Quantifying Raw RF Dataset Similarity for Transfer Learning Applications

Abstract

Transfer learning (TL) has proven to be a transformative technology for computer vision (CV) and natural language processing (NLP) applications, offering improved generalization, state-of-theart performance, and faster training time with less labelled data. As a result, TL has been identified as a key research area in the budding field of radio frequency machine learning (RFML), where deployed environments are constantly changing, data is hard to label, and applications are often safety-critical. TL literature and theory shows that TL is generally successful when the source and target domains and tasks are similar, but the term similar is not sufficiently defined. Therefore, quantifying dataset similarity is of importance for analyzing and potentially predicting TL performance, and also has further application in RFML dataset design. This work offers a dataset similarity metric, specifically designed for raw RF datasets, based on expert-defined features and χ 2 tests, and systematically evaluates the proposed metric using synthetic datasets with carefully curated signal-to-noise ratios (SNRs), frequency offsets (FOs), and modulation types. Results show that the proposed dataset similarity metric intuitively quantifies the notion of similar signal sets, so long as the expert-features used to construct the metric are well suited to the application.INDEX TERMS dataset similarity, deep learning (DL), machine learning (ML), radio frequency machine learning (RFML), transferability, transfer learning (TL)

An Analysis of Radio Frequency Transfer Learning Behavior

“…Then, for each experiment performed herein, subsets of the data were selected from either the synthetic or captured master dataset using configuration files containing the desired metadata parameters. The synthetic master dataset is publicly available on IEEE DataPort [35], and the captured master dataset is described in greater detail in [32].…”

A Real-World Dataset Generator for Specific Emitter Identification

“…More specifically, BURP is capable of producing data suitable for a wide range of RFML use-cases, including signal detection, AMC, and SEI, with varying channel conditions, center frequencies (CFs), data rates, sampling rates, burst contents, burst lengths, transmit powers, and modulation schemes. Currently of interest are experiments related to evaluating RF transfer learning (TL) performance under such changes in channel conditions, transmitter/receiver hardware configuration, and use-case, extending recent research [47], [48] from synthetic to captured data. Such experiments aim to identify how the channel, transmitter/receiver hardware, and use-case impact learned behavior and facilitate or prevent successful TL.…”