Cooperative UK Twin Located Auroral Sounding System Finland HF radar observations of E region coherent spectra are presented for several events. The Burg spectrum method is applied to study microstructure of the spectra. It is discovered that many spectra are two-peaked, and quite often there is a systematic pattern in their occurrence; the double-peak echoes are typically observed at intermediate and far ranges of the E region echo band and at ranges farther than the power maximum in the range profile. The typical separation between the peaks is about 150 m/s and hardly changes with the azimuth of observations. Velocities of two components in a double-peak spectrum are typically larger and smaller than the velocity of the unresolved spectrum and single-peaked velocity of echoes at shorter ranges. It is hypothesized that the two components of the echoes occur because of signal reception from the top and bottom of the electrojet layer.
 Many researchers assume that the observed differences in the echo characteristics at HF and VHF are simply a scale effect of the classical Farley-Buneman (FB) and gradient-drift (GD) plasma instabilities [e.g., Milan and Lester, 1998; Jayachandran et al., 2000]. In a series of papers by Uspensky et al. [1994, 2001, 2003] and Makarevitch et al. [2001, 2002a], HF echoes were also related to the FB and GD plasma instabilities, but the observed differences between HF and VHF echo characteristics were attributed to significantly stronger radio wave refraction at HF. For example, Uspensky et al.  expected systematically increased spectral width of HF echoes at the far edge of an E region echo band since signals here can be simultaneously received from two different heights, from the top and the bottom of the electrojet layer. One should note that Villain et al.  reported such a spectral width jump but attributed it to the onset of the ion-cyclotron instability. Whether refraction indeed complicates the studying of the plasma physics of irregularity formation at HF frequencies as compared to VHF frequencies needs further exploration.
 In this paper we study the E region HF spectra with better than the standard SuperDARN spectral resolution. This is achieved through employing the Burg spectral analysis method, similar to Schiffler et al.  and Huber and Sofko , to resolve features not currently visible in a standard Fourier transform. We concentrate on three E region events observed with the Cooperative UK Twin Located Auroral Sounding System (CUTLASS) in Northern Scandinavia. Two of the events were previously studied by Danskin et al.  and Koustov et al.  by considering only the standard SuperDARN FITACF velocities. In the course of the data analysis, we attempt to understand, from various aspects, the reasons for the microstructure of the observed HF spectra.
2. Experiment Setup and Event Description
 We consider CUTLASS Finland radar observations on February 12, 1999 (1400–1600 UT), February 11, 1999 (1400–1800 UT) and September 3, 1997 (0200–0400 UT) with the first two events being our main focus. The reason for this selection is not accidental. For the first two events, simultaneous VHF (STARE) data were available, and the EISCAT incoherent scatter radar monitored plasma convection in the area 200–400 km poleward of the HF area of observations. Koustov et al.  compared HF velocities, VHF velocities and the EISCAT convection data for the February 12, 1999 event and concluded that HF velocities were much smaller than the cosine component of E × B plasma convection. The authors suggested that perhaps HF echoes were significantly affected by the scatter from the bottom of the electrojet where the decameter irregularity velocity can be significantly depressed by collisions and, in addition, they anticipated a possibility of other neutral wind related instabilities. Our analysis of February 11, 1999 event in a manner similar to Koustov et al.  confirmed their major conclusions. For both events no additional evidences on the importance of the lower ionosphere were found. It is natural to make a closer look at the HF Doppler spectra for these specific events. The third event was chosen just to have significantly different period of observations.
 Details of the experiment for the first two events can be found in Danskin et al.  and Koustov et al. . As a brief summary, we repeat that the CUTLASS Finland radar operated in the fast common mode with a full scan through 16 beam positions being completed in 1 minute. The HF echo power, Doppler velocity and spectral width were determined in 45-km range bins from 180 km to ∼3500 km. Data from beam 5 were mostly explored in this study, since it overlooks the region (at a range of 900 km from the HF radar site) where the EISCAT electron density and electric field measurements were performed. Even though the typical ranges of E region echoes were ∼600 km, the EISCAT information is valuable for various estimates. We concentrate in this study on observations between 1400 and 1800 UT, equivalent to 16–20 MLT at the location of EISCAT. On September 3, 1997 the CUTLASS Finland radar worked in the standard 2-min scan mode.
 To give an example of typical slant-range distribution of short-range CUTLASS echoes, we show by dots in Figure 1 the times when ionospheric echoes were observed for the February 12 event, beam 5. The echoes were detected almost all the time although in quite different bands. Between 1000 and 1400 UT two echo bands are evident, one around 1000 km (band I) and the other one at 1800–2000 km (band II). These are F region echoes received through direct and 1 and 1/2 hop propagation paths, respectively [Danskin et al., 2002]. This mode identification is supported by general echo statistics [Milan et al., 1997], by the angles of echo arrival estimates, and by ray tracing based on electron densities profiles measured by EISCAT [Danskin et al., 2002]. Band II corresponds to scatter near the cusp/cleft region. After 1400 UT, another band (III) of echoes was predominantly seen at much shorter ranges, typically 400–700 km with a slow displacement of the echo band in the north-south direction. According to the interferometer measurements, these were E region echoes [Danskin et al., 2002]. Similar mode identification has been performed for two other events.
 In Figure 1, we indicate the times for detection of multipeak spectra by small vertical lines. The length of the lines reflects the separation between two major peaks in each spectrum (note the scale at the bottom of the diagram). Consistent with Schiffler et al.  and Huber and Sofko , the far-range cusp/cleft related echoes (band II) have a significant number of double-peaks. There were some, not significant, double peak presence within the band I of the F region echoes. For band III, E region echoes after 1400 UT, the most remarkable feature is double-peak nature of many of these echoes and this feature is the main goal of the present study.
 Next we will describe the method of spectra analysis giving better than usual spectral resolution and present examples of E region spectra consisting of two (and more) spectral components.
3. Methods of Spectral Analysis
 Throughout the paper we refer to 3 types of estimates for the Doppler velocity. The traditional way in the determination of velocity for a SuperDARN radar is through the complex autocorrelation function (ACF). An 18-lag ACF with a lag separation of 2400 μs is derived from the 7-pulse pattern that is generated by the radar. In the high time resolution mode of CUTLASS, a ∼3 second averaged ACF is used in the analysis.
 The first method, called the FITACF method, is used routinely with CUTLASS/SuperDARN data. For this method, one assumes that there is only one peak in the spectrum; hence the rate of change of the phase of the ACF is the velocity [Villain et al., 1987]. Also from this method, the spectral width can be estimated by making the assumption that the ACF decays exponentially (or according to the Gaussian law). The width is then related to the rate of the decay of the magnitude of the ACF.
 The second method is the Fourier transform (FFT) of the ACF, which results in the power spectrum. This method has the advantage that it gives the power spectral density (PSD) as a function of velocity, so a multiple component spectrum can easily be seen. The velocity of the strongest peak of the FFT is referred to as the FFT peak velocity. The FFT method has two major drawbacks. First, it is computationally intensive (for real-time data processing) and secondly, with the current pulse pattern and lag separation, it gives a velocity resolution of ∼100 m/s which we refer to in this paper as the usual spectral resolution.
 The third method that we consider is the Burg spectrum method [see Kay, 1988; Naidu, 1996]. This is an autoregressive method that fits a statistical model to the data. This model is designed for short data sequences with the number of solutions equal to the order of the model. For the current analysis, an 8-th order Burg spectrum was calculated. The order of the model must be less than one half the number of data points in the ACF. The Fourier transform of the model can be computed to give an apparent spectrum, which is substantially smoother looking than the standard FFT of the ACF. We identified the existence of the second peak in the Burg spectrum only if its power was not less than 8 dB below the power of the main peak. We also did not consider all subsequent peaks to focus on two major ones. Schiffler , Schiffler et al. , Huber , and Huber and Sofko  have applied this method to SuperDARN data to illustrate the presence of double peaks in the cusp/cleft footprint region of the ionosphere. In general, the peaks of the Burg spectrum are the multiple velocity components in the radar data.
 The FITACF and Burg methods will be used extensively in this paper. The FFT method is used as a check on the validity of the estimates from the other methods. In the next section all three methods are used to illustrate how they work for the estimate of the velocity.
4. Range Changes of HF Spectra (February 12, 1999, Event)
Figure 2 gives an example of spectra observed at 1430 UT in several range gates of beam 5, using the previously described methods. The thin line shows the standard FFT spectra while the smoother heavier line represents the Burg spectra. The vertical dotted line is the velocity estimate from FITACF. The spectra are normalized so that the maximum peak has a value of one. In the upper left corner of each panel the range of the gate is given. In the bottom right corner from top to bottom are the power in dB, the Doppler velocity in m/s and the spectral width in m/s as determined by FITACF. In the upper right corner, Doppler velocities for one or two peaks of the Burg spectrum are given.
 Looking at the spectra at various gates, one can see that they are single-peaked at the near ranges of 450, 495, and 540 km. Their mean Doppler shift increases with range with all three methods giving consistent results. The spectra at the next three ranges are distinctly double peaked. Clearly, the low-velocity (high-velocity) peaks have a velocity below (above) the typical mean velocity of shorter-range single-peak spectra and below (above) the mean Doppler velocity determined through the FITACF method (FITACF velocity). The shifts of both the low- and high-velocity component changes with range; the shift of the high-velocity component increases while the shift of the low-velocity component does not have a clear trend. The double-peaked structure of the Burg spectrum is not recognizable at 720 km, although two velocity values are given in the top right hand corner of the relevant panel. Also, the FFT spectrum has an additional peak at −400 m/s. This peak was not reported by the Burg method because its power was well below the power of the main peak (at small velocities). The spectrum is single-peaked at far range of 765 km. The Doppler shift of this spectrum is of the order of the low-velocity components detected at shorter ranges. Comparing the FFT and Burg spectra presented in Figure 2 one can conclude that, generally, the Burg method reports two peaks whenever an FFT spectrum is strongly asymmetric.
Figure 3 presents further examples of the spectra, on this occasion observed at 1450 UT. For this period, the echoes were seen at larger ranges and the power of echoes maximized around 630 km as opposed to 540 km in the case of Figure 2. The morphology of the spectra change with range is very similar to the data presented in Figure 2. One can see single-peak spectra at short ranges of 540 and 585 km, followed by double-peak spectra at 630–765 km. The very last spectrum at 810 km also has 2 peaks, but the second one is of low intensity and is way off typical velocities one would expect for the period under study (1372 m/s is comparable to the F region plasma convection at this time, as measured by EISCAT). Again the Doppler shift of the low-velocity (high-velocity) components is below (above) the FITACF shift and below (above) the Doppler shifts of short-range echoes.
 Other spectra observed between 1400 and 1600 UT show a similar pattern of the double-peak spectra onset. Close inspection of band III in Figure 1 shows that indeed double-peak spectra occur typically in the middle part of the echo band and quite often they may be seen at the farthest ranges as well. We also analyzed Hankasalmi data obtained with 15-km resolution for one event and found the presence of the double-peaked echoes consistent with those of this study for which the range resolution was 45 km.
5. Azimuthal Distribution of Double-Peak Echo Occurrence
 The data presented in Figures 2 and 3 refer to one beam/direction of observations. A natural question is whether the above morphology of double peak occurrence is typical for other directions. In Figure 4 we present data for several Hankasalmi radar scans. Figure 4a shows the scan whose data from beam 5 were reported in Figure 2. Figures 4b and 4c show data from February 11, 1999 event. Figure 4d presents data for the September 3, 1997 event. All four diagrams indicate that double-peak echoes occur predominantly in the middle part of the echo band in all beams; echoes are typically single-peaked at the closest and farthest ranges of the bands. In Figure 4a, most of the echoes are within the ranges of 450–620 km with an absence of double peaks in the western corner of the echo band at farthest ranges. In Figure 4b the echo band stretches in the southeast direction with double peaks at far ranges of the most western beams and at lower ranges in the eastern part of the scan. In our opinion, such a pattern in echo occurrence is formed because of the nonuniform density distribution over the scanned area so that the electron density is enhanced in the eastern part of the radar's field of view. If one looks at ranges ∼585 km one can see that single-peak echoes in the western part of the scan are changed to double-peak echoes in the eastern part of the scan. Both Figures 4a and 4b seem to hint on the absence of L-shell alignment for double-peak occurrence. The effect is more obvious in Figures 4c and 4d where as one goes in the clockwise direction for the farthest ranges, the double peaks occurrence in western beams is changed to the single peaks occurrence in central beams.
 To assess how often the double-peak spectra occur, we examined the data in all beams for the three events of February 11 and 12, 1999 and September 3, 1997. The latter event had an integration time of 7 s. A total of ∼50,000 spectra for echoes at ranges below 855 km were considered. We found that double-peak spectra present in about 35% of cases independent of the integration time. This value varies slightly from one radar beam to another.
6. Some Statistics on the DP Echo Parameters
 In terms of the velocity separation between the peaks, Figure 1 has minor deficiency in that one cannot recognize cases with small peak separation. This diagram also does not indicate the exact velocities associated with double peaks. Data in Figures 2 and 3 have this information but present only two individual events for one beam. To give more quantitative ideas on what is typical for double-peak spectra in terms of the occurrence rate, mean velocity and peak separation, we present some statistics. We consider for this purpose the 12 February 1999 data.
 In Figure 5 the power of short-range echoes (Figure 5a) and their FITACF velocity (Figure 5b) as well as the double peak occurrence (Figure 5c) and the separation between the peaks (Figure 5d) are given for all Hankasalmi beams during two hours of observations, 1400–1600 UT. Echo occurrence is limited to a band of ranges. The slant ranges for the band is closer to the radar at larger beam numbers as one would expect from the variation in aspect angle. The echoes were of moderate power, with the exception of the closest and farthest ranges. On average, the velocity magnitudes were low (<100 m/s), and as the beam number increase, they changed from negative to positive. The positive velocities were observed in beams 13–15. The velocity variation with the azimuth of observations (beam number) reflects the fact that the plasma convection component changes with flow angle, as reported in other studies [e.g., Jayachandran et al., 2000]. An implication would be that the mean flow was roughly along the magnetic L shells (beam 9 is almost perpendicular to the L shells). The double-peaked echoes were observed in all beams. From the occurrence rate (Figure 5c), one can conclude that the double-peak echoes are more frequent in the central beams where the occurrence rate was as high as 70%. Interestingly enough, the ranges of DP occurrence were slightly higher than the ranges of maximum echo power (Figure 5a). Double peaks occurred for both positive and negative velocities (Figure 5c). If one considers average DP echo occurrence over all ranges of observations (<845 km), the occurrence rate would be around 35% as we reported in Section 5. Figure 5d gives an idea on a typical peak separation and its dependence on the azimuth of observations. In preparing this diagram, those peaks that were separated by more than 400 m/s were not included in the statistics (these are not frequent events and their origin requires a special consideration). The typical separations were 150–200 m/s. Clearly, there is no azimuthal effect. An obvious trend is an overall increase in the peak separation with the range (within the band of echoes) for almost all radar directions.
7. FITACF Velocity and Velocities of the Peaks
 An important question is how the velocities of the peaks are related to the FITACF velocity. Figure 6 gives the distribution of low- (blue crosses) and high-velocity (red diamonds) Doppler shifts versus the corresponding FITACF Doppler shift for several radar beam positions. Figure 6b confirms statistically for beam 5 that the “resolved” components of the spectra have shifts above and below Doppler shifts of unresolved spectrum (as was concluded from Figures 2 and 3 for individual measurements, Section 4). Figures 6a, 6c, and 6d demonstrate that the low- and high-velocity components change in accord with a general decrease of the FITACF velocity as the beam number increases. Once again, the FITACF velocity decrease is expected due to the orientation of the beam being more perpendicular to the direction of the electrojet. With increase in beam number the low-shifted components assume more frequently an opposite polarity as compared to the high-velocity components. One can conclude that typically the FITACF velocity is somewhere in between the double peaks of a resolved spectrum.
8. Double Peaks and Spectral Width
 The presence of double peaks in the spectra of coherent echoes implies an increase in their spectral width, and our analysis does support this expectation. Figure 7 indicates that the spectra are broader for stronger peak separation. What is more interesting for the HF echoes is that, quite often, the low- and high-velocity peaks have a tendency to diverge with an increase in range. Such a tendency can be noticed in Figures 2, 3, and 5. To illustrate further, in a more statistical fashion, we looked at spectral width distribution versus range for all CUTLASS Finland radar beams on February 12, 1997.
Figure 8 presents range distributions for the averaged echo power (Figure 8a) and spectral width (Figure 8b) for the period of 1430–1440 UT for all beams as determined by the FITACF technique. We focus on ranges from 180 to 945 km. The width and power are represented by colored contours with the respective scales shown on the left. The red color in Figure 8 indicates that echo power and width were above the highest level, 30 dB or 250 m/s, respectively. In Figure 8c we superimpose on the width contour plot (Figure 8b) the two highest contours of the power emphasized in Figure 8a by thick black solid lines.
Figure 8a shows a very smooth variation of power with both beam number (azimuth of observations) and slant range, in agreement with 2-hour statistics presented in Figure 5a. The stripe of E region echoes covers almost the entire field-of-view. The slant ranges for the echoes are shifted closer to the radar at larger beam numbers. For beams 0 and 15, the echo range spans are 360–810 km and 315–720 km, respectively. Similarly, the slant range of power maximum along each radar beam was smaller at larger beam numbers (495 km for beam 0 versus 630 km for beam 15). One should note also that the intensity of the power maximum increases with the beam number from ∼25 dB for beam 0 to >30 dB for beam 15.
 The width distribution (Figure 8b) also shows a stripe-like structure although a bit more “ragged.” One can clearly see that the width maximizes at ranges 550–650 km. Unlike the power, there is no width maximum increase toward high-number beams; instead width seems to have a maximum somewhere between beams 3 and 12. The other prominent and important feature of Figure 8 is that width maxima along radar beam are shifted farther from the radar than the power maxima. This feature is illustrated in Figure 8c where we have superimposed power contours over the background of width contour plot. The white areas corresponding to maximum power contours do not cover the red areas where width is at maximum; the power maxima are closer to the radar by ∼100 km. The range shift between locations of strongest and broadest echoes is reminiscent of the range shift between strongest echo occurrence and DP echo occurrence (Figures 5a and 5c).
9. Double-Peak Echoes and Types of HF Scatter
 The double-peak nature of HF echoes can be related to two mechanisms of irregularity formation, the Farley-Buneman and the gradient-drift plasma instabilities, as was done in a number of VHF studies [e.g., Greenwald et al., 1975]. In this scenario, one believes that the double-peaked spectra are resulted from azimuthal convolution of Type 1 and Type 2 scatter. Milan and Lester [1999, 2001] considered a large number of E region HF CUTLASS echoes and sorted them according to their characteristics, including azimuthal variation of the velocity. These authors found more than just classical Type 1 and Type 2 scatter in the body of HF echoes. Since only FITACF velocities were considered, the double-peak nature of many of the echoes was not taken into account.
 One may wonder whether the idea of Type 1 and Type 2 scatter convolution can be used as an explanation of the double-peak echoes presented in this study. Unfortunately, there is no definitive answer to this question. First of all, the standard spectral resolution of the SuperDARN measurements is not good enough to make analogies with the previous VHF measurements. One of the problems is that the width parameter for CUTLASS echoes is about the same for Type 1 and Type 2 scatter [Milan and Lester, 2001]. One can attempt to look statistically at the azimuthal characteristics of the spectra. Unfortunately, the data-processing technique of the Burg method that we used in this study cannot be easily employed for mass processing of the data to study the flow angle/azimuthal variations of double-peak echoes. For this reason, we leave such comprehensive analysis for future work.
 The considered events were limited to observations at large flow angles, as known from the electric field measurements over EISCAT (or STARE convection estimates in closer areas) for the three events. It is not a surprise that among considered spectra there were very few events with a high velocity component. As we mentioned, the high-shifted component of double peaks was below 400 m/s, a typical ion-acoustic speed in the auroral E region.
 In Figure 9 we present an example of rare CUTLASS-Finland observations when a high velocity component was observed in low number beams. Figure 9a shows the FITACF velocity map for the scan of 15:59 UT on February 11, 1999, and Figure 9b shows the Burg spectra in beams 0–14 at a range of 630 km. The velocity map shows a clear and consistent velocity variation with the azimuth with the strongest velocities in beams 0–1, as one would expect for westward convection. In Figure 9a we show the EISCAT convection vector. One can fit the velocity's azimuthal variation with a cosine curve with a maximum of ∼600–700 m/s. These values are slightly less than the convection magnitude observed by EISCAT (800 m/s). One might expect such a reduction in the convection magnitude for the area ∼300 km equatorward the EISCAT spot and, in fact, DMSP measurements do show a smooth decrease of the convection at lower latitudes [Danskin, 2003].
 Data in beam 1 (Figure 9b) show the presence of a high-velocity component with the Doppler velocity of the order of −500 m/s. Such a peak is not seen in the higher numbered beams. Simultaneously with the high-velocity component one can see the low-velocity component for this spectrum such that it is double-peaked. However, the separation between the peaks here is 428 m/s, well above the typical values that we reported for the double-peak structures (Figure 6). The beam 0 spectrum also indicates additional energy around velocities of −600 m/s, but this enhancement was of low intensity so that the software did not identify the spectrum as double (perhaps even triple)-peaked. For beam 2, the high velocity component has a shift of only −300 m/s, the spectrum is double-peaked with the peak separation of 262 m/s, close to the typical separation between the double peaks reported in Figures 2, 3, and 6. The spectra in beams 3–14 are also similar to the ones reported in Figures 2 and 3 with peak separation of the order of 150–200 m/s. One cannot ascertain a trend in the peak separation with the beam number (azimuth) change; in Figure 6 we demonstrated this feature on a statistical basis.
 It is important to realize that the high-velocity component existed in the beam closest to the direction of the plasma flow. For this reason and because of unusually strong velocity of the peak we think that this component of the spectrum corresponds to radar detection of the Type 1 echo. This means that, generally speaking, some double-peaks reported in this study could actually be interpreted in terms of the azimuthal convolution of Type 1 and Type 2 scatter. However, the number of such events is low, especially if one would like to have a distinctly larger separation between the peaks than the typical values. Certainly more work is needed on this aspect by involving SuperDARN radars observing along the L shells where Type 1 echo detection is more likely.
 The major discovery of this study is that many short-range HF echoes are double-peaked. Typically, the peak separation was of the order of 100–200 m/s and there was no obvious change in the peak separation with the azimuth of observations. Quite often we have seen that the peak separation was larger at ranges just farther than the range of echo power maximum. Statistically a more frequent occurrence of broad E region echoes at ranges slightly farther than the range of the echo power maximum was reported by Koustov et al. [2001b] for the Syowa SuperDARN radar.
 One can think about several potential explanations of the double peaks. In the past, the double-peak nature of VHF E region echoes was quite often related to the effects of the FB and GD instabilities producing two independent echo components [e.g., Greenwald et al., 1975]. In Figure 9 we presented data where the observation of two types of echoes was quite probable. However, we did observe double-peaked echoes almost perpendicular to the L shells with almost zero FITACF Doppler shifts (perpendicular to the electrojet flow). For such directions, it is hardly possible to expect Type 1 echo contributions.
Greenwald  explained VHF double-peak echoes with small peak separation in terms of plasmaphysical properties of secondary irregularity excitation. One can think of other purely plasmaphysical explanation of our observations. For example, one can involve the effects of strong plasma gradients in the region of strong electric field [St.-Maurice et al., 1994]. This sounds reasonable for our events since the electric field was quite strong. Unfortunately, we do not have enough data to further explore this option.
 DP echo occurrence in the E region can also be associated with quickly changing electric field conditions within the scattering volume so that separate peaks would simply reflect two different “streams” of plasma flow at two different moments or at two different places (within the scattering volume) during the integration time [Huber and Sofko, 2000]. To address this possibility, we studied Hankasalmi data obtained with 3-s and 7-s integration time for completely different events (unfortunately, there were no special experiments for which these measurements were performed sequentially). We also studied Hankasalmi data obtained with 15-km and 45-km resolution, again for completely different events. For both comparisons, no significant differences in DP echo occurrence, azimuthal distribution or separation between the peaks were found. These findings cast doubt on the possibility of DP occurrence due to temporal/spatial changes in the plasma convection.
Huber and Sofko  also considered a possibility of HF echo reception from both the main and the side antenna lobes (in the horizontal plane) as a reason for the DP occurrence. This explanation sounds reasonable for our observations, especially for measurements almost perpendicular to the flow because then one would be able to explain coexistence of peaks with different polarity of the velocity (Figure 6). Huber and Sofko  rejected this explanation for the F region echoes on the grounds that DP echoes occur only in one-two ranges and for one-two successive scans. In our view, this argumentation is not valid for both F region and E region echoes in our observations. In Figure 1 one can see frequent F region DP occurrence in quite a few radar gates and in many scans. The same conclusion can be drawn from the E region data presented in Figures 1–4. However, we see a different problem with this interpretation. Assume that the velocity of an HF echo is proportional to the line-of-sight component of the plasma convection. Then, as radar scans, one would expect a decrease in the peak separation for observations more perpendicular to the flow directions (L shells) since the side lobe is separated from the main lobe by a fixed azimuthal angle and the cosine function changes a lot at angles near ∼45 degrees but very little near 90 degrees. Our observations do not show such a trend (Figure 5); the peak separation is about the same for all beam positions. Preliminary analysis of Stokkseyri radar observations along the flow revealed about the same values of the DP separation, on average. Another concern with the sidelobe explanation is that if this were the main reason, one would expect the DP occurrence almost always, especially when the signal power is strong; however, the DP echoes had relatively moderate power (Figure 5). We think that the antenna side lobe effect in the horizontal plane is not the prime reason for the E region DP echo occurrence.
Huber and Sofko  concluded that the DP echoes in the F region occur when convection vortices with the scale of 20–30 km are set up in the ionosphere. If one assumes that these vortices are mapped down to the E region heights, then they will produce a circular “streams” of electrons, and our observations seem to fit this hypothesis. Especially attractive is the fact that the circular streams would lead to the same separation between the echo peaks irrespective of the azimuth of radar observations, the main problem with other explanations. In attempt to find signatures of the convection vortices, we carefully examined the DMSP ion drift data (6 samples per second data) for the pass over the Hankasalmi viewing zone on February 12, 1999 (around 1430 UT). We did not find any signatures of such vortices even though DP echoes were detected in nearby areas. One can argue that the small-scale vortices are difficult to detect with the DMSP satellites, but this would contrast significantly with frequent occurrence of DP echoes. We believe that more SuperDARN/DMSP comparisons and other studies are needed to justify the interpretation of E region DP echoes in terms of vortical flows.
Huber and Sofko  also considered the explanation of F region DP echoes as a result of simultaneous echo detection from the F and E regions. They argued that the special propagation conditions, needed for this scenario, should be long lasting, once they are set, which was not supported by the sporadic nature of DP echoes in the F region. We believe that in a case of the E region DP echoes, this argument is not so valid, and double peaks could indicate simultaneous HF echo reception from the top and the bottom of the electrojet layer where the irregularity velocity is slightly different. Uspensky et al.  have considered such an explanation for interpretation of Schefferville HF radar observations of the E region echoes.
 To illustrate how this scenario works for the E region observations we performed the ray tracing modeling for the observations presented in Figures 2 and 3. Figures 10a and 10c show two electron density profiles that are 10-min averaged profiles near the time of the observations. The density values were scaled by a factor of 1/2. As noted by Danskin et al. , the scaled electron densities are more consistent with the Sodankyla ionosonde measurements made equatorward of EISCAT. Also, the predicted ranges of echoes from the ray tracing agree with the observations. We also tried ray tracing with the IRI model selected for the periods under study (the time separation between the events does not give much difference in terms of density) (Figure 10e). The IRI model shows about 2 times smaller densities and lower height of the profile maximum than the ones in Figures 10a and 10c. There are similar discrepancies between the model profiles and the actual measurements at the EISCAT spot.
Figures 10b, 10d, and 10f show the corresponding ray tracing solutions. Only 4 rays corresponding to the elevation angles of 5, 10, 15, and 20 degrees are shown. The dots indicate those ranges (including the rays that are not shown here) for which the aspect angle is within ±1 degree of orthogonality with the magnetic field so that HF echoes are expected. In Figure 10a with somewhat stronger electron density, one can see that the electrojet-related echoes (in the height range of 95–125 km, horizontal dash lines in Figure 10b) can be received from ranges between 400 and 650 km. Figure 10b clearly shows that for the first 50 km of the scatter band, there is only one scattering layer. At farther ranges, between 450 and 600 km, scatter from two distinct layers is possible. We note that with an increase in range, the separation between the scattering layers increases. There is a possibility of echo reception from above the electrojet layer at ranges >600 km simultaneously with echoes from the bottom of the electrojet. For ranges >700 km echoes can only be received from above the electrojet.
 In Figure 10c where there is a less dense ionosphere, the electrojet-related echoes can be received from ranges of 500–750 km (see Figure 10d), about 100 km farther than in Figure 10b. One can recognize the similar crescent-like structure in the height profile of possible scatter but it is more “diffuse” so that the echo region is expected to be more extended. One can get electrojet echoes simultaneously from two completely separated layers at 600–650 km. There is a chance to get echoes from the above the electrojet and simultaneously from the bottom of the electrojet at ranges 650–750 km.
 In Figure 10e where there is even less dense and lower (in height) ionospheric E layer, the echoes can be received from about the same ranges. The lower scattering layer in this case is centered below the typical electrojet boundary heights so that scatter from the bottom layer is expected to be of different origin than from the top layer.
 For the following discussion, it is important to realize that the velocities of echoes coming from the top and the bottom of the electrojet layer are expected to be different. For estimates, one can consider calculations by Koustov et al. [2002, Figure 10d]. If backscatter is orthogonal at all heights, the velocity ratio Vtop/Vbottom can be as large as 2. If backscatter is nonorthogonal, the difference can be stronger or weaker, with the exact value depending strongly on specific heights of scatter and corresponding aspect angles. For the convection velocity of 1000 m/s and the flow angle of 75 degrees, one can expect the separation between the peaks as strong as 130 m/s which is within the range of observations.
 After illustrating a possibility of E region echo detection from two electrojet heights and estimates of the peak separation, we would like to stress those experimental findings that support the idea that the observed two peaks of HF velocity correspond to scatter from the top and bottom of the electrojet layer.
 First of all, the results of the ray tracing analysis and DP occurrence in terms of range are in reasonable agreement. Consider the morphology of DP echo appearance in Figures 2 and 3. For the first case of 1430 UT, the low velocity peak was present all the way to the farthest ranges where the signal was still available, consistent with the ray tracing predictions. The high-velocity peak was detected at several range gates well inside the echo band. It was strongest only at one gate indicating perhaps that the electrojet irregularities were not strong everywhere at the top of the electrojet layer. The high-velocity peak at 675 km had a Doppler shift of −334 m/s which is below the E × B component of ∼500 m/s measured by the EISCAT (appropriate data were presented by Koustov et al. ) but in rough agreement with the DMSP measurements of the plasma convection at the latitudes of echo detection as reported by Danskin , who noticed ∼1.5 times electric field decrease at latitudes ∼300 km equatorward of the EISCAT spot. Thus, in terms of velocity, the high-shifted peak at 675 km can be associated with scatter either from the top of the electrojet layer or even above it (the ray tracing in Figure 10a shows good aspect conditions above as well as below 125 km). We should mention that in the past, Milan and Lester  envisioned a possibility of echo detection from above the electrojet at short ranges.
 The magnitudes of the low-velocity components at 1430 UT were all below 60 m/s which is significantly less than the expected values of the E × B drift of >100–150 m/s. This perhaps indicates that the low-velocity peaks can very likely be coming from the bottom of the electrojet where the velocity depression is expected to be the strongest. It well might be that some of these low-shifted peaks were originated from below the electrojet layer; the ray tracing allows such a possibility (Figure 10b). Such an interpretation means that some peaks have nonelectrojet origin. In this respect, a peculiar feature of measurement at 720 km is that both DP velocities were well below the E × B drift. This might be indicative that one of the peaks was coming from the bottom of the electrojet region while the other one was coming from even lower heights.
 For the second case of 1450 UT, the far-range echo at 810 km had the Doppler shift of −245 m/s. The E × B velocity component measured by EISCAT at this time was about −250 m/s so this peak is very likely scatter from above the electrojet layer. This interpretation agrees with the ray tracing diagram in Figure 10d.
 The other feature in the data that agrees with the interpretation of double-peak echoes as scatter from the bottom and top of the electrojet layer is the non-L-shell alignment in DP echo occurrence, Section 5. Since the electrojet was more or less L-shell aligned (EISCAT measurements), one would expect the double-peak spectra to occur in a region stretched along the L shell direction if the echoes were originated from pure plasma physical effects.
 We also demonstrated that the HF spectra are broader at the far ranges (e.g., Figure 8), farther than the ranges of the echo power maxima, as expected from the Uspensky et al.'s  model. Diagrams similar to Figure 8 were obtained for a number of other periods and a similar feature is quite often evident. The feature is not always present for understandable reasons; one needs nearly uniform density distribution to detect it clearly. A statistically more frequent occurrence of broad E region echoes at ranges slightly farther than the range of the echo power maximum was reported by Koustov et al. [2001b] for the Syowa SuperDARN radar.
 An important implication of the two-height interpretation of E region DP echoes is that quite a few SuperDARN E region echoes are expected to come from the bottom of the electrojet layer or even from below it. At these heights, the irregularity production and their properties can be influenced by the neutral wind. The observed small velocities of low-shifted peaks (<100 m/s) is an additional evidence for this possibility. Makarevitch et al. [2002b], by considering 50- and 12-MHz echo statistics at the Antarctic Syowa station, showed that HF echoes occur more frequently not along the electrojet flow, unlike the VHF echoes, but some tens of degrees off the flow. They also showed that HF echoes consist of 2 groups, the low-velocity and high-velocity groups. The low-velocity group had velocities (FITACF) below ∼150 m/s, and the authors suggested that these echoes were related to the electrojet instabilities significantly affected by the neutral wind or even generated due to neutral wind-related plasma instabilities [Dimant and Sudan, 1997; Kagan and Kelley, 1998, 2000]. In a case of the significant neutral wind effect, one would expect different flow angle variation for the low-velocity component as compared to the high-velocity component. This might lead to the violation of the cosine law for the E region velocity, and Makarevitch et al. [2002b] presented evidence that this might be the case for one event.
 Another expected effect of the low-height backscatter is the significant velocity depression as compared to the E × B component. Koustov et al.  and Makarevitch et al.  analyzed E region HF echo observations for which the FITACF velocities were several times smaller than the plasma convection velocities, and these authors also speculated on the role of neutral wind effects in the HF Doppler velocity.
 To be fair, the interpretation of E region DP echoes as scatter from two heights in accordance with the Uspensky et al.'s  model has some problems. For example, one would expect that as the direction of observations approaches perpendicularity with the L shells, the velocity of both peaks should be approaching zero as well as the peak separation. Experimentally, this was not the case (Figure 5). We also noticed that for observations almost perpendicular to the L shells, the peaks can be of different polarity. The simple approach by Uspensky et al.  does not work in explaining this result, but the situation could be improved if the ion motion is considered as suggested by Uspensky et al. .
 From the above discussion, it is clear that in order to improve our knowledge on the nature of HF E region DP spectra more analysis of events is needed. This work requires significant data processing, and it is in our future plans. Further testing of interpretation of the DP velocities can be done if nearly simultaneous data on electric field and neutral winds are available. Unfortunately, it is a very seldom occasion this can be achieved with the EISCAT facility. In the CP-1 mode, observations are performed at ranges where F region echoes are more typical [Davies et al., 1999; Danskin et al., 2002]. Comparison with DMSP measurements of plasma convection is a more real possibility.
 The obtained results on the microstructure of E region echoes have two potential implications for SuperDARN-related research. One of the questions is how the velocity of each peak is related to the E × B component. One would expect that the high-shifted component is closer to the E × B component whether it is received from above the electrojet layer or just at its topside. Since typically this component is ∼100 m/s larger than the FITACF velocity (Figure 6), an implication is that the currently in use FITACF approach should give small underestimation of the plasma convection if the map potential technique is employed. This is, in our opinion, not significant deficiency. More alarming is the fact that the FITACF E region velocity itself can be much less than the E × B component, as one can conclude from the comparison of VHF and HF velocities by Koustov et al.  and E/F region velocity comparison by Makarevitch et al. . More work is needed in this respect, for example, through a comparison of SuperDARN and DMSP data. As for the F region DP echoes, we should mention that for the February 12, 1999 event (Figure 1), we found that their FITACF velocity was located somewhere in between velocities of the peaks and the typical peak separation was around 200 m/s (in agreement with Schiffler et al.  and Huber and Sofko ). This means, again, that a minor error in the convection (and cross-polar potential) estimates can be made for the case of DP echo involvement.
 The other implication of the obtained results is for the SuperDARN width measurements. According to the understanding gained in this study, the HF spectra should quite often be double (multi)-peaked in the “transition region” from pure E region and pure F region scatter, ranges 600–1000 km for the Hankasalmi radar. On width maps this should be seen as a stripe of echoes with enhanced width. The exact range location of such a stripe can vary in accordance with the distribution of the electron density. The effect of enhanced width can also “work” at far ranges where echoes can be received from both F region and E region by the 1 and 1/2 propagation mode. Moreover, one can expect some enhancement in the spectral width of pure F region echoes if they are received from two heights and the electric field is strongly nonuniform. This might happen due to the 45-km range gate of SuperDARN such that echoes at different heights might correspond to different magnetic flux lines and different electric fields. Thus, simultaneous echo reception from several heights can be a reason for broader than usual echoes. The effect should be stronger near the edges of particle precipitation. It is not a surprise then that HF radars can detect some magnetospheric boundaries through imaging the spectral width distribution [Woodfield et al., 2002]. More work is needed to assess whether this “radiophysical” effect is really occurring.
 New results obtained in this study are as follows:
 1. While E region HF echo spectra are single-peaked at shortest and farthest ranges of an echo band, there are quite often two (or more) spectral components at the intermediate and farther ranges. This “splitting” of spectra starts near the ranges of the power maximum in the power range profile. The typical separation between the peaks is about 150 m/s and is hardly changed with the azimuth of observations.
 2. Splitting of HF spectra is accompanied by an increase of the FITACF spectral width. The effect can be seen in range profiles for the width. The range of maximum width is shifted farther away from the range of the power maximum indicating that increased width is resulted from the presence of 2 (and sometimes more) spectral components with differing velocities.
 3. Velocities of two components in a double-peak spectrum are typically larger and smaller than the velocity of the unresolved spectrum (FITACF velocity) and single-peaked velocity of echoes at shorter ranges. The high-velocity component is closer to the E × B velocity drift along the radar beam than the velocity of the unresolved spectrum. The low-velocity component has values below 200 m/s, the maximum expected wind velocities at the bottom-side E region.
 4. Among possible explanations of the observed E region DP echoes are nonlinear plasma physical processes, excitation of convection vortices, and echo detection from the top and the bottom of the electrojet layer. Though definitively further investigation is needed to decide on relative importance of these effects, we favor echo detection from two heights as the major factor.
 CUTLASS Finland radar is supported by PPARC, the Swedish Institute for Space Physics, Uppsala, and the Finnish Meteorological Institute. This research was supported by NSERC (Canada) grant to A.V.K. The authors are grateful to F. Rich for providing the DMSP data and S. Nozawa for providing the EISCAT data.