next up previous contents
Next: DBBC: Up: Wide Band Observing: RDBE, Previous: Wide Band Observing: RDBE,   Contents

The RDBE system

The RDBE (Roach Digital Backend, where ROACH is the core board containing a large FPGA) is a module that takes in 2 analog IF signals, applies an anti-alias filter that passes 512 to 1024 MHz, sets the power levels, samples the signals at a 1024 MHz sample rate (8 bit samples at this stage), digitally filters the data to the final basebands, resamples the data to 2 bits, and formats it for recording. It takes the place of the IF distributers, baseband converters, samplers, and formatter (including pulse cal detectors) in the old VLBA system. Sometime in 2013, the VLBA antennas will have two RDBE systems each, allowing an increase in the number of channels, at least with the DDC personality (see below). In addition to allowing increased numbers of channels, the use of 2 RDBEs allows simultaneous access to all 4 VLBA IFs. That is useful for the S/X system and for the new 4-8 GHz system, for which two polarization pairs of output data are available.

Control of the RDBE and Mark5C recorders is handled by a new VLBA control system running on a standard Linux computer. The new system software is based on the EVLA Executor. Schedule information is given to this computer by way of the VEX file, which is converted by operations to a Python script that is read by the Executor. All new hardware installed at the VLBA for the next few years will use this control system and, probably slowly, the old hardware will be switched over to it. In the meantime, both crd files and VEX files are needed to control the VLBA sites. When the new 4-8 GHz receiver was installed on the VLBA, a new RF switch controller was installed that affects all observing bands. Because of this, both the new and old control systems must be used to support observations with either the new or old recording systems. Note that the VEX file is also used by field system stations (EVN and others) for antenna control.

The new control system at the site, and the DiFX correlator have a slightly different idea of when scans should start compared to the old systems. With the old system, the media are commanded to start at a time that is the same at all stations and is set as the nominal scan start time minus an offset given by PRESTART. The Vex file shows that time as the uncommented ``start'' time of the scan. But the Vex file also has a station dependent offset for the start of good data. With DWELL scheduling, that is usually zero. But if DURation scheduling is used, it can be significant. That time is often referred to as the good-data time or on-source time. It depends on SCHED's concept of slew times and settling times. The new systems do not start the media, or the correlation until that time.


Note that the terminology for the various signals has become rather confused. For backward compatibility in SCHED , we call the final analog signal sent to the sampler at 512-1024 MHz the ``IF''. That is broken into narrower bandwidths called ``subbands'' by a polyphase filter regardless of RDBE personality. There is no flexibility to move those subbands around in frequency. The final signal that is resampled to, usually, 2 bits and recorded is called the ``baseband channel'' for purposes of SCHED. The baseband channel might be a subband (PFB personality) or might be further frequency shifted and filtered from within a subband (DDC personality). This terminology differs somewhat from EVLA practice where a baseband is the final analog signal and the final filtered signal is a subband.


The FPGA in the RDBE supports multiple personalities that can be swapped as needed. The first developed, and only one available in early 2012, is the PFB personality that uses a polyphase filter to break each of the two 512 MHz IFs into 16 basebands of 32 MHz each, all lower sideband. Exactly 16 channels must be recorded, selected from the total of 32 provided from both inputs. This personality is selected setting the DBE parameter to RDBE_PFB in the setup file. Of the 16 subbands of the polyphase filter from each IF, 15 provide good data. The other is really 16 MHz from each end of the 512 MHz, is probably not useful. It is made available for selection in cases where it is desired to have all 16 required channels in one polarization. More typically, 8 channels, constituting polarization pairs, will be selected from each IF input. This personality can only provide 32 MHz basebands at fixed frequencies within the IF for a total of 2 Gbps. Other than selecting the desired subbands, there is no tuning flexibility. Note that the PFB personality cannot be used on both RDBEs at a VLBA station (once those are available) because the required VDIF output is not available and because the required 2 Gbps per RDBE is beyond the capacity of the recording system.


The second personality that is available is the DDC (Digital Down Converter - under final testing in late 2012). It is selected using the DBE parameter set to RDBE_DDC in the setup file. This personality will provide a few 4 (possibly 8 later) filters per RDBE that can provide arbitrary offset frequencies and can provide any bandwidth at the factor of 2 steps between 1 and 64 MHz. (That may be extended downward some day). There is a complication forced by the use of a polyphase filter first step of filtration to get the clock rate down to values the FPGA can support. Such filters do not have flexible frequency ranges. This one splits the band into 3 segments, 512 to 640 MHz, 640 to 896 MHz, and 896 to 1024 MHz. Each baseband must be confined to one of those ranges. The ``crossover'' frequencies at the filter edges have a range of something like 4-10 MHz (to be determined) that is not really usable. SCHED will issue a warning if an attempt is made to have a baseband cross one of these boundaries. Note that the polyphase filter to use will be determined by the frequency of the LO sum. It is possible that users of the 128 MHz bandwidth will want to offset slightly for better pulse cal performance and this will cause a tiny fraction of the band to get aliased. SCHED will issue warnings, but this can be tolerated.

The frequencies for the band edges in the DDC personality can be set to any multiple of $256 / (2^{32})$ MHz = $0.0596046$ Hz in principle. But values that are not integer Hz would cause problems elsewhere - mainly with returning to phase after changes. The smallest allowed value that qualifies is 15.625 kHz. One way to look at the allowed values is that they are N*125 kHz plus 0, 15.625, 31.250, 46.875, 62.500, 78.125, 93.750, or 109.375 kHz. For the moment, it may be best to stick to multiples of 10 kHz (the step in the BBCs of the old system) until we determine that there is adequate precision everywhere. The only way to do the is with multiples of 250 kHz. However finer tuning (multiples of 15.625 kHz) is now (SCHED version 10.2) supposed to work, but has not yet been tested. /schedb will warn if the frequency is not a multiple of 250 kHz and will abort if it is not a multiple of 15.625 kHz.


The overall LO/IF/RDBE system on the VLBA will have some significant tuning flexibility issues. The RDBE is an addon to the older system where the baseband converters, which could take only a small portion of the 500 MHz IF, provided the necessary flexibility. The LO/IF system that creates those IFs is based on synthesizers that have set points at N*500+-100 MHz. Now, with the ability to take all of an IF, that restricted tuning ability will become an issue, especially in conjunction with the lack of tuning ability for the PFB personality and the crossover points for the DDC. Essentially all frequencies can be reached using more than one LO setting, so no cases have been identified where a particular spectral line cannot be observed. But full tuning flexibility that might be desired is not there. Eventually we hope to upgrade the front end synthesizers to designs with more tuning options, and in fact design and prototyping of such a system has started, although deployment is not yet funded (Feb 2012).

Note that, for the initial /schedb implementation in place (versions 9.4, through at least 10.1) the code to provide default channel frequencies based on the band has not yet been written. It is necessary to give the frequencies in the setup file. See the simple examples. The defaulting capability will be added eventually. But for now, the upper-edge baseband frequencies with PFB personality must be from the following list: 1040.0, 1008.0, 976.0, 944.0, 912.0, 880.0, 848.0, 816.0, 784.0, 752.0, 720.0, 688.0, 656.0, 624.0, 592.0, 560.0. These can either be selected directly using the BBSYN setup file parameter, or values of FREQREF and FREQOFF can be selected so that the difference between the desired baseband frequency and the signed sum of all other LOs is one of these values.


Example SCHED input (.key) files for observations using the new systems are:

egrdbe2.key which is a reasonably simple case with the VLBA and GBT.

manual_1.key shows a lot of SCHED inputs instead of taking the defaults. It uses the RDBE/DDC on the VLBA and WIDAR on the VLA1 (VLA1 not really offered yet, but VLA27 would be similar).

rdbepfb.key which is an even simpler case with just the VLBA and using a standard setup.

egrdbe.key which demonstrates more fully specified setups and demonstrates scheduling of piggyback Mark5A and Mark5C observations.

eg3mm_rd2.key which shows how to do referencing pointing at 3mm, including using narrow band data on masers, when the RDBE is used with the PFB personality. It uses the crd input parameters to control the VLBA legacy hardware which produces the power data used for reference pointing.

egddc.key which uses the DDC personality of the RDBE.

egddc2.key which uses the DDC personality with two RDBEs and 8 baseband channels.

egcwide.key which uses the PFB personality and observes using the new 4-8 GHz VLBA receiver with one dual polarization setup and one single polarization, widely split frequencies setup.

jvla.key is an example that uses the PFB personality of the RDBE for joint observations with the GBT and VLA. There are only 512 Mbps at the VLA in this mode.

vladdc.key is an example that uses the DDC personality with the VLBA, VLA, and GBT with a full 2 Gbps on all three.

hsaddc.key is an all-singing, all-dancing HSA example with the VLBA, VLA, GBT, and Effelsberg. It exercizes reference pointing at all sites, array phasing at the VLA, and Doppler tracking.

pfbsettst.key is a vehicle for testing all the new RDBE/MARK5C standard setup files that use the pfb personality. These are the setups that start with rdbe_pfb.

n2227.key is a sample USNO Earth Orientation observation using PT and MK.


See the Reference Pointing section for much more information on how to do reference pointing.

Reference pointing on the VLBA is handled by the legacy system using power measurements made by the old baseband converters (BBCs). When the main observations are using the RDBE, SCHED tries to set the BBCs as closely as possible to the settings of the RDBE baseband channels. For bandwidths below 16 MHz, this can be done well for 8 channels, as long as the requested frequency is below 1000 MHz in the IF. SCHED will select the center 8 channels if there are more from the RDBE (the PFB always gives 16). For wider bandwidth baseband channels, 16 MHz will be used and it will be centered on the RDBE channel. All this means that reference pointing, including using Doppler for setting frequencies and setting narrow bandwidths for masers can be done normally for the DDC personality. If such reference pointing must be done when using the PFB personality with it's fixed 32 MHz baseband channels, it is best to set up the old system using the piggyback scheme as demonstrated in the example egrdbe.key. That requires separate runs of SCHED for the MARK5C and legacy systems.


Normally when scheduling a project that uses the RDBE/MARK5C system, SCHED creates control files for the old system (.crd files) that drive the telescope and other systems, but do not cause MARK5A recordings to be made. Since SCHED does not have adequate bookkeeping to allow independent specification of frequencies for both systems in one pass, a reasonable choice of frequencies and bandwidths for the old system is made based on the capabilities of that system and the settings for the new system.

During testing of the RDBE/MARK5C system, it is useful to have a parallel Mark5A recording. If the default choices of frequencies and bandwidths for the old system is adequate, SCHED can be told to make MARK5A recordings using parameter switch /htmlrefDOMKAMP:DOMKA. The only way to check what those settings are is to look at the output .crd files. Because of the limited bookkeeping, that information does not appear in the .sum file.

If the user does not want to take the Mark5A setups provided by SCHED with DOMKA, then the run can be set up as a piggyback with separate setups for each system. The scheme for doing was mentioned above, and is described and demonstrated in example egrdbe.key.

next up previous contents
Next: DBBC: Up: Wide Band Observing: RDBE, Previous: Wide Band Observing: RDBE,   Contents
Craig Walker 2014-06-17