Atmospheric Noise: Data Collection and Analysis C.O. Lee Boyce Jr., Sherman C. Lo, J. David Powell, Per K. Enge Stanford University Abstract system since it is a navigation system. To help evaluate the usefulness of the ITU at- With its wide bandwidth and large amplitude spikes, mospheric noise model, the authors proposed collect- atmospheric noise can dominate the Loran band (90- ing atmospheric noise data in the Loran band. We 110kHz). Data collection efforts over the spring and collected data during the peak storm seasons of the summer of 2005 in Norman, OK and over the summer spring and summer months within 2005 in New Mex- at Langmuir Laboratory outside of Socorro, NM have ico, and Oklahoma. We obtained amplitude, fre- captured some of these large amplitude signals as well quency, and timing data of atmospheric noise using as envelope data from the nearby Loran stations. The a 35 kHz bandwidth receiver centered at 100 kHz. data were captured using a receiver with a front end This receiver encompassed the entire Loran band of that had 35kHz bandwidth centered at 100kHz and 90-110 kHz and we will use its data to refine the at- used a Locus monopole antenna. A PC with a high mospheric noise model. speed A/D card recorded in-phase and quadrature This paper covers the results of data collected in data of Loran signals and atmospheric discharges at 2005 from Norman, OK and expands on the data pub- 50kS/s. With proper processing, the data collected lished in [3]. We found that our 35 kHz data corre- compares well with the existing atmospheric model sponded well to the ITU model provided we processed from ITU P.372-7. the data in a similar fashion. Also, we found correla- tion between the voltage deviation, Vd,andthenoise envelope which was only glossed over in the original 1Introduction model. While additional data were captured with a 200 Hz wide receiver, only some of these results will 1.1 Overview be discussed here. The Loran signal present in the Over the past five years, researchers have investi- band corrupts the 200 Hz data more than that of the gated Loran as a potential backup navigation system 35 kHz and makes many of the results less insightful. for aircraft to mitigate the effects of a GPS outage. In December 2004, the Loran Integrity Performance 1.2 Atmospheric Noise, CCIR 322, Panel (LORIPP) proposed [1] that the current Loran and ITU P.372-7 system with some enhancements may be used as a backup navigation system capable of supplying Re- Atmospheric noise generated by cloud-to-cloud and quired Navigation Performance of 0.3 nautical miles cloud-to-ground discharges is a wide bandwidth and to the aviation community. large amplitude noise that corrupts the Loran sig- As part of that work, the LORIPP found a limiting nal. These electrical discharges are sporadic and are factor in the availability of Loran across the cotermi- very non-Gaussian in both their amplitude distribu- nous United States stemmed from the worst case es- tion and in their time of arrival. Hence, the noise is a timations of atmospheric noise. They based the noise non-stationary process which introduces more para- estimates on a model described in the recommenda- meters that what would be required from a stationary tion by the International Telecommunications Union one. (ITU) P.372-7 [2]. The recommendation’s radio noise In order to model this noise process the Inter- model applied directly to radio communications, but national Radio Consultative Committee (CCIR) be- the LORIPP questioned its applicability to the Loran gan in 1957 a four year data collection survey of at- Page 1 mospheric noise and recorded atmospheric noise data Atmospheric Noise 50% Worst Case of All Given Times and Seasons [dBμV/m] from 10 kHz to 20 MHz. Their aim was to help in 50 designing communication radio systems subjected to 45 55 such noise. Over time, CCIR added more data to 55 the report which became CCIR 322-3 [4]. In 1992 65 40 CCIR merged with the International Telecommuni- 65 cations Union (ITU) and the CCIR version of the radio noise document was superseded by the current Latitude [deg] 35 60 60 version, Recommendation ITU P.372-7[2]. Since the 60 original document was produced by CCIR, such will 30 be the designation of that report for the remainder of this article. 25 -120 -110 -100 -90 -80 -70 -60 Lightning varies in intensity throughout the year Longitude [deg] and throughout the day. In order to characterize such variability, CCIR created data models parameterized 50 55 60 65 70 75 80 85 90 Noise [dB μV/m] by four seasons and six 4-hour time bins that corre- RMS sponded to the local time. Using 16 receiver stations around the world, CCIR measured the median noise levels across eight frequency bands from 13 kHz to 20 Figure 1: Median 50% noise level over all times and MHz. With their data, they mapped out the spatial seasons for a 35kHz wide receiver centered at 100kHz. variation of atmospheric noise across the globe. The primary data measurement made was external quency of 100 kHz and a 35 kHz bandwidth during antenna noise factor, Fa, the power received through a loss-free antenna averaged over a 15-minute period. the summer time block from 1600-2000h. We will re- At each of the eight frequency bands, they mixed the fer to this season and time block as our worst-case signal down and passed it through a 200 Hz bandpass period throughout this paper since CCIR predicts it filter. With further analysis, CCIR developed tables to be the most sever for atmospheric noise. to extrapolate the noise data across this frequency The discontinuity at 51 dB μV/m is the result of range and for any arbitrary bandwidth receiver. An using two different log-normal distributions of differ- example of the median 50% noise level converted to ing variances on either side of the median value. The root-mean-square (rms) electric field (E-field) values median value also has an additional standard deriva- for a 35 kHz bandwidth centered at 100 kHz during tion on top of it of 6.5 dB, that is assumed to be the worst time block across all four seasons is shown normally distributed. Thus we may need to shift in Figure 1. This choice of bandwidth and center thewholecurvetotherightbysomanysigmasto frequency characterizes the receiver we used to gather improve our confidence in bounding a given level of our noise data and therefore should correspond to the noise. Thus the prefixofmedian in the term “me- noise found within the Loran band. dian 50% noise level” indicates that we are using the Note that Figure 1 gives the expected value or the expected value of the 50% noise level and have not median of the 50% noise level. From their research, adjusted it by some factor of 6.5 dB, so we have 50% CCIR found that atmospheric noise follows a log- confidence that the 50% noise level will be at some normal distribution over the course of a year. Since specified value or below. they collected many years of data, they could deter- In order to get values larger than the median 50% mine the variation in the median value of the distri- noise level, we can extrapolate out along the upper bution. Therefore, CCIR provides a median value as part of the log-normal distribution. An example of well as a variance on the median itself and a stan- doing so to the median 99% level is shown in Figure dard deviation for the log-normal distribution itself 3. To appreciate the severity of atmospheric noise, to describe the noise. Figures 4 and 5 show the anticipated coverage of the A log-normal distribution looks like the normal dis- Boise City tower if we use the median 50% and me- tribution when plotted on a log scale. CCIR found dian 99% noise levels respectively and estimate the that the rms noise values may be modeled using two cut-off SNR below which we cannot track at -10 dB different log-normal distributions, one to account for SNR. values below and one for above the median rms noise The 15-minute averaged rms E-field measurement level. Figure 2 shows the expected noise probability represented the value assigned to that particular distribution function for a receiver with a center fre- hour in a given time bin. In order to relate the Page 2 PDF of E-Field RMS over 15 min Intervals, BW 35000Hz, Time Block 1600-2000 SNR for #16 Boise City at 50% Level 0.035 50 Upper Distribution Lower Distribution 0.03 -5 45 -5 0.025 -10 V/m] 40 -10 μ 5 15 0.02 5 15 20 35 20 0 -10 0.015 20 0 10 -10 Probability Density [1/dB 0.01 30 10 0.005 25 -120 -110 -100 -90 -80 -70 -60 0 10 20 30 40 50 60 70 80 90 100 Envelope E-Field [dB μV/m] -10 -5 0 5 10 15 20 SNR [dB] Figure 2: Probability Density Function for Summer Figure 4: Boise City coverage given the median 50% 1600-2000h with 35kHz BW at 100kHz. noise level with a cut-off at SNR = -10dB. Atmospheric Noise 99% Worst Case of All Given Times and Seasons [dBμV/m] SNR for #16 Boise City at 99% Level 50 50 45 45 80 40 80 40 85 80 85 5 5101520 -50 1020 70 -50 Latitude [deg] 35 85 70 5 101520 35 -50 75 75 70 -10 30 30 60 65 75 -10 60 65 25 -120 -110 -100 -90 -80 -70 -60 Longitude [deg] 25 -120 -110 -100 -90 -80 -70 -60 60 65 50 55 60 65 70 75 80 85 90 -10 -5 0 5 10 15 20 Noise [dB μV/m] RMS SNR [dB] Figure 3: Median 99% noise level over all times and Figure 5: Boise City coverage given the median 99% seasons for a 35kHz wide receiver centered at 100kHz.