CASPER Workshop 2025

8 Sept 2025, 06:45 → 12 Sept 2025, 17:00 Europe/Amsterdam

Auditorium (ASTRON Netherlands Institute for Radio Astronomy)

CASPER Workshop 2025 | ASTRON

The CASPER workshop is a semi-annual workshop where FPGA, GPU, and general heterogeneous system programmers get together to discuss new instruments in radio astronomy, as well as the tools and libraries for developing and manipulating these instruments.

The conference will take place September 8-12, 2025, at ASTRON Netherlands Institute for Radio Astronomy, The Netherlands.

This workshop is best attended in person. Though remote attendance is possible, in person talks will be scheduled on the program /timetable first (unless the remote presenter has limited resources and is eligible for discounted registration - see registration form).

Monday 8 September

1 Introduction/Announcements

2 Introduction ASTRON, Research & Development

Speaker: Roelien Attema-van Waas (ASTRON)

3 CASPER: Past, Present and Future

Designing, building, and upgrading real-time, in-situ, digital signal processing (DSP) instrumentation poses distinct challenges. These include the manipulation of data at several terabits per second (or higher) though a parallel computing environment. Such challenges can be solved by the development of specialized computing hardware, employing processing architectures that depart from those of traditional, general-purpose systems. Such specialization has traditionally meant that these systems have been developed independently for each project, with substantial investment and reinvestment of time and money. The Collaboration for Astronomical Signal Processing and Electronics Research (CASPER) works toward providing solutions to the processing requirements of DSP instrumentation, and reducing both the development time, and cost, of such systems.

Speaker: Mitch Burnett

4 Is Anybody Out There ?: SETI in the BC and AC Epochs

I'll review SETI Instrumentation BC and AC (in the eras BeforeCasper and AfterCasper); I'll discuss what’s new in the search for extraterrestrial civilizations, concentrating on the CASPER based PANOSETI telescope arrays at Lick and Palomar Observatories.
The Panoramic Search for Extraterrestrial Intelligence (PANOSETI) experiment searches largely unexplored parameter space, observing a large field of view simultaneously for nanosecond to second time scale transients at visible and near-IR wavelengths. PANOSETI will also be utilized to search for dark matter annihilation and explore sources of ultra-high-energy gamma-rays.

Speaker: Dan Werthimer (University of California, Berkeley)

09:55 Break (Location: Central Hall 2012)

5 SKA-Mid Signal Processing Overview

SKA-Mid Signal Processing Overview

Jayashree Roy
Signal Processing Engineer
Square Kilometre Array Observatory (SKAO)
Email id: Jayashree.roy@skao.int

Abstract:
Introduction: The Square Kilometre Array (SKA) project is an international effort to build the world’s largest radio telescope with an unprecedented sensitivity because of its huge collecting area. SKA will be a distributed observatory with radio telescopes at two different radio-quiet sites, SKA-MID in South Africa and SKA-LOW in Australia. The first phase of SKA-MID telescope will consist of ~150-kilometre diameter array of 197 offset Gregorian antennas (dishes). It will be a mixed array of 133 15-m SKA (phase 1) dishes and 64 13.9-m diameter dishes from the MeerKAT telescope, with total ~33,000 m2 collecting area.

Figure 1: SKA science goals, observation categories, and electromagnetic spectrum coverage. SKA-Low frequency range is shown in Green and SKA-Mid in Red (at the top of the figure).

Figure 2: Full SKA-Mid array configuration (Left); The inner 2 x 2 km region of the array with blue dots are the MeerKAT dishes (Middle). Simulated view of the inner part of the array (Right).

SKA Mid signal chain overview

The SKA (phase1) - Mid telescope will be able to achieve high sensitivity over a wide frequency range starting from 350 MHz to 15.4 GHz, in 6 selectable bands. Figure 3 shows the SKA1-Mid signal chain sub-systems. This summary paper describes an overview and progress on design of the Single Pixel Feed Receiver (SPFRx), and Correlator & Beamformer (CBF) system of SKA-Mid telescope.

Figure 3: Major sub-systems of SKA-Mid: The black or red lines indicate signal or data flow; the dash-dot lines for timing and synchronisation signals; dashed green lines for control and monitoring flow

Single Pixel Feed Receiver (SPFRx):

The SPFRx system which receives the RF signal from feed, implements conditioning, and performs digitization and packetization of the input signal before forwarding to the Correlator & Beamformer system. The SPFRx is a key performance-determining sub-system, because of its impact on telescope sensitivity, through signal-to-noise ratio (efficiency) and linearity of analog and digital components.

Figure 4: SPFRx simplified block diagram (left), Summary of SPFRx digitization parameters (right), frequency offset index ‘k’ ranges from 1 – 2222

Synchronization: The entire SKA-Mid telescope is synchronized by distributing phase-locked reference signals to each antenna, it also implements the “Sample Clock Frequency Offset (SCFO)” scheme to supress digitiser-generated self-interference and out-of-band interference by employing different sample clock frequencies at each dish. Subsequently, these sampled sequences are digitally re-sampled to a common sample rate prior to correlation/beamforming.
SPFRx system consists of two EMI shielded enclosures at the dish indexer for Band123 and Band45, and one shielded unit in the pedestal of the dish. The Optical Digital Link (ODL) architecture used for the receiver offers the best compromise between RF performance and EMI suppression by distributing the system components between the feed indexer and the shielded compartment in the dish pedestal, few major components of this module are:
• High performance ADC – TI ADC12DJ5200RF.
• Optical transceiver for transmission of sampled data, and monitor & control signal
• High-speed SERDES transceiver to perform serial-to-parallel and parallel-to-serial data conversion.
• TALON-DX Board with Intel Stratix-10 SX280 FPGA + HPS used at the receiver pedestal unit for Digital processing and packetizing.

Figure 5: SPFRx123 analog signal chain (left), SPFRx123 digital signal chain at the indexer (right)

Correlator & Beamformer (CBF) system:

The Central Signal Processor (CSP) is responsible for accepting data from the receptors, generating visibilities, generating VLBI tied-array voltage beams, performing Pulsar Search and Pulsar Timing processing in real-time, and forwarding the resulting data to the Science Data Processor (SDP).

Figure 6: Processing flow of SKA Mid CSP signal chain

CSP Capabilities: When performing the correlation, CBF calculates full-polarisation cross-correlation spectra with a configurable number of channels per polarization (up to ~380K) for every pair of receptors in each sub-array, each sub-array may be observing in a different Band, and spectral zoom could be employed to provide finer spectral resolution over a range of narrower bandwidths.
The Pulsar Search Engine (PSS) accepts (up to) 1500 Pulsar Search beams from CBF and searches each individually for pulsars and transient sources over a range of dispersion measures (DM), accelerations, and periods, at ~40 to ~800 MHz/beam, channelized into 107,520 Hz channels.
The Pulsar Timing Engine (PST) times up to 16 known pulsars, one in each of the up to 16 coherent beams produced by the CBF beamformer. The PST performs high-fidelity, high precision timing observations of known pulsars. The primary task is phase-coherent dispersion removal, a computationally intensive algorithm that requires performing many large Fast Fourier Transform operations in real time.

The CBF system within any sub-array, can perform correlation and beamforming independently and simultaneously. CBF adopts the “Frequency slice architecture” approach, for better parallel development and integration, and lesser hardware required for full capability - which Splits functionality into two parts:
- Band-dependant processing: generates “VCC Products” – Frequency Slices and Search Window Channels and distributes VCC Products to FSPs for further processing
- Band-invariant processing: - made up of several independent Frequency Slice Processors (FSP), having Four “Function Modes” – (1) Correlation, (2) Pulsar Search Beamforming, (3) Pulsar Timing Beamforming, (4) Very Long Baseline Interferometry

Figure 7: Simplified functional breakdown of Mid.CBF into VCC and FSP processing (Left), An illustration of RF band is split into frequency slices, where each slice is processed independently of the others (Right).

CBF Technology: Some of the key components of CBF hardware:

• Bitware IA-860m FPGA PCIe Card
• Intel Agilex M-Series AGM039 FPGA, 3 x QSFP-DD cages for 3 x 400GbE links (Hard MACs), x16 Gen5 PCIe interface (Hard IP), Board management controller (via USB)
• Bittware Terabox 1501B FGPA Server
• 1U host server with support for up to 4 x IA-860M cards, AMD EPYC 9354 CPU (3.2 GHz, 32C/64T), 512GB (VCC) or 128GB (FSP) DDR5-4800 (8 banks), 2 x 1TB M.2 SSD, Redundant Power Supplies, 16-lane PCIe 5.0 bus for each FPGA (512 Gbps), Mellanox ConnectX-7 200GbE NIC
• Edgecore network switch
• Cisco Nexus 9332D-GX2B network switch, 32 x 400 GbE QSFP-DD ports, 25.6 Tbps / 4.17B packets per second.
CBF’s reliability and robustness is increased using measures such - Selection of high reliability components, Design to have the fewest possible variations of equipment types and configurations, advanced diagnostics, use of components having plug and play modularity for rapid part replacement, etc.

Figure 8: Annotated Correlation (for imaging) end-to-end signal flow diagram

Conclusion: A brief overview of the SKA Mid signal chain architecture of the Square Kilometre Array radio-telescope has been provided. The latest technology developments on the design and prototyping of the Single Pixel Feed Receiver (SPFRx) and Correlator & Beamformer (CBF) system have been discussed. Prototype design and extensive qualification tests are being carried out at present. The consolidated effort from all the teams involved demonstrates that the SKA project is certainly on the path to success.
We would like to acknowledge National Research Council Canada (NRC), MDA Space (Canada), Qamcom Research and Technology (QRT), University of Western Australia (UWA), South African Radio Astronomy Observatory (SARAO) for their invaluable contribution and collaborative efforts to these works.

Speaker: Jayashree Roy (SKAO)

6 Processing the Square Kilometre Array data deluge

The SKA Low and Mid telescopes are next-generation flagship radio astronomical facilities that will revolutionize radio astronomy for decades. The data rates anticipated from the telescopes require that signal processing and data reduction are integrated into the telescope design and operational model. In this talk, I will discuss details of the SKA telescopes and their specifications, highlighting the expected data rates, supported data products, signal processing solutions, and challenges. I will also briefly discuss our operational model and provide details of how and when the community might expect to access the SKA.

Speaker: Danny Price (SKAO)

7 DSA-2000 Engineering update

The Deep Synoptic Array 2000 (DSA-2000) is a 2000-element radio telescope optimized for wide-band surveys from 700 MHz to 2 GHz. Designed as a next-generation survey instrument, the DSA-2000 will deliver transformative science in multi-messenger astronomy, cosmic evolution, pulsar timing, and the dynamic radio sky. Enabled by cryogen-free wideband receivers, a “radio camera” imaging backend, and a chronoscope transient backend, the array will achieve unprecedented survey speed and excellent sensitivity, detecting over a billion radio sources, millions of galaxies in HI, and tens of thousands of fast radio bursts and pulsars. Recent progress includes site development in Spring Valley, Nevada, key antenna and receiver design optimizations, and successful demonstration on a test array at Owens Valley. With construction planned to begin in 2026 and first light in 2027, the DSA-2000 is on track to become a world-leading facility for fully public, commensal radio sky surveys.

Speaker: Mr Francois Kapp (Caltech)

11:30 Lunch

8 LOFAR Status Update

The LOw Frequency ARray (LOFAR) is currently undergoing a major upgrade designed to substantially enhance its scientific capabilities, particularly at the lowest accessible frequencies. This comprehensive effort encompasses nearly the entire signal chain, from the receivers in the field to the central processing facilities. The presentation will provide an overview of the upgraded signal path, outlining the key improvements and the current status of the deployment. Plans for the near future will also be discussed, including the anticipated performance gains and the new opportunities for low-frequency radio astronomy that the upgraded LOFAR will enable.

Speaker: Andre Gunst

9 Arthropod

Speaker: Aditya Parthasarathy

10 Scalability and computational efficiency in self-calibration pipelines

With the development and deployment of telescopes like LOFAR2.0, SKA and ngVLA, radio interferometry is entering the Big Data Era, in which a single observation will produce tens of TB of data. To manage this data deluge within an acceptable and sustainable costing envelope for computing hardware and energy consumption, at least an order of magnitude improvement in computational performance is needed. Data processing pipelines need to scale well across multiple nodes and use processing components optimised for energy efficiency, including optimisation by exploiting accelerators.
In this talk, I will introduce the magnitude of these challenges using experience from the International LOFAR telescope, showing why current practices are unsustainable. I will then discuss how these challenges are currently being addressed in the development of the self-calibration pipelines for LOFAR and SKA. The performance of this pipeline is benchmarked regularly to identify processing bottlenecks and define steps to remedy them. These tests indicate that visibility predictions, calibration solvers and imaging are primary targets for performance improvements. One of the key features of the pipeline architecture is modularity, which should enable us to replace individual processing components by newly developed ones, thereby making these pipelines open to new processing approaches explored by the community, of which I will give a number of examples and indicate steps taken to facilitate integration of such community contributions into these pipelines.

Speaker: Stefan Wijnholds

Breakout Session Topic Discussion

14:00 Break

11 Breakout Session

12 Breakout Session

13 Tutorial

Collect your bicycle (parking lot)

16:15 Drinks

Tuesday 9 September

14 Introduction/Announcements

15 High Performance GPU building blocks for radio astronomy and beyond

Modern GPUs with tensor cores deliver exceptional throughput and energy efficiency for digital signal processing. Due to new hardware and software innovations, the computational performance of GPU correlators has improved two orders of magnitude during the past decade. However, to take full advantage from their increasing computing power, GPU systems have to handle Ethernet packets at proportionally increasing data rates, and this is a challenge, as I/O has traditionally been the GPU's Achilles heel.

In this session, we will present "RADIOBLOCKS": a collection of reusable GPU building blocks for radio telescopes that provide high performance and high I/O data rates. We will focus on the following radio blocks:
- the Tensor-Core Correlator (TCC); an efficient GPU library for computing correlations using tensor cores for complex matrix operations. We include a brief update on new features and improved performance.
- the Tensor-Core Beam Former (TCBF); a new GPU beam forming library for radio astronomy and medical ultrasound imaging, built on the ccglib library for fast complex matrix multiplications.
- a new GPU filter library, that implements a polyphase filter bank with fine delay compensation, bandpass correction, and a transpose for seamless integration with the TCC and TCBF. It uses the new cuFFTDx library to embed FFTs in a (large) GPU kernel, significantly reducing memory bandwidth use and improving performance compared to a cuFFT-based filter.
- High-speed Ethernet packet handling; we demonstrate a method, based on the Data Plane Development Kit and its GPUdev extension, with which a Grace Hopper GPU receives, filters, and correlates 1.2 Tb/s of Ethernet packets.

For each building block we will present measured performance and energy efficiency, showing how modern GPUs can address both computational and I/O challenges for future high data rate instruments.

Speakers: Dr Bram Veenboer (ASTRON), John W. Romein (ASTRON (Netherlands Institute for Radio Astronomy))

09:55 Break (Location: Central Hall 2012)

16 Low-pass Filtering in the MeerKAT Extension Narrowband F-Engine

We present our approach to narrowband channelisation in the next-generation GPU-based correlator for the MeerKAT Extension. Unlike the existing MeerKAT narrowband channeliser, the bandpass response is almost completely flat across the output bandwidth. This is achieved by a hybrid approach to reducing bandwidth: a time-domain FIR filter does an initial bandwidth reduction, while the remainder is handled in the frequency domain by discarding channels.

Intuitively this would seem to be more expensive. However, we show that the reduced constraints on the low-pass filter response allow for the number of taps to be substantially decreased.

Speaker: Bruce Merry (South African Radio Astronomy Observatory)

17 A high-time resolution GPU Imager for FRB searches at low radio frequencies

Fast Radio Bursts (FRBs) are mysterious extremely bright millisecond radio pulses of predominately extra-galactic origin. Despite hundreds of FRBs discovered at frequencies above 400 MHz, only a handful have been detected below 400 MHz. One of the reasons for this is the computational complexity of FRB searches at these frequencies. Firstly, dispersion of radio signals causes time delays of the order of tens of seconds, which requires large amounts of computer memory (even of the order of a Tb). Secondly, large fields of views (FoVs) of low-frequency aperture arrays require formation and searchers over multiple beams (or image pixels). We have developed a new high-time resolution imaging pipeline BLINK, which enables high-time resolution imaging and efficient FRB searches with low-frequency radio telescopes.The BLINK pipeline has been implemented in C++ and performs all the operations on Graphical Processing Units (GPUs). For our fast-imaging use case, the BLINK pipeline performs over 3500x faster than the WSCLEAN imaging package. Once the full pipeline is completed, it will be used to process tens of Petabytes of high-time resolution data from the Murchison Widefield Array telescope. We intend to make the pipeline applicable for future FRB, pulsar and transient searches with other radio telescopes.

Speaker: Danny Price (SKAO)

18 Towards fibre-connected radio telescopes: the Ethernet Receiver concept at ASTRON

At ASTRON, the Netherlands Institute for Radio Astronomy, we are developing the next generation of radio telescopes. Traditionally, coaxial cables have been used to transport signals from the antenna to the processing facility. For future systems, we aim to replace these with optical fibre interconnects. Within ASTRON, this concept is referred to as the Ethernet Receiver. This transition requires placing a synchronised ADC close to the antenna, with both the clock reference and digitised data transmitted via fibre.
We are investigating this concept through two projects: MID4Automotive and Future Network Services (FNS). Key challenges include shielding sensitive analogue circuitry from digital noise and achieving high-precision synchronisation. Within MID4Automotive, we are developing a prototype capable of handling up to 100 MHz bandwidth, aimed at a potential LOFAR upgrade. Synchronisation is achieved using White Rabbit technology. In the FNS project, we are designing an Ethernet receiver based on an RFSoC, supporting bandwidths of up to 2.5 GHz.
In this presentation, we will outline our design choices for the ADC, clock distribution, and the integration of White Rabbit for clock synchronisation. We will share initial results from the 100 MHz Ethernet receiver and present measurement outcomes from the RFSoC4x2 evaluation. Finally, we will discuss our roadmap for the next phase of development.

Speaker: Gijs Schoonderbeek

11:30 Lunch

19 AI as a Detector – Real Time AI Fast Radio Burst Search

The Universe’s radio sky is rich with transient phenomena, but conventional surveys often depend on classical filtering and dedispersion techniques, potentially overlooking novel signals. We present a pipeline that bypasses these traditional steps by streaming RF data directly from telescope digitizers into GPU-accelerated AI models, enabling real-time detection directly on raw spectrograms. This end-to-end AI workflow streamlines model training, curation, and deployment, achieving both high performance and scalability. By eliminating classical pre-processing, our approach has the potential to simplify development, enhance discovery, and facilitate scalability.

Speaker: Adam Thompson (NVIDIA)

20 The Hat Creek Observatory: A Facility for Technology Development in Radio Astronomy and High-Performance Digital Signal Processing

The Hat Creek Radio Observatory (HCRO) in Northern California has a long history of radio astronomy innovation. It is home to the first telescope that was especially designed for the search for technosignatures and pioneered the way for modern "Large Number of Small Dish" (LNSD) interferometers. The Allen Telescope Array (ATA) consists of 43 6.1-meter antennas and operates from 1GHz to 11GHz with upgrades underway that will increase the operational bandwidth to 14.5GHz.
We are currently pursuing two major development efforts aimed at advancing the ATA's capabilities. The first focuses on the development of a new GPU-based backend leveraging NVIDIA's Holoscan. By combining the Breakthrough Listen Accelerated DSP Engine (BLADE) with Holoscan SDK, we are building a next-generation data processing pipeline that leverages the full power of the underlying hardware by using Remote Direct Access Memory (RDMA) to facilitate data transfers between the network interface, graphics card and high-speed NVMe storage.
The second development effort focuses on direct digitization, initially trialing Mercury's 64 Gsps EB-RFS1140 digitizer evaluation board. We are aiming to develop a system capable of directly digitizing the telescope's RF signal at each antenna, providing dual-polarization complex digital data over a 12.8 GHz instantaneous bandwidth at 8 - 14 bits.

Speaker: Dr Alexander Pollak (SETI Institute)

21 CASPER Toolflow Walkthrough

Speaker: Jack Hickish

13:45 Break

22 Breakout Session

Breakout Session

23 Tutorial

16:15 Drinks

Wednesday 10 September

24 Introduction/Announcements

25 Update on the Submillimeter Array

I will present an update of the current operational status and future plans of the CASPER-enabled Submillimeter Array SMA. The telescope team has plans for continued operations and for upgrades on both short and long term time scales. A strategic plan in the coming decade focuses on the future of VLBI including the advancement of the EHT and space missions. Non VLBI science focuses include transient follow up and solar astronomy. All of the science goals benefit from expanded bandwidth and simultaneous operation of multiple receiver bands, enabling frequency-phase-transfer (FPT) for improved phase calibration at higher frequencies.

Speaker: Jonathan Weintroub (Smithsonian Astrophysical Observatory)

26 The Establishment of an RFSoC development system for prototyping and system verification

Radio Frequency System on Chip (RFSoC) is a next generation programmable hardware platform (FPGA based) bringing together wide bandwidth, high frequency direct sampling, wideband DSP and high-throughput data interconnect. RFSoC hardware is well positioned to power next-generation RA instrumentation, however, a current shortfall is the lack of means to enclose and deploy such hardware. In addition, suitable supporting hardware is required to enable both datacenter deployment as well as ruggedised field deployable instrumentation. SARAO DSP Hardware team has identified this as an opportunity to lead in the design, development, and testing of a suitable solution to enable the rollout of RFSoC based hardware for compact, next generation RA instrumentation. This presentation will highlight the design and development of a suitable enclosure with supporting hardware, and focus on key technical considerations with intention to gather community support and interest inviting feedback to foster new ideas for development consideration.

Speaker: Andrew van der Byl (SARAO)

27 Digital Backend Development for High- and Low-Frequency Radio Telescopes

This work presents recent advancements in digital backend development for both high-frequency and low-frequency radio astronomy applications. The high-frequency backend development targets the Greenland Telescope (GLT), the Large Millimeter Telescope (LMT), and other telescopes operating above 100 GHz. In the high-frequency regime (>100 GHz), relevant to millimeter and submillimeter observations at GLT and LMT, we leverage Altera’s Stratix 10 AX series FPGAs, which feature embedded Direct RF transceivers. These devices enable direct sampling of intermediate frequency (IF) signals without the need for analog down-conversion, simplifying front-end design and minimizing phase noise. The digital signal processing chain includes polyphase filter banks, fast Fourier transforms (FFT), and data packetization for downstream processing.

For low-frequency observations (300–700 MHz), we utilize the Xilinx ZCU216 platform as part of the BURSTT (Bustling Universe Radio Survey in Taiwan) project. BURSTT is a wide-field, low-frequency radio observatory initiative aimed at exploring time-domain astrophysics, including fast radio bursts (FRBs), pulsars, and other transient phenomena. To meet the demands of high time resolution, wide bandwidth, and large sky coverage, we have developed a high-performance digital backend based on the Xilinx ZCU216 platform. The ZCU216 integrates an UltraScale+ device with high-throughput RFSoC interfaces and flexible programmable logic, enabling real-time data acquisition, channelization, and streaming for transient detection. The backend is designed to support 16 simultaneous input channels, with firmware optimized for multi-beam, low-frequency array configurations.
Currently, 256 antennas have been installed and are operating at the main site. Additionally, 64 antennas have been deployed across two outriggers located in central Taiwan and on Green Island. These arrays have been actively observing for some time, with known pulsars being detected on a daily basis.

Speaker: Homin Jiang (Institute of Astronomy and Astrophysics, Academia Sinica)

09:55 Break (Location: Central Hall 2012)

28 Universal High-Bandwidth SDR Transceiver based on RFSoC4x2

Open‑source rfsoc_qsfp_offload [1] unlocked real‑time, direct‑RF capture on RFSoC4x2 by streaming up to 2.4576GHz of bandwidth over 100GbE into a host GNU Radio pipeline, with sustained ~80Gb/s demonstrated using GPU‑accelerated spectrometry. Building on that foundation and its growing community uptake this work‑in‑progress talk shares the current status toward our progress toward a universal high‑bandwidth SDR for RFSoC4x2 that serves both radio‑astronomy and instrumentation use‑cases while remaining fully open and reproducible.

New capabilities:

Full transmit path: GNU Radio → 100GbE → FPGA fabric → RFDC DUC → RF‑DAC for arbitrary waveform transmission.

Coherent multi‑channel operation: phase‑aligned capture across RF‑ADC tiles for beamforming/correlation; calibration and SYSREF handling.

Runtime‑selectable rates: flexible decimation/interpolation to operate at lower sample rates when NICs, storage, or downstream tooling impose constraints-without rebuilding the bitstream.

Software control & usability: reusable PYNQ‑based control APIs for fabric/ADC/DAC; host‑side RPC for center‑frequency and DUC/DDC configuration;

[1] https://github.com/strath-sdr/rfsoc_qsfp_offload

Speaker: Marius Siauciulis (University of Strathclyde)

29 Spectral Lines Spectrometer with 100 GbE Data Transfer and GPU Processing Implemented on RFSoC4x2

In San Juan, Argentina, the new China–Argentina Radio Telescope (CART) is currently under construction. This will be the biggest single-dish radiotelescope in south america with 40 m diameter. As part of this project, a new digital backend is being developed using the RFSoC4x2 platform. Taking as a base CASPER Simulink blocks, the team is implementing a spectrometer for H I spectral line observations. Data is acquired with the RFSoC4x2, transferred to a server via a 100 GbE network, and processed in real time using GPU-based FFT computation. The processed spectra are then stored on the server for further analysis.

Speaker: Rodrigo Rodríguez (Cart unsj)

Breakout Session: Summary

11:30 Lunch

Bus trip: Group 1 - ASTRON -> WSRT

Bus trip: Group 2 - ASTRON -> LOFAR

Site visit: Group 1 - WSRT

Site visit: Group 2 - LOFAR

Bus trip: Group 1 - WSRT -> LOFAR

Bus trip: Group 2 - LOFAR -> WSRT

Site visit: Group 1 - LOFAR

Site visit: Group 2 - WSRT

Bus trip: Group 1 - LOFAR -> ASTRON

Bus trip: Group 2 - WSRT -> ASTRON

Social Dinner (Location: Grand Café de Brink, Brink 30-31, 7991 CH Dwingeloo)

Location: Grand Café de Brink, Brink 30-31, 7991 CH Dwingeloo

Thursday 11 September

30 Introduction/Announcements

31 Low-Latency RF-ADC Pipeline with SmartNIC Integration in Radio Telescope Systems

This presentation explores the design and implementation of a low-latency RF-to-ADC data pipeline integrated with a SmartNIC for next-generation radio telescope systems. By leveraging FPGA-based platforms such as RFSoC and SmartNICs, the solution minimizes data movement delays and enables real-time digitization, preprocessing, and transport of high-bandwidth RF signals. The talk highlights architectural optimizations for deterministic latency, efficient packetization, and seamless interfacing with backend compute systems—crucial for large-scale, time-sensitive radio astronomy arrays

Speaker: Sheik Abdullah (VP Engineering iWave Systems Technology Pvt Ltd)

32 Keeping up with industry

As radio astronomy data rates continue to increase due to the desire for more bandwidth and larger radio telescope arrays, the need to process data in a timely and efficient manner correspondingly increases. While the computing and networking industries continue to push hardware to higher and higher levels of performance, reaping the benefits of these improvements can be non-trivial and often requires expertise that is not ubiquitous in the astronomy community. This talk will describe some of the reasons for these challenges and explore ways to surmount them.

Speaker: Dave MacMahon (University of California, Berkeley)

33 Designing for 400G/800G Systems: High-Speed Architectures with MPSoC, RFSoC, and Agilex 9

Modern radio telescope arrays demand ultra-high-speed data acquisition and transport to handle the explosive growth in raw RF bandwidth and spatial resolution. This presentation explores the architectural strategies for enabling 400G and 800G throughput in radio astronomy systems using advanced platforms such as MPSoC, RFSoC, and Agilex 9.

Speaker: Sheik Abdullah (VP Engineering iWave Systems Technology Pvt Ltd)

09:55 Break (Location: Central Hall 2012)

34 Transient Radio Interference Detection Using MeerKAT's D-Engine

Transient radio frequency interference (RFI) detection, classification and mitigation is becoming a more prevalent undertaking due to the increasing sensitivity of radio astronomy observatories. While monitoring RFI using stand-alone stations has benefits, there are shortcomings such as high maintenance and the need for stringent shielding. Thus, a transient detection system within the main signal chain at the digitisation stage is proposed due to the potential of better resource management and earlier detection.

This project presents a time domain transient detection firmware using moving average filtering implemented on the MeerKAT D-engine which performs digitisation at the feed of the MeerKAT antennas. The firmware was designed and implemented using the CASPER toolflow. Thereafter, simulation testing and lab testing were performed. Lab testing included the hardware to capture and to generate transients from typical transient emitting devices. Both lab and simulation testing showed positive results. Lastly, the test setup was performed in an anechoic chamber where transients were recorded successfully. The data set captured would form the basis for further classification algorithms in the future.

Speaker: Zubair Mohamed-Fakier (South African Radio Astronomy Observatory)

35 BVEX: Balloon-borne VLBI Experiment: Technical Challenges and RFSoC based backend

Very Long Baseline Interferometry (VLBI) combines radio telescopes across vast distances to image astrophysical objects with exceptional angular resolution. Ground-based VLBI is inherently limited by the Earth's diameter and suffers from increased atmospheric opacity and phase fluctuations at higher frequencies. Balloon-borne stations, which operate above more than 99% of the atmosphere, offer an innovative intermediate step between ground-based observatories and costly space missions, facilitating observations at higher frequencies. Additionally, since balloon-borne VLBI stations move, they provide superior uv coverage by sampling regions of the uv plane that would otherwise remain empty due to the fixed positions of ground-based telescopes.

The Balloon-borne VLBI Experiment (BVEX), scheduled for launch from Timmins, Ontario in August 2025, aims to address these limitations by demonstrating a viable VLBI station operating at the K-band frequency range (21-23 GHz). BVEX incorporates a state-of-the-art Radio Frequency System-on-Chip (RFSoC) backend, developed using the CASPER toolflow, which includes a high-resolution spectrometer and a low-power 100 GbE data storage system, crucial for handling the demanding data rates of VLBI.

A significant technical challenge for balloon-borne VLBI lies in ensuring precise position and attitude tracking at millimeter accuracy while operating under the extreme environmental conditions of the stratosphere. To address synchronization and phase stability, BVEX employs an ultra-stable Oven-Controlled Crystal Oscillator (OCXO) as its primary timing reference, housed within a temperature-controlled pressure vessel to maintain optimal operating conditions.

This talk will outline BVEX's instrumental design, including a single-sideband heterodyne receiver with a particular focus on the RFSoC-based backend system. I will also detail the specific engineering challenges encountered and solutions devised for high-altitude operations, timing synchronization, and position tracking. BVEX serves as a critical proof-of-concept for balloon-borne VLBI, demonstrating the feasibility of stratospheric radio astronomy and paving the path for high resolution and uv coverage capabilities in future VLBI arrays such as ngEHT or the GMVA(Global mm-VLBI Array).

Speaker: Mayukh Bagchi (Queen's University)

36 MeerKAT X-Band Digitisation and beyond

The MeerKAT radio telescope currently has UHF, L & S-Band Receivers and Digitisers installed. Although X-Band was fully provisioned for, this band was never developed as part of the MeerKAT project. The Italian National Institute for Astrophysics (INAF) secured funding and contracted the South African Radio Astronomy Observatory (SARAO) to design, develop and produce a high frequency Digitiser for MeerKAT. This Digitiser is based on the Square Kilometer Array Observatory (SKAO) Band 5B Receiver. Building on the success and legacy of the MeerKAT Digitiser, the SARAO Receivers Team are developing a new generation Digitisation Engine. The system level design, CASPER toolflow implementation and performance achieved will be presented. Looking beyond the scope of MeerKAT and MeerKAT+, other collaborations, possible deployments and applications will also be discussed.

Speaker: Henno Kriel (SARAO)

11:30 Lunch

37 SKARAB digital spectrometers for the BINGO single-dish radio telescope

We present a dedicated spectrometer and its corresponding firmware developed for the BINGO radio telescope—the first large-scale project in South America designed for 21-cm intensity mapping of neutral hydrogen (HI) across cosmological distances. Built on the CASPER FPGA-based SKARAB platform, the firmware performs real-time power spectrum analysis over the 980–1260 MHz band, supporting cosmological studies in the redshift range $0.127 \leq z \leq 0.449$, with the spectral resolution and data throughput needed for long-duration surveys.

Adapted to BINGO’s “single-dish, many-horns” configuration, the firmware processes real-time data from 28 horn antennas through SKARAB ADC modules, each digitizing four RF inputs—two orthogonal polarizations and two ColFET reference loads—while employing Dicke switching to mitigate gain fluctuations. The digital signal processing chain consists of pre-processing, FFT-based channelization, and Stokes parameter computation, with decimation and accumulation parameters optimized for the telescope’s observing strategy. In the pre-processing stage, a decimation factor of 8 narrows the effective bandwidth to 375 MHz, matching the BINGO observation band. A 32-bit, 8,192-point Polyphase Filter Bank (PFB) and FFT pipeline provides a spectral resolution of 45.7 kHz every 21.8 $\mu$s, ensuring superior channel isolation and reduced spectral leakage. Resulting spectra are transmitted over a high-speed QSFp+ 40 GbE Ethernet link and stored in FITS format, with an accumulation length of 45,776 producing around one averaged spectrum per second—equating to 10.8 GB of data per day for each horn.

This design meets BINGO’s needs and can be scaled for future HI mapping experiments around the world.

Speaker: Rafael Alves Batista (Federal University of Campina Grande)

38 LOFAR2.0 Station: Gateware overview

The Station Digital Processor (SDP) in a LOFAR2 station consist of a digital beamformer and a subband correlator. This year we have added a transient buffer function that can store 4 seconds of ADC data for all antennas in DDR4 memory, and that will be used to investigate the behavior of lighting. The SDP gateware runs on FPGAs of the UniBoard2. The SDP gateware that we use for LOFAR2 builds upon HDL libraries that were also used and extended during previous projects that we did in the last 20 years. In this presentation I will highlight some of the HDL libraries and explain how we maintain and improve the quality of our code.

Speaker: Eric Kooistra (ASTRON)

39 A New Fast-cadence, FPGA Based RFI Monitor for Real Time RFI Excision

Radio Frequency Interference (RFI) is a major source of data loss and contamination in radio astronomy experiments. Flagging this RFI in real time, pre-integration, is a powerful tool for salvaging frequency channels containing fast timescale RFI. I will present the preliminary design and development efforts for an FPGA based RFI monitor intended for real time tracking and excision of horizon based RFI sources, as well as early results from a prototype system. While the monitor is designed for general purpose use, the first iteration is optimized for use in the upcoming Canadian Hydrogen Observatory and Real-time transient Detector (CHORD). The monitor will provide full 360 degree horizon coverage with higher gain than the full CHORD array, producing RFI flags that will be fed in real time to CHORD. These flags will allow RFI excision on timescales as short as the cadence of the onboard FFT.

Speaker: Libby Berkhout (McGill University)

13:45 Break

40 Breakout Session

41 Tutorial

16:15 Drinks

Friday 12 September

42 Introduction/Announcements

43 PYNQ.remote and Versal RF: Next-Generation RFSoC Collaboration Opportunities with CASPER

PYNQ is an open-source framework from AMD Research and Advanced Development that enables Python-based control of AMD's adaptive SoCs. It has been widely adopted across multiple domains, including quantum control, wireless communications, and radio astronomy, where it lowers the barrier-of-entry for platforms such as the RFSoC4x2 development board.

We introduce PYNQ.remote, an extension to PYNQ which moves the Python API to the host, while reducing the target device software to a gRPC server and a PYNQ C++ backend. Familiar PYNQ operations such as loading overlays, accessing registers, and managing DMA streams are executed remotely from the host, making it easier to scale from a single board to multi-device systems. This enables new workflows such as distributed spectrum analysis, remote laboratory access, and integration with AI pipelines.

We also provide an update on the new Versal RF devices, highlighting their relevance for next-generation DSP, and explore collaboration opportunities where PYNQ’s runtime can coexist with existing CASPER workflows.

Speaker: Josh Goldsmith (AMD)

44 The Open-Source Analog Signal Path of DSA-2000

The DSA-2000 is a large-scale radio telescope currently under development at Caltech, funded by Schmidt Sciences. Aligned with our funder's mission to “democratize access to astronomy resources by supporting global usage and open science,” we are committed not only to open data, but also to openly share reusable hardware and software designs as much as possible.

In this talk, I will present the analog signal path of the DSA-2000, focusing on the ambient-temperature low-noise amplifier (LNA) and the RF-over-fiber (RFoF) link design. Every part of the design is released under a strong copyleft license, ensuring broad accessibility and long-term reuse. This approach closely aligns with the CASPER vision of building a global ecosystem of open-source, modular hardware for radio astronomy.

Speaker: Dr Kiran Shila (Caltech)

45 Scilab: A new open-source frontend for CASPER toolflow

The CASPER toolflow is a widely used framework for designing and implementing digital signal processing (DSP) systems, particularly in the field of radio astronomy. It provides a set of tools and libraries that enables researchers to create custom hardware and software solutions for processing astronomical data. The CASPER toolflow has been instrumental in the development of various radio telescopes and instruments, enabling real-time data processing and analysis. However, the current frontend tool that CASPER uses for high-level FPGA design is based on Model Composer integrated into MATLAB/Simulink, which is proprietary software. In this talk, I introduce Scilab, an open-source software platform for numerical computation and data visualization, as a new frontend tool for the CASPER toolflow. Scilab offers a similar environment to MATLAB/Simulink for designing CASPER blocks, generating FPGA IP cores, and simulating DSP systems. I will present on integrating Scilab into the CASPER toolflow and demonstrate its capabilities by creating an FPGA based spectrometer. Our results show that Scilab can successfully be used as an alternate frontend for CASPER-based designs.

Speaker: Wei Liu (University of California, Berkeley)

09:55 Break (Location: Central Hall 2012)

46 Deploying the CASPER Tool-Flow on an Intel Development Board Using a Scilab Front-End

The CASPER philosophy has long championed an open-source, broad scope philosophy with the goal of minimizing time-to-science and engaging as many users in the radio astronomy space as possible. Since CASPER’s inception, this has entailed creating modular, block-based FPGA designs using Simulink and Xilinx System Generator, abstracting away the need for chip-specific and board-specific HDL. Starting in 2020, AMD (Xilinx) released Model Composer, an alternative to System Generator for high-level-synthesis and HDL generation, before completely phasing out System Generator in 2023. This necessitates development of a new front-end for the CASPER toolflow, and presents a crossroads: Either update the toolflow to be compatible with Model Composer or transition to a non-Matlab-based front-end. The latter is appealing because it adheres strongly to the CASPER mission of low-cost, open-source tooling. As such, we have begun developing an alternative front-end for the CASPER toolflow, using Scilab, a free and open-source software package, that still maintains the same modular, block-based design flow as Simulink. In addition to being open-source, Scilab also opens up the possibility of CASPER support for FPGAs from vendors other than Xilinx, such as Altera. Scilab has the added benefit that there are no licensing costs other than for the FPGA vendor's tool. Here-in, I present on initial work to demonstrate the CASPER toolflow on an Altera (Intel) DE10-Nano development board using Scilab as the front-end. I will discuss the steps that needed to be taken to get the CASPER toolflow working as well as future plans for expanding out CASPER support for Altera devices.

Speaker: Ben Godfrey (UC Berkeley)

Breakout Session: Summary

47 Oversampled Polyphase Filterbank for CASPER

Decomposing a wideband signal into smaller frequency components is required for many radio astronomy signal processing systems. A standard digital method involves using the efficient FFT algorithm. However, this often provides an inferior channel response in terms of passband flatness, rolloff etc. A polyphase filterbank (PFB) can be used to improve these. The channel overlap can be controlled by using an oversampled version of this filter. Relevant theory as well as details of a streaming, wideband oversampled PFB implementation targeting FPGAs will be provided. The implementation will be an addition to the CASPER HDL DSP library (based on a library released by Astron).

Speaker: Andrew Martens (SARAO)

48 Final remarks / Conference close

12:15 Lunch

Return your bicycle (parking lot)