Robust projective measurements through measuring code-inspired observables

Quantum measurements are foundational to quantum algorithms including sampling, learning, channel and parameter estimation, and universal computation. However, measurement errors limit performance. Near-term algorithms typically use only the classical outputs of measurements, and while quantum error mitigation can reduce statistical errors via repeated experiments and post-processing, directly correcting measurement errors without full quantum error correction (QEC) remains open. In fault-tolerant contexts, stabilizer-based approaches (data-syndrome codes, single-shot QEC) address measurement errors, but extending robust measurement correction to non-stabilizer codes (e.g., bosonic codes) is unresolved. This work asks: can we implement any projective POVM robustly against classical readout errors without encoding the data into a QEC code? The goal is to design a measurement scheme that is equivalent to the target projective measurement in the noiseless case and that corrects a prescribed number of classical outcome errors.

The paper situates the problem within broad applications of quantum measurements and reviews error mitigation techniques for near-term devices, which reduce statistical errors but do not directly correct readout errors. Prior work on correcting measurement errors within stabilizer codes includes data-syndrome codes, single-shot QEC, and general fault-tolerant frameworks. However, these techniques typically assume stabilizer structure and encoded data, leaving non-stabilizer/bosonic codes less addressed. The authors also note fundamental measurement theory via Naimark dilation to justify focusing on projective POVMs. This context motivates a code-based, measurement-level error correction applicable to arbitrary projective POVMs and non-stabilizer codes.

Core construction: Given a projective POVM P = {P1,..., PM} and a classical q-ary code C with M distinct codewords xk ∈ {0,...,q−1}^n, define n commuting q-observables Qj(C,P) = Σ_{k=1}^M x_k^{(j)} P_k for j=1,...,n. Because the projectors are orthogonal, the Qj commute. The measurement of all Qj yields a classical vector y ∈ {0,...,q−1}^n and a post-measurement state τ. Consistency and error correction: In the noiseless setting, the measurement outcomes form the codeword x_k associated with the projector P_k, making the Q-observables measurement equivalent to measuring P. With errors on at most t components of y, if C has minimum distance d(C) ≥ 2t+1, a decoder D of C maps y back to k, ensuring τ = P_{f(y)}. Theorem 1 formalizes this: given a t(C)-decoder, the scheme recovers the correct projector index despite up to t(C) classical outcome errors. Example: Using the shortened Hamming code C6 (q=2, n=6, d=3, M=8), the authors explicitly construct six binary observables Q1,...,Q6 as sums of subsets of the P_k, illustrating redundancy that enables single-error correction on classical outcomes. Combinatorial bounds: Define n_q(M,d) as the smallest length n for a q-ary code with ≥ M codewords and distance ≥ d. Corollary 2 shows the minimal number of q-observables needed to correct t errors when measuring a POVM with M projectors is n_t^{qM} = n_q(M, 2t+1). Bounds relate to A_q(n,d) and standard coding bounds (Hamming, Gilbert–Varshamov, Johnson, linear programming). For large M, q^n / V_{q,t}(2t) ≤ M ≤ q^n / V_{q,t}(t). Applications to QEC (Knill–Laflamme recovery): Any QEC recovery can be cast as measuring a projective POVM Π′ labeling correctible error subspaces, followed by a conditional unitary. For a code correcting a set Λ of errors, |Π′| ≤ |Λ|+1. The scheme replaces direct Π′ measurement by measuring n observables derived from a classical code with sufficient distance, enabling correction of t readout errors on the syndrome without encoding data in an auxiliary QEC. Implementation 1 (dispersive coupling with bosonic modes): Couple the system to n ancillary bosonic modes initialized in coherent states |α_j⟩ via W_j = γ Q_j n̂_j. Evolution e^{-i W_j τ} imprints eigenvalue-dependent phase-space rotations distinguishable by homodyne detection, projecting nondestructively onto eigenspaces of Q_j. Repeating for all j yields the outcome vector z; classical decoding corrects up to t errors. Implementation 2 (robust stabilizer measurements for qubits): For binary observables on an m-qubit system, extend the last qubit to a four-level system. Construct a unitary U_j mapping eigenstates of Q_j to computational basis states of the extended system so that measuring the first qubit in {|0⟩,|1⟩} reports the eigenvalue. Applying U_j, measuring, and undoing U_j implements Q_j. This enables measuring the set Q1,...,Qn in a setting where measurement errors dominate over gate errors. Performance modeling and comparison: Baseline implements a 16-outcome POVM via four binary observables, with overall error 1−(1−p)^4 if each has error rate p. A repetition scheme measures each observable thrice and majority-votes, giving per-observable error p1 = 3 p^2 (1−p) + p^3 and overall error 1−(1−p1)^4 using 12 measurements. The proposed protocol uses n=8 (t=1) or n=12 (t=2) binary observables constructed via a code; model per-measurement error as p+(j−1) r to capture degradation. Compute probabilities q0 (no errors), q1 (one error), and q2 (two errors) to estimate residual error as 1−q0−q1 (t=1) or 1−q0−q1−q2 (t=2), and compare numerically (Figs. 3–4).

• General scheme: For any projective POVM P and q-ary code C with distance d≥2t+1, measuring the n=q-observables Q_j(C,P) yields the same outcome as P in the noiseless case and corrects up to t errors on the classical outcomes via a classical decoder (Theorem 1).
• Minimal number of observables: The least n required equals n_q(M, 2t+1), where M is the number of projectors in P (Corollary 2). For large parameters, classical coding bounds give q^n / V_{q,t}(2t) ≤ M ≤ q^n / V_{q,t}(t).
• Concrete example: With M=8 and the shortened Hamming code C6 [n=6, d=3], six binary observables suffice to correct one classical readout error.
• QEC application: For the optimal non-additive 9-qubit code (distance 3) with |Π′| ≤ 29, the scheme uses 10 binary observables to correct one measurement error, compared with 15 observables for a triple-repetition approach of a 5-observable noiseless decoder. For a 17-qubit distance-3 surface code with Π′ ≈ 53, selecting codewords from a [12,8,3] code yields 12 binary observables to correct one measurement error, fewer than data-syndrome approaches (e.g., [21,16,3] implies 21 observables).
• Performance vs repetition: For a 16-outcome POVM measured via four binaries, baseline error is 1−(1−p)^4. Majority-of-3 repetition per observable gives p1=3p^2(1−p)+p^3 and overall error 1−(1−p1)^4 using 12 measurements. The proposed protocol with n=8 (t=1) slightly underperforms the 12-measurement repetition when measurement degradation r>0 (Fig. 3). For t=2 (n=12), the proposed protocol can outperform the 12-measurement repetition for sufficiently small r and moderate p (Fig. 4).
• Implementations: Feasible routes via dispersive coupling to bosonic modes with homodyne detection, and via unitary control with a four-level extension for qubit stabilizer contexts, enabling nondestructive eigenvalue readout.

The scheme addresses the central question of correcting classical readout errors in quantum measurements without encoding data into a QEC code. By mapping projector labels through a classical error-correcting code and measuring commuting observables derived from the POVM and code, the method ensures equivalence to the target projective measurement in the noiseless case and enables deterministic correction of up to t outcome errors using standard decoding. This unifies robust measurement design with classical coding theory and applies to any projective POVM, thereby enabling robust syndrome extraction even for non-stabilizer and bosonic codes. Applications show potential reductions in the number of required measurements relative to repetition or data-syndrome approaches, particularly when measurement degradation is small. The method complements error mitigation techniques and could enhance near-term algorithms sensitive to readout errors, as well as inform measurement design in fault-tolerant architectures for non-stabilizer codes.

The paper introduces a general, code-inspired construction of commuting observables that implement any projective POVM robustly against classical outcome errors. The minimal number of observables equals the shortest length of a q-ary code with ≥ M codewords and distance ≥ 2t+1, connecting the design directly to classical coding bounds. Practical implementations are outlined via dispersive couplings to bosonic modes and via controlled unitaries with an extended-level system for qubits. Applications to QEC, including non-stabilizer and bosonic codes, show resource reductions over repetition and, in some regimes, over data-syndrome approaches. Future work includes evaluating the scheme in realistic experimental settings and extending it to broader families of non-stabilizer codes such as concatenated cat codes, rotation- and permutation-invariant codes, codeword-stabilized, error-avoiding, and ground-space codes.

• Focus on correcting classical outcome errors; corruption of the post-measurement quantum state is not directly addressed by this scheme.
• Resource overhead: Requires measuring n commuting observables; although often fewer than repetition or some data-syndrome schemes, n may still be substantial depending on M, q, and t.
• Performance depends on measurement degradation across repeated measurements (modeled by r); advantages over repetition may diminish when r is large or p is very small (as shown in Figs. 3–4).
• Practical requirements: Assumes access to dispersive couplings and high-fidelity homodyne detection or precise unitary control with state space extension, and that measurement errors dominate over gate errors in some implementations.
• Relies on availability and efficient decoding of suitable classical codes with required length and distance; suboptimal code choices can increase n.
• No experimental demonstration; results are theoretical with numerical comparisons under simplified error models.