CASPER_Development_at_Caltech - david-macmahon/wiki_convert_test GitHub Wiki

Introduction

We are implementing a digital back end for a novel wide band radio astronomy receiver with an instantaneous sky frequency coverage of either ~0.5-4 GHz or ~2-14 GHz. This reconfigurable back end will be able to perform spectroscopy, polarimetry, and pulsar observations. The radio telescope is the 34m DSS-28 dish at Goldstone and will be operated by the Lewis Center for Educational Research (www.lcer.org) as part of the GAVRT science curriculum for K-12 students.

New!: Picture of the receiver

Contributions

BEE2 Interchip Interconnects

DRAM Notes

  • Hints for interfacing with the BEE2 DRAM: DRAM Notes.

gtkWaveCapture

GAVRT/Caltech Blockset

I have developed a number of useful blocks, some of which will eventually end up in the CASPER library. The blocks are available in the CASPER SVN: 1. Note: these blocks are mostly tested, but certainly not guaranteed. Please email jjoonneess__ggll@@ccaalltteecchh..eedduu if you find any issues.

(Right-click and save to disk and rename to .zip. Wiki will not allow most file types) Media:Gavrtlib_Dec_19_2007.pdf

Notable blocks

Vector Accumulator: The vacc, Serialize, and Vacc Readout Control blocks together make a very clean and functioning vector accumulator. They are reasonably well documented except for Serialize. (Note: the Readout Control does not work as well as hoped in hardware because the PPC misses the BRAM_WE signal that signifies a valid spectrum has been recorded. A latch is needed; this will be updated soon.)

cram and uncram: These blocks make passing several parallel values around much easier.

nCasts: Just what it sounds like. Saves a lot of time. Trick: if you cascade a nCasts with a cram and connect the output to a gpio configured for Unsigned data, you can quickly wire a number of boolean or other signals to LEDs or general gpio without having to create a separate gpio for each signal.

SuperSlice: Simple but invaluable. I find it much more useful than the normal slice block. Great for breaking a software register out into control bits.

Status

' April 14, 2009': Deep capture buffer is working well on BEE2. This design buffers a run-time configurable number of samples (from 2^22 to 2^30) streaming in from a set of iBOBs over XAUI. The ADCs can be clocked at 1 GHz (2Gsps). In the interim implementation, two iBOBs can be connected to each of FPGAs 2 and 4, and the data will be passed to FPGA 3 via interchip links to be transmitted over a 10 GbE port on FPGA 3. In the final implementation, each user FPGA will connect to a pair of iBOBs for a total of eight 1 GHz bands. The data will then stream into the Control FPGA to be sent out via it's two 10 GbE ports. The next step is to run the incoherent trigger system on the iBOBs simultaneously and then pass the trigger over XAUI to the BEE2 logic. This is well underway.

' February 24, 2009': First giant pulses detected from the Crab Pulsar with DSS-28! Media:2009.02.23_-_First_Giant_Pulses_with_DSS-28.pdf This incoherent dedispersion trigger will be used at Arecibo in March to capture giant pulses from the Crab and B1937+21.

' September 02, 2008': Feeds and receiver has been installed on DSS-28. Both feeds have been successfully cooled and tested on the dish. The LFF is currently running and feeding an iBOB spectrometer under remote control. The HFF will be running this week pending an additional cable installation. The tertiary mirror is currently fixed in place, but should be under computer control this week. Dss28vertex.jpg ' July 22, 2008': Working on capturing and decoding packets from BEE2 capture buffer. Started Gigabit_NIC_Notes to document notes on what works well for getting data from BEE2 to PC. DSS27.JPG July 18, 2008: Successfully tested receiver box and feeds at Goldstone. Initial iBOB is set up, ready to take spectra. GAVRT_RSS1.JPG July 15, 2008: The Phase I GAVRT Receiver will be going to Goldstone tomorrow for initial testing. In addition to getting ready for the trip, I have gotten a basic BEE2 DRAM transient buffer working. In doing so, I started to document how to interface with the DRAM: DRAM Notes.

June 30, 2008: The GAVRT Receiver Box is now on the roof at Caltech for final testing before installation on the telescope. In the past few months I have been working on incoherent dedispersion on the iBOB to allow for more sensitive giant pulse triggering. This seems to be working well. Now I am focusing on the BEE2 DRAM buffer to store the time domain data and transfer it to a cluster of PCs for processing. While investigating the effectiveness of my I/Q imbalance compensating spectrometer, I came across a problem of excess leakage from a CW tone. I am still working to understand and if possible, mitigate this problem. So far, I have fully tested these spectrometer designs, each fitting on an iBOB:

  • 4096 channel, 2 tap, single input, up to 128 MHz BW "Hi-Res" mode
  • 1024 channel, 2 tap, single input, up to 1024 MHz BW "Wide-Band" mode
  • 512x2 channel, 2 tap, dual input, up to 2x512 MHz BW "Dual-Pol" mode
  • 512x2 channel, 2 tap, I/Q input, up to 2x512 MHz BW "I/Q correcting" mode
  • 256 channel, 2 tap, single input, up to 1024 MHz BW on-chip dedispersion "Trigger" mode

May 16, 2008: Continuing receiver tests. I added a test vector generator to my spectrometer designs and then spent about a week understanding the deviations from ideal behavior. Conclusions: the CASPER DSP and GAVRT block sets are robust. Keep in mind that while the power spectrum of a signal with a period that divides the FFT length should not depend on time shifts of the signal, rounding errors can make the spectrum fluctuate in the LSBs depending on the time shift.

Apr. 30, 2008: We have been testing the initial receiver hardware in the lab for the past two weeks. The initial hardware consists of a dual-polarization receiver tunable from 0.5 to 18 GHz with up to 2 GHz IF bandwidth, along with noise calibration and fiber optic transmission. Eventually there will be four such receivers in the box. I have begun end-to-end tests from the input of the receiver, through the fiber optic outputs into some basic iBOB based spectrometer designs. Most CASPER work has been on backend software to support the tests.

Apr. 05, 2008: Receiver hardware should be arriving this week. Thus I will be focusing on refining the fast dump spectrometers to use a common back end infrastructure for readout and data capture on the PC. Have been working on a DRAM capture buffer, to allow capturing large amounts of baseband voltage data. Eventually this buffer will be triggered by an incoherent dedispersion system for transient capture. For now it will be manually triggered. I have learned a lot about the DRAM block, and will write a supplementary document to clarify the existing documentation.

Mar. 29, 2008: Made new versions of PFB_FIR and PFB_FIR_REAL blocks that allow runtime configuration of the FIR coefficients. Thus various windows can be tested without having to compile many seperate designs. Email me if you want a copy. Working on fast dump spectrometer to do stability testing when the GAVRT receiver arrives. Hardware IQ correction seems to be working well.

Mar. 19, 2008: Took a short diversion to write a tool, gtkWaveCapture that captures data from the Simulink environment to allow viewing in the gtkWave waveform viewer.

Mar. 15, 2008: Have been busy preparing for candidacy exam. Now that the exam is successfully out of the way, back to work on IQ correction, which seems to be working in hardware. Updated vacc block to allow setting MUX latency to improve timing.

Feb. 20, 2008: Initial documentation of iBOB to BEE2 link completed: Media:IBOB-BEE2 XAUI Overview.pdf (actual PDF, do not rename). Updated Model files: Media:Xaui align Release Feb20 2008.pdf (rename .zip)

Feb. 19, 2008: Beginning to document iBOB to BEE2 connection. Have spent the past couple of weeks working on an IQ imbalance correction scheme which seems to be working well. Report to come soon. Spent a while trying to figure out why designs with DRAM were failing to meet timing with strange errors. Turned out that I had not applied the 7.0.4 patch to ISE on the build machine. Preview of iBOB-BEE2 XAUI link: Media:xaui_align_PreRelease_Feb19_2008.pdf (rename .zip)

Jan. 24, 2008: Alignment between iBOB and BEE2 seems to be working well at 250 MHz (actually managed to overclock to 300+ MHz, so it seems stable). Saw no glitches in 13 hours of continuous operation. The 'traditional' approach of generating clk/2 on an iBOB SMA output and connecting this to the usr_clk input SMA on the BEE2 and running the BEE2 design off of usr_clk2x did not work at 250 MHz. Instead, I had to generate clk/4 and hack the XPS base system to lock the BEE2 DCM to usr_clk * 4. This seems to work well. All of my CASPER designs will soon be available by anonymous FTP.

Jan. 15, 2008: Continuing work on aligning data from iBOB to BEE2. Original design worked OK but fails to compile at 250 MHz. After analyzing timing reports and trying various pipelining locations, it appears there will need to be a latency of at least 3 cycles between an rx_get being issued to the XAUI block and receiving OOB data, as this feedback is a critical path. Currently figuring out how to get the design to work with this much latency.

Dec. 23, 2007: Traced down trouble with iBOB lwIP refusing connections after a reset connection. Modified version of /drivers/xps_lwip/lwiputil.c:Mod_lwiputil.c

Dec. 19, 2007: Wiki page created. Updated FAQ with information about using the DCMs. Currently working on streaming data from iBOBs to the BEE2 over XAUI in synchrony. This seems to be working well. Current design: Media:Xaui_align_Dec_19_2007.pdf (rename .zip)

People Involved

The overall GAVRT DSS-28 project involves many people, however the people most involved with the back-end development are:

Glenn Jones: CASPER hardware development

Robert Karl: Software development

Tom Kuiper: Science requirements

Sandy Weinreb: Overall system design

Acknowledgments

We are very grateful of Xilinx's generous donations enabling development of this project.