Background

The Microsoft Spectrum Observatory was created with the purpose of providing an intuitive presentation of the usage of the wireless spectrum. The project is sponsored by Microsoft's Technology Policy Group and the data is made freely available to the public. Data is recorded through monitoring stations and is stored and processed for visualization through the Windows Azure cloud. Eventually, the goal is to have spectrum monitoring stations set up all over the world. Microsoft believes that the data gathered on spectrum utilization will help inform policy discussions and decisions implicating various forms of spectrum management – ranging from reallocation of spectrum for exclusive use licensing or unlicensed access to dynamic access to allocated but unassigned spectrum. To support this effort by collecting spectrum usage data, measurement units have been installed by Microsoft in various locations around the world. Microsoft philosophy in developing this program has been to keep costs low and to keep the solution flexible and open to others.

Measurements and Base Stations

Each measurement station consists of a radio frequency spectrum measurement device that is connected to an antenna and mounted in a public space (typically at the top of a building). The device continually takes power spectral density measurements within the range of the RF sensor using a FFT (Fast Fourier Transform) processing technique. The device is programmed to record measurements for a number of frequency bands about every 3 seconds. These power spectral density measurements are stored locally on a computer connected to the base station and then aggregated to determine the average, minimum, and maximum values for 1-minute time intervals. The aggregated data is then automatically transferred to a cloud repository for storage and further processing.

Data Processing

As new data is uploaded, it is then processed even further. First, the data is normalized in both time and frequency. For time normalization, this means that measured data is grouped by hour into 60-minute segments and any minutes which do not have any measured data are encoded with a special value to represent NaN (not a number, or "no measurement"). For frequency normalization, this means that the power spectral density values at associated frequency points on a linear scale are interpolated on a logarithmic scale with a spacing of 200 points/decade. This allows more efficient data storage that is optimized for fast query and presentation.

After normalization is performed, multiple aggregations are computed. Daily, weekly, and monthly values are calculated as available. The values for average, minimum, maximum versus frequency are aggregated over time.

Data Visualizations

At this time the Microsoft Spectrum Observatory provides three different ways to visualize the spectrum data that has been measured and processed: Power Density, Utilization, and Spectrogram.

The Power Density visualization displays a two-dimensional line chart with frequency (in MHz) on the x-axis and power spectral density (in dB/Hz) on the y-axis. It is possible to view any combination of minimum, maximum, and average values by checking the corresponding boxes at the bottom. As a point of reference for interpretation, if no signals are being received, the average values will be a jagged line at or below -120 dB/Hz. This is often times referred to as the "noise floor", since the RFeye is not measuring any signal, only noise. As a general rule of thumb any data points at or above -90 dB/Hz can be considered "occupied", since there is likely a signal present at that frequency. This is, however, not a general rule since occupancy is influenced by other factors (transmitter power, receiver sensitivity, and many others). Finding good metrics for determining whether a particular frequency band is occupied is presently an active area of research.

For RF sensor setups that have been calibrated the occupancy visualization displays a two-dimensional line chart with frequency (in MHz) on the x-axis and occupancy percentage on the y-axis. The occupancy percentage is calculated as the percentage of data points over the specified time range that is greater than -90 dB/Hz. So, for example, if a measurement at 203.45 MHz finds a power spectral density value of -80 dB/Hz for the first 30 minutes of the hour and a value of -100 dB/Hz for the last 30 minutes of the hour, then the occupancy percentage would be 50% since half of the values in the time range (1 hour) are above -90 dB/Hz. The occupancy visualization also allows the selection of "All Frequency Bands" which performs an average across all frequencies in a given frequency band in addition to the calculation of occupancy percentage across all time values, and presents the results in a bar chart. The occupancy percentages presented in this bar chart tend to be lower because all frequency bands are included in the average, including guard bands (unused frequency bands intentionally allocated between adjacent bands to prevent interference).

For RF sensor setups that have been calibrated the spectrogram visualization displays a color image which represents a three-dimensional dataset: power spectral density (in dB/Hz) presented as a color versus both frequency (in MHz) on the x-axis and time on the y-axis. A color legend is displayed on the right. Gray values represent NaN (no measurement). Time progresses from the top of the spectrogram image towards the bottom of the image. One way to think about the spectrogram image is that each row of pixels represents a single Power Density visualization for that instant in time. As an example, if a Power Density visualization contains a section from 800 MHz to 850MHz that is above -60 dB/Hz, then there will be a bright yellow/green vertical bar at 800 MHz to 850 MHz, while the rest of the image is dark green or black. These vertical bars will show up if the signal at that frequency is constant over time. A good example of this is the TV bands which are constantly transmitting. If, however, the signal at a frequency is transient and varies over time, it will show up as alternating bright green and dark green dashes that go from the top of the image to the bottom. An example of this might be a frequency used by emergency response vehicles where there are brief periods of activity followed by periods of silence.

Contact Us

For more information about the Spectrum Observatory, please feel free to contact us using the contact information located here.