Quantum measurement · Link budget
Quantum measurement links
Loss accounting & verification
Squeezed light · Atomic spin · Optomechanics · Superconducting
The same dB number, asked differently
They report at the source; we report at the detector. They give a verdict; we give a ledger you can recompute.
| The same question | Common answer | Our answer |
|---|---|---|
| How much did we get? | dB of squeezing at the source | dB surviving at the detector |
| Is it enough? | Compared with the theoretical limit | Compared with this link's loss |
| Where is the gap? | Raise the squeezing | Which loss term to fix first |
| What do we get? | A conclusion | A ledger with a reproduction pack |
| When it cannot be computed | An optimistic range | “Not applicable”, with the reason |
This way of asking does not rank technology routes and does not predict device performance. It answers one question: given the numbers you report, what should come out.
Skip the ledger, and the error outgrows the design
Squeezing leaves the source and passes optical components, mode matching, fibre, end faces and the detector. Each stage takes a cut. The cut is not linear: the more loss, the less each remaining dB is worth. Three numbers make the point.
Measure → compute → judge
A table squares the books, a set of criteria gives the ruling, a certificate lets someone else check it.
Put the losses on the table
Input squeezing, mode mismatch, aperture, end faces, detection efficiency — item by item. Terms that geometry can compute need no measurement.
Carry the books to the detector
Convert each item to a share, sum, then convert back to effective squeezing. Along the way, the floor of this link and the distance to it.
Deliver the ruling and the ranking
Whether it is enough, which term to fix, what total loss the target requires. The conclusion lands as a certificate with a reproduction pack.
Six questions, each with a definite answer
How many dB are left
Given the input and the loss items, what is the effective squeezing at the detector — not the nominal figure at the source.
Which term to fix
Ranked by recoverable amount: spend on coating, on the detector, or stop touching the geometry.
How much loss is allowed
For 9 dB, total loss must fall below 2.88%. That sentence can go straight into a requirements document.
How tight the tolerances are
Offset, angle, aperture ratio, facet reflection — each limit, and how to split the budget when three combine.
Can someone else recompute it
Data, formulas, deviations, boundaries, reproduction pack, three levels of conformity. If it cannot be recomputed, it does not count.
Which terms need no measurement
Mode mismatch, finite aperture and facet coupling can be computed from geometry directly.
Drag two parameters, see what is left
Defaults are 10 dB in, 30% loss — an ordinary working point.
Checked against published data
Three cases: two close, one does not — and the one that does not marks the edge of the method.
| Case | Input | Observed | Inferred | Conclusion |
|---|---|---|---|---|
| LIGO O3 squeezed-light injection | 6.0 dB | 3.0 dB | L = 0.334 | Closes (consistent with ~30% link loss) |
| Squeezed-light source link (Vahlbruch 2016) | source 23.35 dB | 15.3 dB | η = 0.975 | Closes; three further independent matches (contrast, anti-squeezing, loss table) |
| SiN micro-ring + on-chip gain | not given | 4.6 dB | negative | Not applicable (pre-amplification breaks the expression) |
The third case is not a computational error; its preconditions fail. The amplifier sits before the loss, the noise floor is raised, and the original transfer expression no longer describes that link. Hence one clause in the specification: a pre-amplifier must be reported with its gain and position.
Six things you can take now
The specification sets convention, the budget sheet the arithmetic, the certificate the format, the sample a reference.
Who these are for
Before design freeze
To split the budget and set tolerances. The sheet shows the total and the bottleneck in one pass.
Someone hands you a number
You cannot simply believe it, and cannot compute it yourself. The certificate is decidable and states its boundaries.
A referee asks about loss
You want to know whether your number holds under someone else's convention. We recompute with the same expressions.
What we do not do
Half the value of a method comes from the cases it refuses.
Gaussian states, linear loss, single-parameter estimation only
Non-Gaussian states with photon counting, and joint multi-parameter estimation, are out of scope. Forcing them in gives wrong answers.
We do not replace metrology
We compute what should follow from the numbers you report — not what your device actually is. That requires measurement.
We do not predict device performance
Material absorption, detector response and the like are inputs. We do not estimate them for you.
We do not recommend technology routes
We account costs and benefits. Whether to take the route stays your call.
Contact: sunke@xingyimeiya.cn