CONCERTO : Optimization of readout electronics

TL;DR

The paper presents a scalable FPGA-based readout architecture for the CONCERTO instrument, enabling increased detector count while maintaining signal quality through digital twin modeling and firmware optimization.

Contribution

Developed a Python-based digital twin to optimize FPGA firmware, reducing resource usage and enabling higher detector counts without degrading performance.

Findings

Reduced FPGA resource usage by up to 39% in LUTs.

Successfully identified and mitigated spurs in data.

Supported over 800 MKIDs per feedline on the same hardware.

Abstract

The CONCERTO millimeter-wave spectral-imaging instrument was deployed on the Atacama Pathfinder EXperiment (APEX), where it acquired science data between April 2021 and May 2023. The instrument features two focal-plane arrays, each composed of 2400 Microwave Kinetic Inductance Detectors (MKIDs). Each array is divided into six feedlines containing 400 MKIDs each, with each feedline read out by a dedicated FPGA-based board, KID_READOUT. The next-generation instrument aims to double the detector count per feedline, increasing it from 400 to 800 MKIDs. Achieving this requires a substantial scaling of the readout architecture and poses two key challenges for KID_READOUT: maintaining readout signal integrity and constraining firmware resource usage, as a direct upscaling of the existing design would exceed the available FPGA capacity. To overcome these limitations, we developed a Python-based, cycle-and bit-accurate digital twin of the full FPGA digital signal processing chain. This model enabled a detailed investigation of internal signal behavior and provided quantitative guidance for firmware optimization. Leveraging these insights, we identified the source of two spurs present in CONCERTO data and significantly reduced their amplitudes. At the same time, we achieved substantial reductions in firmware resource usage-39.0%pt in LUTs, 20.3%pt in flip-flops, and 28.98%pt in DSP slices-without degrading readout performance. The resulting architecture supports more than 800 MKIDs per feedline on the same hardware platform while preserving readout signal quality, offering a scalable and resource-efficient solution for future high-resolution millimeter-wave astronomical instruments.