Treffer: An energy-efficient, high-precision SFP LPFIR filter engine for digital hearing aids
CC BY 4.0
Sauf mention contraire ci-dessus, le contenu de cette notice bibliographique peut être utilisé dans le cadre d’une licence CC BY 4.0 Inist-CNRS / Unless otherwise stated above, the content of this bibliographic record may be used under a CC BY 4.0 licence by Inist-CNRS / A menos que se haya señalado antes, el contenido de este registro bibliográfico puede ser utilizado al amparo de una licencia CC BY 4.0 Inist-CNRS
Weitere Informationen
The main contribution of this study is the development of an area-/energy-efficient cascaded direct-truncation (DT) datapath with the so-called static floating-point (SFP) arithmetic to realize a low-delay analysis filter bank (AFB) for digital hearing aids. In the proposed SFP LPFIR (linear-phase finite impulse response) filter engine, lower silicon area and lesser power consumption facilitate better SNR performance than that achieved with the conventional post-truncation (PT) datapath with integer arithmetic. Moreover, in the proposed LPFIR filter engine, a cascaded 16-bit SFP A-M-S-Acc datapath is used that consists of two embedded 1-bit shifters to improve hardware usage and parallelism, one 16-bit DT adder (A), one 16-bit DT multiplier (M), one 16-bit barrel shifter (S), and one 16-bit DT accumulator (Acc). The operations per cycle (OPC) of the proposed SFP LPFIR filter engine reaches 6, which enables efficient fabrication of the low-latency AFB for hearing aids. To verify the effectiveness of the proposed 16-bit SFP LPFIR filter engine, a 10-ms 18-band quasi-ANSI S1.11 1/3-octave AFB for digital hearing aids was implemented using UMC 90-nm CMOS technology. The AFB was operated at 792 kHz to process, in real-time, 24 kHz audio, with the power consumption being approximately 80.6 μW (at 1 V). Compared to the previous design in which the conventional PT datapath with integer arithmetic was used, approximately 9.6% of total power and 8.3% of silicon area were saved and almost the same SNR (signal-to-noise ratio) performance was achieved with the new system, when evaluated by a 3.96-s sequence of Mandarin speech.