Return to NAIC Home
Last updated 1996 March 25. New version, 1996 March 25.
This document is based upon a Radar Interface discussion, 1995 November 1, with Don Campbell, Eddie Castro, Alice Hine, Mike Nolan, Phil Perrilat, Bill Sisk, Mike Sulzer, and Bob Zimmerman present. This is a dynamic document intended to reflect current plans, not a meeting record.
The ``Radar Interface'' data acquisition system has three primary modes of operation for planetary radar: hardware decoding of a coded signal, direct sampling of a coded signal, and CW spectrometry. Construction of the hardware to receive the first two is underway, and can probably be finished by first light. At this meeting, a new purely digital scheme for CW data acquisition was blocked out. This new system will not work by first light, but, since much of the pre-upgrade hardware can probably be made to work, albeit inelegantly and not as well, it was concluded that it was better for the new system to be done well rather than soon.
Note that I have used the term "Radar Interface" loosely, and it really refers to two different things: a specific fast sampling system, and all of the hardware used to do Planetary Radar. I should fix that discrepancy, but haven't yet.
The initial input for this system is a 260 MHz IF (presumed down-converted from 2380 MHz upstairs). This signal has not been Doppler corrected. This signal comes down on an optical fiber, then out of the downstairs IF/LO system on the .2-.4 GHz channel. The analog gain is set in the IF/LO. There are two independent channels in this hardware, but there is only one set of frequency synthesizers. The 260 MHz signal is passed through an analog filter, either 20 MHz or 5 MHz bandwidth, for rejection of out-of-band signals. This filtered 260 MHz signal is then down-converted to baseband by a Doppler-correcting mixer, then analog low-pass filtered for sideband rejection. The signal power is measured at this point for setting the gains to get good dynamic range into the digitizers.
At this point, the signal is fed to two separate 8-bit digitizers running at 80 MHz, referred to here as set A and set B. The digitizers feed the software-selectable digital sinc filters, which provide separate 9-bit I and Q outputs. The set A filters provides 4-bit input to the hardware decoder. Which 4 bits is software selectable. The set B filter outputs are converted to analog for input to the Radar Interface, or for whatever use is desired.
The baseband signal will be sampled at 12 bits by a digitizer running at 40 MHz or 10 MHz, depending on which bandpass filter was selected. The digital signal then goes through up to 8 half-band filters, and comes out as 12-bit I+Q. There will then be hardware to select and pack the bits, and send them to some device that can in principle write the numbers down. Note that the minimum output sample rates are 80 kHz or 20 kHz (depending on the initial filter and sample rate) in each polarization and in each of I and Q. At the 40 MHz sampling rate, the accumulators in the filters could overflow if a low frequency tone were present, but not at the 10 MHz sampling rate.
The data acquisition system for the hardware decoder and direct sampling using the Radar Interface are similar and connected, and are discussed together. Note that the term "Radar Interface" has just become a misnomer, as with this digital system, there are now two: One for coded (windowed noise-like) data, and another for CW (infinite tone-like) data.
The initial input for this system is a 260 MHz IF (presumed down-converted from 2380 MHz upstairs). This signal has not been Doppler corrected, though that may be possible and useful in the future. This signal comes down on an optical fiber, then out of the downstairs IF/LO system on the .2-.4 GHz channel. The analog gain is set in the IF/LO. There are two independent channels in this hardware, but there is only one set of frequency synthesizers. The system will be described as if there were only one channel, for simplicity.
The 260 MHz signal is passed through a (tunable ?) analog filter, either 20 MHz or 5 MHz bandwidth, for rejection of out-of-band signals. (What kind of filter?). The 20 MHz filter was chosen because it is a substantial fraction of the transmitter bandwidth (26 MHz at -1 dB), and because there are cellular allocations in the 2390-2417 MHz band (but possibly not in Puerto Rico). The 5 MHz filter is simply a narrower one in case there are problems, which serendipitously turns out to be useful for the CW interface.
This filtered 260 MHz signal is then down-converted to baseband by a Doppler-correcting mixer, then analog low-pass filtered for sideband rejection. (What kind of filter?) The signal power is measured at this point for setting the gains to get good dynamic range into the digitizers. The importance of this power monitor was debated: In principle the information is available from the Radar Interface, but that has already been digitally filtered, so a birdie within the 20/5 MHz prefilter band, but outside the selected digital filter band could give problems in setting the gains and choosing the correct bits to finally sample. With both monitors, it is possible to detect and compensate for this problem.
At this point, the signal is fed to two separate 8-bit digitizers running at 80 MHz (only??) (allowing 2 samples per baud at 20 MHz), referred to as set A and set B. The digitizers feed the software-selectable digital sinc filters, which provide separate 9-bit I and Q outputs. The set A filters provides 4-bit input to the hardware decoder. Which 4 bits is software selectable. The original scheme was to select the bits based upon a formula from the measured power either from the power counter or the RI. This system amounts to an AGC, and, while convenient for some things, makes the system uncalibratable. The general feeling at the meeting was that, while the option of this AGC would be nice, a purely manual system would be better than a purely automatic one. DC pointed out that, since the antenna gain should be more constant after the upgrade, the need for an AGC should be reduced. "Manual system" in this context means a user-entered value, not a big black knob. Phil suggested that the software would compare the power counter reading to the I output, and flag a mismatch to the user. If this problem were due to a signal out of the (final) bandpass that was not so large that the digitizer dynamic range was exceeded, the output bits could be selected manually and observing could proceed. Note that this procedure may be dangerous with sinc filters because of their sidelobes.
The set B 9-bit I+Q filter outputs are converted to analog for input to the Radar Interface, or for whatever use is desired.
Phil asked whether we will ever need to use both the Sampler and the Decoder at once at all or even at different frequencies. He'd rather not program to allow those options, except perhaps to allow the Sampler outputs to be used for monitoring. The scientists all thought that it seemed like a good idea for some undefined future use, but no champion arose to defend it.
There was a discussion of how the various digitizers should be clocked. It is necessary for the Decoder, and convenient for the Radar Interface, for the digitizer clock to be synchronized with the transmitter, including Doppler-induced time dilation. Since the digitizer for the decoder is the one described above, the clock for filter set A must be shiftable. Since the Radar Interface has its own digitizer, in principle, the clock for filter set B needn't be shifted. Doing so amounts to a slight frequency shift of the filters. Another issue is that we sample the transmitter output for some experiments, and don't want the samples to happen on code transitions. In practice, it's probably easier if all of the clocks are the same, both for implementation and for monitoring, and nobody came up with a good reason not to. If the clocks are all synced, what's the point in going back to analog, except for slower sampling if desired? Is there a reason the digital outputs of set B shouldn't head straight for the digital inputs of the new CW RI (when it exists)?
The "interim" (analog) CW system will consist of parts of the new Decoder / Sampler, combined with cabled analog hardware. While there is always the danger that "interim" + "budget cut" => "permanent", this system isn't really sufficient for the needs of the atmospheric or pulsar groups, and the proposed system seems like it will be.
Again, we have a 260 MHz IF into the Downstairs IF/LO. This will be run through a 5 MHz bandwidth filter, to a Doppler-shifted mixer down to a second IF (probably 30 MHz). The 30 MHz signal will be filtered at something like 10 x the desired baseband width to clean it up and cut out the other sideband. This 30 MHz signal is then mixed and quadrature separated into baseband I and Q. These latter are then low-pass filtered at the desired bandpass and sampled by the Radar Interface.
The filtering and mixing to 30 MHz could be done in the downstairs IFLO, if one of the two remaining synthesizer MUX channels was dedicated to it. (Does this mean the channel can never be reclaimed, or that it's a job? Even we are leery of using up what seems like a scarce resource for a temporary system.)
Whether or not it's done in the IF/LO, the feeling is that this setup can be working by first light, but that it's a hack, and that the available filters aren't optimal. Thus began the digital CW interface:
A digital CW filter set would eliminate the cabling nightmare and give us nice clean filters. It seems like the Sampler could be pressed into service, but there are several issues that suggest we need another system. It would be nice if this were a general-purpose system, as well. The following issues were raised:
Dynamic Range. Mike Sulzer says the 9 bits provided by the sampling system aren't enough; they really need 12 for ionospheric work.
Filters. The CW system should have square filters, not sinc filters. Mike Sulzer agrees for his work. Nobody comes up with a reason to need sinc filters as well. Digital filters have accumulators in them, which can overflow in the presence of a strong tone.
Speed. Planetary Radar only needs a relatively narrow bandwidth, typically a few to a few hundred kHz. Aeronomy routinely needs several MHz, and Phil suggests pulsars would probably like 20 MHz (probably should ask). The problem with sampling that fast is that it's difficult to record the data.
After some discussion, a scheme is blocked out. Again, the signal will come down at 260 MHz, and go through the 20 MHz / 5 Mhz bandwidth filter, be mixed down to baseband and low-pass filtered to remove the other sideband, all as in the Decoder / Sampler (probably the same hardware). The baseband signal will be sampled at 12 bits by a digitizer running at 40 MHz or 10 MHz, depending on which bandpass filter was selected. (Bill Sisk has seen ads for a 40 MHz 12 bit A-D by a reputable manufacturer). The digital signal then goes through up to 8 half-band filters, and comes out as 12-bit I+Q. There will then be hardware to select and pack the bits, and send them to some device that can in principle write the numbers down. Note that the minimum output sample rates are 80 kHz or 20 kHz (depending on the initial filter and sample rate) in each polarization and in each of I and Q, because there are only 8 filters, and 20 MHz / 2^8 = 78125 Hz. Adding more filters would be complicated. Programmable filters exist, but have lower dynamic range, which was the point in the first place. Most believed that this limit was acceptable because it is slow enough that, if desired, additional filtering could be done in software, and in any case is easily recorded on tape. At the 40 MHz sampling rate, the accumulators in the filters could overflow if a low frequency tone is present, but not at the 10 MHz sampling rate. Bill asked if real (rather than I+Q) outputs would be acceptable, as it would cut then number of output lines in half (and run twice as fast), and an 8-octave real filter board has already been designed. The scientists preferred I+Q if possible, because it keeps the math neater. Since the design must be changed anyway to allow packing, it will be I+Q.
D-A converters will be provided for monitoring. Because the half-band filters decimate the frequency (unlike the sinc filters), the output will have a messy frequency response, but an analog filter will be provided to smooth it (is some interpolation necessary anyay?). Because this output is only designed for monitoring, the relatively poor behavior of this analog filter is considered acceptable.
This system is not expected to be on line at first light. Bill was optimistic that it wouldn't take too long, as the boards are similar to some that are already designed. The current VME boards can only go at 15 MB/s (or perhaps 14), and the feeling was to wait until they can take this full data rate.
Some other misc. issues were discussed. We need to monitor the leakage from the transmitter, and to sample the outgoing code for random codes. There's 22 dB of available attenuation before the fiber coming downstairs, which may be enough, but it's not certain. Also, the specifics of sending an arbitrary code to the transmitter are not set.
The "old" CW system and the "Sampler" both seem to need the "old" RI. Can these be switched in a reasonable amount of time in an observing session?
The meeting was called at least in part to discuss what changes to the Downstairs IF/LO were required to do this project, because Eddie Castro and Bob Zimmerman want to close off vague "little changes" to that system, as they get expensive at this point. We discussed using one of the MUX outputs, but there was no concrete discussion of what was to happen and who was to do it, so this goal may not have been met.
We may be being bad frequency-citizens by square-modulating with the code, and could we filter the modulation to keep our bandwidth more contained? It would require a different filtering in the decoder, I think, if it's true. Note that the transmitter has a fairly narrow bandwidth, so it may not be too bad.
By coming down at 260, we bypass some of the niceties of the downstairs IF/LO, like phase adjustment and circular <-> linear converters. Is that a problem?
MD also thought that restricting the Decoder/Sampler hardware to only one of the two in use at a time was a bad idea. If it's to be a software limitation that could be changed if there was a future reason for it, that would be better. How does this question relate to having the Sampler, Decoder, and RI synchronously clocked?
The half-band filters give an extra 1 bit per octave, so do we really need (presumably expensive) 12 bit A-D?
How does the sampling rate affect whether or not the accumulators overflow?
Mike Nolan <nolan@naic.edu>
Last updated 1996 March 12