Homodyne detection is considered as a way to improve the efficiency of communication near the single-photon level. The current lack of commercially available infrared photon-number detectors significantly reduces the mutual information accessible in such a communication channel. We consider simulating direct detection via homodyne detection. We find that our particular simulated direct detection strategy could provide limited improvement in the classical information transfer. However, we argue that homodyne detectors (and a polynomial number of linear optical elements) cannot simulate photocounters arbitrarily well, since otherwise the exponential gap between quantum and classical computers would vanish.PACS numbers: 03.67. Lx, 42.50.Dv, 89.70.+c The fundamental limitations to classical communication in optical channels are due to the quantum nature of the signals being transmitted. These limitations have been well understood for ideal optical communication channels [1,2]. The capacity of a communication channel is defined to be the maximum mutual information (optimized over the choice of source alphabet used by the sender and the detection strategy used by the receiver) across the communication channel with respect to some physical channel constraint -such as the mean energy throughput. This characterization of a communication channel is important because Shannon's noisy coding theorem proves that any attempt at communication beyond this capacity necessarily fails due to unrecoverable errors [3]. A corollary to this theorem states that even for a non-optimal source alphabet or detection strategy, the mutual information of the communication channel is (asymptotically) achievable using error correction [3,4].For a single-mode optical communication channel the optimal capacity, under a mean energy constraint, is achieved with a source alphabet of photon-number states and ideal photon-number detectors [1,2]. In this ideal case the orthogonality of the signals and hence their perfect distinguishability makes error correction unnecessary. Unfortunately, however, such ideal operation is currently impractical, since neither ideal photon-number state preparation nor ideal photon-number detection is achievable.The detection of weak signals (few photons) is especially difficult at communications wavelengths (1.3 − 1.55µm). Ideally, we would wish to achieve this by simply counting the photons. Now the process of photocounting is often synonymous with using an avalanching device with saturated gain, since each photon produces a strong and standard signal at the output. Unfortunately, it is a technological fact that both at infrared and optical frequencies, the best avalanche photodiodes never have as high a quantum efficiency as the best available linear detectors (having linear gain, such as PINphotodiodes). In fact, currently, no commercial photocounters are available at communications frequencies. Thus, at these frequencies PIN-photodiodes are routinely used despite their high dark count rates. Partly because of this,...