Radio Explorations: Computing Identities of Transmissions

Abstract

The SNSF-funded Radio Explorations project engages with a digital archive of radio signals (SIGID Wiki) collected by radio enthusiasts. Radio Explorations operate within the continuum of societal and technological concerns, addressing the onto-epistemologies of radio signals: the process of their categorization and identification. Radio transmissions are hard to characterize because most signals do not have a static representation: especially when transmitting data, signals have different modes, phases, and other temporal variations. Starting from an unordered collection of recordings of different transmissions and their meta-data (frequency, bandwidth, mode, location), the aim of this project is to articulate signals' identities in terms of their own characteristics (rather than pre-existing ontologies). To this end, I examine the capacity of machine learning techniques to support identification of environmental radio transmissions. With artificial neural networks (ANN) of the self-organising map (SOM), I articulate a 'data observatory' that orders data on radio signals based on computable similarity. The 'data observatory' is a digital tool, a navigation apparatus which can be used to orient oneself in the vast landscape of data on radio transmissions. I do not propose to understand these identification processes as world making but, on the contrary, as arbitrary renderings of reality in the eyes of a machine, affirming inherent instability and flexibility of a signal's identity. By rendering signals commensurable in this way, I propose to take an active stance with regards to machine learning algorithms and expose a research interest from which we can learn and tell stories about signals.