Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MUSIC-lite: Efficient MUSIC using Approximate Computing: An OFDM Radar Case Study

Rajat Bhattacharjya, Arnab Sarkar, Biswadip Maity, Nikil Dutt

TL;DR

This work addresses the high computational cost of MUSIC, particularly its SVD step, by applying approximate computing to the SVD/CORDIC pathway within an OFDM radar pipeline. The authors introduce MUSIC-lite, a three-stage methodology—Functional Validation, Hardware Implementation, and Design Space Exploration—to identify optimal accuracy-area-power trade-offs using approximate adders in the CORDIC core. Experimental results on a 512-point IFFT within an OFDM radar setup show average on-chip area and power reductions of 17.25% and 19.4%, respectively, with an average end-to-end error of 0.14% in positive $SNR$ scenarios, and highlight design points that meet strict quality constraints. Overall, MUSIC-lite demonstrates a practical route to low-power, end-to-end MUSIC-enabled radar processing, enabling efficient joint radar-communication systems and adaptable hardware configurations for diverse applications.

Abstract

Multiple Signal Classification (MUSIC) is a widely used Direction of Arrival (DoA)/Angle of Arrival (AoA) estimation algorithm applied to various application domains such as autonomous driving, medical imaging, and astronomy. However, MUSIC is computationally expensive and challenging to implement in low-power hardware, requiring exploration of trade-offs between accuracy, cost, and power. We present MUSIC-lite, which exploits approximate computing to generate a design space exploring accuracy-area-power trade-offs. This is specifically applied to the computationally intensive singular value decomposition (SVD) component of the MUSIC algorithm in an orthogonal frequency-division multiplexing (OFDM) radar use case. MUSIC-lite incorporates approximate adders into the iterative CORDIC algorithm that is used for hardware implementation of MUSIC, generating interesting accuracy-area-power trade-offs. Our experiments demonstrate MUSIC-lite's ability to save an average of 17.25% on-chip area and 19.4% power with a minimal 0.14% error for efficient MUSIC implementations.

MUSIC-lite: Efficient MUSIC using Approximate Computing: An OFDM Radar Case Study

TL;DR

This work addresses the high computational cost of MUSIC, particularly its SVD step, by applying approximate computing to the SVD/CORDIC pathway within an OFDM radar pipeline. The authors introduce MUSIC-lite, a three-stage methodology—Functional Validation, Hardware Implementation, and Design Space Exploration—to identify optimal accuracy-area-power trade-offs using approximate adders in the CORDIC core. Experimental results on a 512-point IFFT within an OFDM radar setup show average on-chip area and power reductions of 17.25% and 19.4%, respectively, with an average end-to-end error of 0.14% in positive scenarios, and highlight design points that meet strict quality constraints. Overall, MUSIC-lite demonstrates a practical route to low-power, end-to-end MUSIC-enabled radar processing, enabling efficient joint radar-communication systems and adaptable hardware configurations for diverse applications.

Abstract

Multiple Signal Classification (MUSIC) is a widely used Direction of Arrival (DoA)/Angle of Arrival (AoA) estimation algorithm applied to various application domains such as autonomous driving, medical imaging, and astronomy. However, MUSIC is computationally expensive and challenging to implement in low-power hardware, requiring exploration of trade-offs between accuracy, cost, and power. We present MUSIC-lite, which exploits approximate computing to generate a design space exploring accuracy-area-power trade-offs. This is specifically applied to the computationally intensive singular value decomposition (SVD) component of the MUSIC algorithm in an orthogonal frequency-division multiplexing (OFDM) radar use case. MUSIC-lite incorporates approximate adders into the iterative CORDIC algorithm that is used for hardware implementation of MUSIC, generating interesting accuracy-area-power trade-offs. Our experiments demonstrate MUSIC-lite's ability to save an average of 17.25% on-chip area and 19.4% power with a minimal 0.14% error for efficient MUSIC implementations.
Paper Structure (11 sections, 2 equations, 6 figures, 1 table)

This paper contains 11 sections, 2 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background
  3. MUSIC-lite Methodology
  4. Functional Validation
  5. Hardware Implementation
  6. Design Space Exploration
  7. Evaluation
  8. Accuracy Analysis
  9. Hardware Evaluation
  10. Design Space Exploration
  11. Conclusion and Future Work

Figures (6)

  • Figure 1: Orthogonal frequency-division multiplexing (OFDM) radar processing pipeline. We approximate the computationally intensive MUSIC block highlighted in red.
  • Figure 2: Approximation using MUSIC-lite
  • Figure 3: Accuracy statistics using various adders (accurate and approximate). Range averaged for 100 runs per SNR for each adder.
  • Figure 4: Target's range profile at SNR = 10 dB with accurate adder; and approximate adders add16se_2U6, add16se_2TN.
  • Figure 5: Area and power statistics for CORDIC core using both accurate and approximate adders.
  • ...and 1 more figures