Modification on vita_tx_chain.v

GNU Radio
#1

Hi all,

I have done some modification on vita_tx_chain.v. I’m trying to get the
data (I&Q) from another FPGA board using debug pins. When I receive the
data, I write them into FIFO. The data are retrieved from FIFO when
there is an strobe_tx signal.

And same sequence of data send using gnuradio and modified version(from
another FPGA board). And look at the spectrum at RX. With modified
version, the noise floor is increased about 20dB. I have attached the
modified verilog file and two spectrum pictures. You can see the
modification on vita_tx_chain.v file. Can anyone suggest me what is
wrong with my modification?

PN

#2

Ok, I glanced at your code quickly, here are my thoughts (some of which
you may already do in files I haven’t seen):

  1. You are using tx_io[15] as a potentially highspeed I/O. This signal
    travels via the radio daughter cards and could easily lead to EMI noise
    issues. You could consider multiplexing I &Q to get more usable pins on
    the debug header.
  2. Your “fifo_not_full” signal may be acted on too late on your external
    FPGA board, and an I/Q data pair may be lost due to FIFO overrun as the
    FIFO becomes full. I suggest using a FIFO “nearly full” signal, and
    adequate pipelining to ensure that AC timing goals are met in passing
    the signal from the USRP2 to the external FPGA board.
  3. Not all bits of the I/Q data bus may be received in the correct clock
    cycle.
    You are using (presumably) a rising clock edge to clock data off chip
    and a falling edge to clock data onto chip. If I presume that your clock
    is 25MHz ( the fastest possible on the USRP2) then that leaves 20nS
    setup and hold margin in an ideal world…it is likely far less. The
    following points will help improve timing.
  • Register all signals passing on and off FPGA’s with registers placed
    in the I/O rather than then core.
  • For more deterministic timing, drive clocks off FPGA’s using a DDR
    registers rather than passing the clock signal directly to the pin. See
    “RAM_CLK_i1” in u2_rev3.v for an example of how this can be done on
    Spartan3.
  • Use IBUFG cells (or equivalent) when bringing a clock into an FPGA. If
    possible use a DLL in “system synchronous” or “source synchronous” mode
    (as appropriate) to give tighter timing control. (You may find that the
    USRP2 has already used many of the available ones.)
  • Use timing constraints in the UCF file for all the signals you are
    passing off chip to achieve AC timing goals. Check the timing results
    after the FPGA image is compiled to ensure timing goals were met.
  1. Try to bring both boards as close to together as possible using a
    minimum length of cabling, ideally using a board to board molex
    connection. Do not use excessive drive strengths are slew rates beyond
    that needed to achieve timing to minimize EMI.
#3

Hi,

  1.  I removed tx_io[15] from my design

  2.  Change fifo_not_full signal to fifo_almost_full

  3.  I tried to follow most of your suggestions to achieve better

timing

  1.  I use minimum cable length.

Now I’m observing strange behavior. I grounded Virtex board and USRP2
board together. Initially I use ground pin at J503 (Test CLK) on USRP2
mother board and observe the spectrum at RX. I see the noise floor of
spectrum goes up and down about 20dB.

And unknowingly, I change ground pin to J101 at RFX2400 daughter board.
Then the observed spectrum at RX is correct. No more noise floor goes up
and down. Can someone please explain what is happening here? Thank you.

PN

From: Ian B. [mailto:[email protected]]
Sent: Tuesday, December 20, 2011 4:16 AM
To: Phone Naing MYINT
Cc: [email protected]; [email protected]
Subject: Re: [USRP-users] Modification on vita_tx_chain.v

Ok, I glanced at your code quickly, here are my thoughts (some of which
you may already do in files I haven’t seen):

  1. You are using tx_io[15] as a potentially highspeed I/O. This signal
    travels via the radio daughter cards and could easily lead to EMI noise
    issues. You could consider multiplexing I &Q to get more usable pins on
    the debug header.
  2. Your “fifo_not_full” signal may be acted on too late on your external
    FPGA board, and an I/Q data pair may be lost due to FIFO overrun as the
    FIFO becomes full. I suggest using a FIFO “nearly full” signal, and
    adequate pipelining to ensure that AC timing goals are met in passing
    the signal from the USRP2 to the external FPGA board.
  3. Not all bits of the I/Q data bus may be received in the correct clock
    cycle.
    You are using (presumably) a rising clock edge to clock data off chip
    and a falling edge to clock data onto chip. If I presume that your clock
    is 25MHz ( the fastest possible on the USRP2) then that leaves 20nS
    setup and hold margin in an ideal world…it is likely far less. The
    following points will help improve timing.
  • Register all signals passing on and off FPGA’s with registers placed
    in the I/O rather than then core.
  • For more deterministic timing, drive clocks off FPGA’s using a DDR
    registers rather than passing the clock signal directly to the pin. See
    “RAM_CLK_i1” in u2_rev3.v for an example of how this can be done on
    Spartan3.
  • Use IBUFG cells (or equivalent) when bringing a clock into an FPGA. If
    possible use a DLL in “system synchronous” or “source synchronous” mode
    (as appropriate) to give tighter timing control. (You may find that the
    USRP2 has already used many of the available ones.)
  • Use timing constraints in the UCF file for all the signals you are
    passing off chip to achieve AC timing goals. Check the timing results
    after the FPGA image is compiled to ensure timing goals were met.
  1. Try to bring both boards as close to together as possible using a
    minimum length of cabling, ideally using a board to board molex
    connection. Do not use excessive drive strengths are slew rates beyond
    that needed to achieve timing to minimize EMI.

On Dec 19, 2011, at 1:43 AM, Phone Naing MYINT wrote:

Hi all,

I have done some modification on vita_tx_chain.v. I’m trying to get the
data (I&Q) from another FPGA board using debug pins. When I receive the
data, I write them into FIFO. The data are retrieved from FIFO when
there is an strobe_tx signal.

And same sequence of data send using gnuradio and modified version(from
another FPGA board). And look at the spectrum at RX. With modified
version, the noise floor is increased about 20dB. I have attached the
modified verilog file and two spectrum pictures. You can see the
modification on vita_tx_chain.v file. Can anyone suggest me what is
wrong with my modification?

PN
_______________________________________________
USRP-users mailing list
[email protected]mailto:[email protected]
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com