by distinguishing tasks carried out at the edge and at the cloud segments of the network at different layers of the protocol stack, with an emphasis on the physical layer.

End-to-end Learning of Waveform Generation and Detection for Radar Systems

“…In order to enable reinforcement learning, a loss metric is measured at the receiver and communicated over a reliable channel to the transmitter. A detailed review of the state of the art can be found in [9] (see also [10] for recent work).…”

Adversarial Deep Learning for Over-the-Air Spectrum Poisoning Attacks

“…T applies a deep learning classifier C T to identify Algorithm 2 T 's training algorithm 1: T collects sensing data over a time period to build its training data D train . 2: T builds a training sample {F t , S t } for each time t ≥ n new , where F t = (p t−nnew−1 , p t−nnew−2 , · · · , p t ), p t is the sensed power at time t, and S t is the busy/idle status at time t. 3 idle time slots. C T is pre-trained using a number of samples, where a sample for time t has the most recent n new sensing results p t−nnew−1 , p t−nnew−2 , · · · , p t as features F t and the current busy/idle status S t as the label.…”

Learning How to Demodulate from Few Pilots via Meta-Learning

“…As discussed in Sec. I, we view demodulation as a classification task (see, e.g., [1], [2]). Accordingly, given set D T , a demodulator p θ (s|y) can be trained by minimizing the crossentropy loss function:…”

“…We now consider a more realistic scenario including Rayleigh fading channels h k ∼ CN (0, 1) and an amplifier's distortion given by (2), where α k = 4 and β k is uniformly distributed in the interval [0.05, 0.15]. We assume 16-ary quadrature amplitude modulation (16-QAM) S and the sequence of pilot symbols in the meta-training dataset D and meta-test dataset D T was fixed by cycling through the symbols in S, while the transmitted symbols in the test set for the metatest device are randomly selected from S. The number of metatraining devices is K = 20; the number of pilot symbols per device is N = 32, which we divide into N tr = 16 metatraining samples and N te = 16 meta-testing samples.…”

“…For channels with an unknown model or an unavailable optimal receiver of manageable complexity, the problem of demodulation and decoding can potentially benefit from a data-driven approach based on machine learning (see, e.g, [1], [2]). Demodulation and decoding can in fact be interpreted as classification tasks, whereby the input is given by the received baseband signals and the output by the actual transmitted symbols, for demodulation, or by the transmitted binary messages, for decoding.…”