Ismaël Cognard - - Instrumentation -

Coherent dedispersion pulsar instrumentation

Through a collaboration with Pr D.C.Backer (University of California, Berkeley sadly prematurely deceased in July 2010) and Paul Demorest, a coherent pulsar dedispersor was built and installed at the Nançay radiotelescope end of 2004.

Pulsed signal coming from radio pulsars makes us sensitive to the dispersion produced on electromagnetic waves by free electrons present in the interstellar medium. Since the beginning of pulsar observations several decades ago, different methods to overcome the limitation introduced by this effect were developped. The first instrumentations were filterbanks without any dedispersion done within the individual channels. As we want to improve sensitivity with larger bandwidths, the intrinsic pulse profile is then unfortunately broadened.

With present computing power available, it has been shown possible to do the dispersion removal (or dedispersion) directly on the voltage received by the telescope (ths is called coherent dedispersion). This is done before the detection of the signal (wave form) and also provide full polarization characteristics.

Several versions of the Berkeley-Orléans-Nançay (BON) instrumentation were built and used over the years. This description goes back in time as you scroll down the page.
An animation of all the observed pulsars week per week on the sky (45184 obs between 2004 and 2016).

BON512 or NUPPI with GPUs (2011-)
This version is a major upgrade of the coherent dedispersion pulsars instrumentation with a 512MHz bandwidth capability.
Based on a ROACH board provided by CASPER (University of California, Berkeley), 4 computers each hosting 2 GPUs (GeForce GTX 280) are able to handle a 16Gb/s data stream. Two computers (sr2 and sr3) are able to send the voltage data to host 'ripata' where raw voltage are used for the LEAP project

BON128 with GPUs (2008-)
Compared to the previous version of the instrumentation (see below), we juste replaced the 4 dataservers and the 70 nodes cluster by just TWO computers each having two of the four high speed DMA parallel boards and two high end graphics cards (Graphical Processing Units or GPUs). As shown below, the two GPUs plugged into each computers were water cooled in order to increase their lifetime (temperature during the real-time processing is around 40C instead of 72C with an air fan).
With the GPUs, we were able to dedisperse a bandwidth of 128MHz for all the pulsars no matter the value of the Dispersion Measure (DM). The previously used CPUs had limitations on FFT lengths and hence on the DM.
Architecture of BON/GPUs
From left to right : picture of the BON128 with the four water-cooled GPUs, scshematics of the data path and a daily status sent by mail every day.

BON128 with CPUs (2004-2008)
An FPGA board (SerendipV, from ATA -Allen Telescope Array-) is used to digitize a 128MHz bandwidth signal over 2 complex polarizations (4 streams over 8 bits). A Polyphase Filter Bank (PFB) running within the FPGA is producing 32 channels of 4MHz bandwidth each. The 32 x4 data streams (4Gb/s or 512MB/s) are transfered through dedicated cables and interfaces to 4 dataserveurs (bi-Xeon) with high speed DMA parallel board (EDT Inc.). Those 4 dataservers feed a 70 nodes cluster (bi-Athlon 1.2GHz) through a fiber Gigabit switch and 1GB/s bonded interfaces. Dedispersion softwares were developped in C language under Linux with socket level code.
A significant fraction of the time, the cluster was only able to dedisperse a bandwidth of 64MHz.

spectrometer ATA 128MHz 2-cplx-polar-8bits Architecture of BON/CPUs
Spectrometer designed in collaboration with ATA at University of California, Berkeley. Two complex polarizations sampled over a 128MHz bandwidth.
Panaramic view of the BON instrumentation
Panoramic view of the BON instrumentation (master with a scientist working, SerendipV within the old NBPP crate, the 70 nodes cluster within the 3 higher crates).

Dr Ismaël Cognard
LPC2E / CNRS - UMR 7328
  3A, Av de la Recherche Scientifique
        Email : icognard at