Advanced ACT Camera System

• Advanced ACT Camera is a state-of-the-art, cryogenic instrument featuring polarization-sensitive TES bolometer arrays for multi-frequency CMB observations.
• It employs a novel two-stage SQUID time-division multiplexing system that efficiently reads out signals from over two thousand detectors per array.
• The design integrates robust noise mitigation techniques—including hardware filtering, digital anti-aliasing, and TES bias optimization—to enhance mapping speed and sensitivity.

The Advanced ACT Camera refers to the focal-plane instrument system deployed on the Advanced ACTPol (AdvACT) receiver, the second-generation, polarization-sensitive upgrade to the 6-meter aperture Atacama Cosmology Telescope (ACT). AdvACT features extensive advancements over the original ACTPol system, with significant increases in detector count, frequency coverage, and noise engineering. Its architecture is defined by its cryogenic camera housing three large-scale, polarization-sensitive TES bolometer arrays operating in broadband, dichroic modes, and by a novel two-stage SQUID time-division multiplexing (TDM) readout system that enables efficient, low-noise scaling to over two thousand detectors per array. The following sections detail the components, readout strategies, noise characterizations, and mitigation procedures integral to the Advanced ACT Camera (Gallardo et al., 2019).

1. Focal Plane Composition: Detector Arrays and Pixel Design

The core of the Advanced ACT Camera is a set of three cryogenic focal-plane arrays—PA4, PA5, and PA6—each implementing broadband, dichroic transition-edge sensor (TES) bolometers:

• PA4: 2,012 detectors, simultaneous observation in 150 GHz and 220 GHz bands.
• PA5 & PA6: 1,716 detectors each, simultaneous observation in 90 GHz and 150 GHz bands.

Each array comprises 32 columns of horn-coupled, polarization-sensitive TES pixels. Multiplexing is realized across either 64 rows (PA4) or 55 rows (PA5/PA6), yielding the total pixel count. Each TES couples to a superconducting microstrip circuit that dichroically separates the two frequency bands to independent TES islands, maximizing sensitivity and coverage in each band (Gallardo et al., 2019).

2. Time-Division Multiplexed SQUID Readout Architecture

AdvACT employs a two-stage SQUID time-division multiplexing (TDM) system, realized in the Multi-Channel Electronics (MCE) platform. Within each of the 32 parallel columns, the TDM protocol sequentially interrogates each row:

• First Stage (SQ1): At each TES, a first-stage SQUID (SQ1) acts as the initial current amplifier. "Row-select" lines address all SQ1s in a row to activate them simultaneously for readout.
• Second Stage (SQ2): The 32 SQ1 outputs per column are summed into a second-stage SQUID array, cooled to approximately 1 K.
• Room-Temperature Readout: The summed signal is transported out of the cryostat, filtered, digitized, and ultimately converted to a time-series data stream.

The achieved multiplexing factors are 64:1 (PA4) and 55:1 (PA5/PA6) per column. Sampling is controlled by the 50 MHz MCE master clock, as:

$$f_{\text{sample}} = \frac{50\,\mathrm{MHz}}{N_{\text{row}}\times\text{row\_len}}$$

After digital anti-aliasing filtering (Butterworth, gain 0.99) and decimation by a factor ($\text{data\_rate}$), the delivered readout frequency is:

$$f_{\text{read}} = \frac{50\,\mathrm{MHz}}{N_{\text{row}} \times \text{row\_len} \times \text{data\_rate}}$$

Typically, $f_{\text{read}} \approx 300.5$ Hz for PA4 (64 rows, data_rate=26) and $f_{\text{read}} \approx 395$ Hz for PA5/PA6 (55 rows, data_rate=23). Fast diagnostic modes enable up to $\sim5$ kHz readout by sampling only four rows, thereby nearly eliminating aliasing in the science band (Gallardo et al., 2019).

3. Aliased Noise: Definition and Characterization

In time-domain multiplexed systems, finite sampling rates cause out-of-band continuous-time noise (above the Nyquist frequency $f_{\text{Ny}} = f_{\text{read}}/2$) to be folded or “aliased” into the science band. The total measured power spectral density (PSD) is then:

$$P_{\text{total}}(f) = P_{\text{inband}}(f) + P_{\text{alias}}(f)$$

where

$$P_{\text{alias}}(f) = \sum_{k\neq 0} S(|f + k f_{\text{read}}|) \cdot |H(f)|^2$$

with $S(f)$ the detector noise PSD and $H(f)$ the anti-alias digital filter response. Given that photon noise dominates in AdvACT, even modest aliasing fractions can degrade science performance (Gallardo et al., 2019).

To quantify this effect, the “aliased-noise fraction” (AF) is introduced as an empirical figure of merit:

$$AF \equiv \frac{\bar{P}_{\text{slow}}}{\bar{P}_{\text{fast}}}, \qquad \bar{P} \equiv \langle \text{PSD}(f) \rangle_{10-60\,\mathrm{Hz}}$$

Here, $\bar{P}_{\text{slow}}$ measures the PSD in standard full-array readout mode ($\sim$300–395 Hz), while $\bar{P}_{\text{fast}}$ comes from a high-speed, low-multiplexing measurement ($\sim$5 kHz). The 10–60 Hz window avoids low-frequency $1/f$ noise and high-frequency filter roll-off. This definition allows purely observational assessment of the in-band excess noise due to aliasing (Gallardo et al., 2019).

4. Measured Aliased-Noise Fractions and Dependence on Bias

Empirical field tests (window-covered, 50% $R_n$ bias) yield the following aliased-noise fractions:

Array, Frequency	AF (mean ± 1σ)	Excess Noise (%)
PA4, 150 GHz	1.04 ± 0.02	4
PA5, 150 GHz	1.03 ± 0.02	3
PA6, 150 GHz	1.05 ± 0.02	5
PA4, 220 GHz	1.10 ± 0.02	10
PA5, 90 GHz	1.07 ± 0.02	7
PA6, 90 GHz	1.10 ± 0.02	10

Distributions are narrow and approximately Gaussian, with quality cuts applied ($0.5 < AF < 1.5$). For the 150 GHz band, AF increases with TES bias (40–90% $R_n$); in 90 and 220 GHz, the trend is flat. A 5–10% readout noise increase corresponds to a proportional sensitivity loss for mapping speed, underlining the importance of aliasing control in these multiplexed systems (Gallardo et al., 2019).

5. Impact on Sensitivity and Mitigation Strategies

With the science noise budget dominated by photon statistics, unmitigated aliasing leads to a direct 5–10% mapping speed penalty. The following strategies have been deployed and proposed to minimize aliased noise:

• Hardware RL Low-Pass Filtering: Each TES bias circuit is fitted with a hardware $RL$ filter, suppressing high-frequency detector noise at $\sim5$–10 kHz.
• Digital Anti-Alias Filtering: A two-stage Butterworth filter in the MCE provides $>$60 dB suppression above $f_{\text{read}}/2$.
• Reducing Multiplexing Factor: Configuring the TDM for the minimum practical $N_{\text{row}}$ increases $f_{\text{read}}$ and suppresses aliasing, at the expense of field coverage.
• TES Bias Optimization: Operating at $\sim$50% $R_n$ biases optimizes detector time constant $\tau$ and minimizes high-frequency noise.
• Future Modalities: Microwave SQUID multiplexing architectures, if adopted, offer the potential for fundamentally lower aliasing by avoiding the time-domain protocol (Gallardo et al., 2019).

6. Engineering Significance and Future Developments

The Advanced ACT Camera demonstrates the scalability of TDM SQUID architectures to multi-kilopixel, multi-frequency CMB instrumentation, and sets benchmarks for aliased noise control using a strictly observational figure of merit. The results confirm that carefully engineered multiplexed readout, anti-alias filtering, and TES biasing can confine sensitivity degradation to acceptable levels—even under the photon-dominated noise regime of contemporary CMB surveys.

Future efforts will extend this characterization under true sky loading conditions and explore the adoption of alternative readout schemes that may further suppress aliasing below the current percent-level. The continuing development of multiplexed cryogenic camera architectures remains critical for increasing mapping speed and frequency coverage in next-generation cosmic microwave background experiments (Gallardo et al., 2019).