Low Power Digital Filter Implementation in FPGA

Abstract

Digital filters suitable for hearing aid application on low power perspective have been developed and implemented in FPGA in this dissertation.
Hearing aids are primarily meant for improving hearing and speech comprehensions. Digital hearing aids score over their analog counterparts. This happens as digital hearing aids provide flexible gain besides facilitating feedback reduction and noise elimination. Recent advances in DSP and Microelectronics have led to the development of superior digital hearing aids. Many researchers have investigated
several algorithms suitable for hearing aid application that demands low noise, feedback cancellation, echo cancellation, etc., however the toughest challenge is the
implementation. Furthermore, the additional constraints are power and area. The device must consume as minimum power as possible to support extended battery life and should be as small as possible for increased portability. In this thesis we have made an attempt to investigate possible digital filter algorithms those are hardware configurable on low power view point.
Suitability of decimation filter for hearing aid application is investigated. In this dissertation decimation filter is implemented using ‘Distributed Arithmetic’ approach.While designing this filter, it is observed that, comb-half band FIR-FIR filter
design uses less hardware compared to the comb-FIR-FIR filter design. The power consumption is also less in case of comb-half band FIR-FIR filter design compared to
the comb-FIR-FIR filter. This filter is implemented in Virtex-II pro board from Xilinx and the resource estimator from the system generator is used to estimate the resources.

However ‘Distributed Arithmetic’ is highly serial in nature and its latency is high; power consumption found is not very low in this type of filter implementation.
So we have proceeded for ‘Adaptive Hearing Aid’ using Booth-Wallace tree multiplier. This algorithm is also implemented in FPGA and power calculation of the whole system is done using Xilinx Xpower analyser. It is observed that power consumed by the hearing aid with Booth-Wallace tree multiplier is less than the hearing aid using Booth multiplier (about 25%). So we can conclude that the hearing aid using Booth-Wallace tree multiplier consumes less power comparatively.
The above two approached are purely algorithmic approach. Next we proceed to combine circuit level VLSI design and with algorithmic approach for further possible reduction in power.
A MAC based FDF-FIR filter (algorithm) that uses dual edge triggered latch (DET) (circuit) is used for hearing aid device. It is observed that DET based MAC FIR filter consumes less power than the traditional (single edge triggered, SET) one (about 41%). The proposed low power latch provides a power saving upto 65% in the FIR filter. This technique consumes less power compared to previous approaches that uses low power technique only at algorithmic abstraction level.
The DET based MAC FIR filter is tested for real-time validation and it is observed that it works perfectly for various signals (speech, music, voice with music). The gain of the filter is tested and is found to be 27 dB (maximum) that matches with most of the hearing aid (manufacturer’s) specifications. Hence it can be concluded that FDF FIR digital filter in conjunction with low power latch is a strong candidate for hearing aid application.