At Dartmouth the audio cassettes were played back into a computer from a standard tape deck through a postemphasis analog filter to restore the signal response. A prerecording deemphasis was used to avoid tape saturation at high frequencies, caused by the normal audio preemphasis. The system then provides a flat response up to 12 kHz with added dynamic range at high frequencies at the expense of a poorer signal to noise ratio. A 32-kHz sample rate was used to digitize the analog tapes. Survey sonograms of the digitized data were made for each cassette tape. Furthermore, the resulting large digital data files were split into 20-Mb sections containing roughly 5 min (312.5 s) of data each, and more detailed survey plots of these files were created to look for interesting events. From these survey plots, interesting times were selected. Owing to the large quantity of data, only the most interesting tapes were processed based on notes made during the recordings while at Churchill, resulting in ~600 min of digitized data.
From this initial survey it became apparent that a major classification could be made based on the appearance of the fine structure. That is, the auroral roar, when viewed at high-time and high-frequency resolution, falls into one of two broad categories: structured or unstructured. An unstructured event appears as a uniform enhancement of noise, although sometimes it appears patchy in time and/or frequency. No amount of frequency or temporal magnification can resolve spectral features. Structured events contain at least some spectral and temporal features which distinguish them from noise or unstructured events.
Figure 1a shows a spectrogram of an auroral roar recorded during the campaign with the PSFR. The horizontal white lines running through Figure 1a indicate the frequency setttings of the DCR. The bandwidth of the DCR is roughly the vertical width of the PSFR pixel. Figure 1b shows DCR output starting at 0218:56 UT as indicated by the second set of vertical lines in Figure 1a. Figure 1b illustrates an example of unstructured roar fine structure. The record begins at a time of relative radio quiet at 2.97 MHz and then switches into unstructured roar at 2.86 MHz 10 s into the record. Figure 1c shows DCR output starting at 0208:16 UT as indicated by the first set of vertical lines in Figure 1a. Figure 1c shows an example of structured roar. This example illustrates that auroral roar is often composed of many different types of features: rising, falling, stationary, wavy, short, and long.
In a random survey of a subset of the digitized data, 18 1-min intervals contained structured auroral roar while 12 of these 18 intervals also contained unstructured roar. Nowhere in this survey did unstructured roar occur without at least one structured feature. In somewhat rare circumstances, roar can be unstructured but patchy in nature as opposed to the more common broadband phenomena in which the entire ~12-kHz bandwidth of the DCR contains no structure.
Structured auroral roar comprises many spectral and temporal features as shown in Figures 2a-2h. These examples were chosen from expanded survey plots to show a particular spectral or temporal feature. Each frame contains 3 s and 11 kHz of DCR data displayed at identical contrast levels to aid in distinguishing different types of features. Note that the center frequencies are not identical in all plots.
The structured features can be classified according to their duration, frequency drift, and grouping with like features. Within these classifications a nearly continuous spectrum of features were recorded.
The duration of a particular spectral feature is an obvious distinction. The time durations range over 3 orders of magnitude from the minimum measurable limit of tens of milliseconds to tens of seconds. Features lasting tens of seconds, a few seconds, roughly a second, hundreds of milliseconds, and tens of milliseconds are shown in Figures 1c, 2c, 2f, 2b, and 2a, respectively.
To quantify the distribution of durations of these features, a random subset of the digitized recordings was surveyed and analyzed. The study set consisted of 18 1-min intervals. Each of these intervals was plotted at several different time and frequency resolutions to ensure that all features present were identified. The durations of features were then scaled from these plots. One to ~40 features were measured during each interval based on the number of features present. The resulting data set, consisting of ~400 time duration measurements, is shown in Figure 4 as a histogram. Figure 4a shows the number of features observed as a function of duration on a linear scale grouped into 50-ms bins, while 4b shows the same data on a logarithmic scale grouped into 0.1-dB bins. Figure 4 shows that roughly 95% of the features last less than 1 s, 3% between 1 and 2 s, and 2% last longer than 2 s.
The spikes in the log distribution near the lower limit (-1.5 and -1.8) are most likely a result of quantization error due to the resolution of the measurement technique. In reality, these spikes are dispersed into the adjacent gaps giving a more uniform distribution which rolls off below -1.5(~0.03 s) at the low end. The apparent lower limit near -2.0 (10 ms) is due to the bandwidth of the signals which require at least a ~10-ms fast Fourier transform (FFT) for sufficient resolution to be identified. It cannot be excluded that features of shorter duration exist.
The apparent upper cutoff in the distribution is somewhat exagerated in part because features tend to drift through the bandwidth of the DCR and as a result are truncated in duration. In addition, the length of the random survey plots (1 min) may have truncated longer features. Also, the difficulties in predicting at what frequency roar would occur forced the operator to tune into the frequency band of an ongoing roar thus shortening some features. Perhaps the most influential aspect of the subjectivity involved in the time duration survey is the contrast level of the survey plots. It was necessary to adjust the contrast on some survey plots to discern the features from the noise. In some cases the intensity of a particular feature appeared to be intermittent thus making it difficult to judge whether a feature was, indeed, a single event or several. Despite these experimental difficulties the data establish that the majority of auroral roar fine structure features are typically less than a few seconds in duration and predominantly less than 1 s.
Another characteristic feature is the frequency drift of fine structures. Figure 2c shows features which drift in a variable manner that is nearly sinusoidal. Other variable drifting features are shown in Figure 1c and Figure 5.
Many groups are composed of individual features with a constant frequency drift. For example, Figures 2b and 2f show constant upward slopes, Figures 2d and 2h show constant downward slopes, and Figure 2e shows constant zero slopes.
The features with constant drifts vary greatly from ~-800 kHz s to ~+100 kHz s. Figures 2b, 2d, and 2h illustrate features with extreme frequency drift magnitudes > 10 kHz s, Figures 2a, 2d, 2f, and 2h show features with more moderate drift magnitudes < 10 kHz s, and Figures 2a and 2f depict roughly stationary features that drift very little. It became apparent from inspection of the survey plots that the slope of extremely steep drifting features is much more commonly negative than positive. To quantify this observation, the initial ~5-min survey plots were carefully analyzed for features with constant drifts, in particular those with drift magnitudes ~10 kHz s. The slopes of features were not randomly selected in order to ensure that we included the largest observed drifts. In this survey all features with drifts exceeding ~+/- 10 kHz s were measured, but for lesser slopes only a few representative events were measured.
Figure 6 shows a histogram of the number of features with constant drifts versus frequency drift. Because only a few representative features were measured from the large number of slopes <~10 kHz s in magnitude, the middle of the distribution is undersampled. The negative drift tail in the distribution in Figure 6 is longer than the positive drift side, supporting the notion that more negative steeply drifting features exist and that the steepest tend to be negative. Of the 139 features measured with slopes > 10 kHz s in magnitude, 107 were negative, and the largest negative and positive slopes observed were -790 kHz s and +88 kHz s, respectively.
Solitary features, those occurring by themselves during a time much longer than the feature duration, are observed in less than 1% of the structured examples of the digitized data. Features more often occur in groups with similar characteristics. These groups contain up to several hundred features. Figures 2h and 2c show groups containing several like features, Figures 2b, 2e, and 2f show groups with many more features, and Figure 2a shows a group with nearly 100 within 3 s. The spacing of individual features in the 18 random 1-min survey plots ranges from ~150 Hz to several kilohertz with most of the features spaced closer than 1 kHz. The median spacing of these features is ~450 Hz. Figures 2a-2f, 2h, and 5 show at least some features spaced closer than 1 kHz.
The minimum bandwidth of auroral roar fine structure is an important clue in determining the generation mechanism. Any model proposed must provide explanations for the coherence of the emissions. Some mechanisms may be elimated on the basis that they provide no feasible explanation of the observed fine structure. To improve the upper bound of the minimum bandwidth of roar fine structure, it is necessary to analyze stationary features. Figure 7a is a spectogram showing several relatively stationary features observed at 0207:23 UT on April 15. In particular, the feature beginning 2 s into the record and highlighted with a box is stationary in frequency for roughly 1 s. The inset in Figure 7b shows a series of spectra each of which represents a 0.256 second FFT, implying a full width at -3 dB of ~5.5-Hz resolution. The measured bandwidth of 6 Hz is the resolution limit of the FFT. The actual bandwidth of the signal may be less than 6 Hz but not greater.
Difficulties in locating a feature which remains stationary for long enough to give better frequency resolution prevent further refinement of the minimum bandwidth. Tape flutter and wow in the recording equipment are ~5 Hz which prevents further refinement even if more stationary features are found thus allowing longer FFTs. However, a new digital DCR is being designed which addresses these difficulties.