Due to increasing diversity and complexity of applications in embedded systems, accelerator designs trading-off area/energy-efficiency and design-productivity are becoming a further crucial issue. Targeting applications in the category of Recognition, Mining, and Synthesis (RMS), this study proposes a novel accelerator design to achieve a good trade-off in efficiency and design-productivity (or reusability) by introducing a new computing paradigm called “approximate computing” (AC). Leveraging from the facts that frequently executed parts of applications (i.e., hotspots) are conventionally the target of acceleration and that RMS applications are error-tolerant and often take similar input data repeatedly, our proposed accelerator reuses previous computational results of similar enough data to reduce computations. The proposed accelerator is composed of a simple controller and a dedicated memory to store limited sets of previous input data with corresponding computational results in a hotspot. Therefore, this accelerator can be applied to different and/or multiple hotspots/applications only through small extension of the controller, to achieve efficient accelerator design and resolve the design-productivity issue. We conducted quantitative evaluations using a representative RMS application (image compression) to demonstrate the effectiveness of our method over conventional ones with precise computing. Moreover, we provide important findings on parameter exploration for our accelerator design, offering a wider applicability of our accelerator to other applications.

The write-margin of the iterative decoding systems for the binary and non-binary low-density parity-check (LDPC) codes are studied in the magnetic recording system using a bit-patterned medium (BPM) R/W channel with write-errors caused by the write-head field gradient, the media switching field distribution (SFD), the magnetization fields from adjacent islands, and the island position deviation. The performances of iterative decoding systems using the binary and non-binary LDPC codes are evaluated by computer simulation at an areal recording density of 2 Tb/inch2. The results show that the non-binary LDPC coding and iterative decoding system improves the write-margin compared with the binary LDPC system. Furthermore it is clarified that in order to obtain the sufficient write-margin, it is important to suppress the influence of the demagnetization fields from adjacent islands.

-In this paper a simple expression is derived for the formula of variable quantizing step size. This formula can be used to improve the following ability of the discrete adaptive delta modulation codec. Then it is shown that the previously published formulas for video signal can be reduced to this expression. Next, clarifying the relation between the first-order difference\xiof input signal and the mechanism of appearance of the slope overload noise by using this expression, the quantizing noise is classified into granular noise, the first class slope overload noise and the second class slope overload noise. Then these noises are simply expressed as a function of\xi. Finally, using this result, the signal-to-noise ratios for sinusoidal input and Gaussian input are obtained and in consequence the utility of our simple analytical method is confirmed.

An analysis is presented on the output signal-to-noise ratios for the FM correlation systems having an FM detector in each input channel of a conventional correlator. The input consists of a frequency-modulated signal combined additively with stationary narrow-band Gaussian random noise. A general expression is derived for the output signal-to-noise ratio. A detailed calculation is made for the output signal-to-noise ratios when each input signal is a carrier frequency-modulated by a Gaussian random process and the integrating filter is of RC low pass. The dependence of the output signal-to-noise ratios on several parameters is discussed.

By the term "FM correlation system," we mean the analog correlation system that consists of two channels, each containing an FM detector. Expressions are derived for the output of the FM correlation system for the following cases. Each channel input to be correlated is a conventional FM signal plus Gaussian noise; and each input is an amplitude-limited or a nonlimited narrow-band Gaussian random signal plus Gaussian noise. The effect of the input carrier-to-noise power ratios is made clear. The features of the FM correlation system are shown in comparison with a conventional correlation system. In addition, a detailed analysis of the effect of the envelopes of the narrow-band Gaussian random signals on the output is presented.