The Influence of Signal Presentation Factors on Performance of an Immersive, Continuous Signal Detection Task

Abstract

Conventional independent trial signal detection tasks are representative of many real-world tasks, such as identifying flavors or spotting abnormalities on an x-ray film; however, there is also a wide range of task types that are not accurately modeled by the standard yes/no decision making paradigm. Active monitoring and driving hazard detection tasks, for example, do not occur as a set of independent events but are executed in response to a continuous flow of incoming stimuli. Identification of a potential threat is therefore dependent on a multitude of factors that may be influenced by not only the characteristics of the threat (the relevant signal) but also by ongoing events (dynamic noise) and one’s environment. Although signal detection theory is typically applied to trials of static signal-noise constructs, it may be extended to many more task types by relaxing the temporal independence of trials and allowing both noise and signal stimuli to be presented dynamically. The primary drawback of these modifications is the reduction of control over the noise that accompanies any given signal. Fortunately, a reduction in that control does not necessitate a loss of control; instead, it changes the tactics that must be employed to ensure the production of usable data. Rather than attend to signal-noise characteristics between trials, designers of a continuous signal detection task must focus on those characteristics as a function of time and also consider the potential lifespan and evolution of any given signal. This paper considers the presentation of signals in a continuous monitoring task which was administered in two experiments conducted in immersive virtual reality. The factors that were found to most strongly influence participants’ detection tendencies are discussed with regards to their impacts and the means by which their effects may be mitigated or controlled. Both signal and noise characteristics were manipulated across the two experiments; the influence of each is evaluated in the context of the resulting signal-noise construct which was presented to participants. Additionally, standard signal detection task parameters such as event rate, signal salience, signal origin type, signal-to-noise ratio, etc. are discussed with regards to their extension into the continuous domain.