Full-System Joint Readout

• Full-system joint readout is a holistic strategy that integrates sensing, transduction, transport, and state reconstruction into one cohesive instrument.
• It employs techniques such as frequency-division multiplexing and deterministic synchronization to coordinate diverse subsystems across quantum, cryogenic, and high-energy physics applications.
• This integrated approach enhances scalability and precision by reducing wiring complexity and allowing centralized calibration and coordinated error inference.

Full-system joint readout denotes a readout strategy in which the sensing element, front-end transduction or multiplexing stage, transport layer, warm electronics, control logic, and often the downstream reconstruction or event-building path are designed, modeled, or validated as one instrument rather than as isolated subsystems. In cryogenic detector work this explicitly means operation of the entire cryogenic-to-room-temperature chain; in high-energy physics it can denote a globally synchronized timing, trigger, and data-flow system spanning front ends, readout boards, and event building; in circuit QED it denotes a single measurement channel whose response depends on the combined state of multiple qubits; and in recent distributed tomography it extends to the joint estimation of quantum states and readout errors (Neidig et al., 9 Sep 2025, Alessio et al., 2018, 0908.1955, Taherpour et al., 15 Apr 2026).

1. Terminological scope and defining characteristics

The defining feature of full-system joint readout is integration across interfaces that are often treated separately in component-level work. For magnetic microcalorimeters, the emphasis is that the result is not a SQUID chip alone or electronics alone, but a complete instrument comprising the MMC array, a microwave SQUID multiplexer chip, and a custom SDR readout chain operating together across the intended bandwidth and multiplexing target (Neidig et al., 9 Sep 2025). For the ATLAS New Small Wheel prototype, the same idea appears as a complete quadruplet test platform incorporating detector planes, pFEBs, sFEBs, a custom DAQ board, and an external cosmic-ray trigger system (Miao et al., 2018).

The literature also uses the term “joint” in a measurement-theoretic sense. In superconducting-qubit experiments, a single resonator transmission line serves as one detector whose response depends on the two-qubit computational basis state, so the readout carries both single-qubit polarization information and a two-qubit correlation term (0908.1955). In the earlier joint dispersive tomography formulation, the same resonator nonlinearity yields a measurement operator containing identity, single-qubit, and $\sigma_{z1}\sigma_{z2}$ terms, making full two-qubit tomography possible without separate single-shot readout of each qubit (0812.2485).

A broader systems interpretation appears in distributed quantum tomography robust to readout errors. There, the “joint” aspect refers to simultaneous estimation of regional density operators and regional confusion matrices, with neighboring regions coupled by reduced-state consistency constraints on overlaps (Taherpour et al., 15 Apr 2026). This suggests that full-system joint readout is not restricted to hardware topology; it also includes coordinated estimation problems in which measurement imperfections are part of the state-reconstruction model.

2. Canonical architectural patterns

Across domains, full-system joint readout usually replaces many dedicated point-to-point readout paths with a shared transport or shared control fabric. In the MMC implementation, the architecture is explicitly

$$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$

with each MMC pixel inductively coupled to an rf-SQUID, each rf-SQUID modulating a quarter-wave microwave resonator, and many resonators sharing one common feedline (Neidig et al., 9 Sep 2025). At room temperature, the feedline is connected to an analog RF front-end, an analog/digital conversion board, and an FPGA processing board, which together generate probe tones, digitize the return, and perform real-time processing.

A different but structurally analogous pattern appears in dexterous robotic hands. Each sensing unit contains a D flip-flop, a bilateral analog switch, and the base sensor. A single high pulse propagates through a chain of flip-flops, one analog switch closes per clock slot, and exactly one analog-output sensor is connected to the shared ADC line at a time. Between modules, only clock, pulse, and analog output lines are used, while power and ground are shared separately. The full readout pipeline is given as sensor module with analog-output base sensor + D flip-flop + analog switch $\rightarrow$ pulse/clock propagated through finger chain $\rightarrow$ one sensor connected to ADC per clock slot $\rightarrow$ MCU performs synchronized ADC acquisition $\rightarrow$ PC/ROS converts raw voltages to joint angles (Kim et al., 2 May 2026).

In large HEP systems, the shared resource is not a feedline or analog line but a timing and event-identity fabric. The upgraded LHCb detector uses a fully trigger-less, full-40 MHz architecture in which Front-End electronics, Readout Boards, the event builder network, and the event-filter farm are synchronized by one Readout Supervisor and a distributed optical control network. The central firmware distributes the main bunch clock, timing signals, control commands, and rate regulation across roughly 15,000 destinations, making synchronization rather than local readout autonomy the core architectural primitive (Alessio et al., 2018).

JUNO places the integration boundary at the DAQ data flow. The front-end electronics generate detector data; ROS processes collect fragments and perform first-level assembly into event fragments; EB processes gather fragments for a given event ID and assemble a full event; and the EBM schedules assignments and cleanup. The collaboration sequence is explicitly coordinated through L1id propagation, round-robin EB assignment, EoE notification, and clear commands back to the ROS layer (Zeng et al., 2018).

3. Multiplexing, synchronization, and shared-resource coordination

A full-system joint readout almost always depends on multiplexing, deterministic synchronization, or both. In microwave SQUID multiplexing for MMCs, the principle is frequency-division readout. Each resonator has a distinct resonance frequency, the dedicated $\mu$MUX chip contains 18 resonators spanning $4.25\,\mathrm{GHz} \le f_r \le 7.75\,\mathrm{GHz}$, and the SDR electronics is designed to support up to 400 readout tones across the full $4\text{–}8\,\mathrm{GHz}$ band. In operation, the FPGA performs digital down-conversion, polyphase channelization/filterbank separation, flux-ramp demodulation, and triggering and metadata extraction (Neidig et al., 9 Sep 2025).

Time-division multiplexing introduces a different systems constraint: the relevant bandwidth is set by row switching and settling, not only by the science-band sample rate. The CMB-S4 end-to-end TDM model treats the TES $\rightarrow$ SQ1 $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$0 SSA $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$1 warm-electronics chain as a linear small-signal system, sums noise contributions in quadrature as

$$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$2

and then explicitly models multiplexing aliasing by generating broadband noise over the full bandwidth, resampling at the multiplexed readout rate, and calculating the folded spectrum. The stated target multiplexing factor is 80, and the model identifies the SQ1 loop inductance, cryocable parasitics, and warm amplifier noise as key limits on the row-switching regime (Goldfinger et al., 2023).

The robotic-hand SIPO architecture is again time-multiplexed, but the switching fabric is digital pulse propagation rather than row-select SQUID addressing. Each sensor receives one clock cycle per scan, the fully integrated hand includes 20 sensor units total, and the full-scan rate is controlled simply by changing the clock frequency (Kim et al., 2 May 2026).

In HEP applications, multiplexing is frequently subordinate to timing determinism. The sTGC prototype readout uses a Timing-Trigger-Control module to fan out global clock, trigger, and synchronization signals across 8 FEBs so that data from different layers can be associated with the same event (Miao et al., 2018). LHCb extends this into fixed-latency command delivery through programmable pipelines, BXID reset distribution, and centralized event metadata generation, so that the whole detector behaves as one coherent readout machine at the full 40 MHz bunch-crossing rate (Alessio et al., 2018).

4. Measurement models, calibration, and joint inference

Joint readout requires an explicit model for how shared hardware or shared control variables encode the underlying physical state. In $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$3MUX MMC readout, the detector signal changes the magnetic flux in the SQUID loop, which changes the SQUID inductance and shifts the resonator frequency. The underlying response is described as the usual flux-dependent resonator response $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$4, and flux-ramp modulation is applied to linearize the nonlinear SQUID transfer. The paper specifies a sawtooth flux signal with amplitude $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$5 and repetition rate $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$6, with an effective demodulation amplitude of about $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$7 after removing ramp transients (Neidig et al., 9 Sep 2025).

In circuit QED, the joint measurement model is explicit at the operator level. For a tone placed at the cavity peak corresponding to $$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$8, the ideal measurement operator is

$$\text{MMC array} \rightarrow \text{rf-SQUID} \rightarrow \text{microwave resonator} \rightarrow \text{shared feedline} \rightarrow \text{SDR electronics},$$9

so the detector is directly sensitive to a constant term, two single-qubit polarization terms, and the two-qubit correlation term $\rightarrow$0. The actual measured homodyne voltage is modeled as

$\rightarrow$1

Rabi experiments on the left qubit, right qubit, and both qubits validate the model and show direct evidence of mixing from the joint $\rightarrow$2 sensitivity (0908.1955).

The earlier two-qubit dispersive tomography formulation makes the same point from resonator-response nonlinearity. Although the dispersive shift operator is linear in $\rightarrow$3 and $\rightarrow$4, the measured cavity transmission is nonlinear in the shift, so the effective measurement operator becomes

$\rightarrow$5

which is sufficient for full two-qubit tomography once local pre-rotations are applied (0812.2485).

Distributed tomography robust to readout errors generalizes calibration into a constrained bilinear estimation problem. Each overlapping region has a local density operator $\rightarrow$6, a local POVM, and a local confusion matrix $\rightarrow$7, with noisy regional probabilities $\rightarrow$8. Neighboring regions are coupled by the exact overlap constraint

$\rightarrow$9

The proposed distributed alternating method uses ADMM for the state-update step and local parallel confusion-matrix updates, so readout error becomes part of the end-to-end inverse problem rather than a separately calibrated nuisance parameter (Taherpour et al., 15 Apr 2026).

5. Representative implementations and reported performance

The MMC realization provides a hardware-centered demonstration of full-system joint readout. The electronics are designed for up to 400 channels; at cryogenic temperature, 14 $\rightarrow$0MUX channels were simultaneously usable, 7 were connected to MMCs, and the system successfully acquired simultaneous time-stream data from 14 channels. The full operating band is 4–8 GHz, measured channel resonances span 4.25–7.75 GHz, the mean white noise at $\rightarrow$1 is $\rightarrow$2 in open-loop mode and $\rightarrow$3 in FRM-demodulated mode, a typical event shows an exponential decay with time constant $\rightarrow$4, and the system ran stably enough to collect 10 s of simultaneous time-stream data from all 14 active channels (Neidig et al., 9 Sep 2025).

The robotic-hand implementation demonstrates that a minimalist three-wire inter-module interface can remain compatible with heterogeneous analog-output sensors and still preserve useful bandwidth and estimation fidelity. The integrated hand contains 16 joint sensor modules and one four-channel tactile sensor module, enabling acquisition of 20 sensor channels at a full-scan rate of 1 kHz, with stable operation up to 1.5 kHz. For joint sensing, the maximum slope APE is 0.446%, RMSE values are 0.233° to 0.368°, and the STD of error ranges from 0.227° to 0.343°. During interaction, direct joint sensing outperforms motor-based estimation: MCP joint RMSE is 0.388° for the joint sensor versus 2.344° for motor-based estimation, and PIP joint RMSE is 0.598° versus 2.600° (Kim et al., 2 May 2026).

Prototype HEP systems highlight coordinated triggering and event transport. The sTGC platform instruments one full-size quadruplet using 4 pFEBs, 4 sFEBs, 1 custom DAQ board, and a cosmic trigger formed from scintillators and PMTs. The DAQ board configures and controls up to 8 FEBs, decodes 8b/10b event data, services eight separated FIFOs in round-robin order, and transfers event data to the PC via Gigabit Ethernet. Oscilloscope observations show a two-pad-layer coincidence and a four-pad-layer coincidence for cosmic muons, and charge spectra were measured for pad and strip channels under a high voltage of 32 kV (Miao et al., 2018).

JUNO reports performance at the distributed DAQ layer. Under fixed $\rightarrow$5 fragments and 1 kHz rate, the stated readout relation is

$\rightarrow$6

Best single-node performance is reported for 8 ROSs per node, supporting up to 1024 channels and achieving about 17.7 Gb/s in the summary; in a 14-node integrated test with 4 readout nodes, 4 EB nodes, 2 sending nodes, and 1536 channels, the overall integration event rate is 1.1 kHz (Zeng et al., 2018).

Integrated microwave SDR platforms illustrate another full-system realization. An RFSoC Gen3-based bolometer readout targeting approximately 1,000 bolometers in the 4–8 GHz band characterizes a comb of $\rightarrow$7 tones centered at 7.5 GHz, with average spacing $\rightarrow$8 MHz and standard deviation $\rightarrow$9 kHz. In RF loopback, the average white phase noise is around $\rightarrow$0 dBc/Hz, better than $\rightarrow$1 dBc/Hz, which the authors describe as totally in agreement with the state-of-the-art readout systems (Redondo et al., 2023). The CRS extends the full-stack approach to microwave KIDs by combining RFSoC-based boards, firmware, open-source control software, and subrack synchronization. A single CRS can operate 4,096 KIDs over 2.5 GHz of complex bandwidth between 0–10 GHz, and each fully populated subrack supports arrays of more than 65,000 KIDs (Montgomery et al., 2024).

Quantum-joint-readout experiments emphasize ensemble-level estimation rather than throughput. With a single joint measurement channel, entanglement metrology reaches $\rightarrow$2 accuracy and $\rightarrow$3 precision despite $\rightarrow$4 fidelity of single measurements; measured CHSH operators reach a maximum value of $\rightarrow$5, and entanglement witnesses give a lower bound of $\rightarrow$6 on concurrence (0908.1955). In joint dispersive two-qubit tomography, reported fidelities are 95% for a separable product state and 74% for a Bell state (0812.2485).

6. Scalability, limitations, and research directions

The principal motivation for full-system joint readout is scaling detector or sensor count without proportional growth in wiring, parasitic heat load, control complexity, or cost. The MMC paper states this motivation directly for large cryogenic arrays and presents the demonstrated system as a scalable path toward future large-scale, high-resolution MMC experiments (Neidig et al., 9 Sep 2025). The CRS makes the same systems claim at observatory scale through common clocking, IRIG-B timing, backplane mesh networking, and synchronized operation across up to 16 boards per subrack and 16 subracks through the mesh/network architecture (Montgomery et al., 2024).

Several papers also make clear that scalability is limited by nonidealities outside the nominal multiplexing concept. In the MMC system, the SDR platform was successfully operated at the target 400-tone configuration, but the available cryogenic attenuation and DAC output power were not yet sufficient to deliver optimal power to all 400 channels simultaneously; the stated remedies are an amplifier at the SDR output or reduced cryogenic input attenuation (Neidig et al., 9 Sep 2025). In JUNO, the receiver CPU is identified as the bottleneck in network readout, single receive buffer outperforms custom ROS Buffer Management, and balanced readout/EB deployment is necessary for system-wide gains (Zeng et al., 2018). In the CMB-S4 model-validation study, measured spectra contain MHz-scale lines not in the model and a discrepancy in the 100 kHz–1 MHz range is attributed to likely cryogenic common-mode pickup in the SQ1 loop, motivating differential cryogenic readout under development (Goldfinger et al., 2023).

The literature also shows that “jointness” does not remove the need for calibration; it relocates calibration to shared transfer functions and shared inference layers. In qubit readout, low single-shot fidelity does not preclude high-quality ensemble tomography because the joint detector is calibrated as a linear response model to Pauli correlators (0908.1955). In regional tomography with readout errors, joint estimation improves state recovery over fixed-readout reconstruction, recovers a substantial portion of oracle performance, and reveals a clear tradeoff between state estimation performance, communication, and computation across Ring, Ladder, Torus, and Hub geometries (Taherpour et al., 15 Apr 2026).

A plausible implication is that future full-system joint readout work will continue to converge on three coupled design principles already visible across these domains: shared physical transport, explicit end-to-end modeling of distortion and noise, and centralized or consensus-based coordination of timing, state estimation, or event identity. The sources differ sharply in hardware substrate and operating regime, but they consistently treat successful readout as a property of the entire coupled chain rather than of any single component in isolation (Neidig et al., 9 Sep 2025, Alessio et al., 2018, Taherpour et al., 15 Apr 2026).