We have investigated the noise characteristics in the intensity-modulated and phase-modulated signals and estimated how much the single-page storage density and data transfer rate increase by using the multivalued signals. In our simulation, the diffracted image was calculated considering two important aspects in actual holographic data storage systems. One is the aperture inserted in the signal optical path, and the other is the oblique hologram shape formed in the recording medium. Our numerical simulation revealed that in the case of the intensity-modulated signal, the single-page storage density was almost unchanged even if the number of multivalued signals was increased. In contrast, when an 8-level phase multivalued signal was employed in the system, the single-page storage density was estimated to be increased by a factor of 2.4 -3.0 compared to the intensity binary signal.
We investigated the use of a spatial filter and a combination of multiplexing methods to obtain a data recording density of 1 Tbit/inch2. A small aperture diameter of the spatial filter can increase recording density, but the signal-to-noise ratio of the reproduced data decreases. We then investigated the minimum size of the aperture diameter under the condition where the data could be decoded. As a result, we found that the aperture diameter could be decreased to 1.2 times as large as the Nyquist width of a data page. To obtain a recording density of 1 Tbit/inch2, 800-data-page multiplexing was needed. We then investigated the recording conditions of angle and peristrophic multiplexings. When we recorded and reproduced data pages with 200 angle multiplexings and 4 peristrophic multiplexings, the bit error rates of the data were less than the permissive error rate, and a recording density of 1 Tbit/inch2 could be obtained.
In this study, we propose an effective data-decoding method for holographic data storage (HDS) by combining convolutional neural network (CNN) and spatially coupled low-density parity-check (SC-LDPC) code. The trained CNN provides output class probabilities and accurately demodulates the reproduced data from HDS. We focus on these probabilities, wherein only the untrainable noise components such as white Gaussian noise remain. These are used for calculating the log likelihood ratio in the sum-product decoding for the SC-LDPC code. We demonstrate an improvement of approximately 10 dB in the required signal-to-noise ratio for an error-free decoding in numerical simulations.
This paper develops a system to visually inspect cutlery based on a simple machine learning algorithm using image features that are robust against overexposure. First, we develop an image acquisition apparatus comprising a laser and a screen that produces speckle images of unique shapes depending on the degree to which the photographed cutlery has been polished. The contribution of this study is to produce speckle images in this way. This enables accurate classification without newly deriving a sophisticated machine learning algorithm in the subsequent processing. We use the speckle images to develop moment-related features that represent the unique shapes and avoid the problem of overexposure. Second, we apply the extreme learning machine, a simple but representative machine learning algorithm, to the obtained features. Experimental results using real cutlery show that our developed system achieved good accuracy and precision regardless of exposure time.
We are developing an advanced Integrated Services Digital Broadcasting-Terrestrial (ISDB-T) system for the next generation of digital terrestrial television broadcasting. The advanced ISDB-T provides 4K8K terrestrial broadcasting service for fixed reception and 2K service for mobile reception simultaneously within one channel. New technologies such as Low-Density Parity-Check (LDPC) code are used for expanding the transmission capacity and for improving the spectral efficiency. The LDPC codes designed for the advanced ISDB-T have two code lengths and 13 code rates for each code length. The code length and code rate can be selected in consideration of the transmission latency requirement or the link budget. Meanwhile, although the LDPC codes have good bit error rate (BER) performance approaching the Shannon limit, a small number of error bits cause an error floor even if the Es/N0 is high enough. The error floor may cause serious issues such as block noise in video and mute in audio because broadcasting is a real-time service without any feedback. To deal with this problem, Bose-Chaudhuri-Hocquenghem (BCH) code is concatenated as outer code to the LDPC codes as inner codes. We conducted a simulation using a field programmable gate array (FPGA) instead of a computer to evaluate the BER performance. An FGPA simulation is 1000 times faster than a computer simulation, so the BER performance can be evaluated quickly with an adequate number of measurement bits. As a result, it was confirmed that LDPC codes perform as designed both in the water-fall and error-floor regions and that the BCH codes correct the small number of error bits after the LDPC decoder.