Model-based qubit noise analysis improves accuracy and speed

Researchers at the Johns Hopkins University Applied Physics Laboratory have developed model-based qubit noise spectroscopy (QNS) approaches inspired by classical signal processing techniques. The work addresses limitations of existing QNS methods, which estimate noise power at set points, by using structure in the noise spectrum to improve statistical accuracy and potentially achieve improvements. These new estimation approaches, detailed in a recent paper, maintain computational benefits while offering a path toward more effective qubit control and adaptive feedback for quantum systems. The team anticipates the flexibility of these models will also prove useful in quantum sensing.

SchWARMA Model Adapts Classical Noise Spectroscopy to Qubits

Classical signal processing techniques are now being directly applied to the analysis of qubit noise, with the Schrödinger wave autoregressive moving-average (SchWARMA) model leading this effort. Unlike existing qubit noise spectroscopy (QNS) methods that estimate noise power at specific points, these new approaches assume underlying structure within the noise spectrum to enhance accuracy. This adaptation allows for the potential of identifying finer details in qubit noise than previously achievable with non-parametric methods.

The team’s work centers on adapting established model-based approaches from classical noise spectroscopy, including autoregressive moving-average (ARMA) models and line spectra, to the challenges of QNS. Beyond simply translating existing algorithms like Yule-Walker and multiple signal classification, they introduced a flexible new approach based on the SchWARMA model itself. This model allows for direct fitting of parameters and uses classical model-selection rules to prevent overfitting and improve the precision of estimations.

The researchers demonstrated the efficacy of SchWARMA through simulations using white noise, bandlimited noise, and complex spectra, further validating the approach with experimental data. A key advantage of these model-based approaches is their ability to produce estimates at all frequency points, a contrast to the limited band resolution of non-parametric techniques. The adaptability of SchWARMA extends beyond estimation. The team showed it can resolve much finer spectral features than non-negative least squares (NNLS)-based estimation. They also successfully “closed” the SchWARMA modeling and estimation loop, confirming its effectiveness in simulating, injecting, and estimating noise.

Yule-Walker and MUSIC Algorithms Applied to Qubit Noise Spectroscopy

Classical algorithms, Yule-Walker and multiple signal classification, now offer a pathway to refine qubit noise spectroscopy beyond traditional methods. These techniques, originally developed for signal processing, provide estimates at all frequency points, a departure from non-parametric approaches that resolve noise power only at set bands. This expanded frequency resolution promises potentially revealing finer spectral details within qubit noise than previously attainable. The team’s work demonstrates that these model-based approaches maintain the statistical and computational benefits of their classical counterparts, offering a powerful new toolkit for quantum systems.

Beyond improved estimation, the flexibility of the SchWARMA model suggests potential applications in adaptive feedback control for quantum systems. This parallels the role of similar models in classical adaptive signal processing and control, indicating a broader applicability beyond mere noise characterization.

Filter Function Formalism Relates Control Signals to Noise Spectra

The filter function formalism provides a direct link between the sequences used to probe qubit noise and the resulting noise power spectrum, offering a new perspective on how to interpret quantum noise spectroscopy (QNS) results. This framework, borrowed from classical signal processing, clarifies that QNS fundamentally involves estimating the overlap between a control signal’s filter characteristics and the spectrum of dephasing noise. Existing QNS methods typically estimate noise power at specific frequencies, resembling classical filtering techniques with limitations on the allowable filter shapes.

To address these limitations, new approaches directly adapt classical model-based techniques to the quantum setting, moving beyond simply estimating power at set points. By fitting model parameters directly to experimental data, researchers can avoid common pitfalls like overfitting, a challenge in many statistical analyses.

Model-Based QNS Offers Superresolution and Reduced Error Variance

Existing qubit noise spectroscopy (QNS) techniques, categorized in classical terms, rely on estimating noise power at discrete points or bands within a spectrum. This contrasts with newly derived model-based QNS approaches that assume underlying structure in the noise spectrum itself, offering potential gains in statistical accuracy and the ability to resolve finer spectral details. Classical signal processing offers established methods for spectrum estimation, including techniques like the periodogram and multitaper methods, but these often struggle to reduce variance as data increases.

Model-based approaches, however, directly estimate parameters defining a presumed functional form of the spectrum, reducing the number of variables and mitigating overfitting through established model-selection rules. The team’s work demonstrates that applying this principle to QNS yields benefits analogous to those seen in classical signal processing, including a reduction in error variance.

A key advantage of these model-based QNS methods is the ability to discern spectral features at a scale finer than the effective Nyquist rate. Simulations using SchWARMA show it can resolve spectral peaks between standard non-negative least squares (NNLS) power bands, a feat previously unattainable. Analysis of fast Fourier transform power spectral density (FTTPS) dephasing spectra revealed that SchWARMA-based estimators produced more accurate power spectra than their NNLS counterparts.

The paper reports, highlighting the direct improvement in spectral fidelity. The use of the Bayesian information criterion (BIC) for model selection helps to minimize the impact of overfitting, ensuring robust and reliable results.

Dephasing Noise Characterization via Single-Qubit Hamiltonian Dynamics

The ability to accurately map qubit dephasing noise extends beyond simply characterizing a qubit’s environment; it directly informs the development of more effective control strategies to improve qubit fidelity. These model-based QNS techniques draw inspiration from classical signal processing, specifically using the filter function formalism to understand how a control signal’s “power” overlaps with the dephasing noise spectrum. This connection allows researchers to move beyond the limitations of conventional QNS, which constrains the forms of filters used in analysis.

The work introduces two distinct methods for adapting these classical model-based approaches to the quantum setting, expanding the toolkit available for noise analysis. A key element of this adaptation is the application of the Schrödinger wave autoregressive moving-average (SchWARMA) formalism, previously developed for modeling correlated noise in classical systems.

The researchers validated their approach using both simulated and experimental data, demonstrating the ability to finely resolve spectral features. “SchWARMA could accurately predict survival probabilities of QNS experiments given a native noise model and a desired injected spectrum,” as demonstrated in a related study. The techniques are not limited to single-qubit dephasing noise, with potential applications extending to multi-qubit noise, control noise and even non-Gaussian noise scenarios where filter-function based approximations remain valid.

Non-Parametric vs. Parametric Approaches to Qubit Noise Spectroscopy

The benefits of adopting a parametric approach extend beyond simply refining the noise power estimate. Model-based methods offer reduced error variance compared to their non-parametric counterparts, a consequence of estimating fewer parameters for a given dataset. This advantage is particularly important when dealing with the limited data often available in quantum experiments. The flexibility inherent in model selection allows researchers to mitigate overfitting, a common challenge in complex data analysis.

Analysis of low-frequency noise suggests fluctuations in the 1/f^α component of the native noise spectrum are a contributing factor to distortions in the estimated spectrum. Distortions also appear at higher frequencies, but these appear to stem from the noise injection process itself, where the output power diminishes noticeably. Understanding these limitations is critical for accurate interpretation of QNS results. The ability to accurately characterize these noise sources is not merely an academic exercise; it directly impacts the efficacy of quantum control strategies aimed at minimizing the effects of time-correlated noise and improving qubit fidelities.

SchWARMA Validation with Simulated and Experimental Bandlimited Noise

The team demonstrated SchWARMA’s ability to accurately reconstruct noise spectra in these simulated environments, a key step toward real-world application. These reconstructions were then compared against those generated using traditional non-parametric methods, establishing a baseline for performance evaluation. Further validation involved analysis of experimental data originally presented in a separate publication. This data, stemming from noise injection experiments, allowed researchers to test SchWARMA’s performance across simulation, injection and estimation stages. The analysis revealed a potential drift between experiments measuring native and injected noise, a nuance previously difficult to discern.

Applications of Model-Based QNS in Quantum Sensing and Control

The flexibility inherent in the recently developed Schrödinger wave autoregressive moving-average (SchWARMA) formalism extends beyond spectral estimation, offering potential benefits for adaptive feedback control of quantum systems. The ability to generate noise models using SchWARMA allows exploration of circuit impacts and direct application in model-based optimal control protocols, a capability not previously available with non-parametric methods. This generative modeling aspect provides a pathway to proactively mitigate noise effects on quantum computations.

Beyond direct application to qubit noise spectroscopy (QNS) and quantum sensing, these model-based approaches may address challenges in more complex noise scenarios. This broad applicability stems from the underlying framework’s ability to accommodate diverse noise characteristics, exceeding the limitations of conventional QNS methods restricted to specific filter forms.

Researchers envision these methods as complementary to existing approaches, such as those using line or bandlimited signal models. The development of SchWARMA-based QNS approaches also provides a computationally efficient route to noise characterization, offering an alternative to traditional non-parametric least squares (NNLS) estimation. The resultant SchWARMA model can be directly employed as a generative model of noise within a quantum system, streamlining the process of predicting noise impacts on circuits. This combination of speed and precision is key for scaling quantum technologies and implementing real-time control strategies.

More like this

Quantum Research NewsMonitoring quantum jumps lets feedback control precision

Quantum Research NewsResearchers from Harvard Quantum, Anthropic and QuEra show new circuits cut qubit counts for quantum calculations

Quantum PhysicsNeural networks identify quantum phases without post-selection

Quantum PhysicsHow Quantum Source turns quantum physics into engineering