Elsevier

Integration

Volume 36, Issues 1–2, September 2003, Pages 41-54
Integration

Low-power VLSI synthesis of DSP systems

https://doi.org/10.1016/S0167-9260(03)00031-2Get rights and content

Abstract

In recent years, power consumption has become a critical design concern for many VLSI systems. Nowhere is this true expect for portable, battery-operated applications, where power consumption has perhaps superceded speed and area, which are the overriding implementation constraints. This adds another degree of freedom and complexity to the design process, and mandates the need for design techniques that address power, as well as area and speed. This paper proposes several low-power synthesis techniques for VLSI implementation of DSP systems both at the algorithmic and architectural levels. Low-power implementation of the Viterbi decoder using register-exchange is proposed. Conceptually, the register-exchange method is simpler and faster than the trace-back method. Reduction in switching activity at one or both inputs of the multipliers is a key to the reduction of power consumption in FIR filters. The switching activity can be reduced by the use of a transpose structure and by time multiplexing of an unfolded filter. The retiming approach is then used to facilitate the constrained operational schedule of different functional units. In the proposed techniques, about 20% in the overall implementation of Viterbi decoder and up to 80% in the implementation of folded FIR filters saving in power consumption has been reported.

Introduction

Recently, reduction of power consumption has become a very critical issue in the design of high-performance VLSI synthesis of DSP systems. The techniques used to achieve reduced power consumption range from algorithmic and architectural levels to gate, switch and device levels. In this paper, we propose low-power algorithmic and architectural design methodologies for several signal-processing systems. There are several approaches for power reduction in high-speed VLSI implementations. Pipelining and parallel processing can be combined for lower power consumption; where pipelining reduces the capacitance to be charged/discharged in one clock period, while parallel processing increases the clock period for charging/discharging the original capacitance. Both pipelining and parallelism can therefore be used to reduce power consumption by reducing the supply voltage. Power consumption can be reduced by reducing the number of gates or by algorithmic strength reduction, where the number of operations in an algorithm is reduced. Reducing the memory accesses can also reduce power consumption. The most effective clock gating technique of reducing power consumption switches-off the functional units that do not compute any useful outputs by using gated clocks [1]. The power reduction approaches proposed in this paper apply to dedicated or FPGA systems.

Designing low-power decoders has become crucial in modern spread-spectrum wireless communication systems. Code division multiple access (CDMA) spread-spectrum technology was introduced in the second generation of mobile systems, and was shown to be the technology for third-generation mobile systems [2]. In CDMA systems, the Viterbi decoder (VD) consumes about one-third of the power dissipation, which is used to optimally decode convolutional codes. In this work, the algorithmic approach is adopted to design a low-power VD with the focus on minimizing the number of operations [3], [4]. Usually, two methods are adopted to realize the Viterbi algorithm; the register-exchange (RE) and the trace-back (TB) method. In the literature, the RE technique is acceptable only for small state trellis structures, whereas the TB approach is acceptable for large state trellis structures (the number of states in CDMA systems is 256). In Section II of this paper, a novel approach to implement the RE method is proposed, which makes it feasible for use with large constraint length VDs, in addition to being simpler and lower in power consumption than the TB method.

FIR filtering is another important operation in DSP systems. In high-performance digital VLSI implementations, FIR filters often require long lengths or large number of taps. An implementation, where every multiply-add operation is mapped to a multiplier-adder in hardware, becomes prohibitively area expensive. Therefore, for area-efficient FIR filters, folding or time multiplexing is used to map multiple algorithm operations (multiplication and addition operations) on a single functional unit, resulting in saving of the silicon area. Unfortunately, maximal resource sharing while minimizing area can lead to a large increase in power consumption. In Section III of the paper, we propose a low-power technique for designing folded FIR filters where storage area is traded-off for reducing power consumption. The paper is concluded in Section IV.

Section snippets

Low power design of Viterbi decoder

In 1967, Viterbi developed the Viterbi algorithm (VA) to decode the convolutional codes to the optimum level [5], [6]. A simple convolutional encoder with K=3 with code rate 1/3, and the generator polynomials G0=58, G1=78 and G2=78, has the trellis diagram shown in Fig. 1. A sample input data to the encoder are tracked until they are decoded at the decoder. The encoder encodes the input sequence (1011010) and generates the codeword (111,011,000,100,100,000,011). This stream is transmitted over

Low power folded FIR filter

Folding [12] is a technique to reduce silicon area by time-multiplexing many algorithm operations into single functional units (such as adders and multipliers). This technique ensures efficient resource sharing for area-constrained behavioral synthesis from a data flow graph (DFG). The throughput requirement in folded architectures is achieved by pipelining the hardware functional units to an appropriate level. In this manner, folded architectures can be used to meet both throughput and

Conclusion

Several power-reduction approaches are proposed in the high-level VLSI synthesis of DSP systems. An algorithmic approach is proposed with the aim of minimizing the number of algorithm operations to reduce power consumption in VLSI implementation of Vietrbi decoder. It has been shown that appropriate unfolding and transpose forms of FIR filter structures can lead to reduction of power by reducing switching activity. The technique employed gives a formal approach to an intuitive notion that

References (21)

  • K.K Parhi

    VLSI Digital Signal Processing Systems

    (1999)
  • I Kang et al.

    Low-power Viterbi decoder for CDMA mobile terminals

    IEEE J. Sel. Areas Commun.

    (1998)
  • A.P Chandrakarsan et al.

    Low Power Digital CMOS Design

    (1995)
  • A Bellaouar et al.

    Low Power Digital VLSI Design

    (1995)
  • A.J Viterbi

    Error bounds for convoltuional codes and asymptotically optimum decoding algorithm

    IEEE Trans. Inf. Technol.

    (1967)
  • G.D Forney

    The Viterbi algorithm

    Proc. IEEE

    (1973)
  • H.L. Lou, Viterbi decoder design for the IS-95 CDMA forward link, Proceedings of VTC, Atlanta, USA, April 1996, pp....
  • S.B Wicker

    Error Control Systems for Digital Communication and Storage

    (1995)
  • J.A. Ryu, S.C. Kim, J.D. Cho, H.W. Park, Y.H. Chang, Lower power Viterbi decoder architecture with a new clock-gating...
  • G Fettweis et al.

    High-rate Viterbi processor: a systolic array solution

    IEEE J. Sel. Areas Commun.

    (1990)
There are more references available in the full text version of this article.

Cited by (2)

  • Variable wordlength soft-decision Viterbi decoder for power-efficient wireless LAN

    2012, Integration, the VLSI Journal
    Citation Excerpt :

    However, the conventional Viterbi decoder uses significant amounts of power to run the large number of operations involved in the decoding process. Various power-reduction approaches for Viterbi decoders have been presented in [1–3]. Soft-decision Viterbi decoding is an attractive approach for OFDM.

  • FPGA based design of Reed Solomon codes

    2009, Indian Journal of Science and Technology
View full text