Leveraging Power Spectral Density for Scalable System-Level Accuracy Evaluation

Barrois, Benjamin; Parashar, Karthick; Sentieys, Olivier

doi:10.3850/9783981537079_0204

Cited by 3 publications

(2 citation statements)

References 10 publications

(14 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This accuracy can be evaluated through the bounds of the quantisation errors [114,115], the number of significant bits [116], or the power of the quantisation noise [117,118,119]. The shape of the power spectral density (PSD) of the quantisation noise is used as metric in [120] or in [121] for the case of digital filters.…”

Section: Analytical Approachesmentioning

confidence: 99%

Fixed-point refinement of digital signal processing systems

Ménard

Caffarena

López

et al. 2019

Digitally Enhanced Mixed Signal Systems

Self Cite

View full text Add to dashboard Cite

For the digital signal processing (DSP) part of mixed-signal systems, fixed-point arithmetic is favoured due to its high efficiency in terms of implementation cost, namely energy consumption, area, and latency. Fixed-point arithmetic provides low-cost operators at the expense of an unavoidable error between the ideal precision values and the finite precision ones. With a careful data word-length optimisation process, the designer can safely profit from the inherent trade-off between implementation cost and computation accuracy. In this chapter, we describe in detail the different stages comprising the fixed-point conversion process (i.e., dynamic range evaluation, word-length optimisation, and accuracy evaluation) and propose practical solutions to address each stage.

show abstract

Section: Analytical Approachesmentioning

confidence: 99%

Fixed-point refinement of digital signal processing systems

Ménard

Caffarena

López

et al. 2019

Digitally Enhanced Mixed Signal Systems

Self Cite

View full text Add to dashboard Cite

show abstract

“…This accuracy can be evaluated through the bounds of the quantization errors [43,2], the number of significant bits [24], or the power of the quantization noise [102,126,18]. The shape of the power spectral density (PSD) of the quantization noise is used as metric in [7] or in [31] for the case of digital filters. In [20], a more complex metric able to handle several models is proposed.…”

Section: Error Metricsmentioning

confidence: 99%

Analysis of Finite Word-Length Effects in Fixed-Point Systems

Ménard

Caffarena

López

et al. 2018

Handbook of Signal Processing Systems

View full text Add to dashboard Cite

Systems based on fixed-point arithmetic, when carefully designed, seem to behave as their infinite precision analogues. Most often, however, this is only a macroscopic impression: finite word-lengths inevitably approximate the reference behavior introducing quantization errors, and confine the macroscopic correspondence to a restricted range of input values. Understanding these differences is crucial to design optimized fixed-point implementations that will behave "as expected" upon deployment. Thus, in this chapter, we survey the main approaches proposed in literature to model the impact of finite precision in fixed-point systems. In particular, we focus on the rounding errors introduced after reducing the number of least-significant bits in signals and coefficients during the so-called quantization process.

show abstract

Approximate FPGA Implementation of CORDIC for Tactile Data Processing Using Speculative Adders

Franceschi¹,

Camus²,

Ibrahim³

et al. 2017

2017 New Generation of CAS (NGCAS)

View full text Add to dashboard Cite

Abstract-In most robotic and biomedical applications, the interest for real-time embedded systems with tactile ability has been growing. For example in prosthetics, a dedicated portable system is needed for developing wearable devices. The main challenges for such systems are low latency, low power consumption and reduced hardware complexity. In order to improve hardware efficiency and reduce power consumption, approximate computing techniques have been assessed. This strategy is suitable for error-tolerant applications involving a large amount of data to be processed, which perfectly fits tactile data processing. This paper presents the first case study of applying Inexact Speculative Adders (ISA) to the FPGA implementation of a Coordinate Rotation Digital Computer (CORDIC) module within the Machine Learning algorithm of a tactile data processing system. The design has been synthesized and implemented on a Xilinx ZYNQ-7000 ZC702 device. Preliminary results have shown dynamic power reduction up to 40 % and delay latency reduction up to 21 % compared to a conventional CORDIC module, at the cost of a negligible average relative error of 0.049 % for sine and 0.003 % for cosine computations.

show abstract

Leveraging Power Spectral Density for Scalable System-Level Accuracy Evaluation

Cited by 3 publications

References 10 publications

Fixed-point refinement of digital signal processing systems

Fixed-point refinement of digital signal processing systems

Analysis of Finite Word-Length Effects in Fixed-Point Systems

Approximate FPGA Implementation of CORDIC for Tactile Data Processing Using Speculative Adders

Contact Info

Product

Resources

About