%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % % % C P - 1 - J - T : D E S C % % % % Gudmund Wannberg, EISCAT HQ, % % Updated June 19, 1992 (MTR) % % Last updated July 30, 1992 (UGW): PP ranges corrected % % % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% IN THE EXPERIMENT THE ELAN,TLAN AND CLAN FILES ARE "MATCHED" TOGETHER. CARE SHOULD THEREFORE BE EXERCISED IF THE EXPERIMENT WERE GOING TO BE MODIFIED EVEN SLIGHTLY. 1. General information The CP-1-J experiment is the first of a next generation of codes. It makes extensive use of the new signal processing hardware which has been installed in the EISCAT system over the past two years. The E region data rate is increased by about 20 % in the power profile part and by as much as 100 - 200 % in the spectral data. This is accomplished at the cost of a minor reduction in Troms| F region statistics (about -15 %), which is caused mainly by the transmission of only one long pulse/IPP rather than two, as in the previous CP-1 versions. So as not to compromise the tristatic velocity statistics, the remote sites are programmed to receive both the long pulse and also the alternating code transmission which produces the E region spectral data in Troms|. This causes a net increase of the number of spectral estimates per unit time at the remote sites of about 70 % relative to the CP-1-I. The integration time is set to 5 seconds, which may be useful in highly dynamic situations (auroral substorms etc.). IT MUST NOT BE CHANGED NEITHER FROM THE EROS CONSOLE NOR THROUGH A CHANGE TO THE 'START-RADAR' COMMAND IN THE :ELAN FILE; ANY CHANGES WILL CAUSE THE ALTERNATING CODE PART OF THE EXPERIMENT TO FAIL. IF A DIFFERENT INTEGRATION TIME IS DESIRED THE TARLAN CODE GENERATOR PROGRAM "ATC-1" MUST BE USED TO CREATE A NEW :TLAN FILE ADAPTED TO THE NEW INTEGRATION TIME !! 2. System configuration This experiment uses all eight channels of the receiver. The PDF filters should be installed so: Used for CH1 and CH2 Linear phase, 25 kHz BW, lower crate F region PP CH3 and CH4 Butterworth, 25 kHz BW, upper crate F region LPACF CH5 to CH8 Linear phase, 25 kHz BW, lower crate D and E reg PP + alt. codes The three channel groups should be internally balanced for gain to within one decibel. It is permissible to completely attenuate out faulty channels within a group in an emergency. This causes a data loss of 50 % per channel in the F region groups but only a 25 % per channel loss in the D,E region group. In every IPP pulses are transmitted on four different frequencies, i.e. one frequency each for the FPP and FACF groups and two frequencies for the D,E region group (one for the PP and the other for the alternating code). In the next cycle, all frequencies are changed and the pattern is repeated, now of course using different receiver channels. Towards the end of each cycle, all channels which are due for use in the next cycle are calibrated. A new member of the 16 baud strong condition alternating code set is transmitted every two IPPs, so it takes 64 IPPs to complete the set. At the end of such a set, the frequency/channel pair used for the D,E region PP is interchanged with the pair used for the alternating code (this happens about twice a second). The result of this operation is that the D,E PP and the ACFs measured with the alternating code are automatically intercalibrated in an average sense. Decoding of the alternating code data is fully automatic and performed by special purpose hardware in the correlator, which is partly controlled from the correlator microprogram and partly from the radar controller (the :TLAN and :TCOD files contain special 'bit' commands to this effect). CAUTION: Because of the intricate commutations of channels and frequencies used in the experiment, it is inadvisable to attempt to change a frequency or channel by editing the :TLAN file. A special "TARLAN generator" program has been written for this purpose. It is still not quite debugged and should be used only in cooperation with the author, UGW at EISCAT HQ, who will be pleased to offer help and advice. 3. Data dump structure, CP-1-J-T The data dump contains five different blocks of data, namely i) An ungated power profile for the D and E regions, also serving as zero lags for the ACFs in block (iii), ii) A "variance profile" computed from the undecoded samples taken from the alternating code modulation, iii) Incomplete ACFs derived from the alternating code, iv) A gated, coarse resolution power profile for the F region, v) Conventional long pulse ACF data. The dump is 2036 complex double integers long ; the SCANCOUNT value is stored in dump address 2035. 3.1 D and E region power profile block This block contains ungated power profile data which also doubles as the zero lags for the ACFs stored in block (iii). Specifically, there are 90 (S+B) gates at addresses 0000-0089 60 B gates at addresses 0090-0149 10 (B+C) gates at addresses 0150-0159 Gate no. 9 at address 0008 coincides in range with A/C ACF gate no. 1 in block (iii). Data from two frequencies/channels are added; this is the only block in the dump that employs channel adding! 3.2 Variance profile data block This block contains 76 power estimates at addresses 0160-0235. Its use is further described below. 3.3. Alternating code ACF block This block is very straightforward. It contains 61 S gates (lags 1-15) at addresses 0236-1150, IT weighting, Ns = 16 2 B gates (lags 1-15) at addresses 1151-1180, IT weighting, Ns = 16 IT = inverse triangular, w(l) = Ns - l where l is the lag index No channel adding is applied (one channel only per IPP). NOTE: these gates contain NO zero lag - it cannot be unambiguously estimated from the alternating code. Note also that the mean of the back- ground is zero if the algorithm works properly, so background subtraction is neither necessary nor recommended in this block (it would just increase the variance !). The RTGRAPH has been upgraded so that if :=-1 a zero lag is extrapolated from the data and the ACF displayed with this included. This allows you to display also power spectra with a proper baseline level. The present :GDEF file uses this feature. 3.4. F region power profile block This block contains (S+B), B and (B+C) gates that exactly match those of the CP-1-I-T low resolution power profile in gate separation and range extent. The start range, the P-P resolution and the order of the subblocks are different: 80 (S+B) gates at addresses 1181-1260 40 B gates at addresses 1261-1300 6 (B+C) gates at addresses 1301-1306 Gate adding is used; three neighbouring estimates sampled 10 us apart are added to form each gate in the output. No channel adding is applied. 3.5. Long pulse ACF block The gating parameters used in this block are identical to those of the CP-1-I-T LPACF, which should facilitate the adaption of existing analysis software to the new program. Specifically, the block contains 21 S+B gates (lags 0-25) at addresses 1307 - 1852, GS weighting, Ns = 15, 6 B gates (lags 0-25) at addresses 1853 - 2008, GS weighting, Ns = 15, 1 B+C gate (lag 0 ) at addresses 2009 - 2034, GS weighting, Ns = 15. GS = Gen-System , w(l) = Ns + l where l is the lag index NO adding of channels is used in this block. NOTE: the only real difference between this data block and the output of a CP-1-I-T experiment is the ORDER of the sub-blocks, plus the fact that only one channel is received per IPP. 4. Data properties 4.1. D/E region power profile This is a very simple power profile data vector. It is derived from two 21 us pulses transmitted immediately after each other on two different frequencies in the first half of each IPP. The frequency used for the first pulse is later used for the alternating code which is transmitted in the second half of each IPP. The frequencies are commutated according to a relatively complicated scheme from one IPP to the next, but this ensures that all frequencies used for the PP pulses are eventually also used for the alternating code; hence the power profiles and the ACFs computed from the code are automatically gain equalised and the PP can be used as a zero lag profile for the ACF series (which does not contain lag 0). The PP channels are sampled at a 1/21 us rate and, as mentioned above, two channels are added in the dump. No gate adding is performed. The filters are 25 kHz linear phase ones. The range coverage is 63.70 - 344.05 km and the range increment/gate is 3.15 km. The start range has been computed using the actual measured impulse response of the 25 kHz LP filters used and it is thus a few hundred meters less than what one would have deduced by using a simple model where the impulse response of the filter maximises at t=(1/BW) (the start range computed from the simple model is 64.65 km, which has been used in the :GDEF and :DESC files prior to 1992-07-30 - please note the change). There are two things worth mentioning in connection with this power profile. Firstly, the background is derived only from samples taken on one channel, so the weight of the background estimate is nominally equal to unity - this arrangement was chosen so as to avoid getting any signal from far out ranges into the background gates. However, the actual weight of the signal is slightly less than two, because in the very first cycle after a data dump, the estimates from the second PP channel are overwritten by those from the first channel. If the scancount is N, there are thus (2*N-1) (S+B) estimates and N background estimates per gate per dump. For the case of a 5 seconds integration time, N is equal to 640 and there are 1279 (S+B) estimates and 640 (B) estimates per gate. If the nominal weights of 2 and 1 are used (as e.g. in the RTGRAPH), the background is overestimated by some 0.75 10**-3. In a careful analysis, this must be taken into account ! Secondly, the background estimate is stored as a time series of 60 points, rather than as a compacted estimate. The background sampling starts at a point in time which is immediately before the point at which the first D- region (S+B) point is sampled, and it then continues for 1260 us. During this time, the receiver gain drops by several percent, although most of the gain drop occurs within a very short time immediately following the RXPOFF command. The gain variation is a consequence of the receiver switching method used in Troms| at present and cannot be suppressed much more than to its present magnitude by any hardware changes. Incidentally, there has probably always been a cyclic gain variation of this kind in the Tromso receiver, but prior to the front end change in 1988 it used to go in the opposite direction, i.e. the gain increased with time into a receive cycle. This is not as notice- able on a raw data display as the present effect, because it tends to bias the data into a physically "expected" direction, i.e. rapidly reducing density when going downwards below the E region. The present effect can result in something which appears as a residual density below 80 km, and negative densities at far ranges, if the background which is subtracted from the (S+B) gates is a plain average over the entire IPP. Now, since the background is available as a time series, one can use different strategies for its removal. One possibility is to use local averages, centered in time on each (S+B) gate, another possibility is to use a fitted time-varying background estimate. If the first ten background points are excluded, the rest can be fairly accurately represented by a straight line fit. Since the gain variation is a function of elapsed time after the receiver has been switched on by the RXPOFF command, everything must be related to this point when the background estimate is calculated. If the RXPOFF command is executed at some time t, the background sampling starts at (t+260) us and the first signal sample is taken at (t+370) us, i.e. about five samples later in relative time. 4.2. Variance profile This data block is included in the hope that it may become useful in the analysis of the alternating code ACF data. It contains a series of raw power estimates derived from all the samples used by the decoding process (there are 76 of them in this case of a 16 bit code and 61 gates) and so it can be understood as an "ambiguous lag-0 profile". An advanced analysis routine might attempt to estimate the true variance of each lagged product in the decoded ACFs by starting from this ambiguous power profile. The first point in the variance profile corresponds to a sample taken as the leading edge of the coded pulse passes 90 km range. As the modulation is 336 us long, the power in this point is an average over 50.4 km, and the lower boundary of this first point is just below 40 km range. Some ground clutter may be expected to occur in the first few points. This, if present, will also affect the first few ACF gates. It remains to be seen to what degree this corrupts the data. 4.3. Alternating code ACF The alternating-code ACF data is derived from a 16 baud, strong condition alternating code set using a baud length of 21 us. A total of four different frequencies and four receiver channels are used for this modulation, and each of the 32 sequences that make up the code set is repeated on two different frequencies. It takes slightly less than one-half second to complete the full set; this is consequently also the time during which the target (ie the ionosphere) must be assumed to be stationary in an average sense, for the technique to produce dependable results. Sampling is at 21 us/sample, filters used are 25 kHz linear phase. The lag extent is from 21 - 315 us and the lag increment 21 us. Fitting for velocity should be straightforward - the zero lag is not used anyway in the velocity estimator. Fitting for density requires an estimate of the zero lag. This can be taken from the power profile in block (i), which is measured through the same receiver channels and on the same frequencies as the alternating code - it is consequently automatically intercalibrated with the nonzero lags and can be patched into the zero lag locations of the ACFs after proper scaling. The relative weight of the PP is 2, but remember that it must be background subtracted also ! PP gate no. 9 corresponds to ACF gate no. 1, which is at 89.00 km range. The range increment/gate is 3.15 km. 4.4. F region power profile This power profile is mainly intended as a replacement for the 2*29 us power profile in the previous CP-1-I-T. The number of gates, range coverage and gating all are identical to the CP-1-I version. The modulation is slightly different - as there is more freedom with respect to how long the transmit pulses can be made in CP-1-J, we have decided to use just one, 40 us pulse per IPP rather than the two, 29 us pulses of CP-1-I. The p-p range resolution of this modulation is thus slightly coarser than in the previous version, about 9 km, but the FWHM is only marginally in- creased. Only one channel per IPP is used. NOTE: The starting range of this PP is 70.35 km. This is also computed using the actual impulse response of the 25 kHz LP filters used. 4.5. Long pulse ACF The signal carrying part of this block is arranged to behave exactly the same as the CP-1-I-T, i.e. the series of gates is computed with a volume index of 15 and a max. lag of 25. The first gate is centered on 150 km range. Six background gates are computed instead of five; they are all fully independent. The noise calibration part of this data is quite different from that of CP-1-I - it contains a computed zero lag with the proper weight relative to the other gates, but all higher lags are forced to zero in the output. There is a reason for this: If one computes only the zero lag, only 15 noise samples are required, but if one were to compute the full ACF from samples taken during the noise injection period, 73 samples would be required! Omitting everything but the zero lag computation thus enables us to save a considerable amount of time, namely (73-15)*10 us = 580 us, in the calibration phase of the experiment. This represents a significant increase in the effective duty cycle of the experiment. At the same time, the output format is consistent with that of the other ACFs in the block. Note, however, that in RTGRAPH it is impossible to display the calibration gate using background subtraction - all lags except lag 0 will come out negative in this case...... The sampling increment is 10 us, hence the lag extent is from 0 - 250 us. 25 kHz Butterworth filters with a nominal risetime of 24 us are used, which leads to oversampling and a substantial correlation between neighbouring lags. REMOTE SITE ************ 5. Data dump structure, CP-1-J-R The data dump contains three blocks of data with somewhat different structures, namely i) Conventional long pulse ACF data, ii) Incomplete ACFs derived from the alternating code, iii) Boxcar weighted full ACFs for calibration of the channels used for the alternating code. The dump is 394 complex double integers long ; the SCANCOUNT value is stored in dump address 393. 5.1. Long pulse ACF block The basic structure of this block is identical to that of the previous CP-1-I-R data dump, which should facilitate the adaption of existing analysis software to the new program. Specifically, the block contains 5 S+B gates (lags 0-29) at addresses 0000 - 0149, IT weighting, Ns = 32, 2 B gates (lags 0-29) at addresses 0150 - 0209, BC weighting, Ns = 320, 1 B+C gate (lags 0-29) at addresses 0210 - 0239, BC weighting, Ns = 320. IT = inverse triangular, w(l) = Ns - l where l is the lag index BC = boxcar, w(l) = Ns for all l. NO adding of channels occurs in the (S+B) part of the data. Two channels are added in the calibration and background, but this adding is implied in the assigned weight Ns = 320, so no action is necessary in the analysis routines. NOTE: the only real difference between this data block and the output of a CP-1-I-R experiment is the WEIGHTING of the background and calibration gates, plus the fact that only one channel is received per IPP. 5.2. Alternating code ACF block This block is very straightforward. It contains 7 S gates (lags 1-15) at addresses 0240-0344, IT weighting, Ns = 16 No channel adding is applied (one channel only per IPP). NOTE: these gates contain NO zero lag - this cannot be unambiguously estimated from the alternating code. Note also that the mean of the back- ground is zero if the algorithm works properly, so background subtraction is neither necessary nor recommended in this block (it would just increase the variance !). 5.3. Boxcar ACF block This block contains background and noise injection gates necessary for a proper analysis of the data in the alternating code ACF block: 2 B gates (lags 0-15) at addresses 0345-0376, BC weighting, Ns = 80, 1 B+C gate (lags 0-15) at addresses 0377-0392, BC weighting, Ns = 80. 6. Data properties 6.1. Long pulse ACF The signal carrying part of this block is arranged to behave exactly the same as the CP-1-I-R, i.e. the series of gates is computed with a sample space overlap of 10 samples, and the third gate should be properly illuminated if the second and fourth gates are of about equal magnitude. All analysis should be performed on the third gate only. This is also the one to use for SNR and real time velocity estimates in RTGRAPH. The sampling increment is 10 us, hence the lag extent is from 0 - 290 us. 25 kHz Butterworth filters with a nominal risetime of 21 us are used, which leads to oversampling and a substantial correlation between neighbouring lags. 6.2. Alternating code ACF The seven ACFs in this block are arranged to fall symmetrically around the nominal intersection of the Troms| beam with the remote site beams. As the sampling rate is 1/(21 us) and the filter impulse response is matched to this, the spatial response is such that the gates are separated along the scattering bisector by (3.15 cos @/2), where @ is the full scattering angle. This distance is typically of the order of 3 km in the F region, which is only about one half of the FWHM of the bistatic illumination distribution, defined by the spatial overlap of the two beams, so there will be useable signal power in at least the centermost three gates under most conditions. Furthermore, the properties of the alternating code algorithm ensure that the illumination is the same at all lags for all gates - there is no skewness as is the case in the long pulse ACF. There may therefore be some advantage in simply adding a few gates together be- fore fitting the data. Fitting for velocity should be straightforward - the zero lag is not used anyway in the velocity estimator. Fitting for density requires an estimate of the zero lag. This may be extrapolated from the measured ACF and intercalibrated with the long pulse ACF data using the calibration values in data block 3 (see below). Sampling is at 21 us/sample, filters used are 25 kHz linear phase. The lag extent is from 21 - 315 us and the lag increment 21 us. 6.3. Boxcar background/calibration ACF block An estimate of the channel gain (ie its numerical output per noise injection sample) of the channels used for the alternating code modulation is necessary for a proper density and temperature analysis. Since the decoding algorithm suppresses the mean of any stationary, uncoded signal component in the data, it also suppresses the background noise and the calibrated noise injection, so no gain estimates can be derived from block 2. This is the reason for including block 3 - it is computed using conventional boxcar-weighted ACFs with no decoding applied and so it provides zero lags both for the background noise as well as for the (background + calibration) noise. Sampling rate and filters are identical with those used in block 2. Since the alternating code channels are fitted with linear phase filters and sampled at exactly the Shannon rate, the passband shape as derived from an FFT applied to the ACF data should appear totally flat. Any deviation from a flat spectrum indicates trouble and must be investigated. 7. Additional caveats As indicated above, the data dump contains two types of data representing the state of the same target volume. They cannot be added before analysis, but it should be possible to add the fitted parameters in a weighted manner to reduce their variances substantially - this is especially true of the velocity, whose mean does not depend on the absolute power received. The gain in statistics depends directly on the assumptions made in the adding algorithm and cannot be specified here, but for velocity it may approach 50 % above the previous CP-1-I-R. IT IS ULTIMATELY THE END USER WHO MUST DECIDE HOW TO ADD AND WEIGHT THE TWO DATA SETS AND ESTIMATE THE GAIN IN STATISTICS IN A GIVEN SET OF CIRCUMSTANCES ! 8. GEOMETRICAL PROPERTIES ********************** THE GEOMETRICAL PROPERTIES ARE ONLY SLIGHTLY MODIFIED FROM THE CP-1-F, ADAPTED TO THE NEWEST IGRF MODEL 1985, EXTENDED TO 1990, AND CALCULATED FOR NOVEMBER 1986. THE TROMSO ANTENNA IS KEPT FIXED PARALLEL TO THE MAGNETIC FIELD AT THE CENTER OF RANGE 7, 285.0 KM CORRESPONDING TO AN ALTITUDE OF 278.6 KM. THE REMOTE SITES MAKE MEASUREMENTS IN A 10 MINUTE CYCLE, WHICH CONSISTS OF 6 E-REGION AND 1 F-REGION POSITION. THE F-REGION POSITION OCCURS EVERY 5 MINUTES: APPROX. ACTUAL MOVE TIME POS. ALT/KM NOMINAL MOVE+DWELL TIME/S KIR SOD 1 (F) 278.6 30 + 60 22 18 2 (E1) 90.5 40 + 50 32 22 3 (E2) 101.1 10 + 50 4 (E3) 116.9 10 + 50 5 (F) 278.6 30 + 60 26 19 6 (E4) 95.8 40 + 50 29 7 (E5) 109.0 10 + 50 8 (E6) 124.8 10 + 50 FOR MORE DETAILS SEE THE TABLE AT THE END OF THIS SECTION. THE MOVING TIMES WHICH HAVE BEEN ALLOWED FOR IN THE DESIGN OF THE EXPERIMENT ARE 40 SECS FROM F-LAYER POSITION TO THE LOWEST E-LAYER POSITIONS, 10 SECS BETWEEN E-REGION POSITIONS AND 30 SECS FROM THE HIGHEST E-LAYER POSITIONS TO THE F-LAYER POSITION. SOME MAGNETIC-FIELD INFORMATION, BASED ON THE IGRF 1985-1990 MODEL: GEOGR ALT/KM LAT / LONG L-SHELL INV-LAT DIP DECL B/N B/E B/D B/ABS 278.60 69.06 19.14 6.228 66.378 77.54 2.57 0.101 0.0044 0.459 0.470 101.13 69.39 19.19 6.233 66.387 77.74 3.65 0.107 0.0067 0.494 0.506 109.05 69.37 19.18 6.230 66.382 77.73 3.59 0.107 0.0066 0.492 0.504 119.60 69.35 19.18 6.229 66.381 77.72 3.53 0.107 0.0064 0.490 0.502 132.80 69.33 19.18 6.231 66.384 77.71 3.45 0.106 0.0062 0.488 0.499 REMOTE SITE GATES ARE CENTERED ON THE FOLLOWING ALTITUDES: ALTITUDE OF THE VOLUME : 90.5 95.8 101.1 109.0 116.9 124.8 278.6 RANGE FROM TROMSO : 92.7 98.1 103.5 111.6 119.7 127.8 285.0 CORRESPONDING CLOSEST MONOSTATIC GATE NUMBER (INDEXED FROM 1) E-LAYER POWERPROFILE : 10 12 14 17 20 23 E-LAYER ACF : 2 4 6 9 12 15 LOW RESOLUTION POWERPROFILE : 5 6 7(8) 9 11 (12)13 (47)48 LONG PULSE ACF : 7 THE LOW RESOLUTION POWERPROFILE DOES NOT HAVE GATES WHICH NOMINALLY COINCIDE WITH THE REMOTE STATION GATE. THE NEAREST ONE IS GIVEN WITHOUT PARENTHESIS. IN ALL THE OTHER CASES GIVEN IN THE TABLE THE TRISTATIC VOLUME AND THE NOMINAL MIDDLE POINT OF THE GIVEN TROMSO VOLUME COINCIDE.