How to use ATLOD

          

ATLOD has a long list of adverbs, and I will discuss them all and in the order in which they list in AIPS. Note that if you have a large number of sources to load in one go (i.e. more than 400), please alert either myself or Henrietta May so that we can ensure that there is adequate internal storage in ATLOD for your needs.

1. Input file specification adverbs
   • intape should indicate which tape drive you mounted and plan to read your RPFITS files from (intape is irrelevant if infile is set).
   • You can skip files at the beginning of the tape with the nskip adverb. It is a little tricky to skip files on a VMS `COPY' tape with the AIPS verb AVFILE because the ANSI format actually has three files (a header, the data, and a trailer) per RPFITS file. You have to remember to skip three for every one. If you get lost you can get really lost, because the verb TPHEAD, which generally lists the header of the next file on a tape, does not recognize ANSI file headers. Note however, that ATLOD now skips files as fast as AVFILE and you don't have to be able to multiply by three (nskip is irrelevant if infile is set).
   • Select the number of files that you wish to load with nfiles=10 (say, for 10 files). All the input files selected on the tape will be concatenated into the one output file, so it helps if they are in time order on the tape - if not, you will need to use UVSRT\ and INDXR as described in § 5). The default is that ATLOD loads one file (nfiles is irrelevant if infile is set).
   • bcount and ncount give you control over which scans you would like to load from the file. Leave them at zero for all scans, or specify the first scan to load with bcount, and the number of scans to load with ncount. These apply to all files loaded, and are mainly useful for disk file input (when infile is set).
   • If infile is blank, ATLOD will try to load data from the specified tape device (intape). Otherwise it will look for disk RPFITS file input. You should set this adverb to something like infile='MYAREA:BIGFILE' where `MYAREA' is an upper-case environment variable pointing to the directory where the upper-case file `BIGFILE' resides (see § 4.3). If you don't include the environment variable, ATLOD will look for the specified file in the /DATA/FITS area.

2. Output file specification adverbs
   • Fill in the output file name with commands such as outname='multi', outclass= 'uv', outseq=0, and outdisk=1. Select the local disk with the most free space (use the AIPS command FREE to see which). Note that if you completely specify these four adverbs, ATLOD will always try and put the data it is reading into that file. If you leave, say, outseq at zero, every application of ATLOD will try to make a new file with an increasing version number. One application of ATLOD reading multiple input files will always put the output in one file (even if outseq=0). This is the usual mode of use.

     DBCON can be used to combined multi-source files from the same observation. However, it has some problems with multi-source files, so please try to do this concatenation with ATLOD.

     Note that data from different observing runs with different configurations should not be combined into one multi-source file with ATLOD. This is largely because baselines with the same number but from different configurations would become confused. In principle, the subarray capability of AIPS could be used to deal with this, but ATLOD does not currently take advantage of this. There are other practical problems too. Multi-configuration data can be combined after calibration with DBCON. This is discussed in § 14.

3. Action selection adverb
     
   • ATLOD has the capacity to just list summaries of the contents of the file rather than actually load it into AIPS. If you would like to see summaries of each scan, set optype='list'. For very brief summaries of each file, set optype='summ'. Otherwise, to load the data, put optype='load' or leave it blank.

     A further option, optype='sysc', is available to write three text files containing the on-line measurements of ancillary system calibration information. These text files can contain (see also cparm(9) the antenna-based XY phase differences (), the system temperatures and sampler statistics for the X and Y feeds of each antenna. These files are written into the /DATA/FITS data area. The only other selection adverbs that are active in this mode are freqsel, ifsel, sources and timer. You can use the program pltsys to plot the contents of these files. Just enter the command pltsys at the Unix prompt and answer the questions. This program has frequency discrimination, so you can use this option of ATLOD and select all frequencies when generating the text files rather than running ATLOD once for each frequency.

4. Data selection adverbs
   • You can select a subset of the available sources in your RPFITS file to load by specifying a list to load with the sources adverb. If you prefix any of the sources in the list with a minus sign, such as sources='-0032+11',' ', then all sources except the named sources are loaded. The default (blank) means load all sources.
   • Load sources from a specified time range with the timerange adverb. The first day in the data is day zero, so, to select data from 12 hours and 33 minutes into the first day (day zero), until 2 hours, 12 minutes and 53 seconds into the second day (day one), use timer=0,12,33,0, 1,2,12,53. To load all data, leave it at zero.
   • Select data at the desired frequencies (this refers to the frequency at the band centre) with a command such as freqsel=4740,8500, where the frequency must be in MHz. Load data at all available frequencies by leaving freqsel at zero. The tolerance for matching frequencies is 1 KHz.
   • Select data from the desired IFs with a command such as ifsel=1,2. For example, consider an observation where IF 1 was set to the 2300 MHz band and IF 2 was set to the 1400 MHz band and you cycled in time through different pairs of frequencies in these bands. If you wanted to load just the 1400 MHz frequencies, you could list them all with freqsel but easier would be to just set ifsel=2.
   • You can load a subset of the available channels in each spectrum with the chansel adverb. Leaving it at zero ensures all channels are loaded. You can specify up to 10 groups (start, end and increment) of channels; all others will be loaded, but flagged. In the special case of one group of channels (such as chansel=7,25,2) ATLOD\ will actually drop the unselected channels and adjust the channel spacing in the file header to reflect this.

     For example, to load (unflagged) channels 1 to 320, and 322 to 513, use chansel=1,320,1, 322,513,1.

     This option would be useful if you are forced to use data compression (see below) and there is a channel with strong interference in it. The unselected channel is not used when scaling the visibilities into integers.

     You may also find it useful to discard a group of channels at the beginning and end of the spectrum where the filter response is poor. In this way also, should you have chosen data compression, you can increase the potential dynamic range of the data, as the range of visibility amplitudes from the central portion of the band is likely to be smaller than the range from the entire band. This also helps with any disk space problems.

     You should also consider discarding every second channel if you have invoked Hanning smoothing (see dparm(2)).

5. Output file control adverbs
     
   • If you have data with 2 or more simultaneous frequencies, the IF axis of the AIPS visibility file structure can be used for the contemporaneous data. It is a regular axis, so the IFs are numbered 1 to N. The actual frequency of the IF axis location is tracked by the use of offsets in the FQ table (see also Appendix C). Each entry (FREQID) in the FQ table is in fact an array of length equal to the length of the IF axis. Thus, each FREQID delineates a different collection of simultaneous frequencies. For example, you might observe at 4400/4600 for a while, and then at 8400/8600 for a while. Each pair of simultaneous frequencies would be described by a separate FREQID.

     Setting ifmap=1 will cause ATLOD to map the on-line IF number to the AIPS IF axis location. For example, if you observed at 4400/4600, selected only frequency 4600, and set ifmap=-1 (default) then ATLOD would put this in IF axis location one. However, if you set ifmap=1, then ATLOD would put it on IF axis location 2, and zero fill location 1 (this would be wasteful of disk space of course).

     Generally, I suggest that you set ifmap=-1.

   • NIF allows you to tell ATLOD how long the IF axis should be. Normally, ATLOD will work out the length of the IF axis based upon the information present in the first scan header that is consistent with the user's selection crteria. However, it may be that you know more than ATLOD, and you need to tell it the length of the IF axis.

     Consider the case where you have an observation such as

                    IF chain 1       IF chain 2
        Scan 1:        4700              8300
        Scan 2:        4600              8000

     Let us say you set, for some reason known only to yourself, freqsel=4700,4600,8000, thus omitting the 8300 frequency on IF chain 2. Now, you MUST set nif=2 to get all these frequencies loaded. ATLOD notes that you want one frequency from the first scan, and that it would go to IF axis location 1. But this is the only information it has when it reads the header for the first scan, which is when it must build the attributes for the output file. It doesn't know that the other specified frequencies are is succeeding scans and that they will need an IF axis of length 2. Thus, you must set nif=2. IF axis location 2 for the first scan will get filled with dummy visibilities of zeros.

     Note, that if you set freqsel=8300,4600,8000 instead, and ifmap=-1, then the 8300 frequency would be put onto IF axis location 1. However, if ifmap=1 then it would go to location 2, and location 1 would be zero filled instead.

     Generally, I suggest you set nif=0.

   • If you plan on loading a very large visibility data base, then it may be advisable to set the data compression adverb, douvcomp=1. However, note that you sacrifice some dynamic range (each real and imaginary part of the correlation is stored as a 16 bit integer rather than a 32 bit floating point value) in doing this, so it's a trade off. If your data suffer from severe interference, then loading them straightforwardly with data compression would compromise the good data, as all the dynamic range would be expended in coping with the interference. However, each visibility is compressed independently, so if a visibility has strong intereference and you compress it, you only sacrifice dynamic range in that visibility, not all the others as well.

     For continuum work you should set douvcomp=-1. For spectral-line work you must make your own decision.

6. The APARM array
   • aparm(1) controls an important part of the procedure which interacts with a number of the other ATLOD inputs. This is what form, if any, the correlations are to be converted to. That is, are they to be converted to Stokes parameters (see § 3.2), or should they be left in their original form.

     Note that a full polarimetric calibration of ATCA data cannot be performed with AIPS; you must do this with MIRIAD (see the MIRIAD User's Guide).

     Recall that raw ATCA correlations (visibilities) are in terms of linear polarizations, and that our ultimate goal is to convert them to the Stokes parameters, I, Q, U, and V which we can then image. A nominal conversion to Stokes parameters is achieved if you put aparm(1)=1.

     This nominal Stokes conversion makes three assumptions (the need for such assumptions is obviated by a correct calibration in MIRIAD).

       
     1. It assumes that the X and Y gain amplitudes were equal when observing. Such an equalization is generally a part of the observing set-up chores (ATCA users will be familiar with the program CACAL and earler DELCOR that does this) so that this should be a good assumption. ATCA data from the first year or two of operation may not have equalized gains. Any mismatch causes Stokes I, Q, and U to be corrupted by a term approximately equal to

        
        
        

        where is the error in the amplitude of the gain a (for the X and Y feeds, and i and j are the antennas involved in the interferometer). This is a first-order effect.

     2. It assumes that the feeds are perfectly linearly polarized. Errors in this assumption cause an unpolarized source to appear polarized by an amount given by the instrumental polarization. This is about 3%. It affects I only as a second order effect.
     3. It also assumes that the measurements are good (see § 3.5). The quality of these measurements is quite good now, although you should still inspect them first (with the optype='sysc' and pltsys combination) rather than use them blindly. They are generally worse at 13 and 20 cm than at 3 and 6 cm.

     If you don't convert to Stokes parameters, then you must calibrate the XX and YY (remember 2I=XX+YY) polarizations separately under the assumption that your calibrators are not linearly polarized. Although this means the gains are wrong by amounts given by the linear polarization of the source, the errors cancel out on application to a target source (see § 3.3). So you are left only with uncalibrated second order leakage terms.

        If you set (aparm(1)=-1), then no Stokes conversion is performed, but AIPS is tricked into believing that the linear polarizations are circular (you could also use PUTHEAD on the header to effect this trickery if you don't do it with ATLOD). It does not affect the actual calibration in any deleterious way. Leaving aparm(1)=0 means load the polarizations just as they are and label them as linear; I do not recommend doing this.

     I suggest you set aparm(1)=-1, thus loading the correlations in their natural form, whilst labelling them as circular. You should use MIRIAD to make the best possible calibration, whether you are interested in Stokes I only or all Stokes parameters.

   • If you set aparm(2)=1 then the data that the on-line system flagged as bad are loaded to your AIPS file. You should have a good reason for doing this.

     Generally I suggest you set aparm(2)=0.

   • Auto-correlations are dropped by default. Put aparm(3)=1 to retain these. It is sometimes useful to retain the auto-correlations, as is the phase of the auto-correlation between the X and the Y feeds. The that is used for the Stokes conversion is an average across the central part of the band, and is stored in a different place. If you were particularly conscientious, you might care to examine as a function of frequency by examining the relevant auto-correlation.

     ATCA data with 128 MHz bandwidth and one IF taken after February 1991 contain a full set of meaningful auto-correlations. 128 MHz bandwidth data from earlier periods do contain the XY correlations, but they are not labelled in an obvious fashion. Please consult with me should you wish to extract them from the older style data. Spectral line data do not contain auto-correlations, so aparm(3) is unimportant in this case. 128 MHz data with 2 IFs do not contain auto-correlations as there is not enough correlator to form them.

     Generally I suggest you set aparm(3)=0.

   • For a compact array like the ATCA, the problem of shadowing may arise. That is, one antenna may be blocked by another when the source is low and of declination close to the latitude. ATLOD will drop shadowed antennas by default. If you want to change the shadowing criterion, then set aparm(4) in metres. An ATCA dish is 22 m in diameter. To relax the shadowing criterion, make aparm (4) smaller. Note that the shadowing calculation assumes that the source is at the pointing centre. For wide-field imaging, you might keep in mind that a source far from the pointing centre will be shadowed slightly differently.

     Initial tests indicate that at 20 cm, cross-talk begins before geometric shadowing. At 6 cm, it appears to occur soon after the onset of geometric shadowing. I suggest you set aparm(4)=0 at 3 and 6 cm, and perhaps aparm(4)=25 at 13 and 20 cm if you are feeling conservative.

       

   • If you load your data in time order, ATLOD will create the index (NX) table (see Appendix C). You have control over when index records are written. Basically, you want to start a new index record when there are no data for some period, and after data have been acquired for some contiguous time. The defaults of aparm(5)=10 and aparm(6)=60 minutes should be adequate.

     Generally I suggest you set aparm(5)=0 and aparm(6)=0.

       

   • ATLOD will also build the pristine calibration (CL) table if you load the data in time order. There is a CL table entry as often as you want interpolated gain solutions (see Appendix C). The default of 3 minutes (aparm(7)) should be acceptable for most situations. Note, however, that the CL table entry time interval sets the time scale on which you can self-calibrate a multi-source data base. You can however, split a source into a single-source data base, whereupon the fundamental integration time of the observation (generally 15 seconds for ATCA data) sets the minimum possible time scale for self-calibration.

     Generally I suggest you set aparm(7)=0.

   • An on-line correction for the system temperature is generally made (although you can turn it off when observing). This is used to scale the visibilities as the correlation coefficient gives the ratio of correlated power to uncorrelated (noise) power. However, the hardware that measures the system temperature is not ideal, so that this measurement is noisy, which contributes to the noise in your final image. aparm(8) allows you to average the measurements for a few integrations so that you apply less noisy measurements. If the on-line correction was already applied, this is first undone before application of the averaged correction.

     Set aparm(8) to the number of measurements that you wish to average (I would suggest 10 or so). If you leave aparm(8) at zero (or unity), then no additional scaling is done. If you set aparm(8)=-1 then the on-line correction is undone, and redone with an assumed of 50 K. This is equivalent to not turning on the on-line correction.

     Because the improvement is small, and it slows ATLOD down, I suggest that in general you leave aparm(8)=0.

   • If you require the on-line values (see discussion of aparm(1), cparm(4) and § 3.5) for Stokes conversion or for application to the Y gains (neither of which is recommended) then you may consider the use of aparm(9). Even if the current hardware allowed us to make a perfectly accurate measurement of , it would still be a measurement of a noisy quantity. aparm(9) specifies how many measurements should be averaged together to form the applied . If using this option, I would suggest averaging over 5 or so points. Leave it at zero to use the on-line values without any averaging.

     When this averaging option is active, wildly discrepant values are replaced by the running mean of the current stack of aparm(9) measurements. A point is considered discrepant if it is more than a few times the width of the current stack from the stack mean.

     Because we do not recommend general application of the on-line values, and because we now prefer all application of to be in MIRIAD, I do not recommend that you use this option. Set aparm(9)=0.

   • Dropouts in the visibility data generally occur in both the cross- and auto-correlations simultaneously. Thus, we can use the measurements (an auto-correlation) as a monitor and discard cross-correlations when the auto-correlation is obviously discrepant. If aparm(10)=0 then no flagging based on the quality of values is done.

     We can detect discrepant values in two ways. First, if aparm(9)>1 then the current can be compared to the mean of the previous aparm(9) measurements. Second, if you are not using the on-line measurements of (as recommended) directly but have input mean values into the xyphase array then a can be noted as bad if it is discrepant from the value in the xyphase array by more than some value.

     In the first case, you should have set aparm(9)>1 so that a stack of aparm(9) values can be accumulated for comparison with the current value. If aparm(10) >0 then visibilities associated with the antennas with the discrepant are dropped. If aparm(10)<0, then in addition to ATLOD's own internal criteria a point is not treated as discrepant unless it is abs(aparm(10)) degrees different from the running mean of the current stack of aparm(9) measurements. You need to inspect the values via the optype='sysc' and pltsys combination to be able to set this value meaningfully.

     In the second case, you should have set aparm(9)=0. Then, if you set aparm(10) < 0, a will be deemed bad if it is more than abs(aparm(10)) degrees from the relevant value in the xyphase array. Visibilities associated with the bad antenna will then be dropped.

     This is a useful editing option. However, because we do not usually recommend direct use of the on-line values, I suggest you only use it only if you input the mean values into the xyphase array and set aparm(10)<0 (and set cparm(4)=1 indicating use of the xyphase array).

7. The BPARM array
   • The bparm array allows you to tell ATLOD not to load specific baselines or antennas. You might have a priori knowledge that specific ones are bad and not flagged by the on-line system. You would want to be very certain of this before doing it, as every baseline is precious. See the HELP file for details.

     Generally I suggest you set bparm=0.

8. The CPARM array
   • cparm(1) allows you to print out some useful information on the workstation screen.

     If cparm(1)=1 then some of the system calibration group information is listed. You get the system temperatures, the values and the parallactic angle. You get the chance to stop the listing after a while (answer `Q' when prompted) and continue loading the data. Control is then returned to AIPS. This is meant for a quick inspection rather than an exhaustive listing. Use option optype= 'sysc' and examine the output text files for more extensive examination.

     If cparm(1)=2, then you are informed when ATLOD detects that the values or the system temperatures have jumped. By this, I mean jumped to a new value and stayed there, not just a dropout. ATLOD accumulates its averages for the for each frequency, and the system temperatures for each frequency and source. It resets its accumulations if one of these accumulations jumps unexpectedly. This option is active only if aparm(8) or aparm(9) is greater than 1.

     If cparm(1)=3, both the previous two options above are active.

     Generally I suggest you set cparm(1)=0.

   • cparm(2) should be left at zero.
   • If cparm(3)=1 then the visibility phases are negated. This can be useful for some data taken in 1990. Refer to the communiqué issued on the Epping bulletin board to see if any old data that you have needs this treatement.

     Generally I suggest you set cparm(3)=0.

   • If cparm(4)=1, then the values in the xyphase array are used rather than the on-line measurements.

     I suggest you only do this if you wish to use the editing option (see aparm(10)).

   • can be thought of as the phase part of the gain for either the X or Y feeds. Our convention is that they are allocated to the Y gains. If cparm(5)=1 and you have not asked for Stokes conversion, then the values are allocated to the Y gains which are then applied to the visibilities. There are two reasons for doing this.

     First, it causes the XX and YY visibilities to be in phase. Thus, you could ask an AIPS task to form Stokes I by summing XX and YY before calibration. This would allow you to edit on only one quantity, instead of doing it for XX and YY separately. However, generally the correct observing setup done on-line will ensure that XX and YY are already roughly in phase anyway.

     The second reason is now deprecated. We used to recommend that for those of you who were going to calibrate in MIRIAD that you apply the values in ATLOD. However, our preferred path now is to do this in MIRIAD.

     I suggest you set cparm(5)=0.

   • cparm(6) offers the prospect of either listing (cparm(6)=1), deleting (cparm(6)=2), listing and deleting (cparm(6)=3), and correcting (cparm(6)=4) data for which the sampler statistics are not at their optimal values. When these statistics are bad, it is likely that either something is wrong and the data are useless (such as strong interference for which there was insufficient range in the attenuators to bring the sampler statistics into the servo loop) or that the attenuators were slightly wrong as the servo loop bought them back into range for a source of significantly different flux density from the previous one; basically, it means that the visibilities have been scaled with the wrong factor.

     If you choose the option to correct the visibilities when the sampler statistics are wrong, then this works only for small errors (), and data with larger errors in the statistics are dropped.

     Note that since the sampler statistics are antenna based, all visibilities involving the bad antenna are dropped.

     It has been noted in some data prior to March 1992 that the sampler statistics were not getting updated every integration. This means that the off-line correction would make the wrong correction for some visibilities. To assess whether you have this problem or not, you should use the optype='sysc' option with cparm(9)=2 or 3 (see below) to create text files containing the sampler statistics which you can then plot with utility program pltsys. Look for successively duplicated entries.

     For data prior to March 1992, I suggest you set cparm(6)=1 (just providing you with warnings). For more recent data taken up until 11 December 1993, I suggest you put cparm(6)=4 and correct the data. For data taken since then, the corrections are done in the correlator. ATLOD will refuse to correct such data.

   • If cparm(6) >0 but you want to use a tolerance different from that of ATLOD for the tolerable error in the sampler statistics, then set cparm(7) to your desired value. ATLOD marks an antenna as bad if the sampler statistics are more than 5% from the correct value (cparm(6)=1,2 or 3), and uncorrectable if they are more than 10% from the correct value (cparm(6)=4).

     Generally I suggest you set cparm(7)=0.

   • If cparm(8)=1 then the file specified by the infile adverb is converted to lower case before being used. This is intended for use at Narrabri where VAX disks are mounted on the SUNs via multinet.

     Generally I suggest you set cparm(8)=0.

   • When you set optype='sysc', ATLOD writes useful sytem calibration information into text files in the FITS area. By default, you get the s, and the X and Y system temperatures. If you set cparm(9)=2 then the text files contain the X sampler statistics (one file for each of the low, mean and high sampler measurements). Similarly, cparm(9)=3 gets you the Y sampler statistics.

     Generally I suggest you set cparm(9)=0 unless you are interested in the samplers.

   • If you set cparm(10)>0, then the system temperature for IF 1 will be applied to IF 2. This is a fudge for data taken in a period of time when IF 2 did not have its own measurement of the system temperature.

     Generally, I suggest you set cparm(10)=0.

9. The XYPHASE array

   This adverb is offered for you to enter one number per antenna and frequency for the values that you choose from the pltsys plots if you require them. We no longer recommend application of in ATLOD.

   However, the editing option (see aparm(10)) is still useful, and this is the only case now for which you should enter values into this array.

   You must enter one phase difference (in degrees) for each of the six antennas in the array, regardless of how many antennas were in your observation. Naturally, the values that you insert for antennas that you do not have are arbitrary. If you have observed at more than one frequency (e.g., 3 and 6 cm), then you must enter these values for all frequencies. First, you enter the six numbers for the first frequency observed, and then the six numbers for the second frequency observed, and so on. They must be in the correct frequency order as encountered and loaded in the observation. You could check this by listing a bit of your data (optype='list') and seeing what order the frequencies arrive in.

   Except for editing, I suggest you set xyphase=0.

10. The DPARM array
    • There was a problem with some data for a short period of time where the sources were mislabelled on-line. If you set dparm(1)=1 then all sources in the scan will be labelled as the first source in the scan based RPFITS source table. This is appropriate only for data containing one source per scan (i.e., not mosaicing data).

      Generally, I suggest you set dparm(1)=0.

       

    • If dparm(2)=1 then each spectrum is three-point Hanning smoothed. The usual reason for Hanning smoothing spectral-line data is to reduce ringing around narrow spectral features. The ringing occurs because the weighting function in the lag domain is a top-hat function so that the frequency spectrum is convolved by a narrow sinC function. If you do this then you should probably consider dropping every other channel (with adverb chansel) as adjacent channels will no longer be independent.

      Of course, at 128 MHz bandwidth, adjacent channels are already correlated so this option is only meant for spectral-line data.

      For 128 MHz bandwidth data you should set dparm(2)= 0. For narrower bandwidths, this option should be considered.

    • Until March 03 1994, ATLOD wrote the weights of the visibilities as 1.0. After this date, and for data taken after February 22, 1994, the default behaviour is for ATLOD to write the visibility weights as the integration time in seconds divided by 15. If you set dparm(3)>0 then ATLOD will write the weights as equal to the value of dparm(3). You could therefore set it to unity to get the old behaviour.

      The integration time that ATLOD now has access to includes times losses such as the on-line hold and blank times and also reflects any averaging done in the correlator. Therefore it should accurately reflect the on-source time for each integration.

      The division by 15 is for two reasons. Firstly, the task AVSPC (see § 6) that averages channels together sets the weight of the averaged channel to the sum of the weights. This means that the weights may get very large and PRTUV which lists the weights (as well as the data) will overflow the weights field. Secondly, most ATCA data taken to this date have an integration of around 15 seconds (usually somewhat larger for spectral-line data). If you concatenate old unity weight data with new data via DBCON you should rescale the weights of one of the data sets accordingly (see 14). If you forget to do this, this normalization will minimize the damage (or possibly obscure it !).

      Generally, I suggest dparm(3)=0.