[Pafgbt] GBT PAF system assumptions
Brian Jeffs
bjeffs at byu.edu
Tue Feb 9 19:15:01 EST 2010
Rick and Karl,
Things are still fluid at this point, but my initial plan for Jonathan
was to build a full bandwidth correlator and beamformer for fewer
antennas channels (4-8). The motivation here is that it should be
easier from this little platform to take the step up to a full 20 or
40 channel system.
The 2 ROACH boards and gateware dedicated to the F engine (data
sampling, FFT, digital receiver) expand to more channels unchanged,
simply by adding more ROACH boards (total of 10). The 2 board X
engine (correlator/beamformer) expands by adding more nested stages of
similarly structured gateware code on the full 10 board engine. Thus
expansion to the full 40 channel system is probably simpler than
expanding a 40 channel narrowband beamformer to broadband. The code
architecture has to change more drastically in the latter case.
Further, the bandwidth and number of antenna channels is really set by
the ADC hardware, and the FPGA clock comes directly from the ADC. We
ordered iADC boards, which give us up to 4 channels per ROACH at 1
Gsamp/s.
We also ordered one 64 channel ADC for a single ROACH to use for
medium bandwidth data acquisition. Building a single board narrowband
beamformer with this is certainly possible, and might be a good
learning experience, but no existing correlator/beamformer gateware
uses this architecture and ADC board. The code would be quite
differently structured from that of the final full bandwidth system,
and we would be starting from scratch. We have access to codes using
the iADC 1 G samp/s boards which is full bandwidth and pretty close to
what we want eventually anyway for the 40 channel system. Thus I
think it will actually be easier for us (Jonathan) to develop a
smaller (4-8 channel) correlator at full bandwidth, than a narrowband
40 channel system.
Of course the problem is that the 40 channel full bandwidth system
requires purchasing many more ROACH boards. A norrowband beamformer
using the 64 channel ADC would get us to the point of real-time
beamforming at a much lower hardware cost, albeit with perhaps a
longer gateware development timeline. If this is a high priority and
if capital equipment funds are limited, we may want to choose this
path anyway.
Brian
On Feb 9, 2010, at 2:48 PM, Rick Fisher wrote:
> I risk suggesting too many options, but a narrowband beamformer is
> worth keeping in the back of our minds. My concern is that we might
> end
> up building a beamformer that's much wider bandwidth than the spectral
> line people need but not wide enough to interest the pulsar folks.
> We'll
> need to iterate on this one a bit.
>
> A replacement for the stream to disk acquisition system sounds
> attractive
> for development purposes.
>
> Rick
>
> On Tue, 9 Feb 2010, Karl Warnick wrote:
>
>> I'll let Brian comment, but I think our near-term goal is a
>> prototype FXB
>> engine using several ROACH boards with standard two channel iADC
>> boards
>> that we would eventually scale to a 40+ channel broadband back end.
>>
>> The narrowband iADC64 board is a side project that could be a
>> possible
>> replacement for our stream to disk acquisition system. If it made
>> sense,
>> we could certainly consider implementing a beamformer on the 64
>> channel
>> system.
>>
>> Karl
>>
>> Rick Fisher wrote:
>>
>> Karl, Brian,
>>
>> Are you thinking of implementing a modest bandwith beamformer on a
>> single
>> ROACH board for starters?
>>
>> Rick
>>
>> On Tue, 9 Feb 2010, Karl Warnick wrote:
>>
>>
>>
>> Rick,
>>
>> We are having an iADC64 board built right now, which has 64 RF
>> inputs and
>> a sample rate of 50 or 64 Msamples/sec per channel. It plugs into one
>> ROACH board. Would this work?
>>
>> Karl
>>
>> Rick Fisher wrote:
>>
>> I was thinking that we might get early HI science by putting a
>> narrow band
>> beamformer in the receiver room, but this may not make sense. It's
>> been
>> pointed out that we'd need at least 10 ROACH boards just to
>> accommodte 38
>> ADCs. I'd be prepared to abandon the idea of any beamformer in the
>> receiver room, but maybe there's a counter-argument. Eliminating an
>> interm solution may very well shorten the time to implement a wider
>> bandwith beamformer.
>>
>> Rick
>>
>> On Tue, 9 Feb 2010, Roger Norrod wrote:
>>
>>
>>
>> I wonder about the wisdom of #5. It sounds like many months of
>> specialized
>> effort to get a limited system in the Receiver Room, and it could
>> be a
>> serious diversion from where we need to concentrate work. The
>> analog links
>> may be considered a diversion too, but at least there's a chance
>> they become
>> part of a long-range solution. If we could manage to get some
>> people to
>> really concentrate for a few months on the analog/digital link
>> comparisons
>> (#7), and leave #5 as a fall-back position, I think it would be good.
>>
>> Roger
>>
>>
>> Rick Fisher wrote:
>>
>>
>>
>> 3. Ultimately we want to digitize the signal from each array element
>> in the front-end box for greatest phase and amplitude stability and
>> lower cable weight of optical fibers. However, the first array will
>> use 38 coaxial cables to carry the element signals into the GBT
>> receiver room. These cables should have sufficiently low loss and
>> outer shield leakage to carry signals frequencies up to 2.3 GHz so
>> that they can transfer either IF or RF signals to the receiver room.
>>
>> 5. The long-range plans are to locate the beamformer electronics in
>> the Jansky laboratory. This offers the greatest room for growth and
>> minimizes the problems of space, weight, and EMI in the GBT receiver
>> room. However, the first beamformer with modest bandwidth will be
>> located in the GBT receiver room so that its implementation is not
>> dependent on transmitting its input signals to the Jansky lab. [Can
>> fewer ROACH boards accommodate 38 lower speed ADCs?]
>>
>> 7. We'll vigorously develop digitizers and digital fiber links that
>> allow signals from the array elements to be transmitted to the Jansky
>> lab on digital fiber links, but we don't want this to be on the
>> critical
>> path to implementing a wider bandwidth beamformer. An alternative
>> solution will be to install commercial 0.9-2.2 GHz analog fiber
>> modems
>> to transmit RF signals directly to the lab. The feasibility of
>> such a
>> solution depends on it being stable enough to be tracked with the
>> phase and amplitude monitoring system. Two modem pairs are in hand,
>> and tests of them on fibers between the GBT and the lab will begin
>> soon. Each modem pair costs about $2K, and a set to handle 38 signal
>> paths will cost about $80K so we need to be certain that it will
>> offer
>> significant scientific pay-off before taking this option. Note that
>> the modems in hand do not work below 900 MHz so they would not
>> transmit
>> low-frequency IF signals from the BYU receiver modules currently
>> under
>> construction. Analog modems that work at lower frequencies are
>> available, but they may be more expensive.
>>
>>
>>
>> _______________________________________________
>> Pafgbt mailing list
>> Pafgbt at listmgr.cv.nrao.edu
>> http://listmgr.cv.nrao.edu/mailman/listinfo/pafgbt
>>
>>
>>
>>
>> --
>> Karl F. Warnick email: warnick at byu.edu
>> Associate Professor Tel: (801)
>> 422-1732
>> Department of Electrical & Computer Engineering FAX: (801)
>> 422-0201
>> Brigham Young University
>> 459 Clyde Building
>> Provo, UT 84602
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> _______________________________________________
>> Pafgbt mailing list
>> Pafgbt at listmgr.cv.nrao.edu
>> http://listmgr.cv.nrao.edu/mailman/listinfo/pafgbt
>>
>>
>>
>>
>> --
>> Karl F. Warnick email: warnick at byu.edu
>> Associate Professor Tel: (801)
>> 422-1732
>> Department of Electrical & Computer Engineering FAX: (801)
>> 422-0201
>> Brigham Young University
>> 459 Clyde Building
>> Provo, UT 84602
>>
>>
>>
>>
>>
>>
> <ATT00001..txt>
More information about the Pafgbt
mailing list