A quiet amplifier at one frequency may be costly across a band

A preprint submitted on October 13 analyzes the noise cost of local reservoir couplings in directional amplifier chains. It compares fixed designs over a frequency band while matching signal response, ports and complete rate budgets, and gives conditional bounds for specified chain families. The microwave array described in the paper is proposed, not reported as built. The result addresses a practical difference between attaining a minimum at one frequency and constructing one physical set of couplings that performs well across the entire band needed by a measurement.^[1]

An amplifier must do more than make a signal larger. It also adds fluctuations, and those fluctuations can obscure the information in a weak input. Two devices with the same gain can therefore have different usefulness for readout. The signal response describes how an input propagates to the output; the noise description also depends on environmental fluctuations and their correlations. Matching the first does not automatically match the second. That is why a fair noise comparison specifies the relevant response and ports instead of allowing one design to change the measurement task.

Frequency creates another constraint. A laboratory device normally has one set of couplings while the input occupies a range of frequencies. A mathematical optimization that chooses a different reservoir independently at every frequency can provide a useful lower reference, but it may not describe one implementable device. The physical design has to compromise across the band. The resulting optimum depends on the weighting assigned to frequencies, the required gain range and which couplings can actually be built or tuned simultaneously.

An original three-frequency example makes the distinction visible. Suppose design A has noise values one, three and three in arbitrary common units, while design B has two, two and two. A wins at the first frequency. Their equally weighted averages are about 2.33 and 2.00, so B wins across this illustrative band. No experimental values are used. A different weighting could change the preferred design, which is precisely why the chosen band objective should accompany any statement that one amplifier is quieter than another.

Correlated environmental inputs can interfere in their contribution to the output. A cancellation arranged at one phase relation can weaken when propagation changes that relation with frequency. This is a general reason that a sharp optimum may not extend across a band. In a quantum device, the permitted correlations also have to satisfy physical constraints; one cannot assign arbitrary fluctuations solely to cancel a measured output. A credible reservoir construction supplies a physically consistent noise model along with the favorable signal calculation.

Locality limits which parts of the chain share those environmental couplings. Extending their range can add flexibility, but it can also require additional wiring, auxiliary modes or controlled interactions. Increasing rates can supply another resource. A comparison that grants one design stronger or more numerous couplings may find a real improvement without isolating the value of nonlocality itself. Holding a complete resource budget fixed makes the question sharper: which spatial arrangement of the same allowed resources gives the useful broadband noise reduction?

Longer chains do not automatically remove a boundary limitation. If the measured output retains a significant sensitivity to fluctuations introduced near a particular port, adding remote elements may leave that contribution largely unchanged. Whether that occurs depends on the response structure, so it should be demonstrated for the chosen family rather than generalized to every amplifier. The engineering implication is to identify where the limiting noise enters and how it reaches the output. More components help only if they change that limiting route under the permitted constraints.

The next useful experimental comparison would hold the response, frequency band and total resources fixed while changing the reservoir arrangement, then measure the corresponding input-referred noise with uncertainty. That would connect the conditional bounds to a physical design choice. Readers should expect the same discipline in any broadband claim: one realizable setting, a defined band objective and a complete accounting of the resources that made the result possible. A single quiet point remains valuable, but it answers a narrower question than reliable low-noise performance across a working measurement band.