Changes between Version 1 and Version 2 of QuadPulserEarlyThoughts


Ignore:
Timestamp:
May 30, 2017, 12:36:59 PM (5 years ago)
Author:
Eric Hazen
Comment:

--

Legend:

Unmodified
Added
Removed
Modified
  • QuadPulserEarlyThoughts

    v1 v2  
    1 == Some Thoughts ==
    2 
    3 32 channels of 1GSPS is probably too expensive.  There is a 12-bit part from TI (ADC12D500)
    4 which is a two-channel 500MSPS part which can interleave to get 1GSPS.  They're about $280 each.
    5 
    6 || '''Bits''' || '''MSPS''' || '''Cost/ch''' || '''PN/Notes''' ||
    7 || 12 || 1000|| $280|| TI ADC12D500 (two 500MSPS channels interleaved) ||
    8 || 12 || 500|| $168|| Intersil KAD5512P ||
    9 || 12 || 250|| $44|| ADS4229 (two 250MSPS) ||
    10 || 12 || 125|| $21|| LTC 2145 (2 ch) ||
    11 || 12 || 105|| $23|| ** ADS6424 (4 ch, 12 bit Serial LVDS) ||
    12 || 12 || 105|| $22|| AD9633 (4 ch, 12 bit serial LVDS) ||
    13 || 12 || 105|| $24|| ADC12DS105 (2 ch, 12 bit Serial LVDS) ||
    14 
    15 Observations about signals:
    16 
    17 * "Normal" case is three short pulses (few uS long at most) at 0, 50us and around 1ms
    18 * Stray discharges are typically quite short (few 100ns), amplitude similar to normal pulses (<1V)
    19 * Sparks are characterized by large amplitude and significant width.  There can be a train of spark pulses spread out over substantial fractions of the 1ms pulse period.
    20 * Beam timing has ~ 10ms between pulses where the should be no signals
    21 
    22 One possible proposal:
    23 
    24 * Digitize 32 channels at 125 MSPS (single gain scale), full scale +/-5 or +/-10V
    25 * Artix XC7A35T has 1.8Mb block RAM.  These are around $35 each.  Propose to use one per Quad pulser (total of 4 in system).  This could store ~ 150us of waveform per channel on all channels, seems fine.
    26 * Transferring entire 1.8Mb in 10ms elsewhere requires a 180Mbps link (only)
    27 * There are 4 MGT transceivers on the !PicoZed 7015 or 7030.
    28 * Store zero-suppressed waveforms with length limit, when limit exceeded record just width or number of zero-crossings or other info to characterize the pulse.  Possibly the absolute-value envelope with low resolution but calculated in the FPGA
    29 * Readout via MGT point-to-point to !PicoZed, thence over Ethernet
    30 * Worst case bandwidth:  1.8Mb * 4 pulsers * 12Hz = 86Mbps (10 MBytes/sec or 2.7 MHz 32-bit words)