An Energy-Efficient Approximate Floating-Point Multipliers for Wireless Communications

Approximate computing has been introduced to reduce the circuit area and power consumption in error-tolerant applications. The wireless communication system can inherently tolerate certain errors at the receiving end due to the noise introduced in the channel. This paper presents an energy-efficient approximate floating-point (FP) multiplier for 5G wireless communication systems. Taking into account the data distribution characteristic in wireless communication systems, a low-complexity accurate mantissa multiplier with the subnormal number approximated to zero is proposed. Based on the proposed mantissa multiplier, a low-power half-precision FP multiplier is proposed with low-weight partial products truncating and the probability compensation. Simulation results show that the truncating 10 columns partial products design reduces area by 66.86%, delay by 57.42%, and power consumption by 65.95% with 99.95% accuracy (MRED is 0.05%). Applying the designed approximate FP multiplier to a beam weight transformation module in 5G communication systems, the deterioration of the bit error rate is less than 0.2dB.