Semiparametric Estimation of a Noise Model with Quantization Errors

The detectors in mass spectrometers are precise enough to count ion events. In practice, the statistics of chemical noise are affected by large quantization errors and overdispersion because of amplification in the detector. The detector signal is modelled as X =floor(t N) where N represents integer-valued ion counts and t represents the gain parameter. When t <= 1, the gain parameter cannot be recovered without a priori information on N. When t > 1 however, we introduce compatible lattices and derive an estimator for t that is optimal, independent of N and enables subsequent analyses of the ion counts.

💡 Research Summary

The paper addresses a fundamental problem in mass‑spectrometry data analysis: the detector signal is subject to both amplification‑induced overdispersion and quantization error. The authors model the observed signal as
(X = \lfloor t N \rfloor),
where (N) denotes the true integer‑valued ion count and (t) is the gain (amplification) parameter. When the gain is less than or equal to one ((t \le 1)), the floor operation destroys information about (t) unless additional prior knowledge about the distribution of (N) is available. The novel contribution of the work lies in showing that for gains greater than one ((t > 1)) the problem becomes tractable without any assumptions on the distribution of (N).

The authors introduce the concept of “compatible lattices.” By expressing the gain as a reduced fraction (t = p/q) (with coprime integers (p) and (q)), the relationship between the observed quantized values and the underlying counts can be written as
(q X \le p N < q X + q).
Consequently, the set of possible observations (X) forms a lattice with spacing (q). This lattice structure provides a sufficient statistic for (t) that is independent of the unknown distribution of (N).

Building on this insight, the paper derives an explicit estimator for (t). The estimator exploits differences between successive observations, (\Delta X_k = X_{k+1} - X_k), and uses integer‑theoretic operations—specifically the least common multiple (LCM) of the observed differences and the greatest common divisor (GCD) of the implied count differences—to construct
\