Efficient design of non-uniform cosine modulated filter banks for digital hearing aids

An Efficient Method for Designing Multiplier-Less Non-uniform Filter Bank Based on Hybrid Method Using CSE Technique

“…Thus the optimal design of 3-channel NUFB having decimation factor (4, 4, and 2), 4-channel NUFB with decimation factor (8,8,4, and 2), and 5-channel NUFB having decimation factor (16,16,8,4, and 2) have been designed and evaluated, in which quantized coefficients are optimized rather than continuous coefficients, and represented in suitable CSD and CSE form, respectively. Obtained results are summarized in Table 5, while simulation results are illustrated in Figs.…”

“…In recent years, non-uniform filter banks have gained significant importance among the various types of filter banks owing to its ability to decompose a given signal into different subbands and to provide rational decimations so as to find numerous applications in the field of signal processing such as in digital hearing aid [16], in multiple channelization of different bandwidth for communication [20], in wideband spectrum sensing of cognitive radio [39], and in trans multiplexers [34] etc. In Mchannel tree-structured non-uniform filter banks (NUFB), the original input signal is first split into two bands of equal bandwidths and then decimated by a factor of two.…”

“…In some applications such as digital hearing aids and in trans multiplexers [16,35], the multirate filter bank is required to be designed with adjustable gain, while a flexible technique is highly desirable in case of software defined radio (SDR) so as to operate as a reconfigurable multi-channel filter bank for different wireless standards [12]. When the filter banks are hardwired, the basic components used in realization of digital filters are adders and multipliers, out of which multipliers are the most expensive and resource consuming components that introduce large complexity in the system [8,11,25,26].…”

Design of shift-invariant nonuniform cosine-modulated filter bank with arbitrary integer sampling factors

“…The existing design methods for oversampled NPR NUFBs can be broadly classified into three categories: joining [14][15][16], frequency warping [17][18][19], and merging [20,21]. The first category obtains an oversampled NPR NUFB by firstly designing several uniform filter banks (UFBs) and then a set of transition filters to join the consecutive sections of these UFBs [14][15][16].…”

“…In [19], Parfieniuk and Petrovsky combined frequency warping and subband merging to offer flexible frequency partition while minimizing the implementation complexity, but at the expense of loss in partition accuracy. The third category obtains an oversampled NUFB by firstly designing a UFB and then directly merging an appropriate number of consecutive subbands of this UFB [20,21].…”