HOME ANNOUNCEMENTS CONTACTS DATA INSTRUMENTS MAIN MENU MISCELLANEOUS PLOTS PUBLICATIONS SCIENCE WINDPROFILER PRINCIPLES For this page: INTERNAL LINKS 
THE NERC MST RADAR FACILITY AT ABERYSTWYTH V3 MST RADAR SIGNAL PROCESSING: SPECTRA


Introduction The figure below shows a typical Doppler spectrum observed by the NERC MST Radar. Power Spectral Densities (PSDs) are shown in both log (i.e. dB) and selfnormalised linear units as each panel emphasises slightly different features. They are stored in dB units in the spectral files (PSD_dB) but must be converted into linear units (PSD_linear_units) before processing is carried out: PSD_linear_units = 10.0^(PSD_dB/10.0) Positive values of Doppler velocity imply motion away from the radar. The number of velocity bins within a spectrum is given in the spectral and radial data files by by the variable number_of_complex_samples_in_discrete_fourier_transform, abbreviated here as N_DFT. Velocity bin numbers (b) run from (N_DFT/2) to (N_DFT/2  1). The corresponding Doppler velocity values (v_Doppler), which are stored by the spectral file variable doppler_velocity, are given by: where the spacing between velocity bins (delta_v_Doppler) is stored by the radial file variable spectral_velocity_bin_spacing. Note that spectra are cyclic about the Nyquist velocity, i.e. about velocity bin (N_DFT/2). Observed spectra typically consist of a single (quasiGaussian) radar return signal component (although there are sometimes additional signal components) superimposed on a background of white noise. The latter implies quasiconstant PSD values at all velocity bins. Removal of dc contamination DC biases in the raw radar samples lead to contamination of the PSD for the zeroDopplervelocity velocity bin, i.e. at b = 0. This must be removed from the recorded spectra before processing can begin. If a Hanning rather than a Rectangular window is used in deriving the spectra (this choice is recorded by the spectral and radial file variable data_weighting_window_index), the contamination is additionally spread to the velocity bins to either side, i.e. it affects velocity bins from (b = 1) to (b = +1). The contamination is typically removed by replacing the PSDs with values linearly interpolated between those at the velocity bins to either side of affected region. The dc contamination has been removed from the spectrum shown above. Determination of (mean) noise PSD The (mean) noise PSD (PSD_noise) is determined by the method of Hildebrand and Sekhon (1974). It is shown by the magenta line in the figure above. It can typically only be seen in the log plots as it tends to be very small compared to the peak signal PSD. The noise power in linear units (P_noise_linear_units) is simply the noise PSD integrated across the width of the spectrum: This is stored in dB units by the radial file variable noise_power (abbreviated here as P_noise_dB). The noise PSD in dB units can therefore be derived from the recorded noise power as: For spectra derived using a Rectangular window, leakage from strong signal components can affect all velocity bins  see e.g. Hooper (1999). This leads to an overestimate of the noise level, which is otherwise expected to be constant at all range gates. A better estimate of noise level can therefore be obtained by taking the median value over all range gates. The noise power is dominated by broadband lowerVHF galactic radiation, which undergoes a distinct quasidiurnal variation of almost 4 dB amplitude. Determination of signal component velocity bin limits A smoothed (linear) spectrum is derived by applying a 5velocitybin runningmean (cyclically) to the observed PSDs. The velocity bin limits of the signal component are determined by first locating the peak value of the smoothed PSD. The ratio of this value to the mean noise PSD, in dB units, is recorded by the radial file variable peak_smooth_psd_to_noise. Note that this is a measure of detectability and is not the same as the signaltonoise ratio, which is typically defined as the ratio of the signal power to the noise power. The smoothed PSD envelope is followed out to either side of this peak location until:
These (inclusive) limits  recorded by the radial file variables first_velocity_bin_number and final_velocity_bin_number, but abbreviated here as b1 and b2  are shown by the blue lines in the figure above. During the process of signal identification, the observed spectrum is first extended with a copy of itself being wrapped around the Nyquist velocity bin to either side, i.e. (b = N_DFT/2) and (b = +N_DFT/2). A partially velocityaliased signal component, i.e. one which is apparent at both Doppler velocity bin extremes of the original spectrum, is indicated in the radial files by having ones of its velocity bin limits being extended beyond the expected range of (N_DFT/2) to (N_DFT/2  1). Calculation of the principal parameters for the signal component The three lowestorder moments (m_0, m_1, and m_2) of the signal component are calculated from the the PSDs (in linear units) within the velocity bin limits which have been:
Following Woodman and Guillen (1974) or Woodman (1985): The signal component power (P_signal), radial velocity (v_R) and spectral width (σ_w) are given by: They are stored by the radial file variables signal_power, radial_velocity, and spectral_width, respectively. These can be used to define the magnitude, the central location and the standard deviation, respectively, of an equivalent Gaussian signal component  shown by the orange feature in the figure below  which give rise to the same values of the three lowestorder moments.


Page maintained by
David Hooper Last updated 12th September 2008 