The laws of quantum mechanics pose stringent constraints on the amplification of a quantum signal. Deterministic amplification of an unknown quantum state always implies the addition of a minimal amount of noise. In principle, linear and noiseless amplification is allowed provided it works only probabilistically [1,2].
The state comparison amplifier  is an approximate probabilistic amplifier that amplifies a coherent state chosen at random from a set of coherent states with known mean photon number. The amplification process works as follows: Alice picks uniformly at random an input state and passes it to Bob. He desires to amplify the state so he mixes it with a guess coherent state at a beam splitter in an attempt to achieve destructive interference in one of the output arms. This output is fed into an APD detector.
The lack of trigger at the detector is an imperfect indication that Bob’s guess is right and that the output contains the correct amplified state. On the other hand, if the first detector fires Bob knows that his guess was wrong but he can still correct the output by changing the input state for a second amplification stage via a feed-forward loop.
In summary, Bob declares success when both the detectors do not fire or when the first detector does fire and state correction is performed. We generalize this mechanism for an arbitrary number of input states and beam splitters, using an on-line learning strategy based on maximum a posteriori probability.
The success probability-fidelity product  of the SCAMP is the joint probability of success and of passing a measurement test on the output comparing it to right amplified state.
Our figures of merit compare favorably with other schemes. The success probability-fidelity product of the SCAMP is always bigger than that of a USD based amplifier  that, when inconclusive, delivers a conveniently chosen random output.
The SCAMP can be realized with classical resources (i.e., lasers, linear optics and APD detectors), the ability to switch between input states on the fly requires delay lines and fast switching but it can still be achieved with classical resources and the loss introduced by the delay can be offset at the second stage. Similar systems, with no state correction, proved to achieve high-gain, high fidelity and high repetition rates, e.g. [4, 5].
Due to its simplicity, the system we propose might represent an ideal candidate either as a recovery station to counteract quantum signal degradation due to propagation in a lossy fibre or across the turbulent atmosphere or as a quantum receiver to improve the key-rate of continuous-variable quantum key distribution with discrete modulation. The system is also suitable for on-chip implementation.
 T.C. Ralph & A.P. Lund, Proceedings of the 9th QCMC Conference 2009.
 S. Pandey, et al., Phys. Rev. A 88, 033852 (2013).
 E. Eleftheriadou et al., Phys. Rev. Lett. 111, 213601 (2013).
 R. Donaldson et al., Phys. Rev. Lett. 114, 120505 (2015).
 R. Donaldson et al., in preparation.